Smart digital auto-guidance device dynamically providing a worker-centric, personalized return-to-work action guidance, and corresponding method thereof

Abstract

Proposed is a digital guidance device based on a dynamically generated worker-centric and personalized return-to-work action guidance allowing to auto-navigating users back to work after beginning of work absence. The digital guidance device comprises a user interface, wherein via the user interface, possible obstacles to a return-to-work are identifiable within a coded cell and transmitted to the digital guidance device. Based on the transmitted obstacles, the smart digital guidance device provides specific solution to each of the obstacles, wherein at least parts of the provided solutions have to be marked as completed by the user before or after his / her return-to-work. The transmitted obstacles and solutions are converged by the digital guidance device automatically listing identified obstacles and solutions in two groups visualized by a corresponding digital template, wherein a first group comprises solution to be completed before return to work and a second group comprises solution to be completed after return to work, and wherein the digital template at least comprises the generated, digital return plan.

Description

[0001] Smart Digital Auto-Guidance Device Dynamically Providing a Worker- Centric, Personalized Return-to-Work Action Guidance, and Corresponding Method Thereof

[0002] Field of the Invention

[0003] The present invention relates to smart digital auto-guidance devices and digital interactive systems for auto-navigating or auto-guiding users in their journey of returning to work after beginning of work absence, in particular to appropriate digital interactive smart widgets and / or smart application. Further, the present invention relates to a forecast system and / or integrated forecast unit and corresponding digital forecast method for forecasting a probability of a return to work (RTW) time period for a person returning to work after beginning of work absence based on person-specific profile data of several return-to-work factors influencing the period of absence and the return time period, respectively. Particularly, the invention relates to a smart, interactive digital auto-guidance device dynamically generating a user-specific return-to-work guidance, wherein obstacles are automatically and user-specific identified by the device based on measured or captured based on user-profile parameter data comprising measured and otherwise captured data of the user, wherein the obstacles are weighed and rated based upon their forecasted impact on the duration of the work absence, and wherein targets to be reached by the user or other guidance are generated based on the mutual weighting or rating of the obstacles. Even more particularly, the invention may also relate to a digital guiding device 1 being realized by adopting a biopsychosocial approach converted in a dynamically adapted decision tree structure to supporting absent workers in identifying their obstacles and solutions to return to work and generating a time-based return to work plan tailored to the individual via a coding system.

[0004] Background of the Invention Occupational injuries and sickness-related absenteeism cause an immense burden on the population and economy. On the example of occupational injuries, though occupational injuries (Ol) vary greatly in severity, they can cause great pain and suffering. Over 30% of Ols typically cause time away from work. They also disproportionately affect under-represented minorities and lead to shortened lifespans. Ols financially impact individuals, directly by causing income reduction, and indirectly by leading to job loss and increasing medical expenditures. Moreover, time away from work can have an economic impact on firms due to reduced productivity.

[0005] According to various reports, the loss of productivity caused by long term work absence due to illnesses or injuries can be substantial depending on the industry, region, and specific workforce demographics. Globally, estimates are ranging into the hundreds of billions of US dollars annually. For example, the Centers for Disease Control and Prevention (CDC) reports that productivity losses linked to absenteeism cost employers US$225 billion annually in the United States. The Federal Institute for Occupational Safety and Health in Germany indicates a total loss of gross value added of €207 billion, or €5,000 per employee in 2022. The costs are typically distributed among various stakeholders, including employers, employees, insurers, and government entities. Employees may face severe financial burdens due to lost or reduced income. In countries with socialized healthcare systems, the government may bear a portion of the costs through healthcare subsidies or other support programs. For employers, factors directly contributing to these losses are wages paid to absent employees, replacement of employees and overtime pay for other employees. Indirect costs may include excess manager time, poor quality of goods and services resulting from overtime fatigue or understaffing, safety issues e. g. related to inadequate training or rushing, and poor morale among employees who have to fill in or do extra work to cover absent coworkers.

[0006] Insurances, such as disability insurance, income protection insurance, insurance against incapacity for work or worker compensation insurance, may cover some costs related to loss of income, healthcare expenses, medical care, re-integration support, and return to work (RTW) programs. For example, the Association of British Insurers (ABI) reports that in 2022 over 15,900 people claimed against their individual income protection policies, which usually cover a percentage of the policyholder's salary and provide support services to help employees back into work. The value of claims totals to £231 million paid to people who were unable to work due to illness or injury. For example, claims for musculoskeletal issues such as neck and back pain were the main cause for an individual income protection claim, accounting for 34%. The US National Academy of Social Insurance (NASI) published estimates of total worker compensation payments, which amounted to approximately US$65.5 billion in 2019. This figure includes payments for medical care, cash benefits, and other expenses associated with work-related injuries and illnesses.

[0007] In summary, the losses and damages caused by workers not returning to work for a long time period represent a significant burden not only for economic development and success but also for social integration and individual contentment. While some prognostic factors are rather obvious, for example the severity of the illness or injury, the individual's occupation, the physical demands of their job, resources and support available to the individual, other factors are less apparent. Therefore, a lot of research focusses on understanding factors and mechanisms impeding or facilitating return to work (RTW) after illnesses or injuries.

[0008] However, returning to work for people with a health problem is a journey. Presenting RTW as a journey allows it to be seen to have a clear starting point and a visible destination, which is a path not a maze where the path can be guided, and the best path can be anticipated and / or optimized. It also allows for stages where sequential tasks are undertaken: education; identifying obstacles and solutions; constructing, negotiating and agreeing a plan; implementing the plan; RTW and monitoring progress. Following the idea of a journey provides a new possibility for a technical approach which e.g. allows to see where the OH practitioners share the tasks with the worker. Before starting the journey, the auto-guidance system can e.g. measure the person's work ability using the work ability question: "Assume that your work ability at its best has a value of 10 points. How many points would you give your current work ability?" This gives an indication of the worker's perceptions of their readiness to start the journey back to work, and it can also be measured again for later comparison to see how this has changed over time. Measuring this at the start of the journey provides an indication that it may not be seen as straightforward and can often appear to be one fraught with uncertainty. That uncertainty can lead to indecision, confusion, and fear of RTW. The way of identifying specific obstacles to RTW is disclosed in more detail below, however, from a technical perspective, the starting point on the journey is to help the person by an automated guidance device to set some realistic goals / expectations, with the recognition that these can evolve. Initially, it may be more about preparing for the journey - that may be easier to contemplate than diving straight in. Setting goals requires typically professional to help the worker understand where they are and what might be needed to move on. Of course, some workers have misunderstandings and fears (often caused by conflicting advice), which result in thinking RTW must be put on hold until they feel fully recovered. Therefore, alongside helping them in the rehabilitation process, helping them dispel those negative myths about work and health is equally important and may be easier achieved by a more neutral device than professionals. For many other workers, the goal may simply be to return to work as soon as possible. The goals should come from the worker, with the auto-guidance device being there to guide and support. It is worth bearing in mind that the initial goals may not be work-related. Initially, it will be useful to find out how important returning to work is for the worker. That can be followed by assessing how confident they feel about returning. The present auto-guidance device has been developing and automatically providing a shared user-specific decision-making guidance for work and health, which includes questions on importance and confidence. If importance is low, it is important to assess for the worker what is important to them, and why they gave a low score for the importance of getting back to work. That can lead on to helping them set non-work as well as work goals that are relevant to them right now. If confidence is low, again it is important to assess why they gave that score and show empathy and understanding while explaining that the autoguidance device can help with figuring out solutions to their concerns and help build their confidence. Of course, this is not sufficient to establish relevant goals, but it can lead on to opening out the path back to work.

[0009] In summary, there is a need for a technical-based, interactive digital system and method for providing a worker-centric, personalized Return-to-Work action plan via interactive user interfaces, as appropriate digital interactive widgets or smart applications. Further there is a need for a technical-based and parameter-based improved approach to providing tailored support to workers with injury-related or sickness-related absences from work, being relied, for example, on an improved forecasting system and method for dynamic prediction of return to work for patients with injury-related or sickness-related absences from work.

[0010] Further, managing the recovery process of injured or sick patients is a difficult task; it involves synthesizing clinical knowledge and practice guidelines with evolving patient recovery information. This process is intense, requiring medical management by highly trained clinicians, including physicians directly guiding patient care and case managers who oversee thousands of simultaneous cases on behalf of healthcare payers.

[0011] The current state of managing sickness absence or supporting workers back to work in the insurance world is at a later stage (3-6 months in absence) due to deferred periods, at which point it becomes more difficult for the absent worker and more costly for the insurer. The herein disclosed digital guidance is, therefore, in line with scientific evidence on the importance of early intervention in preventing avoidable work disability. Further, in the current state of the art for patient recovery prediction are static models that are used at the onset of injury. These models are often used by payers to estimate a patient's Return to Work (RTW) date. The current static models estimate RTW based on information at the time of a patient's initial injury. Thus, providing a quantitative measure on the expected type and amount of resources needed for a patient's recovery, and enabling stratification of the injured patient population. While these static models may provide useful insight into initial severity and potential future resource needs, these estimates rapidly lose value as time progresses.

[0012] A synthesis of reviews about factors affecting RTW after injury or illness published by Cancelliere et al in Chiropractic & Manual Therapies (2016, 24:32) concludes that common factors associated with positive RTW outcomes are higher education and socioeconomic status, higher self-efficacy and optimistic expectations for recovery and RTW, lower severity of the injury / illness, RTW coordination, and multidisciplinary interventions that include the workplace and stakeholders. Common factors associated with negative RTW outcomes are older age, being female, higher pain or disability, depression, higher physical work demands, previous sick leave and unemployment, and activity limitations. Factors related to the specific illness or injury appear not to impact the RTW outcomes. In other words, in many cases, it is likely that the health condition itself is not that important in influencing RTW. That means prognostic factors other than disease-specific factors are associated with RTW positive or negative outcomes. Similarly, a review in the Journal of Occupational Rehabilitation (2019, 29:679-700) presents that positive attitude, self-efficacy, young age, higher education levels and support from line managers or supervisors and co-workers facilitate RTW. Job crafting, economic status, length of absence and job contract / security showed promising results. However, RTW outcomes regarding gender were inconsistent. RTW does not appear to be the result of a single factor. Instead, RTW seems to be influenced by an interplay of multiple factors. Accordingly, companies, healthcare providers and federal organizations developed RTW interventions including essential components like centralized coordination of the worker's RTW, formal individual psychological and occupational interventions, workplace-based interventions, work accommodations and communication between the various stakeholders and interventions. This is what the present inventive digital guidance device technically supports to be achieved with individually auto-driving the process of return-to-work.

[0013] Ultimately, a comprehensive assessment of the various factors that can influence a worker's ability to return to work is needed to determine the likelihood of the worker's RTW. While there are tools and assessments available to help make these predictions, their reliability can vary depending on the accuracy of the information used, the complexity of the individual's situation and the prediction methods applied. The Work Ability Index (WAI) is for example an instrument used in clinical occupational health and research to assess work ability during health examinations and workplace surveys. The index is determined on the basis of answers to a series of questions which take into consideration the demands of work, the worker's health status and health resources. The present digital auto-guidance device allows to provide this index at the start of using the guidance and at the end to see changes in workers perception of their readiness and ability to return to work based on the 2 time point scores.

[0014] The worker completes the questionnaire before an interview with an occupational health professional who rates the responses according to predetermined instructions. Another method is the Functional Capacity Evaluation (FCE) which is a set of tests and observations used to assess an individual's physical abilities as they relate to performing work-related activities. FCEs are most often performed to determine an individual's return to work status after an injury or illness. The results of these tests are used by medical providers to determine whether an individual is able to return to work, what capacity they can return to work in (full or reduced capacity), and whether the individual qualifies for worker's compensation, Social Security Disability (SSDI), or other benefits. Such methods are time consuming, lack accurate reproducibility, are difficult to extend to alternate or additional prediction factors and are inadequate for large quantities of RTW assessments and predictions. Further, these methods do not focus on a probability of a worker's RTW at a future point in time and therefore are not suitable for predicting a RTW timeframe and thus are not useful for claim assessment.

[0015] A computerized method for assessing and processing payments associated with workers' compensation insurance claims is described in US 2016 / 0125544 Al . The method combines different systems to review a workers' compensation insurance claim and includes jurisdiction-based requirements to determine an appropriate benefit amount for a claimant. The system used for the method receives data indicative of a plurality of insurance claims along with an indication of location for each insurance claim. A first workers' compensation benefit rule is then determined for a first workers' compensation insurance claim associated with a first location. Similarly, a second workers' compensation benefit rule is determined for a second workers' compensation insurance claim associated with a second location. A first benefit amount and a second benefit amount are then automatically calculated for the workers' compensation insurance claims based on the rules and application of appropriate waiting period and retroactive period rules. The method fails to include a likelihood for a timeframe when affected workers may return to work.

[0016] US 2014 / 278566 Al shows a system for assessing workers' compensation risk transfer analysis. The data processing comprises transmitting standard workers' compensation questions to a business or an employer respectively, receiving a response to the set of standard workers' compensation questions, selecting a set of supplemental questions based on the received responses to the set of standard workers' compensation questions and transmitted the set of supplemental questions to the user device, receiving a response to the set of supplemental questions and generating a risk score for workers' compensation insurance based on at least the responses to the standard questions and the supplemental questions. The standard questions and supplemental questions relate to aspects of the business, employers' activities, worker environment, human resource aspects, worker surveillance programs, etc. However, the system does not include information of individual workers.

[0017] Further, US 5,712,984 shows a system finding future losses incurred by an insurance carrier on active workers' compensation insurance claims. Statistical models are used to predict future costs and durations of claims. The statistical models are generated by applying statistical analysis techniques to representative historical insurance claim data with the aid of a professional statistician. Forecasting accuracy is improved by generating several models wherein each model represents a predefined type of insurance claim. Each model is applied to active claims of the corresponding type to obtain cost and duration predictions for those claims.

[0018] None of these known methods and systems provide a comprehensive, data-driven, quantitative and individualized approach for predicting a return-to-work likelihood based on various factors related to the individual person suffering from illness or injury and their incapability to work as discussed above. Therefore, the accuracy and reliability of a RTW forecast is unclear and RTW classification is uncertain particularly at an early stage of a worker's illness or injury.

[0019] Summary of the Invention

[0020] It is one object of the present invention to provide a digital auto-guidance device adopting a biopsychosocial approach to supporting absent workers in identifying their obstacles and solutions to return to work and generating a time-based return to work plan tailored to the individual via a coding system. A further object is to provide a digital forecast system and a digital forecast method for forecasting and / or measuring a return time period having a highest occurrence probability value for an individual returning to work after beginning of work absence that allow for an automated assessment of a likelihood of a worker returning to work in a given timeframe based on various prognostic RTW factors, for managing disability income claims or pre-claims, for accurate and reliable evaluation of a worker's RTW situation, for flexible adaption to various individual worker conditions, for supporting a reduction of time absent from work, that diminish work loss and decrease RTW timeframes, while including factors of various stakeholders involved in the RTW process. Finally, it is a further object of the invention to provide parameter-based auto-generation of return to work plan aids providing interlinking of tailored flexible adaptations or accommodations and effective negotiation with the various RTW stakeholders.

[0021] Particularly, the smart digital auto-guidance device and method shall allow for fast and early evaluation of a RTW probability, indicate a most likely return time period, provide trustworthy information about compliance with predetermined timelines, identify factors indicating a deviation of a predicted RTW probability, monitor RTW progress and simplify RTW classification and claims management.

[0022] Especially, it is one object of the present invention to provide technicalbased, improved forecast system and method for dynamic prediction of return to work (RTW) for patients with injury-related or sickness-related absences from work. It is a further object of the invention to provide a technical-based, improved forecast system for dynamic prediction of return to work for patients with injury-related or sickness- related absences from work providing a technical basis for an automated evidencebased claims management system using machine learning structures to predict diverging return to work (RTW) probabilities based on the limited information claims assessors typically have.

[0023] Further, it is an object of the present invention to provide a digital forecast system and a digital forecast method for developing of a return-to-work action plan, which identifies and / or facilitates improvement measures for a faster recovery and a shortened work absence. It is to be noticed that the present smart digital autoguidance device can e.g. also be tailored to allow working while recovering. The digital forecast system and method shall be able to provide information, tools, and links to support early RTW of an employee, particularly ensuring all players are onside and acting, shifting to positive return-to-work culture, overcoming return-to-work obstacles, and promoting work ability. It is especially an object to equip absent employees with all information, knowledge and resources needed to effectively devise and manage their own RTW action plan. The object of the invention is to reduce work loss due to ill health in the early stages of absence, minimize DI claims outcomes (inceptions / recoveries) through proactive facilitation of earlier and sustained RTW rates, and improve individual resilience with overall reduction in claims costs. According to the present invention, these objects ore achieved, particularly, with the features of the independent patent claims. In addition, further advantageous embodiments can be derived from the dependent claims and the related descriptions.

[0024] According to the present invention, the above-mentioned objects for an automated technical smart digital auto-guidance device based on a dynamically generated worker-centric and personalized return-to-work action guidance and / or digital plan navigating users back to work after beginning of work absence, are achieved in that the smart plan builder and digital guidance device comprises a user interface, wherein via the user interface, possible obstacles to a return-to-work are identifiable within a coded cell, and transmitted to the smart plan builder and digital guidance device, in that, based on the transmitted obstacles, the smart plan builder and digital guidance device provides specific solution to each of the obstacles, wherein at least parts of the provided solutions have to be marked as completed by the user before or after his / her return-to-work, and in that the transmitted obstacles and solutions are converged by the smart plan builder and digital guidance device automatically listing identified obstacles and solutions in two groups visualized by a corresponding digital template, wherein a first group comprises solution to be completed before return to work and a second group comprises solution to be completed after return to work, and wherein the digital template at least comprises the generated, digital return plan. Both obstacles and targets / solutions to overcome the obstacles are identified before return to work by the inventive device, which informs the return to work action plan generated. The guidance and / or digital plan is then shared with relevant RTW stakeholders and agreed before absent worker returns to work with agreed accommodations / adaptions suggested on the plan. A phased return to work can e.g. be selectable, if required by the user, wherein the digital template at least comprises the generated, phased or not phased digital return plan for a user's navigating back to work. If a phased return to work is selected, a phase return plan can e.g. be populatable within a cell by the user via the user interface. The cell can e.g. be realized as a digital tick box. The digital template can e.g. further comprise a phased return plan. The user can e.g. be requested by the smart plan builder and digital guidance device via the user interface to add further elements of the generated, digital return plan at least comprising a return-to-work date and / or access authorization for users to share the generated, digital return plan with and / or date of solution attainments and / or responsibilities regarding the solutions to be completed. After the generation of the digital return plan, a simplified version of final return to work actions can e.g. be revealed via a visual interface extracting main details to be shared with a line manager and / or other user and focusing a possible return-to-work conversation between the user and the line manager. The inventive auto-guidance device improves the RTW conversations and facilitates easy transition back to work for the worker towards a sustained return.

[0025] According to an embodiment variant of the present invention, the above- mentioned objects for a smart digital auto-guidance device and method are achieved in that further an automated reliable prediction of diverging return to work (RTW) probability values is provided with the limited information claims assessors typically have. As another embodiment variant, the inventive smart digital auto-guidance device and method can e.g. comprise means to identify high risk individuals through integrating biopsychosocial (BPS) factors capturing into disability income (DI) claims triaging. The inventive device can be based on a specific combination of biological, psychological, and social factors providing a higher statistical significance when predicting RTW likelihood within 6 month or not, as conventional prior art systems do. The biological, psychological, and social factors (BPS) are captured by asking the client a small set of simple biopsychosocial questions. The prediction and validation is performed based on the captured BPS factors using machine learning structures and providing the RTW outcomes in an insured population. The inventive system maps the low-medium-high risk categories to a catalogue of actions for claims assessors. It automatically flags high risk claims that need more targeted claims management efforts and rehabilitation support. Further, it provides the measuring basis for an improved understanding of customer's needs at an individual level, which allows to automate appropriate segmentation of a portfolio in order to focus the required claims management resources.

[0026] According to the present invention, the above-mentioned objects for a smart digital auto-guidance device and method is based on person-specific profile data of several return-to-work factors and can e.g. further be achieved by an integrated digital forecast system or unit that comprises or is connectable to a profile database providing profile data of a personal profile of the person being absent, a return-to-work database providing historical return-to-work data, a communication system, a data processing unit for forecasting a return time period probability and an output unit generating an output signal indicating the return time period probability.

[0027] According to the present invention the return-to-work database can e.g. store a plurality of historical sets of return-to-work data, wherein a set of return-to-work data includes initial data quantifying one or more biological factor, psychological factor and social factor at and / or before the beginning of work absence and later data quantifying return time data indicating a return time period terminating the work absence and / or quantifying the one or more biological factor, psychological factor and social factor at a defined time period after the beginning of work absence. The biological factor, psychological factor and social factor represent return-to-work factors which may influence a return time period. Thus, the historical sets of return-to- work data comprise initial information about the factors characterizing the biological, psychological and social circumstances of the person before the illness or injury and / or at the start of an absence period, i.e. at the beginning of the illness or injury, as well as later information about the same factors after a period of time, when the person already returned to work or when the illness or injury is still ongoing. The profile data and the return-to-work data can be quantified by data values for each of the return-to-work factors characterizing the work absence situation, for the person being absent from work, and for historical return-to-work scenarios of other people having been absent from work due to illness or injury, respectively. The data values of the profile data and the return-to-work data are observable specifications or measures captured by technical frameworks and stored as digital data in the profile data base and the return- to-work database, respectively. The technical set up of the technical frameworks depend on the character of the data values to be captured. The technical frameworks may be designed as sensor devices, audio recording devices, digital recording devices, scaling devices, rating devices, etc. for capturing biological, psychological and social data values, particularly physical, cognitive, situational, descriptive and other technical data values of the person or related to the persons work environment or work situation. The profile data base and the return-to-work database can be hosted in the same or different data repositories.

[0028] As an embodiment variant, the set of biopsychosocial data can be those provided by the user at the point of sign-up. These data can e.g. be used for research purposes to extrapolate (on aggregate level) return to work outcomes across the various biopsychosocial factors, providing insights and / or appropriate threshold values into factors that facilitates early engagement and return to work.

[0029] The communication system can at least comprise a digital network and a signal generator. The digital network may be a network based on electric wires or cables, a wireless network or a combination thereof. The communication system is connected or connectable for example by an interface to the profile database via the digital network. The signal generator generates a digital input signal based on the profile data received from the profile database and is connected to the data processing unit for transmitting the input signal to the data processing unit. The return- to-work database may be connected to the data processing unit or may be hosted in the data processing unit. For connecting the return-to-work database with the data processing unit the communication system may establish a data retrieval channel. The data retrieval channel may comprise a simple interface connecting the return-to-work database and the data processing unit, a transmitter and receiver for transferring the data, a filter arrangement, error detection and correction techniques to identify and correct data errors, communication protocols and / or other technology to ensure fast and safe digital data exchange between the return-to-work database and the data processing unit.

[0030] The data processing unit may for example be realized by a computing device comprising known electronic components responsible for running various processing modules, executing data processing instructions, performing calculations, etc. such as a central processing component, memories, storage components, interface connectors, input / output ports, etc. According to the present invention the data processing unit is connected or connectable to the profile database and the return-to-work database for retrieving the digital input signal and the historical sets of return-to-work data for forecasting a probability of a return time period. The data processing unit at least comprises a validation module for validating one or more historical sets of return-to-work data and a predictor module for determining a statistical return time period probability for a specific return time period. Optionally the data processing unit may comprise a selection module selecting historical sets of return-to- work data for the probability forecast and / or a classification module for classifying a determined return time period probability. The validation module can e.g. comprise a comparison algorithm identifying one or more historical sets of return-to-work data comprising initial data corresponding to return-to-work factors of the profile data. That means the initial data includes data values for one or more return-to-work factors that is also present in the profile data. Such a validated historical set of return-to-work data represents a valid data set for being used by the predictor module for determining a statistical return time period probability. Advantageously, the comparison algorithm identifies historical sets of return-to-work data comprising initial data corresponding to the majority of return-to- work factors of the profile data, i.e. the initial data includes a data value for at least half of the return-to-work factors of the profile data. The valid historical sets of return-to-work data validated by the comparison algorithm are provided to the predictor module. The validation module in a technical manner reduces the large number of historical sets of return-to-work data provided by the return-to-work database to a smaller number of data sets that is related to the specific person being absent and that is statistically relevant for forecasting the probability of a return time period for the specific person.

[0031] The predictor module comprises a predictive modelling algorithm for determining the statistical return time period probability for a specific return time period at least based on the later data of the validated one or more historical sets of return-to- work data. That means the data values of the return-to-work factors captured at a defined time period after the beginning of work absence and / or the return time value of the return time period after which the absentee returned to work form the basis for modelling the statistical probability of the specific return time period. The data values of the return-to-work factors at a later point in time compared to the data values of the initial data at the time of the beginning of the work absence comprise information about a positive, negative or neutral recovery development of the absentee suffering from illness or injury. The information is captured by the technical frameworks observing specifications or measures of the return-to-work data as mentioned before. The predictive modelling algorithm may for example be realized as a regression algorithm or a machine learning structure. The algorithm extrapolates the statistical return time period probability from the interrelation and correlation of the profile data and historical data. The predictor module may determine a series of statistical return time period probabilities for a series of different specific return time periods. The specific return time periods of interest may be defined by a user of the digital forecast system. Further, the predictor module may determine a specific return time period associated with a specified statistical return time period probability.

[0032] Different individual characteristics extracted from the profile data can e.g. be used as input to the predictor module and / or machine learning structure, in particular comprise age, education level, earned income, and self-efficacy. Alcohol consumption showing a significant P value can e.g. be used also. Other occupational and supportive variables may be used as well. For example, the total available datasets of profiles can e.g. be randomly divided into two subsets, a training dataset (e.g. 75%) to make prediction models and a test dataset (e.g. 25%) to measure the prediction ability of each model structure. The random selection can e.g. be done by a function 'sample' of R with setting numbers in order to assure reproducibility. The dependent variable can be working status at a definable number of years after termination of a defined recovery period, x2tests can be performed for each independent variable. To show the efficiency, for example, two MLR model structures can be generated, one with the total available profile data and another with the training datasets. The model structure with the total dataset can e.g. be compared with the model structure from the training dataset, and the odds ratio (OR) from the two model can be assessed. ORs from the total model structure can be used to measure the results, while ORs from the training model can e.g. be used to adjust and optimize the model structure's predictive ability. As in classical analytic model structures, P values below 0.05 can be considered significant. In particular, model structures using machine learning techniques including decision tree (DT), random forest (RF), and support vector machine (SVM) can e.g. be formed with the training datasets. To measure the predictive ability of each training model structure, the structures can be transformed and reconstructed by excluding unimportant variables and / or tuning hyper-parameters to improve the model's performance with automated process. The system 1 can e.g. comprise a digital selector triggered by the best model structure performances by selecting the highest predictive ability for each structure as the representative model structure. The selector can e.g. compare all variables and structure performances by measuring or otherwise generating the accuracy, sensitivity, specificity, kappa value, and area under the curve (AUC), wherein the final predictive values can e.g. be signaled or indicated with mean ± standard deviation from a defined number of iterations of the same procedure after random allocation of training / test datasets with different seed numbers. The inventive system shows that by using the herein disclosed inventive system with its machine learning structures, the probability value of return to work can be technically reliable forecasted by a defined time range e.g. form start of disability or from the end of the official recovery period. The RF model structure can show the best performance, however the other model structures disclosed can e.g. show almost comparable performance. This is the automated system enabled to use machine learning structures for complete automated prediction of return-to-work periods. Though machine learning strucutres have a more complex structure basis than classical modeling structure, since the inventive system does not assume linearity, machine learning based inventive system 1 is more flexible and require fewer variables resulting in a greater energy efficiency and runtime performance by having the same or better forecast precision.

[0033] The data processing unit is connected or connectable to the output unit, for example via the digital network, for providing the statistical return time period probability by the output signal. The output signal may for example be represented on a display or provided to a profile management system as for example used for health care management, human resource management, claim management, etc.

[0034] The digital forecast method according to the invention for forecasting a probability of a return time period for the person returning to work after beginning of work absence based on person-specific profile data of several return-to-work factors in summary comprises the following steps. The profile data of the personal profile of the person being absent of work is provided to the data processing unit via the digital network. Preferably, the personal profile at least comprises profile data indicating data values of biological factors of the person that are not changing due to illness or injury, such as age and / or gender. The plurality of historical sets of return-to-work data is provided to the data processing unit for example via the data retrieval channel. As mentioned before, a set of return-to-work data includes initial data quantifying one or more biological factor, psychological factor and social factor at or before the beginning of work absence and later data quantifying return time data indicating a return time period terminating the work absence and / or quantifying the one or more biological factor, psychological factor and social factor at a defined time period after the beginning of work absence. One or more of the historical sets of return-to-work data for the forecasting of the probability of a return time period are validated by the validation module by comparing the initial data of the historical sets of return-to-work data with the profile data and identifying these historical sets of return-to-work data that correspond to at least one of the return-to-work factors of the profile data. The statistical return time period probability for a specific return time period is determined by the predictor module at least based on the later data of the validated one or more historical sets of return-to-work data by applying the predictive modelling algorithm. The statistical return time period probability for at least one specific return time period for the person being absent is transmitted as an output signal.

[0035] The digital forecast system and method according to the invention provide a solution how to technically realize a holistic approach to assess a return-to-work situation of a person being absent from work that includes diverse factors and various stakeholder aspects influencing the return-to-work capability. These influences are reflected and quantified by the data values of the biological factors, psychological factors and social factors characterizing the return-to-work situation. Biological, psychological and social factors all play a significant role in influencing the likelihood of a person returning to work after an illness or injury. For example, on a biological level the severity of the illness or injury, the person's overall physical health and physical ability to recover can impact their ability to return to work. Social factors such as the support system available to the individual, including family, friends and colleagues, can also play a crucial role in their recovery and ability to resume work duties. Psychological factors such as motivation and coping skills can greatly influence a person's willingness and ability to return to work. For example, individuals with high levels of self-efficacy and a positive attitude towards their recovery are more likely to successfully return to work compared to those with low self-efficacy and negative coping strategies. Overall, the holistic approach including quantified biological, psychological and social factors provides a reliable and reproducible forecast for RTW and assists in supporting employers and employees in the return-to-work journey after illness or injuries.

[0036] The digital forecast system and method according to the present invention enable an automated determination of a likelihood of a worker returning to work and can provide a specified timeframe for RTW based on the validation and predictive modelling of the return-to-work factors by the data processing unit. The digital forecast system and method can quickly and flexibly adapt to various individual work conditions, different illnesses and injuries and diverse worker personalities by validating and sorting the historical sets of return-to-work data relevant for a specific RTW situation. The reliably determined information about the return-to-work timeframe improves the recovery planning, reduces health care costs and costs associated with the work absence, and supports resource management.

[0037] As an embodiment variant, a pre-claim, self-managed user-accessible system suite of the inventive system provides individuals with the information, knowledge and resources needed to effectively devise and manage their own RTW action plan. The return to work (RTW) process frequently breaks down or is prolonged due to poor ownership and communication of coordination between the individual, employer, and healthcare team. The pre-claim, self-managed suite is a practical, accessible, RTW resource encouraging and empowering individuals to be proactive, take informed ownership, and self-manage their own RTW following absence.

[0038] The term software suite, as used herein, denotes a structure holding programs, i.e. a collection of applications, of related functionality, typically sharing a similar user interface and the ability to easily exchange data with each other. The inventive system can e.g. be implemented as a user application implemented as a digital suite providing information, tools, and links to support early RTW of an employee. Thus, situations can be better managed in the early stages reducing work absence.

[0039] In one example embodiment of the digital forecast system the system suite is provided in that the data processing unit comprises a return-to-work management module comprising a planning algorithm for generating a personalized return-to-work action plan for the person returning to work based on person-specific profile data of one or more biological factor, psychological factor and / or social factor. The return-to- work management module defines the personalized action plan by connecting the person-specific profile data with actions represented by digital outputs. For example, the return-to-work management module automatically generates a digital calendar or digital calendar entries, a digital list of action items for the person returning to work and / or of proposed solutions for return-to-work obstacles identified by the one or more biological factor, psychological factor and / or social factor, and assignments of responsibilities to the stakeholders involved in the return-to-work process. The personalized action plan can for example be displayed in a dashboard format on an electronic user device like a computer, tablet or smartphone. In this example embodiment of the digital forecast system the data processing unit serves as a communication means and helps bridging the communication gap between the various stakeholders involved in the return-to-work process of the worker.

[0040] The system suite provided by the digital forecast system comprises (i) key information empowering individuals to address obstacles and identify sources of support, (ii) if ready, the individual is guided by the inventive suite through all stages to build his / her RTW action plan, which provides an agreed timeframe, listing who is responsible for what, how, and when, (iii) the suite allows individuals to monitor their progress by recording completion of the planned actions, and (iv) step (iii) is linked to a feedback loop: a follow-up strategy for individuals to reflect on their RTW journey.

[0041] In a further example embodiment of the digital forecast system the profile data at least includes pre-existing profile data indicating data values at least of biological factors of the person. The pre-existing profile data mainly refers to profile factors that have constant data values, for example age, gender, blood type, genetic defects, birth defects, etc. For example, the pre-existing profile data can be stored in a profile database hosted in a digital profiler system. The digital profiler system is for example a patient monitoring and / or patient data management system as used by health care providers, a digital client management system as used by risk-transfer systems as insurances, or a digital human resource management system as used by business. The digital profiler or profiler system is connected to the communication system for retrieving the profile data. For example, the system is connected by a data transfer interface like an application programming interface that is configured to select and / or download profile data. Using pre-existing profile data that is not subject to change over time facilitates automated data acquisition by the digital forecast system and streamlines data gathering by the data processing unit.

[0042] In a further example embodiment, the digital forecast system comprises a digital data collector hosting a questionnaire application for collecting additional profile data of one or more biological factor, psychological factor and / or social factor about the person being absent from work. The digital data collector may for example be a component of the data processing unit, may be a web space hosting digital application, a hardware component of the system, or the like. The data collector can be connected or is connectable to the communication system for storing the additional profile data in a database, which can for example be the profile database, the return-to-work database or any other data storage solution. The communication system can provide the additional profile data as input signal to the data processing unit. The additional profile data can be included in the input signal providing the profile data as mentioned earlier.

[0043] The questionnaire application allows to conduct a workability assessment for users at 2-time points to establish their ability to work despite the physical and mental demands of their job. As an embodiment variant, the questionnaire application can e.g. comprise defined questions about biological, physical and / or social aspects of the person, for example questions about the return-to-work environment, personal return-to-work needs and expectations, self-assessments, personal fears and limitations, etc. The questions may for example be multiple choice questions, questions with binary response options, questions requiring ranking of response options, questions with answers from a pull-down menu, questions with descriptive responses, etc. The questionnaire application is accessible by an electronic user device like a computer, tablet or smartphone, via the digital network for collecting answers to questions of the questionnaire application. The user may enter responses to the questions using a keyboard, a touchscreen, or the like of the electronic user device. The questionnaire application can for example be accessed by the sick or injured person, a family member of the person, an employer representative, a physician, a care giver, a therapist, etc. More than one person can access the questionnaire application and provide additional profile data. In case several users use the questionnaire application to provide additional data for the same person, i.e. answering the same questions with respect to the same person, the several data values captured for the same return-to- work factor by the same question may be aggregated as one consolidated data value quantifying the associated return-to-work factor the questions refers to. The questionnaire application increases data accuracy and expands data availability for the probability forecasting of the return time period, which supports a precise prediction of the statistical return time period probability. As an embodiment variant, the auto-guidance device only grants access to the worker or user, however, the user can share outputs produced from the inventive device with relevant RTW stakeholders.

[0044] In an example embodiment of the digital forecast system the information gained by the questionnaire application can be used by the return-to-work management module to determine a personalized action plan for the person returning to work based on the person-specific profile data of one or more biological factor, psychological factor and / or social factor gathered by the questionnaire application. Also, the digital outputs generated by the return-to-work management module can be provided as an input signal to the questionnaire application that was using the user device for gathering the additional profile data, which may display the action plan on a monitor of the user device.

[0045] Optionally, the questionnaire application may also be used for collecting additional person-specific profile data of one or more biological factor, psychological factor and / or social factor from the absent person at a later point in time. For example, the questions are being answered at a time when the person is halfway on a planned recovery period or at a time when the person was supposed to return to work. Analog to the initial data validation, the validation module may identify one or more historical sets of return-to-work data comprising later data corresponding to return-to-work factors of the additional profile data captured at a time after the beginning of the work absence. And the predictor module may re-assess the statistical return time period probability based on the later data of the data sets validated based on the later additional profile data. Such forecasting of a probability of a return time period for a person returning to work based on data captured during the work absence can be used to assess the quality or success of return-to-work measures.

[0046] The biological, psychological and social factors reflect various aspects of a return-to-work situation as mentioned earlier. The biological factors used for the forecast of a return time period probability preferably at least include data values for age, gender, diagnosis and treatment. Further biological factors may for example refer to physiological parameters, weight, size, drug consume, tumor size and other illness or injury related biological specifications. Most data values of biological factors can be captured by common medical measuring or imaging devices and may for example be indicated as a numeric value or a numeric range. A social factor is for example provided by a data value indicating a marital status, education level, profession, employment status, a number of children, a number of family members, communication habits and / or a number of memberships, e.g. a membership at sports clubs, social associations, political parties, etc. A psychological factor is for example provided by a data value indicating self-efficacy, recovery expectations, perceived work ability, pain experience, fear experience, etc. The data values of the return-to- work factors can be indicated by diverse data formats. While numeric values are common for biological factors, data values of psychological and social factors can advantageously be captured as a ranking of a psychological or social statement on a numerical scale, alphabetical scale, color scale or descriptive scale, like low, medium and high. Further, the data value can be indicated as a binary value, for example Yes / No or 0 / 1 . Biological, psychological and social information can be provided as a textual description by a few sentences or a short paragraph. Such descriptive text data may be processed to allow for a standardized representation of the information as one or more data values for the corresponding return-to-work factor. For example, the text may be summarized, keywords or key phrases may be extracted, etc. as will be explained in more detail below. The result may be compared to a classification scheme and transformed into a data value indicating for example a ranking on a scale as mentioned above.

[0047] In yet a further example embodiment of the digital forecast system according to the invention the data processing unit may comprise a selection module selecting such historical sets of return-to-work data identified by the validation module wherein a data value of the initial data of the corresponding return-to-work factors matches a data value of the profile data for determining the statistical return time period probability by the predictor module. The selection module narrows down the number of validated historical sets of return-to-work data that are used for determining the statistical return time period probability by excluding such data sets wherein the data values are not representing a similar or identical historical return-to-work data value. It is suggested that the initial data of the corresponding return-to-work factors is matching a data value of the profile data as long as the initial data value deviates insignificantly from the data value of the profile data indicating the same return-to-work factor. For example, a historical data value matches a corresponding profile data value in case it deviates only 10% or less from the profile data value. Selecting such historical sets of return-to-work data simplifies the modelling strategy of the predictive modelling algorithm and refines the determination of the statistical return time period probability by the predictor module.

[0048] In an advantageous variant of the digital forecast system the data processing unit comprises a selection module, which selects one or more historical sets of return-to-work data identified by the validation module as a basis for forecasting the return time period probability such that a probability value (p-value) of the statistical return time period probability determined by the predictive modelling algorithm is below a threshold of 0.1 , preferably below a threshold of 0.05. That means sets of historical return-to-work data rendering the p-values above the threshold are excluded from being selected as the basis for determining the statistical return time period probability. Thus, historical data sets may be excluded from contributing to the determination of the statistical return time period probability despite the fact that at least some of their data values match corresponding data values of profile data as described above. Accepting only statistical return time period probabilities for a specific return time period with a p-value below the threshold further improves accuracy of the RTW forecast.

[0049] Preferably, the predictor module provides a feedback signal to the selection module, wherein the feedback signal provides the probability value resulting from the one or more identified historical sets of return-to-work data providing the basis for forecasting the return time period probability. The feedback signal may further include information about which of the historical data sets caused the probability value to exceed the threshold, for example the predictive modelling algorithm indicates those historical data sets that comprise significant statistical outliers identified by the applied modelling algorithm.

[0050] In case the probability value of the determined statistical return time period probability is not below the threshold of 0.1 , the data processing unit can initiate a request to capture additional profile data using the questionnaire application. The additional profile data will allow for further refining the validation and selection of the historical sets of return-to-work data used for the basis of determining the statistical return time period probability such that the probability value result is below the threshold.

[0051] It is emphasized that a selection module as described above can be used to replace the validation module of the data processing unit. In this case the data processing unit determines the statistical return time period probability based on the plurality of historical sets of return-to-work data stored in the return-to-work database. The number of historical sets of return-to-work data is narrowed down by excluding data sets that prevent the probability value from being below the defined threshold. In a further example embodiment of the digital forecast system of the invention the predictive modelling algorithm comprises a machine learning structure, which is trained based on the plurality of historical sets of return-to-work data stored in the return-to-work database. The machine learning structure may comprise a decision tree structure, a random forest structure and / or neural network structure. In general, the decision tree structure makes predictions based on decision rules, the random forest structure aggregates predictions from multiple decision trees to improve accuracy and robustness, and the neural network structure learns complex patterns in the training data to make probabilistic forecasts. In the decision tree structure, the algorithm recursively splits the plurality of datasets based on the input data values, creating a tree-like structure where each internal node represents a decision based on a return-to- work factor, and each leaf node represents a return time period probability distribution. The decision tree predicts the statistical return time period probability by traversing the tree from the root to a leaf node, where the probability is estimated based on the distribution of training dataset examples that reach that leaf. In the random forest structure, multiple decision trees are trained independently on random subsets of the input datasets. During prediction, each tree in the forest independently predicts probabilities, and the final prediction is obtained by aggregating the individual predictions of all trees, such as taking the average probability across all trees or using a voting mechanism to provide the final statistical return time period probability. In the neural network structure, multiple layers of interconnected nodes process the input datasets through a series of weighted connections and activation functions. Neural networks learn complex patterns and relationships in the training datasets through iterative training using techniques like backpropagation. For forecasting the statistical return time period probabilities, an output layer of the neural network consists of nodes representing different probability distributions. The statistical return time period probability is obtained by passing the input data values through the neural network structure and interpreting the output layer's activations as probabilities. However, other machine learning techniques may be used for the predictive modelling algorithm such as support vector machine structures, gradient boosting machine structures, K-nearest neighbor structures, Gaussian progress structures, logistic regression structures or combinations thereof.

[0052] Further, the machine learning structure may comprise natural language processing techniques for transferring responses captured by the questionnaire application into standardized data values for the associated return-to-work factors, for example for transferring descriptive text into data values as mentioned above. Such natural language processing structures may include various approaches, including text classification, named entity recognition, topic modeling, sentiment analysis, word embeddings, text summarization, and rule-based systems. Text classification algorithms categorize response text into predefined categories or labels, while named entity recognition identifies and extracts entities from text. Topic modeling techniques discover latent topics within the response text. Word embeddings represent words as dense vectors, aiding in capturing semantic similarities. Text summarization generates concise summaries, and rule-based systems use predefined patterns to transform responses. Particularly, a sentiment analysis structure is useful for transferring response text into data values characterizing the return-to-work factors analyzed by the questionnaire application. The sentiment analysis structure quantifies subjective opinions or attitudes by analyzing the textual response and infer the sentiment expressed within the text. By breaking down the text into individual words or phrases and extracting relevant features, a sentiment analysis algorithm classifies the sentiment of the text into predefined categories such as positive, negative, or neutral. The algorithm learns patterns from labeled examples, utilizing supervised learning techniques like support vector machines, Naive Bayes models or deep learning models such as recurrent neural networks or convolutional neural networks, to predict the sentiment of the descriptive response text. The sentiment expressed in the text is then quantified for example through sentiment scores, indicating the intensity or degree of positivity or negativity conveyed in the text. Thus, the predictor module enables the conversion of textual responses into standardized data values, which facilitates analysis, interpretation, and comparison across responses, and allows for transforming the descriptive text into quantified data values of profile data analyzed by the questionnaire application.

[0053] To enhance the quality of the historical sets of return-to-work data as a source of learning data for the predictive modelling algorithm, advantageously the pre-existing profile data and, in case available, the additional profile data of the one or more biological factor, psychological factor and / or social factor collected by the data collector as well as a return time data indicating an effective return time period of the person returning to work after work absence is stored as a historical set of return-to-work data in the return-to-work database. For capturing the effective return time period, the data collector may comprise a return questionnaire that can be completed by the person who returned to work, the employer, a physician, a care giver, etc. Or the questionnaire as discussed above may include one or more questions about the RTW. Over time the return-to-work database will be growing and improve the learning datasets available for the predictor module, which improves the quality of the probability forecast.

[0054] In yet a further example embodiment of the digital forecast system the data processing unit may comprise a classification module for classifying the statistical return time period probability. The classification module assigns a risk factor depending on a deviation of the forecasted return time period associated with the highest statistical return time period probability from a standard return time period as it is for example common for a diagnosis or treatment as indicated in the profile data. The standard return time may be received from a return time statistics for a specific diagnosis, a specific treatment, a specified set of return-to-work factors or other statistical source. Advantageously, the standard return time period can be defined by an average of the return time data of the validated or selected historical sets of return- to-work data. This way, no additional external sources about standard return times are required. The risk factor may be represented as a numerical indication of a degree of deviation from the standard return time period, as a color code depending on the degree of deviation or another categorization. Preferably, a most likely return time period comprising the highest statistical return time period probability is automatically classified by a low, medium or high-risk factor depending on a deviation of the most likely return time period from the standard return time period. It is also possible to provide a risk indicator for each of the return-to-work factors or return-to-work factor categories, particularly a risk indicator for the biological factor category, the psychological factor category and / or the social factor category.

[0055] The risk factor classification helps to identify deficiencies in the recovery progress, a lack of support measures, unmet needs of the absent person, etc. Particularly, the risk factor classification may facilitate any planning of RTW interventions, like actions required for coping with the work absence, hiring additional workers, changing work tasks of the worker, reviewing the treatment strategy, etc. at an early state in the period of absence. Further, it may support the assessment and management of insurance claims and compensation payments. For example, the predictor module or the classification module identifies standard data values of a standardized set of return-to-work factors resulting in a standard return time period, e.g. by determining an average data value of a group of historical data sets including an average return time. Next, the module determines return-to-work factors of the profile data which comprise profile data values deviating from corresponding standard data values of the standard set of return-to-work factors. Depending on the deviation the return-to-work factor can be classified for example as a low, medium or high-risk factor. Depending on the classification improvement measures to improve the profile data values of the return-to-work factors are determined. The digital forecast system may comprise a list of improvement measures and depending on the classification specific measures can be selected and allocated to the return-to-work factor or return-to-work factor category. Thus, a set of improvement measure can be assembled automatically, which facilitates the support of the absent person and improves recovery.

[0056] Finally, the classification module may determine a statistical return time period probability at several points in time during the work absence and provide a trend indicator depending on the change of the statistical return time period probability over time. The trend indicator may be realized as a color scheme indicating a fast, medium or slow trend of RTW. Again, the trend can be predicted automatically and may provide an alert for critical progress of recovery.

[0057] In summary, the digital forecast system and the digital forecast method according to the invention provides an automated forecast of the probability of a worker returning to work in a specific time period which is based on a holistic assessment of various and diverse prognostic return-to-work actors, while including various stakeholders participating in the RTW progress. They provide a detailed and reliable evaluation of a worker's RTW situation and support a reduction of time absent from work. The system and the method flexibly adapt to diverse return-to-work conditions. The digital forecast system and the digital forecast method enable quick evaluation of a RTW probability and a RTW trend right at the beginning of the work absence of a person. They indicate a most likely return time period and provide useful estimates about compliance with expected timeframes. It is possible to identify and classify deviations of a predicted RTW probabilities, monitor RTW progress and simplify absence and cost management. Brief Description of the Drawings

[0058] The present invention will be explained in more detail below relying on examples and with reference to these drawings in which:

[0059] Figure 1 shows a diagram, schematically illustrating an exemplary technical framework for a smart digital auto-guidance device 1 based on a dynamically generated, interactive, worker-centric and personalized, digital, return-to-work action guidance 21 , the digital guidance 21 navigating a user 9 efficiently back to work after a point in time marking a beginning 971 of a work absence of the user 9, according to the invention.

[0060] Figure 2 shows a block diagram, schematically illustrating a procedural flow of an exemplary of a digital forecast method for forecasting a probability of a return time period for a person returning to work after beginning of work absence according to the invention.

[0061] Figures 4 to 6 schematically illustrate embodiments for the automated technical-based, improved forecast system 7 and method for dynamic prediction of return to work for patients with injury-related or sickness-related absences from work are achieved in that an automated reliably prediction of diverging return to work (RTW) likelihood values is provided with the limited information the claims assessors typically have. The system comprising means to identify high risk individuals, in particular also claimants, through integrating biopsychosocial (BPS) factors capturing into disability income (DI) claims triaging. The inventive system 1 is based on a specific combination of biological, psychological, and social factors providing a higher statistical significance when predicting RTW likelihood within 6 month or not, as conventional prior art systems do. The biological, psychological, and social factors (BPS) are captured by requesting response by the individual to a small set of simple biopsychosocial questions e.g. within a decision tree. The prediction and validation are performed based on the captured BPS factors using machine learning structures and providing the RTW outcomes in an insured population. The inventive system automatically maps the low-medium-high risk categories to a catalogue of actions e.g. for case managers. It automatically flags high risk individuals and / or claims that need more targeted individual / claims management efforts and rehabilitation support. Further, it automatically provides the measuring basis for an improved understanding of user's needs at an individual level, which allows to automate appropriate segmentation of a cohort of individuals or a portfolio of individuals to focus the required recovery resources and / or claims management resources.

[0062] Figure 7 illustrates an example embodiment. With reference thereto, some embodiments build a dynamic model structure of an individual's health-state following an Ol using deep learning techniques. In an exemplarily embodiment, the dynamic health state modelling structure can synthesize the individual's entire injury or inability and treatment history (e.g., using input data), where the system 1 provides an effective, reliable and accurate forecast. Contextualizing the modelling structure against all observed historical profile datasets allows to forecast the future work-status of an individual, which can be interpreted as the probability of a given individual returning to work. In the aggregate, such forecast for cohorts of individuals overseen by clinicians can help direct limited resources to individuals at greatest risk of protracted recovery. The forecasts (along with any other data) may be stored in the profile database and accessed by a case manager or the individual etc.

[0063] Detailed Description of the Preferred Embodiments

[0064] Figure 1 schematically illustrate an architecture for a smart digital autoguidance device 1 based on a dynamically generated, interactive, worker-centric and personalized, digital, return-to-work action guidance 21 . The digital guidance 21 automatically provides electronic digital auto-navigation navigating a user 9 efficiently back to work after a point in time marking a beginning 971 of a work absence of the user 9.

[0065] The smart digital guidance device 1 comprises an electronic plan builder engine 2 connected to a plurality of user interfaces 51 . One or more possible return obstacles 96 are generated defined by dedicated obstacle parameters 961 by the plan builder engine 2. The return obstacles 96 are obstacles which hinder or delaying a return-to-work 972 of a user 9, and are selectably indicated to the user 9, i.e. indicated for selection by the user 9, via a user interface 51 within a coded cell 51 1 1 of the interface 51 and a visual representation 51 1 shown at the user interface 51 . Upon selection by a user 9, assigned obstacle identifications 962 of the selected return obstacles 96 are transmitted back to the plan builder engine 2 as a user-specific obstacle parameter pattern 1301 . The auto-guidance device 1 can e.g. use a flaggingframework to allow the device 1 to rely on fundamental biopsychosocial principles. Such a psychosocial digital flags framework can be used to facilitate the identification of obstacles 96 to RTW. The framework can e.g. have three psychosocial 'flag' categories, each given a color: yellow flags are about the person - thoughts, feelings, and behaviors; blue flags are about work - perceptions, expectations and behaviors of both the employer and the worker; and black flags are about the context - the wider environment and its systems. Each of these categories contains factors that can act as an obstacle. For instance, if the worker believes that continuing to work will make their health problem worse, that will be an obstacle to RTW, as will unhelpful processes used by the employer, and misunderstandings about healthcare.

[0066]

[0067] It is important to identify the specific obstacles for each person in his / her job, at this time, with this work-relevant health problem. There may be few obstacles or there may be many and they may be additive, so it is important to identify all those that are relevant. It is to be noted that identifying obstacles requires the right questions to be generated which are correctly weighted and rated mutually. Thus, an appropriate machine-based structure is needed. That structure can be provided by using a "stem question" approach or as a preferred embodiment variant, using the below discussed digital forecast system 7, which can be realized as an integrated part of the digital guidance system 1 or as stand-alone device transferring its output to the digital guidance device 1 . The "stem question" approach is not unlike motivational interviewing in that an initial open question can develop a conversation on a specific topic. A stem-question assessment decision tree has been specifically developed to work with the flags' framework. For the present invention, it has been shown to be both feasible and effective.

[0068] The stem questions may be part of a stem question database or storage. The appropriate stem questions may be selected based on the measured and / or captured profile parameter values. Using the digital forecast system or module 7, the impact of a pattern-based selected stem question or selection of stem questions on the forecasted time of return to work can be predicted by the digital forecast system 7. By varying the selected stem question or selection of stem questions, an optimized, i.e. shortest, predicted time for the user 9 to return to work can be determined by the guidance device 1 . Thus, the selected stem question or selection of stem questions of the stem database can e.g. be varied by the digital guidance device, until the shortest predicted time for a user 9 to return to work is reached. Are the obstacles determined by the digital guidance device 1 form a user 9, as a further variant, the same optimization process can be conducted by based on the obstacles. As a further embodiment variant, the used stem questions are selected based on pattern-matching of the profile parameter values of a user 9 to other users, stored in a database, while the obstacles, determined based on varying the obstacles within the prediction of the forecast system 7, are listed and ranked or prioritized by the digital guidance device 7 for the electronic auto-guidance of a user 9 back to work. The following table 2 provides an example of a flag framework with obstacles: The overarching solution to overcome an obstacle in many instances may rely on information: providing the necessary knowledge and understanding of the principles of obstacles underpins the guidance. Knowledge and understanding are not just for the worker but are also imperative for the other key players if they are to give their all-important agreement to the plan. Some useful evidence-informed educational material has been co-produced with the relevant stakeholders and written in appropriate language for each group. They can be included with the generated digital guidance and / or plan as a means of helping to understand why the guidance is likely to be helpful for the worker. While providing the actions (solutions) and / or targets to be reached by the actions, it is important to get from the worker's agreement to the timelines for actions, the players responsible for each action (including the worker), the players who need to agree the plan, and the dates when the solutions need to be actioned. The following table shows an example of such a digital RTW auto-guidance by the device 1 : The following table showing a modified guidance during the journey back to work:

[0069] Finally, the following table gives a rough overview of a possible journey of a user 9 back to work using the digital auto-guidance device 1 : If required, a phased return to work can, for example, be selectable indicated to the user 9 via said user-interface 51 , wherein the digital template 51 12 at least comprises the generated, phased 21 1 or not phased (if not selected) digital return-to-work action guidance 21 for the user's navigation back to work. If a phased return to work guidance 21 1 is selected, the return-to-work action guidance 21 is populatable within an additional cell 51 13 by the electronic plan builder engine 2 via the user interface 51 . The additional cell 51 13 can e.g. be realized as a digital tick box. The digital template 51 12 can e.g. further comprise alongside a phased digital return- to-work action guidance 21 . When generated properly, the phased return is a structured transitional accommodation; the principle being to start at an achievable level increasing on a regular quota until working normally again. Typically, it refers to a timeframe of increasing work presence or increasing selected work tasks. For example, if it is agreed that the worker can manage two days per week, start at that, with the increase being a day a week for the next three weeks to bring the work time to a full week. The regularity of the quota is important, both for setting expectations and to provide a measure of progress. It should go without saying that any recommendation for a phased return must include a specified duration.

[0070] The user can e.g. be further requested by electronic plan builder engine 2 via the user interface 51 to add further elements to the generated, digital return-to- work action guidance 21 at least comprising a return-to-work date 1225 and / or access authorization for users 9 to share the generated, digital return-to-work action guidance 21 with and / or date of guidance action target attainments 222 and / or responsibilities regarding the guidance action 22 to be completed.

[0071] Based on user-specific parameter pattern 130 comprising the transmitted user-specific obstacle parameter pattern 1301 and user-specific profile parameter pattern 1301 , the plan builder engine 2 generates pattern-specific obstacle guidance 21 1.21 i to each of the obstacles 96 selected by the user 9, and are indicated via the user interface 51 to the user 9 within the coded cell 51 1 1 , an obstacle 96 to be marked upon completing all obstacle guidance 21 1.21 i steps assigned to a specific obstacle

[0072] 96, the steps to be performed by the user 9.

[0073] The obstacles 96 and pattern-specific obstacle guidance 21 1.21 i are dynamically updated upon selection and continuously converged by the electronic plan builder 2 dynamically listing selected obstacles 96 and marked or unmarked pattern-specific obstacle guidance 21 1 21 i in two groups visualized by a corresponding digital template 51 12 of said coded cell 51 1 1 . A first group 51 121 comprises guidance steps to be completed before return to work and a second group 51 122 comprises solution to be completed after return to work. The digital template 51 12 at least comprises the generated, digital return-to-work action guidance. At least parts of the provided solutions have to be marked as completed by the user before or after the user's return-to-work.

[0074] After the generation of digital return-to-work action guidance 21 , a simplified return-to-work action guidance 23 comprising final return to work actions 22 can e.g. be indicated via the user interface 51 extracting main details to be shared with a line or case manager 92 and / or other user 91 , and focusing a possible return-to- work conversation between the user 9 and the line or case manager 92.

[0075] In a further embodiment variant, the smart digital auto-guidance device 1 can e.g. comprise a digital forecast system 7 for forecasting a return time period 733 having a highest occurrence probability value for an individual returning to work after beginning of work absence, the forecasting and / or measuring being triggered by userspecific profile data of the user 9, wherein possible return obstacles are weighted and / or based their impact on the forecasted a return time period, and wherein only return obstacles 96 having an impact to the return time period higher than a predefined threshold value are indicated to the user 9 via the user interface 51 . In particular, the digital interactive system 1 for navigating or guiding users in their journey of returning to work after beginning of work absence, can comprise said forecast system and / or integrated forecast unit and corresponding digital forecast method for forecasting a probability of a return to work (RTW) time period for a person returning to work after beginning of work absence based on person-specific profile data of several return-to-work factors influencing the period of absence and the return time period, respectively. Particularly, this embodiment variant relates to a smart, interactive digital auto-guidance device dynamically generating a user-specific return-to-work guidance, wherein obstacles are automatically and user-specific identified by the device based on measured or captured based on user-profile parameter data comprising measured and otherwise captured data of the user, wherein the obstacles are weighed and rated based upon their forecasted impact on the duration of the work absence, and wherein targets to be reached by the user or other guidance are generated based on the mutual weighting or rating of the obstacles. The smart digital auto-guidance device 1 machine-learning structures, as nearest neighbor or LHS structures, which compare the parameter pattern of a user 9 in question with historic parameter pattern allowing to assign a forecasted and predicted, respectively, time period value for the work absence to said user-specific parameter pattern. As another embodiment variant, the digital guiding device 1 is realized by adopting a biopsychosocial approach converted in a dynamically adapted decision tree structure to supporting absent workers in identifying their obstacles and solutions to return to work and generating a time-based return to work plan tailored to the individual via a coding system.

[0076] For the digital forecasting system 7 plurality of return-to-work factors are extracted from the user-specific data 100. The digital forecast system 7 comprises or is connected to a profile database 12 providing at least parts of the person-specific data 100 by a personal profile of the individual, a data processing unit 70 of the digital forecast system 7 receiving the person-specific data 100 as an input signal 103 and assessing the profile data 100 to deduce the forecasted probability value of the return time period. An output unit 6 generates an output signal 52 indicating the forecasted return time period 733 with the highest occurrence probability value. The digital forecast system 7 further comprises:

[0077] • a return-to-work database 205 storing a plurality of historical sets of return-to-work data 1 10, wherein a set of return-to-work data 1 10 includes initial data 121 quantifying one or more biological measuring factor, psychological factor and social factor at and / or before the beginning of work absence and later data 122 quantifying return time data 1225 indicating a return time period terminating the work absence and / or quantifying one or more biological factor, psychological factor and social factor at a defined time period after the beginning of work absence,

[0078] • a communication system 3 at least comprising a digital network 31 and a signal generator 32. The communication system 3 is connected or connectable linked to the profile database 12 via the digital network 31 for generating the digital input signal by the signal generator 32. The data processing unit 70 is connected to the communication system 3 and the return-to-work database 205 for receiving the digital input signal 103 and historical sets of return-to- work data 1 10.

[0079] The data processing unit 70 at least comprises

[0080] • a validation module 71 comprising a comparison algorithm 71 1 identifying one or more historical sets of return-to-work data 1 10 comprising initial data 121 corresponding to at least one return-to- work factor of the profile data 100, and

[0081] • a predictor module 73 comprising a predictive modelling structure 731 for determining a statistical return time period probability for a specific return time period at least based on the later data 122 of the validated one or more historical sets of return-to-work data 8121 , wherein the data processing unit 70 is connected to the output unit 6 for providing the statistical return time period probability as the output signal 52.

[0082] The profile data 100 can e.g. include pre-existing profile data 101 indicating data values at least of biological factors of the person. The profile database 12 can e.g. be hosted in a digital profiler system 10, which is connected to the communication system 30. The data processing unit 70 can e.g. comprise a return-to-work management module 75 comprising a planning structure 751 for generating a personalized return-to-work action plan for the person returning to work based on person-specific profile data of one or more biological factor, psychological factor and / or social factor.

[0083] The digital forecast system 7 can e.g. comprise or can e.g. be connected to a digital data collector 4 hosting a questionnaire application 41 for collecting additional profile data 102 of one or more biological factor, psychological factor and / or social factor about the person being absent from work, wherein the questionnaire application 41 is accessible by an electronic user device 5 via the digital network 31 . The data collector 4 is connected or connectable linked to the communication system 30 for storing the additional profile data 102 in a database 42 and / or providing the additional profile data 102 as input signal 103 to the data processing unit 70. A data value of a biological factor can e.g. be provided by an indicator for age 101 1 , gender 1012, diagnosis 1013 and / or treatment 1014.

[0084] A data value of a social factor can e.g. be provided by an indicator for marital status, profession, employment status, a number of children, a number of family members and / or a number of memberships 1024. A data value of a psychological factor or social factor can e.g. be measured based on a ranking of a psychological statement or social statement on a numerical scale, alphabetical scale, color scale or descriptive scale.

[0085] The data processing unit 70 can e.g. comprise a selection module 72 selecting and or hashing for these historical sets of return-to-work data 821 identified by the validation module 71 wherein a data value of the initial data 121 of the corresponding return-to-work factors matches a data value of the profile data 100 for determining the statistical return time period probability by the predictor module 73. As an embodiment variant, the data processing unit 70 can e.g. comprise a selection or filter module 72 selecting one or more historical sets of return-to-work data 821 ' identified by the validation module as a basis for forecasting the return time period probability such that a probability value of the statistical return time period probability determined by the predictive modelling algorithm is below a threshold of 0.1 , preferably below a threshold of 0.05. For the automated pattern matching, for example, approximate nearest neighbor (ANN) search, as locality sensitive hashing (LSH) can be used, where the input values are given by the set of the user 9 in question to the one or more historical sets of return-to-work data 821 ' . It is to be noted, that the profile database 12 may containing millions or even billions of dataset samples. The technical problem here arises in "how to compare efficiently all of those samples". Even on the best hardware, comparing all pairs is out of the question. This produces an at best complexity of O(n2). Even if comparing a single query against the billions of samples, this still returns an at best complexity of O(n). according to the inventive selection module 72 and / or the digital prediction and / or auto-forecast system 7, every dataset is stored as a vector, in particular a highly-dimensional vectors, which increases the processing complexity even further, impacting the data processing efficiency and speed. To overcome this technical problem, an approximate search is applied. Rather than comparing every vector in an exhaustive search, the search is approximated and limited by the system 7 using a limited and approximated search scope to only the most relevant vectors. Thus, the system 7 provides a machine-based structure with those sub-linear search times.

[0086] Again, the technical problem arises that when the complexity of finding similar pairs of vectors is considered, the number of calculations required to compare everything is typically unmanageably enormous in the present case and application for any data processing system even with reasonably small datasets. For example, a vector index can be considered. If just one new vector has to be introduced and the system 7 attempts to find the closest match, the system 7 must compare that vector to every other vector in the profile database 12. This results in a linear time complexity, which cannot be scaled technically to fast search in the larger datasets of the profile database 12. The problem is increases if all of those vectors are compared against each other, where the optimal approach sorting structure to be technically achieved is at best log-linear time complexity. The system 7 provides a way to reduce the number of comparisons. Ideally, only vectors should be compared that are believed to be potential matches or candidate pairs. To achieve this, for example, the system 7 can apply a locality sensitive hashing (LSH) structure, in particular shingling, MinHashing, or final banded LSH function structures. At its core, the forecast system 7 segments and hashes the same sample several times. When the system 7 find that a pair of vectors has been hashed to the same value at least once, the system 7 tags them as candidate pairs, that is, potential matches. The process can e.g. be compared to processes that are used in not related technical fields as in Python dictionaries, where a key-value pair is taken to be fed into the dictionary. However, there, the key is processed through the dictionary hash function and mapped to a specific bucket, where the respective value is connected to this bucket. There is a key difference between this type of hash structures and that used in the system 7. With dictionaries, the goal is to minimize the chances of multiple key-values being mapped to the same bucket, here, the system 7 minimizes collisions. Thus, the machine structure used by system 7 is, from its technical object, technically almost the opposite. The system 7 wants to maximize collisions, although ideally only for similar inputs. The important part of forecast system 7 is to provide the appropriate "bucket similar samples through a hash function" logic structure, where the rest can vary according to the concrete application. The inventive approach is based on a three-step process. First, the system 7 converts and extracts, respectively, parameters of the profile database 12 to sparse vectors of an extracted user dataset parameter vector 141 , for example, using a k- shingling (and one-hot encoding) structure. It is to be noticed, that the parameters of a user's dataset 141 can e.g. comprise (i) profile data 100 comprising pre-existing profile data 101 and / or additional profile data 102, and / or (ii) data from historical data sets of return-to-work data 1 10, and / or (iii) data from the return-to-work database 120, and / or (iv) user-specific parameters and / or parameter pattern 130. For the extraction, the system comprises the profiler 10, which is (inter alia) accessible by the user 9 and which steers and operates the conversion and extraction of the user dataset parameter vector 141 from the profile database 12 using the automated extractor 14. The extractor 14 is an auto-operating unit operating extraction and conversion upon initiation with an appropriate request with input parameters, at least comprising a user identification 90. The extractor 14 can e.g. then use minhashing to create "signatures" which are passed onto to a LSH process of the profiler 10 to weed out candidate pairs.

[0087] As discussed the LSH structure can e.g. comprise shingling, MinHashing, and / or banding. If a k-Shingling structure is chosen (herein referred to as shingling), the profiler converts data of the profiler database 12 (it is to be noted that the user data of the profiler database 12 comprise physically measured parameters, e.g. clinical measuring parameters (related to illness or accident) and / or captured data, as e.g. data inputted by the user 9 via the user interface 51 to the profiler 10 and the profiler database 12) consisting of measuring parameter values and / or structure data (by input mask) or even strings of text into a set of ‘shingles' . The present process is similar to moving a window of length k down a string of different data and taking a picture at each step. The system 7 then collates all of those pictures to create the set of shingles. Shingling may also be used to remove duplicate data. The machine-based structure can e.g. be created inter alia using a k-shingling function in Python. With this, the system 7 has the technically required shingles. Next, system 7 creates appropriate sparse vectors. To do this, the system 7 first needs to union all of the sets to create one big set containing all of the shingles across all of the sets, sometimes referred to as vocabulary of the structure. The system 7 uses this vocabulary to create and store the sparse vector representations of each set. First, all the system 7 does is to create an empty vector full of zero values having the same length as the vocabulary. Then, the system looks at which shingles appear in the set. For every shingle that appears, the system 7 identifies the position of that shingle in the vocabulary and set the respective position in the new zero-vector to 1 . It is to be noted that this technical approach can be regarded analogously as e.g. one-hot encoding.

[0088] Applying a minhashing structure is the next step in the inventive data processing, allowing the system 7 to convert the sparse vectors into dense vectors. The system 7 comprises now the generated sparse vectors, and randomly generate one minhash structure and function for every position in the signature (e.g., the dense vector).

[0089] For example, if the system 7 is required to create a dense vector / signature of 20 numbers, it will use and generate 20 minhash structures and functions, respectively. Those minhash functions are realized as a randomized order of numbers, where it counts from 1 to the final number (which is len(vocab)). Because the order of these numbers has been randomized, that number can be 1 in position 57 (for example) of the randomized minhash structure. Looking at a sparse vector, the system 7 checks if this shingle exist at the vocabular (number 1 ) in the set. If it does, the sparse vector value will be 1 , where in this case, it does not exist the value is set to 0. Then the system 7 move to number 2, identifying its position (0) and does the same process etc. This time, if the answer is yes, the minhash output will be 2. That's how the system automatically provides one value in the minhash signature. However, the system 7 needs to produce 20 (or more) of these values. So, the system 7 assigns a different minhash function to each signature position and repeats the process. At the end of this process, the system 7 has produced and generated the minhash signature or dense vector.

[0090] One critical technical question, is if the information is truly maintained between the much larger sparse vector and much smaller dense vector. It's not trivial to identify a pattern in these new dense vectors, however, the similarity between vectors can be calculated or assessed for control purposes. If the information truly has been retained during the technical downsizing, the similarity between the vectors will be similar, too. For testing this, for example, Jaccard similarity measures can be used to measure or calculate the similarity between the input structured and unstructured data from the profile database 12 in shingle-format, and then repeat for the same vectors in signature format. For the present system 7, a much higher similarity can be found, as normally expected. It let assume that the similarity information is maintained between the sparse vectors and signatures.

[0091] In system 7, the final step by identifying similar parameter pattern is the LSH structure itself.

[0092] For example, a banding approach can be taken to LSH. Thus, the system 7 will take the above discussed signatures, then hash segments of each signature, and finally looks for hash collisions, as described above. Through this process, the system 7 produces these vectors of equal length that contain positive integer values in the range of 1 — > len(vocabulary) . These are the signatures that are typically inputted into the present LSH structure.

[0093] If the system 7 is to hash each of these vectors as a whole, it may struggle to automatically build a hashing function that accurately identifies similarity between them. However, the system 7, in the present case, does not require that the full vector is equal, only that parts of it are similar. In most cases, even though parts of two vectors may match perfectly, if the remainder of the vectors are not equal, the function will likely hash them into separate buckets. For the present system 7, this is not desired. It is desirable to have signatures that share even some similarity to be hashed into the same bucket, thus being identified as candidate pairs. The present system 7 using the above discussed banding structure solves this technical problem by splitting the vectors into sub-parts. Then, rather than processing the full vector through the hash function, the system 7 passes each band of the vector through a hash function. To give an example, if a 100-dimensionality vector is split by the system 7 into 20 bands, this results in 20 possibilities to identify matching sub-vectors between the vectors (Note, the system 7 splits the signature into b sub-vectors, each is processed through a hash function (system 7 can be realized using a single hash function, or b hash functions) and mapped to a hash bucket.)

[0094] The system 7 can be extended by adding a more flexible condition. Id est, given a collision between any two sub-vectors, the system 7 considers the respective full vectors as candidate pairs (Note, since the system 7 splits the signatures into subvectors, each equivalent subvector across all signatures must be processed through the same hash function. However, it is not necessary to use different hash functions for each subvector (the system 7 may use just one hash function for them all) . Only part of the two vectors must match for the system 7 to consider them. This also increases the number of false positives (samples that should be marked as candidate matches where they are not similar). However, the structure of system 7 tries to minimize these as far as possible.

[0095] As a further variant, it is possible to optimize the band value b to shift the similarity threshold of the LSH structure. The similarity threshold is the point at which the LSH structure should switch from a non-candidate to a candidate pair. Then the probability-similarity relationship can be given by P = 1 - (1 — sr)b, with the probability (P) of a pair being identified as candidate pairs given a similarity score (s), number of bands (b), and number of rows in each band (r). There is a correlation between the theoretical, calculated probability and the genuine candidate pair results. The system 7 may push the probability of returning candidate-pairs at different similarity scores left or right by modifying b (note that that r will be len (signature) / b) . Because the system 7 is now returning more candidate pairs, this will naturally result in more false positives, where the system 7 returns "candidate pair" for dissimilar vectors. This is an unavoidable consequence of modifying b. However, this allows to efficiently build the LSH structure and further allows technically to tune the similarity threshold, which is not possible with the prior art systems in the field. Finally also note, that in practice, it can be advantageously to realize the LSH structure using digital libraries built specifically for similarity search.

[0096] The predictor or forecast module 70 provides a feedback signal 831 to the selection module 72, wherein the feedback signal 831 provides the probability value resulting from the one or more identified historical sets of return-to-work data 821 ' providing the basis for forecasting the return time period probability.

[0097] The data collector 4 can e.g. collect return time data indicating a return time period of the person returning to work after work absence and stores the collected additional profile data 102 of the one or more biological factor, psychological factor and / or social factor together with the collected return time data 1225 as a historical set of return-to-work data 1 10 in the return-to-work database 205. The machine-based predictive modelling structure 731 can e.g. comprise or be realized as a machine learning structure 732, which is trained based on the plurality of historical sets of return-to-work data 1 10. The machine learning structure 732 can e.g. comprise or be realized as a decision tree structure, a random forest structure and / or neural network structure.

[0098] The data processing unit 70 can e.g. comprise a classification module 74, which assigns a risk factor 742 depending on a deviation of a return time period associated with the highest statistical return time period probability from a standard return time period.

[0099] Finally, it is to be noted that the present invention also discloses a smart digital auto-guidance method for realizing the digital, return-to-work action guidance 21 of a smart digital auto-guidance device 1 , as described above. Further, it has to be noted, that the auto-forecast and / or prediction system 7 can e.g. also be realized as a stand-alone forecast device, where the measured and predicted output parameter value for the predicted return-to-work data or the predicted period for a user 9 to return to work can be used and be important in various ways.

[0100] Parameter-based auto-forecast and forecast / prediction system 7

[0101] Figures 1 to 6 schematically illustrate an architecture for an automated technical-based, improved forecast system 7 and method for dynamic prediction of return to work for patients with injury-related or sickness-related absences from work are achieved in that an automated reliable prediction of diverging return to work (RTW) likelihood values is provided with the limited information the claims assessors typically have. The system comprising means to identify high risk claimants through integrating biopsychosocial (BPS) factors capturing into disability income (DI) claims triaging. The inventive system is based on a specific combination of biological, psychological, and social factors providing a higher statistical significance when predicting RTW likelihood within 6 month or not, as conventional prior art systems do. The biological, psychological, and social factors (BPS) are captured by asking the client a small set of simple biopsychosocial questions. The prediction and validation is performed based on the captured BPS factors using machine learning structures and providing the RTW outcomes in an insured population. The inventive system maps the low-medium-high risk categories to a catalogue of actions for claims assessors. It automatically flags high risk claims that need more targeted claims management efforts and rehabilitation support. Further, it provides the measuring basis for an improved understanding of customer's needs at an individual level, which allows to automate appropriate segmentation of a portfolio in order to focus the required claims management resources.

[0102] In more detail, Figures 1 schematically illustrates the digital forecast system 7 for forecasting a probability of a return time period for a person returning to work after beginning of work absence due to an illness or injury based on person-specific profile data 100 and historical sets of return-to-work data 1 10 of several return-to-work factors as will be explained in the following.

[0103] The example digital forecast system 7 is connected to a digital profiler system 10, which comprises a profile database 12 and an extractor 14. Although the profiler, also referred herein as profiling system 10 is depicted as a single device, the profile database 12 and any modules of the extractor 14 could be distributed entities, for example hosted in a digital cloud environment. The electronic profiler 10 could for example be a component of an automated, electronic health care management system, automated human resource management system or an electronic insurance management system comprising a portal for users to interact with the system. Users are for example individual clients, employees or patients, physicians, therapists, health care service providers, underwriters, insurance agents, employer representatives, etc. The users can access the profiling system 10 to enter or update person-specific information data about a personal profile of a client, employee or patient, as it is for example known for automated managing processes. That means the profile data base 12 stores the person-specific data 100 about an individual or person, which for example includes profile data values for name, birth date, age, gender, health history, diagnosis, treatment plans, treatment costs, and / or clinical, medical or diagnostic measuring parameter values etc. The person-specific data 100 captured by the profiling system 100 commonly includes pre-existing person-specific data 101 that already exists before the person gets sick or suffers an injury and a work absence begins. In the present example, the person-specific pre-existing data can e.g. include an age data value 1 1 1 , a gender data value 1012, one or more diagnosis data value 1013, one or more treatment data value 1014, and one or more clinical measuring parameter values, as e.g. weight, blood-pressure progressions or measuring points, heart rate or other biological measuring factors etc. The pre-existing person-specific data 101 is readily available for the digital forecast system 7. Usually, it mainly focusses on biological factors influencing the return time period of the absent person since this type of factors are the main focus of profiling system. However, the pre-existing person-specific data may also include return-to-work factors in form of psychological factors and social factors. In general, the profile database 12 can at least comprise observational data collected on individuals triggered by specific criteria. The profile data can e.g. comprise a collection of measuring or otherwise captured data related to individual's diagnosis, demographics, exposures, laboratory tests, and family relationships, in particular including electronic health records (EHR), and electronic medical records (EMRs) . Further, the profile data base 12 can e.g. comprises measuring parameters such as parameters of biobanks of individual's materials such as measuring parameters associated with e.g. serum, DNA, and pathology specimens etc.

[0104] Further, the example digital forecast and / or prediction system 7 is connected to a data collector 4, which provides a digital platform for hosting a questionnaire application 41 providing a set of questions for collecting additional profile data 102 of return-to-work factors for the person being absent of work. The questionnaire application 41 is for example realized as a software application which presents questions about return-to-work factors to a user and captures response data and information. For example, commonly used survey and form building software tools may be used, such as Google Forms, Typeform, SurveyMonkey, and Qualtrics. Also, readily available plugins or modules of the profiling system may be used for creating and managing a questionnaire. For more customized and advanced questionnaires, programming languages such as JavaScript, Python, and Ruby can be used along with libraries or frameworks like React, Angular, or Vue.js for front-end development. Backend frameworks like Django (Python), Rails (Ruby), or Expresses (Node.js) can be utilized for handling form submissions and data processing. The questions of the questionnaire application 41 include questions about biological, physical and / or social aspects of the person, for example questions about the return-to-work environment, personal return-to-work needs and expectations, self-assessments, personal fears and limitations, and the like. The questions may for example be multiple choice questions, questions with binary response options, questions requiring ranking of response options, questions with answers from a pull-down menu, questions with descriptive responses, or other commonly used digital questioning formats. The person using the questionnaire application 41 can be the person being absent, an employer of the absent person, a physician or therapist treating the absent person, a work supervisor of the absent person or the like. The questionnaire application 41 is for example accessible by an electronic user device 5, 5', 5" via a digital network. Several users may respond to the same questionnaire. Advantageously, response data of the several users can be aggregated as a single data value of a return-to-work factor corresponding to the one or more questions. The captured response data and information are stored as additional profile data 102 in a database 42 of the data collector 4. Alternatively, the additional profile data 102 could for example be stored in the profile data base 12 of the profiling system 10, in a cloud data storage solution or any other database of the digital forecast system 1 . The digital forecast system 7 may for example send an invite to a user to fill out the questionnaire of the questionnaire application 41 once the work absence begins. The data collector 4 is mainly used to collect information and data that is missing in the preexisting profile data 101 . The questionnaire can advantageously be adapted according to required data about an absent person. Particularly, the questionnaire application 41 focusses on capturing psychological factors and social factors of a return-to-work situation since data about these factors is often neglected by profiling systems. In the present example, the data collector 4 captured person-specific additional profile data 102 that includes a profile self-efficacy data value 1021 , a profile fear experience data value 1022, a profile perceived work ability data value 1023 and a profile membership data value 1024. However, the questionnaire may also capture missing biological data of the absent person or request an update of such data. In general, additional profile data 102 can be captured with the data collector 4 at any time during the work absence and even after termination of the absence.

[0105] Further, the digital forecast system 7 comprises a return-to-work database 205 storing a plurality of historical sets of return-to-work data 1 10. In the present example the database is realized as a cloud storage. However, other storage solutions are possible as long as their capacity is large enough to store the large quantity of digital data presented by the plurality of historical sets of return-to-work data 1 10. In general, the historical sets of return-to-work data 1 10 include information about past scenarios of work absence and a return time period of these scenarios. In particular, the historical set of return-to-work data 1 10 includes initial data 121 quantifying one or more biological factor, psychological factor and social factor at and / or before the beginning of work absence. One example of a historical set of return-to-work data 1 10 comprises initial data 121 providing an initial self-efficacy data value 121 1 , an initial fear experience data value 1212, an initial perceived work ability data value 1213 and an initial treatment data value 1214. Further, the historical set of return-to-work data 1 10 includes later data 122 quantifying return time data by indicating a return time data value 1225 indicating a return time period terminating the work absence of the work absence scenario. Further the later data 122 quantifies the one or more biological factor, psychological factor and social factor at a defined time period after the beginning of work absence. In the present example of the historical set of return-to- work data 1 10 the later data 122 provide a later self-efficacy data value 1221 , a later fear experience data value 1222, a later perceived work ability data value 1223 and a later treatment data value 1224. It is not necessary that a later data value is provided for all the initially quantified return-to-work factors.

[0106] Further, the digital forecast system 7 comprises a communication system 30 comprising a wireless digital network 31 and a signal generator 32. The digital network 31 can be realized as a data transmission network or data transmission line, e.g. comprising a cellular mobile network 33 and / or a satellite transmission line 34 for connection to the internet. The communication system 30 is connectable to the profile database 12 by an interface 35 and to the data collector database 42 via an interface 36 via the digital network 31 for receiving the person-specific profile data 100, including the pre-exiting profile data 101 and the additional profile data 102. The profile data 100 is supplied to the signal generator 32 for generating a digital input signal 103. Further, the communication system 30 creates data retrieval channel 37, which connects to the return-to-work database 205 for retrieving historical sets of return-to-work data 1 10 for example via an interface 38. The digital network 31 comprises common wireless network core components, such as data centers, servers, networking equipment, and storage resources hosted by cloud service providers. The core components provide the foundation for delivering communication services for data transmission. Additionally, the communication system 30 may include features for user authentication, security, and compliance, ensuring that communication data remains secure and meets regulatory requirements. Overall, the components of the communication system 30 work together to enable reliable and scalable communication services for the digital forecast system 1 . Further, the digital forecast system 7 has a data processing unit 70, which comprises several processing modules for forecasting the probability of a return time period for an absent person based on the person-specific profile data 100 and the historical sets of return-to-work data 110 according to the invention. The data processing unit 70 is connected to the communication system 30 for receiving the digital input signal 103 providing the profile data 100 and for retrieving historical sets of return-to-work data 1 10 from the return-to-work database 205 via the data retrieval channel 37.

[0107] The digital forecast system 7 comprises a validation module 71 comprising a comparison algorithm 71 1 for identifying one or more historical sets of return-to-work data 110 comprising initial data 121 which corresponds to at least one return-to-work factor of the profile data 100. Various types of comparison algorithms can be employed for validating the data. For example, a comparison algorithm based on an exact matching check may be used to assess whether data values are identical and useful for verifying categorical or numerical data. A fuzzy matching structure may be used to allow for approximate comparisons, accommodating variations in within a value range, formatting, or insignificant discrepancies. Statistical comparison methods may be used to assess a distribution of numerical data values. Further, the comparison algorithm 71 1 may be designed as a rule-based algorithm for applying predefined rules or constraints to validate the data, ensuring adherence to specific criteria. Also, a combination of different types of comparison algorithms may be used to design the comparison algorithm 71 1 of the validation module 71 . For the present example, the comparison algorithm 71 1 e.g. identifies such historical sets of return-to-work data that comprise initial data 121 corresponding to the age return-to-work factor quantified by the age data value 101 1 and / or corresponding to the gender return-to-work factor quantified by the gender data value 1012 and / or corresponding to the diagnosis return-to-work factor quantified by the diagnosis data value 1013 and so on.

[0108] The validation module 71 may comprise further algorithms to facilitate preparation of the data for further processing. For example, a formatting algorithm 712 may be applied for streamlining the format of the data values characterizing the return- to-work factors provided by the profile data 100 and historical sets of return-to-work data 1 10. Further, in this example of the digital forecast system 7 the data processing unit 70 comprises a selection module 72 selecting these historical sets of return-to-work data, from the datasets that were identified by the validation module 71 , wherein a data value of the initial data 121 matches a data value of the profile data 100 for corresponding return-to-work factors. The selection model 72 may comprise a selection of filter structure 721 for choosing subsets of the identified historical sets of return-to-work data based on specific criteria. For example, a stratified sampling algorithm can ensure that the selected subset includes the same return-to-work factors or data values as the profile data. An adaptive sampling algorithm dynamically adjust the selection process based on feedback or changing factors, optimizing the selection process over time. The selection structure 721 also may be based on a type of comparison method as explained above, particularly a fuzzy matching structure. For example, the selection structure 721 compares the initial self-efficacy data value 121 1 with the profile self- efficacy data value 1021 , the initial fear experience data value 1212 with the profile fear experience data value 1022, the initial treatment data value 1214 with the profile treatment data value 1014 and so on and selects only these historical sets of return-to- work data with matching data values. As explained above, the initial data 121 of the corresponding return-to-work factors is regarded as matching a data value of the profile data 100 as long as the initial data value deviates insignificantly from the data value of the profile data 100 indicating the same return-to-work factor. For example, the initial data value deviates only 10% or less from profile data value.

[0109] The digital forecast system 7 comprises a predictor module 73 comprising a predictive modelling algorithm 731 for determining a statistical return time period probability for a specific return time period at least based on the later data 122 of the validated and selected one or more historical sets of return-to-work data 1 10. The later data 122 represent possible developments of the person-specific profile data 100 and possible outcomes for the return time period. For extracting, an accurate and significant forecast of the statistical return time period probability the predictive modelling algorithm 731 processes these later data 122. The predictive modelling algorithm 731 in this example comprises a machine learning structure 732, which is trained based on the plurality of historical sets of return-to-work data 1 10, particularly on the datasets validated and selected by the validation module 71 and the selection module 72. The machine learning structure for example comprises a decision tree structure, a random forest structure and / or neural network structure as explained above.

[0110] Further, in this example of the digital forecast system 7 the data processing unit 70 comprises an optional classification module 74 comprising a classification structure 741 for assigning a risk factor 742 depending on a deviation of a return time period associated with the highest statistical return time period probability from a standard return time period. In one example, in case of a small deviation the risk factor 742 indicates a low risk, in case of a medium deviation the risk factor 742 indicates a medium risk and in case of a high deviation the risk factor 742 indicates a high risk. The standard return time period may be received from existing external statistics resources for example indicating average return times for specific diagnosis, age groups, gender, etc. Or the standard return time can be derived from the later data 122 using a central tendency machine-structure or algorithm 743. The central tendency machine-structure or algorithm 743 is for example designed to indicate the arithmetic mean, the mode or a type of median from the return time data values 1225 of the historical sets of return-to- work data 1 10 used for determining the statistical return time period probability by the predictor module 73 or any other suitable subgroup of historical sets of return-to-work data.

[0111] Furthermore, in the shown example of the digital forecast system 7 the data processing unit 70 comprises an optional return-to-work management module 75 comprising a planning machine-structure and / or algorithm 751 for generating a personalized return-to-work action plan for the person returning to work based on person-specific profile data of one or more biological factor, psychological factor and / or social factor. The return-to-work management module 75 gathers the personspecific profile data provided by the profile data base 12, the selection module 72 and / or the data collector 4. The planning structure 751 is designed for deriving the personalized action plan by processing the person-specific profile data, particularly the additional profile data and for digitally indicating return-to-work actions. Finally, the data processing unit 70 is connected to an output unit 6 generating an output signal 52 indicating the statistical return time period probability and the risk factor. The risk factor may for example be represented as a color code. For example, a low risk is marked green, a medium risk is marked yellow, and a high risk is marked red. The data processing unit 70 may be connected to the output unit 6 via the digital network 31 . The output signal 52 is for example displayed on a computer or other electronic device.

[0112] Figure 2 illustrates method steps of a preferred digital forecast method for forecasting a probability of a return time period for a person returning to work after beginning of work absence using the digital forecast system 7 as described above. Further, figure 2 provides details of the inventive method based on the self-managed user-accessible system suite, as mentioned above.

[0113] The data processing unit 70 connects to the profile data base 12 of the profiling system 10, the data collector data base 42 of the data collector 4 and the return-to-work database 205 via the communication system 30 and the digital network 31 , respectively. The signal generator 32 generates the input signal 103 which provides profile data of the personal profile of the person being absent of work. The input signal 103 includes the pre-existing profile data 101 and the additional profile data 102 characterizing return-to-work factors related to the work absence of the person. In a profile data transmission step 81 the input signal 103 is transmitted to the validation module 71 of the data processing unit 70. The input signal at least comprises profile data values for biological return-to-work factors, like the age data value 101 1 , the gender data value 1012 and the diagnosis data value 1 13.

[0114] In a historical data retrieval step 81 1 , the data processing unit 70 retrieves historical sets of return-to-work data 1 10 for the data processing unit 70 via the data retrieval channel 37. The historical data retrieval step 81 1 provides a plurality of historical sets of return-to-work data 1 10 to validation module 71 . As explained earlier, a set of return-to-work data includes initial data 121 quantifying one or more biological factor, psychological factor and social factor at or before the beginning of work absence and later data 122 providing the return time data value 1225 indicating the return time period terminating the work absence and / or quantifying the one or more biological factor, psychological factor and social factor included in the initial data 121 at a defined time period after the beginning of work absence. It's not required to provide later data for each of the return-to-work factor quantified in the initial data.

[0115] In a validation step 812, the comparison algorithm 71 1 of the validation module 71 validates one or more historical set of return-to-work data 1 10 for the forecasting of the probability of a return time period by comparing the initial data 121 of the historical sets of return-to-work data 1 10 with the profile data 100 including the pre-existing profile data 101 and the additional profile data 102. The comparison algorithm 71 1 identifies these historical sets of return-to-work data 1 10 that correspond to at least one of the return-to-work factors addressed by the profile data. Thus, a subgroup 8121 is extracted, wherein the profile data 100 and the identified historical sets of return-to-work data 1 10 provide data values for at least one common return-to- work factor. In the present example, the comparison algorithm 71 1 identifies the historical sets of return-to-work data 1 10 comprising the initial treatment data value 1214, because the profile data includes the profile treatment data value 1014. Both data values 1214 and 1014 quantify the return-to-work factor for treatment. Further, the formatting algorithm 712 is formatting the data values characterizing the return-to-work factors in case necessary.

[0116] In a selection step 530, the selection structure 721 of the selection module 72 selects one or more historical sets of return-to-work data 1 10 of the subgroup 8121 as a basis for forecasting the return time period probability. The selection structure 721 can be configured such that data values of the return-to-work factors of the initial data match data values of the same return-to-work factors of the profile data within a value range of less than 10% of the initial data. For example, the initial treatment data value 1214 corresponds to over 90% to the profile treatment data value 1014. Additional or alternatively, the selection structure 721 can be configured such that a probability value of a statistical return time period probability determined by the predictive modelling structure 731 is below a threshold of 0.1 , preferably below a threshold of 0.01 , as explained earlier. Historical sets of return-to-work data 1 10 of the subgroup 8121 that are not selected by the selection structure 721 are excluded from the subgroup 8121 and a sub-subgroup 821 is created including only those historical datasets selected by the selection structure 721 .

[0117] In a forecast step 83, the predictive modelling structure 731 of the predictor module 73 determines a statistical return time period probability for a specific return time period at least based on the later data of the historical sets of return-to-work data 110 that are included in the sub-subgroup 821 . The machine learning structure 732 of the predictive modelling algorithm 731 refers to at least the datasets of the subsubgroup 821 as learning datasets to predict the statistical return time period probability. Additionally, the predictive modelling algorithm 731 may include further historical sets of return-to-work data for improving the learning process. Further, the predictor module generates a feedback signal 831 , which provides the probability value resulting from determining the statistical return time period probability. The feedback signal 831 is transmitted to the selection module 72 for refining the selection of the historical sets of return-to-work data 1 10 that serve as the basis for the probability forecast. As mentioned above, the selection structure 721 for example excludes historical datasets that showed significant outlier data values when applying the predictor modelling algorithm 731 . Thus, the results of the forecast step 83 serve as the basis for improving the selection of the training data for the machine learning structure 732. The feedback step can be repeated until the probability value of the determined statistical return time period probability is below the threshold. With each iteration the selection module reviews the historical sets of return-to-work data of the sub-subgroup 821 and provides a refined sub-subgroup 821 ' to the predictor module 73. The statistical return time period probability may indicate a probability for one or more specified return time periods, for example a first probability for a return period of 6 month, a second probability for a return period of 9 month and a probability for a return period of 12 month. Further, the statistical return time period probability may indicate the highest probability which corresponds to the most likely return time period. Each of these probability indications can be useful for allocating resources for the recovery process, replacing the absent worker, providing compensation payments, etc.

[0118] In a classification step 84, the classification structure 741 of the classification module 74 automatically classifies the most likely return time period comprising the highest statistical return time period probability depending on a deviation of the most likely return time period from a standard return time period. For example, the most likely return time period is classified by a low, medium or high-risk factor 742. The standard return time period can be defined by an average of the return time data of the validated or selected historical sets of return-to-work data as mentioned earlier.

[0119] In a transmission step 85, the statistical return time period probability for at least one specific return time period for the person being absent is transmitted to the signal generator 6, which generates the output signal 52 as a result of the digital forecast method. Advantageously, the digital forecast and / or prediction method may comprise a questionnaire step 86, wherein the data collector 4 of the digital autoforecast system 7 sends an inquiry for responding to the questionnaire application 41 to one or more users. The questionnaire step 86 is for example triggered once the digital forecast system 7 receives a request to forecast a probability of a return time period. Further, the questionnaire step 86 may be initiated by a trigger signal 861 generated by the predictor module 73 in case the probability value determined by the predictor module 73 does not fall below the predefined threshold. The questionnaire step 86 will provide additional profile data 102 that is likely to improve the outcome for the statistical return time period probability.

[0120] Further, the digital forecast method may comprise a return-to-work planning step 87. In the return-to-work planning step 87 the classification module 74 identifies standard data values of a standard set of return-to-work data, which result in a standard return time period. For example, the standard set of return-to-work data may be received from an external statistics source or may be determined as average values from the historical sets of return-to-work data. Next, it is determined which of the return- to-work factors of the profile data of the absent person comprise profile data values deviating from the standard data values of the standard set of return-to-work factors. These deviating return-to-work factors indicate deficiencies in the RTW situation of the absent person. Therefore, improvement measures to improve the profile data values of the return-to-work factors are determined. The improvement measures can be flexibly adapted for each person. They can address deficiencies in all categories of return-to- work factors, i.e. for biological factors, psychological factors and social factors. For example, the improvement measures include person-specific life-style improvement, treatment amendments, workplace or work task adaptation, etc.

[0121] Alternatively or additionally, in the return-to-work planning step 87 the return-to-work management module 75 may generate the personalized action plan based on the person-specific profile data of one or more biological factor, psychological factor and / or social factor. The planning structure 751 may use the preexisting profile data 101 , historical data 121 , 122, subgroup data selected by the selection module and / or additional profile data 102 from the data collector. The return- to-work management module 75 may initiate the questionnaire step 86 for example by a trigger signal analogue to the trigger signal 861 . The planning structure 751 generates the personalized action plan by connecting the person-specific profile data, particularly the additional profile data, with actions represented by digital outputs that may be included in the output signal 52. For example, the work management module 75 automatically generates a digital calendar or digital calendar entries, a digital list of action items for the person returning to work and / or of proposed solutions for return-to- work obstacles identified by the one or more biological factor, psychological factor and / or social factor, and assignments of responsibilities to the stakeholders involved in the return-to-work process. In one example, the output signal is provided to the electronic user device 5 of the absent worker and other stakeholders by the digital network 31 for display on a monitor of the devices. For example, the questionnaire application 41 could be used to generate a digital dashboard version of the personalized action plan. The personalized action plan illustrates the RTW scenario, which provides transparency of the return-to-work steps, can connect all the entities involved or affected by the work absence and facilitates the identification of improvement measures for a faster return to work.

[0122] Just by way of example, a personalized action plan may be based on additional profile data received to the question "What key things about your health problem and your work are affecting your work ability?" The answers may be collected from a set of answers provided by a pull-down menu, by free text entries, by ranking of obstacles provided in a listing, etc. Answers may for example indicate obstacles like: (1 ) I'm getting anxious and that's clouding my outlook, (2) my line manager is busy and not easy to pin down, (3) shoulder pain and weakness means I cannot do my job, (4) I cannot drive just now, and train times don't fit for getting to work on time, etc. Solution options may be represented as action items generated from these data information may include: (1 ) sign up for online therapy to be less anxious and more positive, (2) ask the department secretary to book an appointment with my line manager, (3) book a physio session to learn exercises to strengthen my shoulder, (4) start 30 minutes later to fit with train times and finish later, etc. The action items may be transformed into calendar entries and displayed in a calendar. The identified obstacles, solutions, responsible people and schedules may be displayed as listings in a digital dashboard.

[0123] While Figure 2 illustrates a procedural flow of a preferred digital forecast method, some of the method steps can be omitted without deviating from the digital forecast method as disclosed by the present invention. For example, the predictor module may determine the statistical return time period probability from the subgroup 8121 without sorting out historical sets of return-to-work data 1 10 in a selection step 530. Further, a classification of the result of determining the statistical return time period probability is not necessary but can be helpful for defining further interventions for the RTW scenario. Also, the development of a personalized action plan is optional but supports communication between the absent worker, the employer, the insurance, medical professionals etc.

[0124] The digital forecast system and method according to the present invention provides an evidence-based holistic return-to-work management tool, that is based on predictive modelling structures, particularly machine learning structures, to proactively triage return-to-work situations and tailor return-to-work management plans. The digital forecast system and method pays special attention to individual psychosocial needs of a person suffering from illness or injury. They facilitate inclusion of all stakeholders and allow for a systemic return-to-work assessment. The digital forecast system and method enable continuous improvement not only of the return-to-work measures but also of the forecasting results. The digital forecast system and method allow for the assessment of diverse return-to-work strategies and selecting to most promising for a short return time period. Overall, the consistency and the efficiency of forecasting a return time period probability is improved and the return-to-work rate is increased. Providing a holistic return-to-work approach that includes biological, psychological and social factors of a return-to-work scenario decreases costs related to work absence.

[0125] The digital forecast system and method reveal missed opportunities and lack of actions to support a quick return-to-work process. They assist in finding appropriate return-to-work measures: (1 ) for psychological factors for example interventions to improve self-efficacy, strengthen expectations of recovery, amend the perceived work ability, alleviate experienced pain and fear, and (2) for social factors for example interventions providing RTW coordinators and managers, improving understanding of workers rights, defining employer expectations, promoting health benefits of work, improving the work environment, amending the work place. Overall, the reliable and repeatable forecast of a return-to-work time period according to the invention is beneficial for all stakeholders and saves costs associated with long absence of work. The inventive system and method, inter alia, have the advantage to be able to provide a digital online smart solution for income protection of policy holders. It further allows to empower individuals to avoid long-term sickness absence. The return- to-work forecast system has the further advantage to allows supporting package aims to promote good work which includes early involvement from the employer in supporting health and recovery, since good work is important for good health.

[0126] Further the forecast system supports the goal of working while recovering. The system allows to provide suite of information and technical tools to support early RTW by (i) ensuring all players are onside and acting, (ii) shifting the culture, (iii) overcoming obstacles, and (iv) promoting work ability. The automated "smart" enhancement to an income protection policy is able to provide individuals with what, why, and how to devise / use a RTW Action Plan in the early stages (e.g. within 4 weeks of absence) . A further object of the invention is to equip an individual with a technicalbased understanding, knowledge and resources needed to effectively devise and manage their own RTW action plan.

[0127] For employees, key points of the inventive system are, to put them in driving seat of negotiating good work for their health on RTW: (i) Available to individuals irrespective of industry, condition or job role, (ii) Focuses on absence management tailored to the individual, i.e. individual-specific, automated absence management, (iii) Providing an automated online tool, (iv) Allows targeting individuals within first 4 weeks of absence, and (v) Provides an enhancement, technically improved system. The system is based on the facts that good work is important for good health, prolonged absence leads to work disability and poorer health, and early RTW is health supportive. As figure 3 illustrates, the inventive system is able to provide readiness, action, self-audit, feedback by (i) Key information empowers individuals to address obstacles and identify sources of support, (ii) If ready, the individual will be led through the stages to build their RTW Action Plan, which provides an agreed timeframe, listing who is responsible for what, how, and when, (iii) Individuals then monitor their progress by recording completion of the planned actions, and (iv) This links to a feedback loop: a follow-up strategy for individuals to reflect on their RTW journey.

[0128] In particular, the present inventive system further provides an evidence- measure-based holistic claims management system, that uses machine learning structures to proactively triage disability claims and tailor management plans. The system allows to guide claims assessors to an appropriate path of claims support, tailored to individual psychosocial needs.

[0129] As illustrated by figure 4, the system provides automated biopsychosocial claims triaging by focusing attention on complex claims and reducing it on simple ones.

[0130] Figure 5 illustrates the current technical approach of a typical prior art system. Prior art systems and technologies struggle to predict diverging return to work (RTW) likelihoods with the information their claims assessors have.

[0131] Figure 6 shows the improved technical approach using the present inventive system. The present inventive system allows to support users to automatically identify high risk claimants through integrating biopsychosocial (BPS) factors into disability income (DI) claims triaging.

[0132] Further, figure 7 illustrates how the present inventive system supports users to automatically distinguish between claimants through integrating biopsychosocial factors into claims triage.

[0133] Figure 8 illustrates how the inventive system leverages advanced analytic data processing and data science through the novel and inventive technical approach. The present inventive system provides inclusion of biopsychosocial measures and data increasing the predictive power to determine whether a claimant will RTW in 6 months or not by -20%.

[0134] The system uses data science-based data processing methods to more accurately classify disability claims and identify likelihood measure values to return to work. The advantages are, inter alia, (i) a focused and efficient resource handling by the system. The system allows an improved claims resource utilization through more targeted focus of resource on claims where a claims assessor can have the most impact; (ii) The system provides a consistent, automated claims management approach resulting in 20% RTW classification improvement with BPS questions, thereby ensuring consumers all receive a high standard of claims service and support by providing a driving best practice claims management. Using machine learning allows to provide continuous dynamic improvement, e.g. in the measuring precision etc.; and (iii) The system allows to provide a technically improved claims outcomes by anticipating improvements in claims inceptions, durations and recoveries, and in the overall automated consumer claims journey.

[0135] Finally, the present inventive forecasting system relies on physical and biopsychosocial parameter values to technically and automatically predict whether a claimant is likely to RTW within 6 months or not, thus supporting automated triage of claims support requirements, and can be used for electronic signaling of automated triage systems. Figure 9 illustrates again the above discussed in a summarizing manner.

[0136] In an embodiment variant, the digital forecast system provides predictions of work-statuses of individual patients given by the probability of returning to work within a specific time window, the system comprising, via one or more processors: training a machine learning structure based on: (i) input observation data and measured biopsychosocial (BPS) factor values, and (ii) work-statuses; inputting, into the machine learning structure, the observation data and the measured biopsychosocial (BPS) factor values of the individual patient; and predicting, with the machine learning structure, a predicted work-status measure value of the individual patient based on the observation data and the measured biopsychosocial (BPS) factor values of the individual patient.

[0137] In another embodiment variant, the forecast system predicts a work-status measure value of an individual patient, the forecast system comprising one or more simulation and data processing processors configured to: train the machine learning structure based on: (i) input observation data and measured biopsychosocial (BPS) factor values, and (ii) work-statuses; input, into the machine learning structure, observation data and measured biopsychosocial (BPS) factor values of the individual patient; and predict, with by means of the machine learning structure, the work-status measure value of the individual patient based on the observation data and measured biopsychosocial (BPS) factor values of the individual patient.

[0138] In yet another embodiment variant, the inventive automated forecast system for predicting a work-status measure value of an individual patient, the forecast device comprising: one or more simulation and data processing processors; and one or more memories and / or persistence storage coupled to the one or more simulation processors; the one or more memories and / or persistence storage including executable machine learning structures stored therein where specific processor instructions of the system, when executed by the one or more simulation processors, cause the one or more simulation processors to: train the machine learning structures based on: (i) input observation data and measured biopsychosocial (BPS) factor values, and (ii) workstatus measure values; input, into the machine learning structures, observation data and measured biopsychosocial (BPS) factor values of the individual patient; and predict, by means of the executed machine learning structures, a predicted workstatus measure value of the individual patient based on the observation data and measured biopsychosocial (BPS) factor values of the individual patient.

[0139] In some embodiments, the work-status measure value is a return to work (RTW) status measure value providing a predicted probability measure value for an individual to return to work within a certain time window. The inventive forecast system does not have the disadvantage of the prior art system where at least one of the factors limiting the forecasting process of return to work (RTW) following work disability is the use of measurement tools that are not able to capture a complete picture of workers' RTW experiences. To facilitate the forecasting of RTW, the current inventive system can e.g. be based on a developmental conceptualization structure of RTW, which is based on the assumption for an expanded awareness that encompasses four phases: off work, work reintegration, work maintenance and advancement.

[0140] For the present system, it is necessary to establish some key terms. Because RTW outcomes are often multidimensional (e.g., the act of working can be quantified in term of hours, duties, workplace, and employer), measuring RTW requires strategies that accommodate the capture of multiple dimensions. For the present system, the distinction between RTW outcomes, dimensions, and instruments can e.g. be made as follows: (i) RTW outcomes are measurable characteristics of workers' RTW status or experience, (ii) RTW dimensions describe the relevant parameters of a RTW outcome, and (iii) the present RTW forecast system is an electronic device developed for RTW forecast technology that is able to automatically quantify RTW dimensions. Each identified RTW outcome or RTW measure value is articulated by at least one RTW dimension (i.e., a measurable feature or indicator) . The multitude of RTW measure values i.e. the RTW outcomes and their applicability across various RTW phases can necessitate the grouping of outcomes according to additional criteria. As a starting point, the International Classification of Functioning, Disability and Health (ICF) can e.g. be taken. The ICF can e.g. be chosen since it is a widely approved classification system, which provides a description of situations with regard to human functioning and its restrictions within the context of the real world. Figure 10 illustrate the structure of disability that can serve as the basis and describes the interactions between its various components. RTW can e.g. also be considered within the activities and participation domains, which are said to give an indication of the experience of disability. These domains, which can e.g. be described as tasks and actions, are applied in a way to cover the full range of life areas in which a person can be involved. Tasks and actions are normally neutral, but they can also be used to record positive or negative features of individual's functioning.

[0141] At the highest level, the activities and participation domains can e.g. be coded as follows: Learning and applying knowledge, general tasks and demands, communication, mobility, self-care, domestic life, interpersonal interactions and relationships, major life areas, and community, social and civic life. Employment can e.g. be included within the major life areas. However, since ICF, as a basis, fails to specify the content of some of its main claims about the nature of impairment and disability, the parameterization has to be adapted. Thus, if started by the ICF framework, for the present system, it is not possible to draw RTW task and Action outcomes directly. The ICF states are therefor used in conjunction with a further definition of RTW task and action outcomes within the inventive more general structure.

[0142] Although injured workers' RTW tasks and actions may encompass much of what can be considered as RTW outcomes for the present system, there can be added also additional outcomes. For example, there may be a measure to know if the RTW was managed appropriately, or if a modified RTW option is offered. To classify such outcomes, the present inventive system focuses on contextual factors that represent the background of an individual's life and encompass two components: environmental and personal. Even further components, as e.g. "Body Functions and Structures," can in some embodiment variants be neglected as not relevant to the measurement of RTW outcomes for it focuses on the body's physiological functions and parts. In addition to tasks and actions and contextual RTW outcomes, the inventive system can e.g. use a group of structures that focus on measuring aspects of the RTW process, i.e. the process that workers go through to reach, or attempt to reach, their goals. The inventive system can e.g. group RTW process outcomes based on their primary focus: time, service, human resources, productivity, or cost.

[0143] For RTW tasks and actions, this group of outcomes captures and parameterizes tasks and actions that workers undertake throughout the RTW process. At the highest level, there is interest in determining an individual's vocational participation (i.e., the characteristics of his or her vocational life), which is a concept that is broader than RTW. Several measures of vocational participation can e.g. be used by the RTW the forecast system. These include labor force participation (distinguishing between employed, unemployed but seeking work, and those unemployed but not seeking work), vocational mode (distinguishing between those in positive, i.e. employed and volunteer workers, students and homemakers, i.e. non-positive vocational mode), and vocational status (distinguishing between those employed, unpaid but productive, and unemployed). An individual's work participation characteristics can be technically important for the present system. Work characteristics can be measured in numerous ways. A plethora of instruments exists to assess work-related abilities, facilitating the measurement of different aspects of ability that may be of interest: physical, mental, or functional. Among instruments describing work abilities, a distinction can also be made between those operating outside of a specific job context and those that include specific job requirements. In the RTW forecast system, structures incorporating and parametrizing specific job characteristics are used as functional or work capacity assessments and can be used either just prior to work re-entry or during the time that the worker adjusts to being back at work and include assessments based on self or expert report and examination of physical and environmental factors.

[0144] Several other dimensions can e.g. be comprised to capture RTW outcomes prior to work re-entry. The extent to which a person pursuing RTW is or was ready to resume employment can be evaluated (i.e., work readiness). Another set of dimensions focuses on tasks and actions that workers undertake when they are off work either preparing for resumption of employment by creating a RTW plan or undertaking vocational rehabilitation (i.e., work preparation) and seeking employment opportunities (i.e., job seeking). Once resumption of employment is achieved, the maintenance or durability of such status can be evaluated by reference to dimensions focused on either interruption to work resumption (e.g., work-disability recurrence or number of RTW attempts), or time back at work. Work participation can also be parametrized for the present forecast system in terms of the type of actions undertaken by workers resuming employment. Depending on the measuring precision, work participation can be measured in different ways-. For example, the focus can be placed on simply the act of returning to competitive activity or include such detail as the number of hours worked and type of duties performed (full, light, or modified, i.e., with accommodations). A distinction can also be made between returning to the same or a new job and the same or new employer. A parameter structure can e.g. be used to measure whether workers perform their activities in good health, and evaluations exist to determine if an agreed upon goal has been reached, with goal status e.g. being definable depending on the user specific needs. Perhaps one of the most important indicators of work participation is workers' productivity following their return to the workplace. Specified parameterizations can e.g. comprised in the system and used to assess productivity include both assessments based on recorded productivity data and measures based on individuals' assessment of how they performed their duties. The subjective parameterization structures include various dimensions rated by the worker. For example, a worker limitations questionnaire can incorporate limitations with handling time, physical limitations, mental limitations, and limitations regarding handling work demands, whereas an occupational role questionnaire can e.g. consist of two scales, which are productivity scale and satisfaction with work scale. Another RTW action can be an evaluation of the RTW. Such evaluations can include aspects of RTW important both in the short or long term. From the workers' perspective, their perception of achieved outcomes is an important metric and, depending on the phase of RTW, can include measurement of job satisfaction, satisfaction with RTW, or satisfaction with their status.

[0145] Further, for the contextual RTW outcomes, the inventive system can e.g. classify contextual RTW outcomes as 'environmental' or 'personal.' Environmental outcomes capture RTW-related physical, socioeconomic, and attitudinal data external to the injured worker. Personal outcomes capture detail on the injured worker's personal and non-behavioral contextual RTW experience (by and large focusing on worker's cognitions and emotions). The system can e.g. use measures of personal and environmental contextual RTW outcomes. In most situations, this type of data are collected with the aim of identifying persons at risk of poor outcome; however, this type of information may also be used as an intermediate outcome, giving some indication as to whether the individual is on the path to success. An example of an intermediary outcome is RTW cognitions, for these have been found for the present system to strongly predict long-term success. As such, RTW cognitions are a feature of the injured workers RTW experience that can be measured at any time and used to predict the likelihood of long-term success. RTW cognitions including intentions (i.e., workers' desired outcomes) and expectations (i.e., workers' assessment of probability of desired outcomes) can e.g. be captured using survey instruments. In the case of RTW expectations, these instruments are either single questions asking about an aspect of expectations following work disability (pain, activities of daily living, functional impact, recovery, work resumption), or an expectations measure based on responses to several questions. Along with intention- and expectation-measures, motivation measures can be used both to predict outcomes and as an outcome itself. To measure RTW motivation, the system can e.g. comprise dimensions such as a worker's engagement in his or her job, as well as the role that work plays in his or her social life. A worker's need for income can also influence RTW motivation. It has been found that individuals receiving workers' compensation benefits have longer duration of time off work, which can be explained by lower motivation to return to employment. Consequently, the present system can comprise compensation status as a proxy for workers' motivation to resume employment. Another important RTW outcome can be workers' satisfaction. When forecasting RTW status, workers' thoughts regarding vocational participation can be considered by the system. The fact that a worker is not at work need not be interpreted as negative if this is acceptable within the worker's personal and environmental context. For those who do return to work, satisfaction with working conditions can become relevant for the measuring. At any point throughout the RTW, measurements and quantifying assessments can be made of the interaction between workers and their environment, i.e., environmental contextual factors. The system can also capture parameters determining whether the injured worker is receiving benefits, or whether payments for time off work are being made. In addition, measurements can be made and comprised in the system for capturing the nature of the interaction between the worker and others involved in the RTW process, including payers, peers or supervisors, and providers. For RTW-process Outcomes, several parametrizing structures can be comprised by the system that specifically capture quantified aspects of the RTW process, i.e., the process that workers go through to reach, or attempt to reach, their goal. Process measurement tends to involve either a description of changes over time or a comparison to a standard or expectation. Repeated measurement can be applied; however, in many cases an implied value exists (e.g., when work disability begins) and repeated measurement is not required. This is certainly the case for outcomes that are time orientated — in most situations, previously-used RTW dimensions such as duration of time off work or compensated lost work time have a value of zero when the worker becomes work-disabled. The time it takes the worker to achieve important RTW milestones is a less frequently used dimension but may be as important in the context of the conceptualization (i.e., time to milestone achievement). For example, the system can comprise structures evaluating time to return to workplace and time to maintained RTW. Similarly, RTW patterns can be examined either by focusing on recurrence of work disability or patterns of employment following a work- related injury. In addition to time-orientated RTW-process outcomes, the RTW forecast system can e.g. use various structures to capture other quantifiable aspects of the process. Among those, monetarily orientated outcomes, such as the cost of RTW, can also be of interests. These costs can include health care costs, indemnity (or absenteeism) costs, or vocational rehabilitation costs. Other forecasted aspects of the RTW process focused on lost productivity and utilization of various services during the period of work disability, particularly utilization of health care services.

[0146] It is to be noted, that physicians and payers managing the recovery of patients suffering from occupational injuries have to manage large amounts of longitudinal (temporal) data. The present inventive system provides a new technical framework for dynamic prediction of work-status and return to work, that yields daily predictions for the future work-status of patient's given new information, utilizing a combination of deep learning and a novel data transformation pipeline. In tandem, these methods allow for prediction of patient outcomes with respect to complex clinical information observed over time. In some example implementations, prediction experiments were conducted on a large claims' dataset, with good out-of-sample discriminative performance and calibration. The following will demonstrate that it is possible to use widely available data to predict return to work, allowing physicians and payers to manage large populations of injured patients more efficiently. As discussed above, every year Ols affect millions of patients and cost workplaces and payers hundreds of billions of dollars. The true burden of these injuries is likely to be significantly underestimated. In addition to physical symptoms, patients with Ols often experience complicating psycho-social issues, such as depression; however, these complicating issues are rarely detected or treated. Together, these factors provide an incentive for patients, workplaces, physicians, and payers to understand the amount of time a patient will be away from work, with the goal to eventually help minimize it.

[0147] For the present inventive system, the identification of the factors affecting time of RTW are extensively discussed above. Again, major groups of factors include patient specific, injury related, professional, workplace related, treatment, and psychosocial factors. Examples of a selection of specific factors in each of these groups are shown in table 1 below. Predictors of shortened RTW duration include job control, work ability, perceived (good) health, and high socio-economic status. Some of the predictors of lengthened RTW duration include job strain, anxiety & depression, comorbidities, older age, and low education. The above parametrization illustrates how RTW can be shaped, quantified, and made measurable by a specific technical selection of the measuring factors that vary across patients, workplaces, and injuries.

[0148] The RTW modeling process can e.g. be taken the form of a time-to-event prediction task. Much of the modelling work can be performed treating RTW as a single event, whose occurrence time is predicted when a patient is initially injured. A possible modelling structure used which can be used herein is the Cox proportional hazards model structure. For example, there can be time to RTW prediction using hazard models, with extensions to predict length receiving benefits, and prolonged claim identification. Machine learning (ML) techniques, such as decision trees, Naive Bayes, and gradient boosted machines, can also be applied to situations involving Ols other than predicting RTW. For prediction of RTW, a time-to-event approach can be used focusing on ancillary prediction tasks. Two examples may be appropriate rehabilitation intervention selection and classification of a patient's final work state disposition.

[0149] Some embodiments disclosed herein provide a new technical approach to modeling RTW prediction, one that can be used dynamically, unlike the static time to event prediction methods, which is the industry standard of all known prior art systems. This involves moving from a framework that treats the input and outcome as fixed, to a framework that treats the input as a sequence of information and the output as another, related sequence of information. The input sequence can be thought of as the sequence of all the collected data, or observations of a patient. And the output sequence can take several forms, either directly predicting the time-to-event of RTW, or indirectly predicting RTW by estimating probabilities of work-status at future time points. Thus, there are two sequences; and the desired task is to sequentially predict the outcome sequence given the observation sequence, a sequence-to-sequence prediction task.

[0150] Markov chains can e.g. be applied to achieve the inventive dynamic sequences. A Markov Chain is a stochastic process that enforces a conditional distribution between the current state Xt and the future state Xt+i. A key constraint is that given a current state (Xt), the next state (Xt+i) is independent of all the previous states (Xo, Xi Xt-i), meaning that future states are only dependent on the current state. Markov models are often used in medicine, due to their simple structure and ease of clinical interpretability. An extension of the Markov chain is that of the Hidden Markov Model (HMM), which enables to provide a modeling of a sequence of observed signals while the actual sequence underlying the Markov chain is unobserved. HMMs can be used to study sequences generated from systems with stochasticity, and they can be used in the field of medicine from studying protein sequences, to analyzing human movement, and even pre-dicting treatment decisions. Despite their wide use, Markov chain-based models are limited by their underlying formulation, which restricts the sequential dependence of (Xt+i ) to only (Xt) . More complicated processes (such as depending on the previous 3 time-steps) can be transformed into a Markov chain formulation by redefinition of the states. Thus, fixed length histories can be embedded into the current state, which allows for representation of history by state-space expansion. However, this comes at the expense of a much larger state-space. Another common assumption in Markov chain based mod-els is time-homogeneity, as the probability of transitioning to (Xt+i) depends only on (Xt), independent of the current time-step, unless this dependence is represented in the state-space. The observations for patients returning to work are high dimensional, as they include several types of categorical variables that may take on many possible values. An example of this is diagnosis, where there are thousands of possible diagnoses for patients injured at work, and on any given day a patient may have zero, one, or more diagnosis codes assigned to them. Treatment is another high-dimensional category. The timing and order of treatments may impact the recovery of an injured patient; thus, the history beyond the current observation is important in modeling RTW. This high-dimensionality and history dependence makes Markov chain-based models ill- suited for the present task. Dimensionality could be reduced by either restricting the problem definition to a specific disease (e.g. lower back sprain) or by lowering the dimensionality by grouping. However, both of these require significant effort to create and validate specialized models. In contrast, the present inventive system focuses on a generalizable approach that applies to a broad range of injuries.

[0151] Some embodiments begin by providing a motivating example to help lay the groundwork for the model formulation. When a patient first experiences an Ol, there is very limited data that may include the time and type of injury and basic demographic information about the patient. As time progresses, additional observations are acquired including diagnoses and treatments. These observations con-tribute to defining the patient's health status and may help predict the patient's future work-status. Over the course of an injured patient's case, from the first claim to the last claim, a patient's work-status may transition between not-working and working multiple times. Some embodiments aim to forecast the probability that they will be at work at a given future date (e.g. 30 days) . Therefore, some embodiments treat RTW- status as a binary state variable that is perfectly observable. For the training procedure results in Pn,t, between 0 and 1 can be interpreted as probabilities.

[0152] The inventive system can e.g. use an RNN to build a model structure that maps the input sequences (Xn,t) to the output sequences (Yn,t) , while simultaneously building and utilizing history representations. Some embodiments use the above transformation methods to pre-process the data. To train the model, some embodiments use an objective function that minimizes binary-cross entropy and / or use a special generator function to pad the variable lengths of patient sequences in a set batch.

[0153] List of references Smart digital auto-guidance device

[0154] 10 Profiler or profiling system

[0155] 12 Profile database

[0156] 100 Profile data

[0157] 101 Pre-existing profile data

[0158] 101 1 age data value

[0159] 1012 gender data value

[0160] 1013 diagnosis data value

[0161] 1014 treatment data value

[0162] 102 Additional profile data

[0163] 1021 profile self-efficacy data value

[0164] 1022 profile fear experience data value

[0165] 1023 profile perceived work ability data value

[0166] 1024 profile membership data value

[0167] 103 Digital input signals

[0168] 1 10 Historical data sets of return-to-work data

[0169] 120 Return-to-work database

[0170] 121 Initial data

[0171] 121 1 Initial self-efficacy data value

[0172] 1212 Initial fear experience data value

[0173] 1213 Initial perceived work ability

[0174] 1214 Initial treatment data value

[0175] 122 Later data

[0176] 1221 Later self-efficacy data value

[0177] 1222 Later fear experience data value

[0178] 1223 Later perceived work ability

[0179] 1224 Later treatment data value

[0180] 1225 Return time data value

[0181] 130 User-specific parameters and / or parameter pattern

[0182] 1301 User-specific obstacle parameter pattern

[0183] 1302 User specific profile parameter pattern

[0184] 14 Extractor

[0185] 141 Extracted user dataset parameter vector Electronic plan builder engine

[0186] 21 Digital, return-to-work action guidance and / or digital guidance plan

[0187] 210 Phased return-to-work action guidance

[0188] 21 1 Guidance for overcoming a first obstacle

[0189] 212 Guidance for overcoming a second obstacle

[0190] 21 i Guidance for overcoming a i-th obstacle

[0191] 22 Digital, return-to-work actions

[0192] 221 Digital, return-to-work action targets

[0193] 222 Date of return-to-work action target attainment

[0194] 23 Simplified, return-to-work action guidance Communication system

[0195] 31 Digital network

[0196] 32 Signal generator

[0197] 33 Cellular mobile network

[0198] 34 Satellite transmission line

[0199] 35 Interface

[0200] 36 Interface

[0201] 37 Data retrieval channel

[0202] 38 Interface Data collector

[0203] 41 Questionnaire application

[0204] 42 Data collector database , 5', 5" Electronic user device

[0205] 51 User interfaces

[0206] 51 1 Visual representation

[0207] 51 1 1 Coded Cell

[0208] 51 12 Digital template

[0209] 51 121 First group

[0210] 51 122 Second group

[0211] 51 13 Additional coded cell for phased guidance Output unit / Signal generator

[0212] 61 Electronic output signals Digital parameter-based auto-prediction and / or forecast system

[0213] 70 Data processing unit 71 Validation module

[0214] 711 Comparison structure

[0215] 712 Formatting structure

[0216] 72 Selection module

[0217] 721 Selection structure

[0218] 73 Predictor module

[0219] 731 Prediction modelling structure

[0220] 732 Machine learning structure

[0221] 733 Forecasted return time period based on measured or captured input

[0222] Parameters to the prediction modelling structure 731 or machine learning structure 732

[0223] 74 Classification module

[0224] 741 Classification structure

[0225] 742 Risk factor

[0226] 743 Central tendency machine-structure and / or algorithm

[0227] 75 Return-to-work management module

[0228] 751 Planning structure Processing steps

[0229] 81 Profile data transmission step

[0230] 811 Historical data retrieval step

[0231] 812 Validation step

[0232] 8121 Historical subgroup

[0233] 82 Selection step

[0234] 821 , 821 ' Sub-subgroup

[0235] 83 Forecast step

[0236] 831 Feedback signal

[0237] 84 Classification step

[0238] 85 Transmission step

[0239] 86 Questionnaire step

[0240] 861Trigger signal

[0241] 87 Return-to-work planning step User / Patient

[0242] 90 User identification

[0243] 91 Clinician

[0244] 92 Case or line manager 93 Patient database

[0245] 94 Input

[0246] 941 Health state model

[0247] 95 Prediction 96 Return obstacles

[0248] 961 Return obstacles parameters

[0249] 962 Obstacle identification

[0250] 963 User-specific obstacle parameter pattern

[0251] 97 Work absence 971 Start of work absence - Point in time marking a beginning of a work absence

[0252] 972 Return to work - Point in time marking an end of a work absence

Claims

Claims1 . A smart digital auto-guidance device ( 1 ) based on a dynamically generated, interactive, worker-centric and personalized, digital, return-to-work action guidance (21 ), the digital guidance (21 ) navigating a user (9) efficiently back to work after a point in time marking a beginning (971 ) of a work absence of the user (9), characterized in that the smart digital guidance device (1 ) comprises an electronic plan builder engine (2) connected to a plurality of user interfaces (51 ), wherein one or more possible return obstacles (96) are generated defined by dedicated obstacle parameters (961 ) by the plan builder engine (2), the return obstacles (96) hindering or delaying a return-to-work (972) of a user (9), and are selectably indicated via a user interface (51 ) to the user (9) within a coded cell (51 1 1 ) of the interface (51 ), and wherein, upon selection, assigned obstacle identifications (962) are transmitted back to the plan builder engine (2) as a user-specific obstacle parameter pattern (1301 ), in that, based on user-specific parameter pattern (130) comprising the transmitted user-specific obstacle parameter pattern (1301 ) and user-specific profile parameter pattern (1301 ), the plan builder engine (2) generates pattern-specific obstacle guidance (21 1.211) to each of the obstacles (96) selected by the user (9), and are indicated via the user interface (51 ) to the user (9) within the coded cell (51 1 1 ), an obstacle (96) to be marked upon completing all obstacle guidance (21 1.211) steps assigned to a specific obstacle (96), the steps to be performed by the user (9), and in that the selected obstacles (96) and pattern-specific obstacle guidance (21 1.211) are continuously converged by the electronic plan builder (2) dynamically listing selected obstacles (96) and marked or unmarked pattern-specific obstacle guidance (21 1.211) in two groups visualized by a corresponding digital template(51 12) of said coded cell (51 1 1 ), wherein a first group (51 121 ) comprises guidance steps to be completed before return to work and a second group (51 122) comprises solution to be completed after return to work, wherein the digital template (51 12) at least comprises the generated, digital return-to-work action guidance, and wherein at leastparts of the provided solutions have to be marked as completed by the user before or after the user's return-to-work.

2. A smart digital auto-guidance device (1 ) according to claim 1 , characterized in that, if required, a phased return to work guidance (21 1 ) is selectable via said user-interface, wherein the digital template (51 12) at least comprises the generated, phased or not phased digital return-to-work action guidance (21 ) for the user's navigation back to work.

3. A smart digital auto-guidance device ( 1 ) according to claim 2, characterized in that, if a phased return to work is selected, the return-to-work action guidance (21 ) is populatable within an additional cell (5133) by the electronic plan builder engine (2) via the user interface (51 ) .

4. A smart digital auto-guidance device ( 1 ) according to claim 3, characterized in that the cell is realized as a digital tick box.

5. A smart digital auto-guidance device ( 1 ) according to one of the claims 2 to 4, characterized in that the digital template (21 1 ) further comprises alongside a phased digital return-to-work action guidance (21 ) .

6. A smart digital auto-guidance device (1 ) according to one of the claims 1 to 5, characterized in that the user is requested by electronic plan builder engine (2) via the user interface (51 ) to add further elements to the generated, digital return-to- work action guidance (21 ) at least comprising a return-to-work date and / or access authorization for users to share the generated, digital return-to-work action guidance with and / or date of guidance target (22) attainments (222) and / or responsibilities regarding the guidance action (22) to be completed.

7. A smart digital auto-guidance device (1 ) according to one of the claims 1 to 6, characterized in that after the generation of digital return-to-work action guidance (21 ), a simplified return-to-work action guidance (23) comprising final return to work actions (22) are indicated via the user interface (51 ) extracting main details to be shared with a line or case manager (92) and / or other user (91 ), and focusing apossible return-to-work conversation between the user (9) and the line or case manager (92) .

8. A smart digital auto-guidance device (1 ) according to one of the claims 1 to 7, characterized in that the smart digital auto-guidance device (1 ) comprises a digital forecast system (7) for forecasting a return time period having a highest occurrence probability value for an individual returning to work after beginning of work absence, the forecasting and / or measuring being triggered by user-specific profile data of the user (9), wherein possible return obstacles are weighted based their impact on the forecasted a return time period, and wherein only return obstacles (96) having an impact to the return time period higher than a predefined threshold value are indicated to the user (9) via the user interface (51 ) .

9. A smart digital auto-guidance device (1 ) according to claim 8, characterized in that a plurality of return-to-work factors are extractable from the userspecific data (100), wherein the digital forecast system (7) comprises or is connected to a profile database (12) providing at least parts of the person-specific data (100) by a personal profile of the individual, a data processing unit (70) of the digital forecast system (7) receiving the person-specific data (100) as an input signal (103) and assessing the profile data (100) to deduce the forecasted probability value of the return time period, and an output unit (6) generating an output signal (52) indicating the forecasted return time period with the highest occurrence probability value, the digital forecast system (7) further comprising: a return-to-work database (205) storing a plurality of historical sets of return- to-work data (1 10), wherein a set of return-to-work data (1 10) includes initial data (121 ) quantifying one or more biological measuring factor, psychological factor and social factor at and / or before the beginning of work absence and later data ( 122) quantifying return time data (1225) indicating a return time period terminating the work absence and / or quantifying one or more biological factor, psychological factor and social factor at a defined time period after the beginning of work absence, and a communication system (3) at least comprising a digital network (31) and a signal generator (32), wherein the communication system (3) is connected orconnectable linked to the profile database (12) via the digital network (31 ) for generating the digital input signal by the signal generator (32), wherein the data processing unit (70) is connected to the communication system (3) and the return-to-work database (205) for receiving the digital input signal (103) and historical sets of return-to-work data (1 10), the data processing unit (70) at least comprising a validation module (71) comprising a comparison algorithm (71 1 ) identifying one or more historical sets of return-to-work data (1 10) comprising initial data (121 ) corresponding to at least one return-to-work factor of the profile data (100), and a predictor module (73) comprising a predictive modelling structure (731 ) for determining a statistical return time period probability for a specific return time period at least based on the later data (122) of the validated one or more historical sets of return-to-work data (8121 ), and wherein the data processing unit (70) is connected to the output unit (6) for providing the statistical return time period probability as the output signal (52).

10. A smart digital auto-guidance device (1 ) according to claim 9, characterized in that the profile data (100) includes pre-existing profile data (101 ) indicating data values at least of biological factors of the person.11 . A smart digital auto-guidance device ( 1 ) according to one of the claims 9 or 10, characterized in that the profile database (12) is hosted in a digital profiler system (10), which is connected to the communication system (30) .

12. A smart digital auto-guidance device (1 ) according to one of the claims 9 to 1 1 , characterized in that the data processing unit (70) comprises a return-to-work management module (75) comprising a planning structure (751 ) for generating a personalized return-to-work action plan for the person returning to work based on person-specific profile data of one or more biological factor, psychological factor and / or social factor.

13. A smart digital auto-guidance device (1 ) according to one of the claims 9 to 12, characterized in that the digital forecast system ( 1 ) comprises or is connected to a digital data collector (4) hosting a questionnaire application (41 ) for collecting additional profile data ( 102) of one or more biological factor, psychological factor and / or social factor about the person being absent from work, wherein the questionnaire application (41 ) is accessible by an electronic user device (5) via the digital network (31 ) .

14. A smart digital auto-guidance device ( 1 ) according to claim 13, characterized in that the data collector (4) is connected or connectable to the communication system (30) for storing the additional profile data (102) in a database (42) and / or providing the additional profile data ( 102) as input signal ( 103) to the data processing unit (70) .

15. A smart digital auto-guidance device (1 ) according to one of the claims 9 to 14, characterized in that a data value of a biological factor is provided by an indicator for age (101 1 ), gender ( 1012), diagnosis (1013) and / or treatment (1014) .

16. A smart digital auto-guidance device (1 ) according one of the claims 9 to 15, characterized in that a data value of a social factor is provided by an indicator for marital status, profession, employment status, a number of children, a number of family members and / or a number of memberships (1024) .

17. A smart digital auto-guidance device (1 ) according to one of the claims 9 to 1 6, characterized in that a data value of a psychological factor or social factor is provided by a ranking of a psychological statement or social statement on a numerical scale, alphabetical scale, color scale or descriptive scale.

18. A smart digital auto-guidance device (1 ) according to one of the claims 9 to 1 7, characterized in that the data processing unit (70) comprises a selection module (72) selecting these historical sets of return-to-work data (821 ) identified by the validation module (71 ) wherein a data value of the initial data ( 121 ) of the corresponding return-to-work factors matches a data value of the profile data ( 100) for determining the statistical return time period probability by the predictor module (73) .

19. A smart digital auto-guidance device (1 ) according to one of the claims 9 to 18, characterized in that the data processing unit (70) comprises a selection module (72) selecting one or more historical sets of return-to-work data (821 ') identified by the validation module as a basis for forecasting the return time period probability such that a probability value of the statistical return time period probability determined by the predictive modelling algorithm is below a threshold of 0.1 , preferably below a threshold of 0.05.

20. A smart digital auto-guidance device (1 ) according to claim 19, characterized in that the predictor module (70) provides a feedback signal (831 ) to the selection module (72), wherein the feedback signal (831 ) provides the probability value resulting from the one or more identified historical sets of return-to-work data (821 ') providing the basis for forecasting the return time period probability.

21. A smart digital auto-guidance device (1 ) according to claim 20, characterized in that the data collector (4) collects return time data indicating a return time period of the person returning to work after work absence and stores the collected additional profile data (102) of the one or more biological factor, psychological factor and / or social factor together with the collected return time data ( 1225) as a historical set of return-to-work data (1 10) in the return-to-work database (205) .

22. A smart digital auto-guidance device (1 ) according to one of the claims 9 to 21 , characterized in that the predictive modelling algorithm (731 ) comprises a machine learning structure (732), which is trained based on the plurality of historical sets of return-to-work data (1 10).

23. A smart digital auto-guidance device (1 ) according to claim 22, characterized in that the machine learning structure (732) comprises a decision tree structure, a random forest structure and / or neural network structure.

24. A smart digital auto-guidance device (1 ) according to one of the claims 9 to 23, characterized in that the data processing unit (70) comprises a classification module (74), which assigns a risk factor (742) depending on a deviation of a return time period associated with the highest statistical return time period probability from a standard return time period.

25. A smart digital auto-guidance method for realizing the digital, return-to- work action guidance (21 ) of a smart digital auto-guidance device (1 ) according to one of the claims 1 to 24.

Images