Computer-implemented method for predicting future operating conditions of healthcare laboratory

Abstract

To provide a computer-implemented method for predicting future operating conditions of a healthcare laboratory.SOLUTION: The method comprises: receiving medical sample event data describing events relating to one or more medical samples to be processed by a healthcare laboratory; augmenting the medical sample event data with laboratory data by a cloud system 102; generating, from the augmented medical sample event data, a feature set including time bracketed data derived from the augmented medical sample event data; using the feature set as an input for a trained machine learning model that has been trained on historical feature set to predict a future medical sample event within a prediction time range; and receiving, from the trained machine learning model, prediction of the future medical sample event for the prediction time range as an output.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a computer-implemented method for predicting future performance of a healthcare laboratory. [Background technology]

[0002] Modern healthcare laboratories for processing medical samples are increasingly employing automated processes in their operations. Such laboratories typically have the capacity to process thousands of samples per day, with larger automated laboratories even processing up to 100,000 samples per day. In such automated systems, the process is typically monitored by a central computer system, which is responsible for managing the flow of medical samples through the laboratory. In addition, the central computer system often performs quality control and generates test reports. The system may monitor, for example, the ordering of medical processing of samples by hospitals or general practitioners, as well as the transportation, processing, and return of medical samples. Such central processing systems are typically capable of generating reports regarding the status of the laboratory's medical sample processing.

[0003] These reports are intended to enable healthcare laboratory managers to efficiently allocate laboratory staff and sample processing equipment to process required medical samples. However, conventional systems do not have access to sufficient data to adequately inform these decisions and do not provide anything beyond real-time value regarding the medical samples within the system.

[0004] The present disclosure has been made in light of the above considerations. Summary of the Invention

[0005] Accordingly, in a first aspect, embodiments of the present invention provide a computer-implemented method for predicting future performance of a healthcare laboratory, the healthcare laboratory including one or more in vitro diagnostic laboratory devices configured to process medical samples, the method comprising: receiving medical sample event data describing events relating to one or more medical samples processed by the laboratory; Augmenting medical sample event data with laboratory data; and generating a feature set from the augmented medical sample event data, the feature set including time period data derived from the augmented medical sample event data; using the feature set as input for a trained machine learning model trained on the historical feature set to predict future medical sample events within a prediction time range; receiving, as output from the trained machine learning model, a prediction of future medical sample events for a prediction time range; Includes:

[0006] An in vitro diagnostic laboratory device can be a pre-analytical device, an analytical device, a post-analytical device, or a device used for transporting medical samples. For example, a healthcare laboratory can be an in vitro diagnostic laboratory that includes at least one analytical device.

[0007] Medical sample event data may include, for example, data describing a medical sample order and data describing the transportation of a medical sample between a laboratory and a sample ordering entity. Medical sample data associated with a particular event may, for example, describe the date and time of the event and may optionally include additional event-specific metadata. Medical sample event data may be understood as pre-laboratory data, i.e., data related to a sample before it is delivered to the laboratory and therefore before it is processed by the laboratory. It may further describe the number of samples collected from different healthcare providers, the date and time a given sample was ordered, data available from one or more pre-laboratory sample tracking systems, and data available from one or more systems associated with sample shippers or transporters external to the laboratory. Medical sample event data may, for example, be data from pre-laboratory sample tracking systems and / or systems associated with sample shippers or transporters. Time-of-day data may refer to data in which values ​​are placed into corresponding bins according to their time values. For example, time-of-day data may include bins with time labels indicating time periods. A corresponding counter may indicate the number of data points belonging to a given category for that time period. In one example, the daypart data may be ordered by time periods within a given day, or the daypart data may be ordered by date.

[0008] In one example, the laboratory data includes dynamic laboratory data and static laboratory data.

[0009] Laboratory data may be understood as laboratory data, i.e., data originating from and relating to a laboratory, such as the number of samples arriving at multiple laboratories or at each of multiple laboratories over time as historical data, when the samples arrived at the laboratory, when various processing steps occurred in relation to the samples (e.g., data generated by equipment and collected by middleware), data describing events that occur to samples, analytical, pre-analytical, or post-analytical data from equipment, data from information systems (e.g., laboratory information systems, healthcare information systems, and / or middleware), data available from sample monitoring applications.

[0010] Laboratory data may include static data and dynamic data. Static data can be understood as data that does not vary or change significantly over time (e.g., the average time it takes for a sample to arrive at the laboratory from the clinic). Dynamic data can be understood as data that changes over time (e.g., the number of samples to be processed, the amount of reagents available in an analytical device, etc.). Thus, static laboratory data is laboratory data that is not subject to continuous change and may describe, for example, in vitro diagnostic laboratory equipment throughput data, in vitro diagnostic laboratory equipment availability data, staff scheduling data, workflow definitions (e.g., the process flow of a given test, staff assignments for a given test, the priority of a given test, and / or equipment settings for a given test, etc.), and / or laboratory layout data. Dynamic laboratory data is continuously changing and may describe, for example, medical sample event data for medical samples currently being processed (as opposed to samples to be processed by the laboratory), current reagent levels in the laboratory, and / or the current status of each in vitro diagnostic laboratory device. Static medical sample event data may include, for example, the average time it takes to transport samples from a clinic to a laboratory for processing. Dynamic medical sample event data may include, for example, the location of a medical sample. A feature set may be a dataset including features, where each feature is a measurable characteristic derived from the augmented medical sample event data. For example, the average number of samples ordered may be used as a feature. A value may be calculated for each feature for each time period of the time-period data. Dynamic laboratory data may be from, for example, in vitro diagnostic laboratory equipment, a healthcare laboratory information system (e.g., a laboratory information system or a healthcare information system), and / or middleware (e.g., a system for managing in vitro diagnostic laboratory equipment and interacting with a healthcare laboratory information system). Static laboratory data may be from, for example, a database of laboratory data, and may be from in vitro diagnostic laboratory equipment, a healthcare laboratory information system, and / or middleware.

[0011] The predicted future medical sample events within the forecast time range may enable further predictions regarding the laboratory, such as future laboratory performance or requirements, such as expected future throughput, reagent consumption, and staff needs for the laboratory.

[0012] Herein, medical sample event data and laboratory data may be collected and processed at the event level (ie, the level of an event occurring with respect to a sample or laboratory).

[0013] In this manner, the computer-implemented method can predict future performance and medical sample events of a healthcare laboratory. Advantageously, the method collects both medical sample event data and data related to the laboratory and / or one or more in vitro diagnostic laboratory devices. The medical sample event data may be received from one or more sources external to the laboratory. For example, the data may be received from a healthcare provider, such as a hospital, general practitioner, or other healthcare provider, that requested the medical sample be processed, as well as a shipping company that transports the medical sample. The medical sample event data may be received from one or more sources internal to the laboratory, such as an internal sample tracking system or a laboratory information system (LIS). The medical sample event data may include data generated within the healthcare laboratory, data generated outside the healthcare laboratory, or a mixture of the two. Data generated outside the healthcare laboratory may include, for example, data from an app that coordinates external activities, data entered manually (e.g., into a terminal), and / or data obtained from a local data bank (e.g., an LIS) but originating outside the laboratory, and / or may be derived values, such as calculated, interpolated, and / or empirically based values.

[0014] In this way, the method collates data on multiple events for each medical sample processed by a healthcare laboratory. This advantageously results in a large amount of available event data, which is converted into a feature set and used as input for a trained machine learning model. As a result, the trained machine learning model can accurately predict future medical sample events. This prediction is beneficial to laboratory management because it allows laboratory staff to process medical samples in a more efficient manner. The trained machine learning model can be a linear Bayesian model, a generalized linear model, a tree-based model (e.g., XGBoost, or light GBM), a recurrent neural network / long short-term memory model, a seasonal autoregressive integrated moving average model with exogenous regressors (SARIMAX), or a transformer model.

[0015] The laboratory data may include data from any of the following data sources: in-vitro diagnostic laboratory equipment, healthcare laboratory information systems, and / or databases of laboratory data. In particular, dynamic laboratory data may include data from in-vitro diagnostic laboratory equipment and / or healthcare laboratory information systems. Furthermore, in some examples, static laboratory data may include data from healthcare laboratory databases. Additionally or alternatively, static laboratory data may include data from in-vitro diagnostic laboratory equipment and / or healthcare laboratory information systems.

[0016] In vitro laboratory equipment is used to process medical samples and transmit data about the medical samples, such as the identification of the samples and their processing progress. By augmenting the medical sample event data with this diagnostic equipment data, the quality of the feature set is improved and thus the trained machine learning model may have an improved ability to predict when future medical samples will undergo processing.

[0017] The healthcare laboratory information system, which may be a Laboratory Information System (LIS) or Hospital Information System (HIS), manages information about each medical sample and groups the medical samples by patient. The use of this data in the augmentation improves the accuracy of the predictions.

[0018] The pre-laboratory sample tracking system may obtain input from the system of the healthcare provider who requested the medical sample to be processed (e.g., a hospital information system (HIS), a general practitioner (GP) information system, or a GP entering data directly into the pre-laboratory sample tracking system). Alternatively, or in addition, the pre-laboratory sample tracking system may obtain input from the system of the carrier transporting the medical sample. The use of this data enhances the predictive capabilities of the trained machine learning model in that the pre-laboratory sample tracking system notifies the model in advance that a medical sample has been requested to be processed, allowing the model to predict when the medical sample will arrive at the laboratory for processing.

[0019] The database of laboratory data may include data regarding the layout of a healthcare laboratory and data specific to each in-vitro diagnostic laboratory device. The data specific to each diagnostic device may be data regarding the type of device and the expected throughput of the device. Augmenting the medical sample event data with data from the database of laboratory data allows the predictions of the trained machine learning model to be specific to the layout and context of a particular laboratory. Thus, predictions of future medical sample events may be made with reference to the specific layout and context of a particular laboratory. For example, staff may be assigned to specific areas of a laboratory, or medical samples may be assigned to diagnostic devices with higher throughput.

[0020] Augmenting medical sample event data according to an embodiment of the first aspect may include identifying one or more specific medical samples referenced in data obtained from two or more data sources and combining data from each of the data sources relating to each specific medical sample.

[0021] Inputs for the trained machine learning model according to embodiments of the first aspect may further include current state data regarding the healthcare laboratory.

[0022] This status data may include, for example, the current utilization level and / or status of each piece of equipment in the healthcare laboratory, the current consumables level in the laboratory, and the current staffing level and / or current staff assignments. In this manner, the method may be able to recommend improvements or future guidance related to each of these variables. For example, the method may be able to estimate the number or quantity of consumables needed to process a predicted volume of medical samples. In another example, the method may assign staff to the processing of medical samples.

[0023] Augmenting medical sample event data according to embodiments of the first aspect may include identifying incomplete data fields of one or more medical samples.

[0024] These incomplete data fields may then be filled with standard or expected values. For example, if the incomplete data field is a single event in a chain of events, the data for the incomplete field may be interpolated. As an example, the data may describe the timestamps of when a sample was ordered, received, and processed. For example, if the timestamp of receipt is missing, it may be interpolated by using the time of the immediately preceding event and adding the average expected time difference to the missing event. If the algorithm for interpolation is appropriately selected, it is also possible to mitigate null values. For example, mean imputation or the Amelia II algorithm may be used. For illustration purposes, assume that the timestamps of multiple consecutive events exist along with the associated timestamp of a single sample. In some cases, a sample may be missing one of the timestamps / events. In this example, there are three events in the chain: Event 1, Event 2, and Event 3. The sample is missing Event 2, so only the timestamps of Events 1 and 3 are present. Average imputation can be performed by calculating the average time difference between Event 1 and Event 2 for a set of samples that have the timestamps of Event 1 and Event 2 (for other examples, also the average time difference between Event 1 and Event 3 or Event 2 and Event 3), and for samples that lack Event 2 but have the timestamp of Event 1, calculating a possible expected time for Event 2 by adding the average time difference between Event 1 and Event 2 to the timestamp of Event 1.

[0025] In a further example, if an event class or category is missing (e.g., only data regarding internal events such as timestamps of sample receipt and sample processing is accessible), a variant of the prediction algorithm may be implemented that does not require the missing data, or the value of the incomplete data field may be set to 0. Advantageously, this allows for the creation of a more complete feature set that results in a more accurate prediction.

[0026] Both the augmented medical sample event data according to embodiments of the first aspect and data relating to healthcare laboratories may be used to generate a feature set.

[0027] Advantageously, augmenting the medical sample event data with data about the laboratory allows the predictions of the trained machine learning model to be specific to the layout and circumstances of the particular laboratory.

[0028] The laboratory data according to an embodiment of the first aspect comprises: in vitro diagnostic laboratory equipment availability data; in vitro diagnostic laboratory equipment throughput data; Staff Data, and / or Examination room layout data It may include any of the following.

[0029] The in-vitro diagnostic laboratory equipment availability data is data regarding the availability of a healthcare laboratory's in-vitro diagnostic laboratory equipment for processing medical samples by time slot. Diagnostic equipment may be unavailable because the equipment is not functioning or because it is being used or scheduled to be used to process other medical samples. This data is advantageous for prediction in that, using this data, a model can calculate the laboratory's medical sample processing capacity taking into account non-functioning and in-use equipment. Furthermore, this data allows the model to recommend the use of other available equipment. The in-vitro diagnostic laboratory equipment throughput data is data regarding the amount of medical samples each in-vitro diagnostic laboratory equipment processes within a given time slot. Diagnostic equipment with different specifications and types may have different processing capacities. The staff data is data regarding the number of staff available within a given time slot. Using this data in generating feature sets not only allows for recommended staff allocation but also enables a more accurate calculation of the laboratory's overall processing capacity. The laboratory layout data is related to the number of different types of diagnostic equipment within the laboratory and their specific physical layout within the laboratory. Thus, predictions of future medical sample events by the model and any associated recommendations may be made with reference to healthcare laboratory details.

[0030] A feature set according to an embodiment of the first aspect may comprise fitted Fourier partial sums for a given time window or for given coefficients of the fit. The Fourier transform allows for easy and low computational cost derivation of recurring patterns / periodicity.

[0031] The predicted time range according to an embodiment of the first aspect may be determined at least in part by user input.

[0032] In this way, a healthcare laboratory personnel may select a predicted time range. The predicted time range may be defined via a selected time range or via an event identification. For example, the selected time range may specify two dates and / or times, between which the predicted time range is defined. In another example, a person may indicate to the system that a diagnostic device will be unavailable at a particular time, and a predicted time range may be selected to precede this event.

[0033] While shorter predictions may be more accurate than longer predictions, shorter predictions may provide less time to respond to predictions of future medical sample events by allocating staff or ordering supplies. Therefore, it would be advantageous for laboratory personnel to be able to select a prediction time range so that more useful predictions can be made taking into account the particular circumstances of the laboratory. Furthermore, this selection may be made via a laboratory application.

[0034] The computer-implemented method according to an embodiment of the first aspect may further comprise the step of retraining the trained machine learning model.

[0035] Retraining may be triggered by the expiration of a predetermined retraining period, such as every 30 days for a trained machine learning model. Retraining may also be triggered manually via user input indicating a change in laboratory settings and / or the presence of anticipated future bias in ordered medical samples. For example, a user may have information that a particular medical center will stop requiring a particular sample type, and retraining may be used to prompt an explanation for this change.

[0036] Furthermore, retraining may be triggered by an indication that the accuracy of predictions made by the trained machine learning model is below a threshold accuracy level. For example, the accuracy level may be measured by recording actual medical sample events within a prediction time range and comparing the actual medical sample events within the prediction time range with predicted future medical sample events for the prediction time range. If this accuracy level falls below a threshold accuracy level of, for example, 95% accuracy, retraining may be triggered. Retraining may be advantageous in that it allows the machine learning model to make predictions based on more recent medical sample event data and may utilize more recent patterns to improve the accuracy of the predictions. Retraining may also allow a larger amount of medical sample event data to be used as input to the model as this data becomes increasingly available.

[0037] The computer-implemented method of the first aspect may be used to recommend changes to the operation of a healthcare laboratory, for example, the allocation of staff to areas of the healthcare laboratory.

[0038] Recommendations for staffing each area of ​​the laboratory may include recommendations on how many staff members are needed at a given time, which areas of the physical laboratory should be staffed with different numbers of staff, and the proportions of staff who should operate different types of in-vitro diagnostic laboratory equipment. For example, a laboratory may have a pre-analytical zone, a clinical chemistry zone, and an immunology zone. In this example, the model recommends staffing these three zones with a given number of staff members given a prediction of future processing required. This may be presented as a predicted volume of incoming samples per area per time unit, or as a distribution of processing requirements per time unit.

[0039] Laboratory personnel may be automatically notified when the predicted workload exceeds a threshold workload, and the personnel may act based on further recommendations. This notification may be provided via a user interface application. Recommendations for staffing areas of the laboratory advantageously automate parts of the laboratory management process, optimizing the otherwise error-prone allocation process.

[0040] The computer-implemented method of the first aspect may be used to recommend the amount of supplies that should be made available.

[0041] The amount of supplies to be made available may refer to, for example, the amount of reagents or test kits required to process a predicted volume of medical samples. This may be calculated primarily based on the throughput of available in vitro diagnostic laboratory equipment and the known volume of reagent kits, which is stored in a laboratory database. This calculation may be performed for each area of ​​the laboratory so that each area can be appropriately replenished. The data may be presented as a predicted amount of supply usage per area per time unit, or as a distribution of supply usage per time unit. Recommending the amount of supplies to be made available is advantageous in that it allows supplies to be ordered in advance of sample processing, thereby making sample processing more efficient. This supply ordering may be performed by laboratory management or by a computer system. For example, the method may include automatically ordering additional supplies when it is determined that the estimated number or amount of supplies required to process a predicted volume of medical samples exceeds the number or amount of available consumables. This automated ordering is advantageous in that it may be more reliable than human ordering and reduces the administrative burden on laboratory staff.

[0042] In a second aspect, embodiments of the present invention provide a system for predicting future performance of a healthcare laboratory, the healthcare laboratory including one or more in vitro diagnostic laboratory instruments configured to process medical samples, the system comprising one or more processors and a memory, the memory including machine-executable instructions that, when executed on the one or more processors, receiving medical sample event data describing events relating to one or more medical samples processed by the laboratory; Augmenting medical sample event data with laboratory data; and generating a feature set from the augmented medical sample event data, the feature set including time period data derived from the augmented medical sample event data; using the feature set as input for a trained machine learning model trained on the historical feature set to predict future medical sample events within a prediction time range; receiving, as output from the trained machine learning model, a prediction of future medical sample events for a prediction time range; on one or more processors.

[0043] In a third aspect, embodiments of the present invention provide a computer-implemented method for training a machine learning model to predict future performance of a healthcare laboratory, the healthcare laboratory including one or more in vitro diagnostic laboratory devices configured to process medical samples, the method comprising: receiving historical medical sample event data describing events relating to one or more medical samples; Augmenting historical medical sample event data with laboratory data; and generating a historical feature set from the augmented historical medical sample event data, the historical feature set including time period data derived from the augmented historical medical sample event data; Using the historical feature set as input to train a machine learning model to predict the volume of future medical sample events within a prediction time range; Includes:

[0044] The medical sample event data according to any of the above aspects includes: Ordered medical samples, and / or Medical samples arriving at a healthcare laboratory One or more of the following may be described:

[0045] The present invention may include any one or any combination of the optional features described in relation to the first aspect, provided that they are not inconsistent.

[0046] The present invention includes any combination of the described embodiments and preferred features except where such a combination is expressly not permitted or explicitly avoided.

[0047] Further aspects of the present invention provide a computer program comprising code which, when executed on a computer, causes the computer to perform the method of the first and / or third aspects, a computer readable medium storing a computer program comprising code which, when executed on a computer, causes the computer to perform the method of the first and / or third aspects, and a computer system programmed to perform the method of the first and / or third aspects. [Brief explanation of the drawings]

[0048] [Figure 1] FIG. 1 is a flow diagram of a computer-implemented method according to one aspect of the present disclosure. [Figure 2] FIG. 1 is a flow diagram of a training and retraining process according to one aspect of the present disclosure. [Figure 3] 1 is an example of medical sample event data according to one aspect of the present disclosure. [Figure 4] 1 is an example of a feature set according to one aspect of the present disclosure. [Figure 5] 1 is an example of a prediction of future medical sample events over a prediction time range according to an aspect of the present disclosure. [Figure 6] 1 is an example of a processed prediction of future medical sample events over a prediction time range according to an aspect of the present disclosure. [Figure 7] 1 is an example of a processed prediction of future medical sample events to recommend staffing for areas of a healthcare laboratory according to an aspect of the present disclosure. [Figure 8] 1 is an example of a processed normalized prediction of future medical sample events to recommend staffing for areas of a healthcare laboratory according to an aspect of the present disclosure. [Figure 9] 1 is an example of a processed normalized prediction of future medical sample events for recommending the amount of supplies to be made available according to an aspect of the present disclosure. [Figure 10] 1 is an example of medical sample event data used to generate the amount of supplies to be made available according to one aspect of the present disclosure. [Figure 11] 1 is an example of a prediction of future medical sample events over a prediction time range according to an aspect of the present disclosure. [Figure 12] 1 is an example of a processed prediction of future medical sample events to recommend the amount of supplies to be made available according to an aspect of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0049] Aspects and embodiments of the present invention will now be described with reference to the accompanying figures. Further aspects and embodiments will be apparent to those skilled in the art.

[0050] FIG. 1 illustrates a flow diagram of a computer-implemented method for predicting future operational status of a healthcare laboratory 100 via a cloud system 102, which may include one or more processors and memory. The described functionality in the cloud system 102 may be realized, for example, by software running on dedicated or pooled hardware components. The laboratory includes one or more in-vitro diagnostic laboratory instruments configured to process medical samples according to one aspect of the present disclosure. In a first step, a healthcare provider 104 outside the laboratory requests processing of one or more medical samples. The healthcare provider may be, for example, a health center, a hospital, a general practitioner, or any other healthcare provider. The processing request is sent to a courier 106, which transports the one or more medical samples to the laboratory 110 for processing. Furthermore, all processing requests are monitored by a request monitoring application 108. The request monitoring application 108 may fully or partially fulfill the role of the pre-laboratory sample tracking system described above in that, upon detecting a processing request, it may collect medical sample event data related to the request, such as a timestamp, sample ID, a reference to the medical test requested for the sample, and the sender of the request, as well as other meta-information. The data is encrypted and transmitted to the cloud storage system 112 of the cloud system 102.

[0051] The laboratory 100 includes an information system 114, such as a laboratory information system (LIS), a hospital information system (HIS), or middleware that manages data related to patient-specific medical sample orders. The information system receives data related to the arrival and processing of medical samples. This data, as well as data directly from diagnostic devices processing the medical samples, is sent to an edge device 116. The edge device is a server for securely transmitting data to a cloud storage system 112, which may encrypt the data prior to transmission to the cloud storage system and may be separated from the cloud storage system by a firewall. For communication, both standard interfaces (e.g., HL7 and FHIR) and proprietary interfaces (in the case of interface incompatibility) are used. The information system continuously receives medical sample event data and sends it to the cloud storage system. Additionally, each diagnostic device directly transmits data related to its availability to the cloud storage system.

[0052] The medical sample event data is then augmented 118 with laboratory data. In this example, the laboratory data is data about the laboratory's in-vitro diagnostic laboratory equipment, information, and data from the request monitoring application 108. These data sources are merged into a single dataset, which therefore contains more detailed data about the medical sample than could be obtained from a single source. Additional data is drawn from the laboratory database 119 of the cloud system 102, which includes data about the laboratory's specific layout, the laboratory's specific operating status, and the laboratory's specific in-vitro diagnostic laboratory equipment. As previously mentioned, the laboratory data can be static and / or dynamic laboratory data. For example, the laboratory data can come from any of the following data sources: one or more in-vitro diagnostic laboratory equipment (e.g., connected to the LIS 114), a healthcare laboratory information system (e.g., the LIS 114), and / or a database of laboratory data (119).

[0053] This augmented data set is then used to generate a feature set 120. Each feature in the feature set is an individually measurable characteristic of sample events within a predicted time range. The feature set is used as input to a machine learning model that is used to predict 122 future medical sample events. This prediction is accessible to laboratory personnel via a user application 124. Additionally, this user application may be used to set the predicted time range for predictions by the machine learning model.

[0054] 2 illustrates a flow diagram of a training and retraining method according to one embodiment of the present disclosure. A user inputs a predicted time range 206 into an application 204 within a healthcare laboratory. This input may be entered via direct entry into a command line interface, a graphical user interface, or via configuration of a widget in the application. The input is sent to the cloud 200, where the predicted time range input is then used to create a feature set 208. The feature set is used as input to a trained machine learning model 210 to generate an output of predictions of future medical sample events according to the predicted time range. This output undergoes post-processing 212, such as transforming the data for use in two applications: staffing and supply management.

[0055] The trained machine learning module 210 is retrained according to the training method 202. A user defines a historical training time range for training the machine learning model. The length of this training time range should be selected to be proportional to the intended prediction time range; for example, when predicting the next month, a training time range spanning one year should be used to train the model. This training time range is used to create a feature set 214, which is then optimized for training 216. This optimization includes weighting the feature set to emphasize recent medical sample events. Model parameters are optimized based on the training set and selected based on the validation set. In some examples, the parameters are optimized with a minimizer that attempts to minimize a given scoring metric. The minimizer and metric are selected depending on the underlying model. An example of a suitable minimizer is the limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) algorithm. The validation set comprises a portion of the original dataset, and the original dataset is divided into, for example, a training set, a validation set, and a test set. In some examples, the division is 70% training, 15% validation, and 15% test. Then, the model is trained on the training set, and performance is evaluated on the validation set. Next, the selected model is tested on the test set, and the performance using this test set is reported as the metric of the optimized and selected model.

[0056] The resulting trained machine learning model 218 is integrated into the forecasting methodology, replacing the trained machine learning model 210. Retraining may be triggered by the expiration of a predetermined period of time, or by an indication that the accuracy of predictions by the trained machine learning model has fallen below a threshold accuracy level.

[0057] 3 is an example of medical sample event data according to one embodiment of the present disclosure. This is an example of raw, unenhanced medical sample event data. Each data point includes a timestamp indicating when the event occurred, an event type indicating the event that was monitored, a unique sample ID code to identify each sample, a test field indicating the type of processing that needs to be done to process the sample, a sender field indicating the healthcare provider that requested the sample, and other meta-information related to the sample that may be specific to the sender and event type, for example.

[0058] Figure 4 is an example of a feature set according to one embodiment of the present disclosure. The data from Figure 3 is organized into time slots. In this example, all sample events from 10:00 to 11:00 on June 6, 2023 are collected and represented in this time slot. "Weekdays," or days of the week, are calculated from the timestamps, with "2" corresponding to Tuesday, for example. To account for seasonality in the data, first and second fitted Fourier partial sums are calculated. The number of samples per time slot and the number of each type of test (in this example, Na, GluC, and TSH) are provided.

[0059] Additionally, each time period may represent season, time of day, month, day, average number of samples ordered, average number of tests per sample, average tests per sample, number of specific tests (e.g., Na tests or GLUC3 tests), sender-specific information such as the number of different senders and metrics for a given set of senders (e.g., number of samples or tests per set), order-location-specific data such as the number of different senders and metrics for a given set of order locations, receiving location (if the model represents multiple laboratories), laboratory processing areas (e.g., clinical chemistry, immunology, hematology), number of samples per processing area, and / or total number of staff.

[0060] 5 is an example of a prediction of future medical sample events over a forecast time horizon according to one embodiment of the present disclosure. This is a typical output from the trained machine learning model 210. For each hourly time slot, the data includes the number of medical samples predicted to arrive, the number of medical samples predicted to be ordered, and the number of each process that must be performed in the in-vitro diagnostic laboratory equipment to process the samples.

[0061] FIG. 6 is an example of processed predictions of future medical sample events over a forecast time horizon according to one embodiment of the present disclosure. The data is post-processed 212 to provide recommendations regarding healthcare laboratory staff assignments. This post-processing includes augmenting the prediction data (e.g., that of FIG. 5 ) with laboratory-specific data from the laboratory database 119, such as data specific to the laboratory layout and each in-vitro diagnostic laboratory device. Augmenting the prediction data with this layout data allows predictions of future medical sample events to be made with reference to the specific layout and circumstances of a particular laboratory. In the data of FIG. 6 , each processed test is assigned to either the clinical chemistry or immunology area of ​​the laboratory. This is advantageous in that the recommendations of the trained machine learning model may allocate staff to these areas more optimally than laboratory staff would, thereby improving laboratory efficiency.

[0062] Figure 7 is an example of processed predictions of future medical sample events to recommend staff allocation to each area of ​​a healthcare laboratory according to one embodiment of the present disclosure. The data in this figure has undergone further post-processing 212, such as aggregating the data from Figure 6. Figure 7 provides the total number of medical samples processed and tests processed for each area of ​​the laboratory (clinical chemistry and immunology in this example). This further post-processing may make the suggested data more useful to laboratory staff when assigning staff to laboratories.

[0063] Figure 8 is an example of a processed, normalized prediction of future medical sample events for recommending staff allocation to areas of a healthcare laboratory according to one embodiment of the present disclosure. Thus, the data in this figure represents the data of Figure 7 after a further post-processing step 212 of normalization. Each workload is expressed as a percentage of the total workload, and a corresponding percentage of staff can be assigned to individual areas of the laboratory (clinical chemistry and immunology in this example). This further post-processing can make the suggestion data more useful to laboratory staff when assigning staff to laboratories.

[0064] 9 is an example of a processed normalized forecast of future medical sample events to recommend the amount of supplies to be made available according to one embodiment of the present disclosure. The data in FIG. 9 represents the data in FIG. 8 with a further processing step 212 of multiplying the normalized forecast distribution by the number of available staff for each time slot. For example, if 10 laboratory staff are available and the clinical chemistry area has a normalized workload of 60% for a particular time slot, then 6 laboratory staff will be assigned to the clinical chemistry area for this particular time slot. This further post-processing may make the suggested data more useful to laboratory staff when assigning staff to laboratories.

[0065] 10 is an example of medical sample event data used to generate the amount of supplies to be made available according to one embodiment of the present disclosure. This data represents a dataset that can be used to predict the number of supplies used by a laboratory. In particular, this dataset includes event data related to the processing or analysis of each medical sample on an individual in-vitro diagnostic laboratory device. Providing the specific in-vitro diagnostic laboratory device used to process each test is advantageous because it allows for more effective tracking of supplies used, thus improving the ability of the trained machine model to predict supply usage.

[0066] 11 is an example of a forecast of future medical sample events over a forecast time horizon according to one aspect of the present disclosure. This data represents the post-processing 212 step of aggregating data over an extended period of time for use in ordering supplies. The medical sample forecast data is aggregated into multiple time periods, each one week in length. Thus, the model can predict the total number of each type of processing event per week.

[0067] FIG. 12 is an example of a processed prediction of future medical sample events to recommend the amount of supplies to be made available, according to one embodiment of the present disclosure. This data represents a post-processing 212 step that enhances the prediction with data from a laboratory database regarding known volumes of supply kits. This data can be used by laboratory staff to suggest ordering specific numbers of reagent bottles and / or kits of each type (in this example, ISE NA, GLUC3, and TSH). This further post-processing can make the suggestion data more useful to laboratory staff when ordering supplies for the laboratory. Additionally, the user application 124 can notify laboratory staff when a specific type of supply reaches a certain threshold number or when there are insufficient numbers for the number of predicted processing events in a future time period. Thus, laboratories can be made more efficient by ensuring they have sufficient supplies.

[0068] The systems and methods of the above embodiments, in addition to the structural components and user interactions described, may be implemented in a computer system (particularly computer hardware or computer software).

[0069] The term "computer system" includes hardware, software, and data storage for implementing a system or executing a method according to the above-described embodiments. For example, a computer system may include a central processing unit (CPU), input means, output means, and data storage. A computer system may have a monitor that provides a visual output display. The data storage may include RAM, a disk drive, or other computer-readable medium. A computer system may include multiple computing devices connected by a network and capable of communicating with each other via the network.

[0070] The methods of the above embodiments may be provided as a computer program, or as a computer program product or computer readable medium carrying a computer program configured to perform the above-described method when executed on a computer.

[0071] The term "computer-readable medium" includes, but is not limited to, any non-transitory medium that can be read and accessed directly by a computer or computer system, including, but not limited to, magnetic storage media such as floppy disks, hard disk storage media, and magnetic tape, optical storage media such as optical disks or CD-ROMs, electrical storage media such as RAM, ROM, and memory, including flash memory, and hybrids and combinations of the above, such as magnetic / optical storage media.

[0072] While the present disclosure has been described in conjunction with the exemplary embodiments set forth above, many equivalent modifications and variations will be apparent to those skilled in the art in light of this disclosure. Accordingly, the exemplary embodiments of the present disclosure set forth above are considered to be illustrative and not limiting. Various modifications may be made to the described embodiments without departing from the spirit and scope of the present disclosure.

[0073] In particular, although the methods of the above embodiments are described as being implemented on the systems of the described embodiments, the methods and systems of the present disclosure need not be implemented in conjunction with each other and may each be implemented on alternative systems or using alternative methods.

[0074] The features disclosed in this specification, the following claims, or the accompanying drawings, expressed in a specific form or expressed in terms of means for performing a disclosed function or a method or process for obtaining a disclosed result, may be used individually or in any combination of such features, as appropriate, to realize the disclosure in various of its forms.

[0075] While the present disclosure has been described in conjunction with the exemplary embodiments set forth above, many equivalent modifications and variations will be apparent to those skilled in the art in light of this disclosure. Accordingly, the exemplary embodiments of the present disclosure set forth above are considered to be illustrative and not limiting. Various modifications may be made to the described embodiments without departing from the spirit and scope of the present disclosure.

[0076] To avoid any misunderstanding, any theoretical explanations provided herein are provided for the purpose of improving the reader's understanding, and the inventors of the present invention do not wish to be bound by any of these theoretical explanations.

[0077] Any section headings used herein are for organizational purposes only and are not to be construed as limiting the subject matter described.

[0078] Throughout this specification, including the claims which follow, unless the context requires otherwise, the terms "comprise" and "include," and variations such as "comprises," "comprising," and "including," will be understood to mean the inclusion of the thing or step or group of things or steps stated therein but not to the exclusion of any other thing or step or group of things or steps.

[0079] It should be noted that, as used in this specification and the appended claims, the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. Ranges may be expressed herein as from "about" one particular value and / or to "about" another particular value. When such a range is expressed, another embodiment includes from the one particular value and / or to the other particular value. Similarly, when values ​​are expressed as approximations, by use of the antecedent "about," it will be understood that the particular value forms another embodiment. The term "about" with respect to numerical values ​​is arbitrary and means, for example, + / - 10%.

Claims

1. 1. A computer-implemented method for predicting future performance of a healthcare laboratory, the healthcare laboratory including one or more in vitro diagnostic laboratory devices configured to process medical samples, the computer-implemented method comprising: receiving medical sample event data describing events relating to one or more medical samples processed by the laboratory; augmenting the medical sample event data with laboratory data; generating a feature set from the augmented medical sample event data, the feature set including time period data derived from the augmented medical sample event data; using the feature set as an input for a trained machine learning model trained on the historical feature set to predict future medical sample events within a prediction time range; receiving, as an output from the trained machine learning model, a prediction of future medical sample events for the prediction time range; 11. A computer-implemented method comprising:

2. The computer-implemented method of claim 1 , wherein the medical sample event data is augmented with dynamic laboratory data and static laboratory data.

3. The medical sample event data includes data from a pre-laboratory sample tracking system, and the laboratory data used to augment the medical sample event data comes from the following data sources: in vitro diagnostic laboratory equipment, said healthcare laboratory information system; and / or Laboratory data database 3. The computer-implemented method of claim 1 or claim 2, wherein the data includes data from any of:

4. 4. The computer-implemented method of claim 1, wherein augmenting the medical sample event data includes identifying one or more specific medical samples referenced in data obtained from two or more data sources and combining data from each of the data sources relating to each specific medical sample.

5. 5. The computer-implemented method of claim 1, wherein the input for the trained machine learning model further comprises current state data for the healthcare laboratory.

6. The computer-implemented method of any one of claims 1 to 5, wherein augmenting the medical sample event data comprises identifying incomplete data fields for one or more medical samples.

7. The computer-implemented method of any one of claims 1 to 6, wherein both augmented medical sample event data and data related to the healthcare laboratory are used to generate the feature set.

8. The laboratory data includes: in vitro diagnostic laboratory equipment availability data; in vitro diagnostic laboratory equipment throughput data; Staff data, and / or Examination room layout data 7. The computer-implemented method of claim 6, comprising:

9. The computer-implemented method of any one of claims 1 to 8, wherein the feature set comprises fitted Fourier partial sums for a given time period or for a given coefficient of a fit.

10. The computer-implemented method of any one of claims 1 to 9, wherein the predicted time range is determined at least in part by user input.

11. Use of the computer-implemented method of any one of claims 1 to 10 to recommend staff allocation across areas of the healthcare laboratory.

12. Use of the computer-implemented method of any one of claims 1 to 10 for recommending the amount of supplies that should be made available.

13. 1. A system for predicting future performance of a healthcare laboratory, the healthcare laboratory including one or more in vitro diagnostic laboratory instruments configured to process medical samples, the system comprising one or more processors and memory, the memory including machine-executable instructions that, when executed on the one or more processors, receiving medical sample event data describing events relating to one or more medical samples processed by the laboratory; augmenting the medical sample event data with laboratory data; generating a feature set from the augmented medical sample event data, the feature set including time period data derived from the augmented medical sample event data; using the feature set as an input for a trained machine learning model trained on the historical feature set to predict future medical sample events within a prediction time range; receiving, as an output from the trained machine learning model, a prediction of future medical sample events for the prediction time range; on the one or more processors.

14. 1. A computer-implemented method for training a machine learning model to predict future performance of a healthcare laboratory, the healthcare laboratory including one or more in vitro diagnostic laboratory devices configured to process medical samples, the computer-implemented method comprising: receiving historical medical sample event data describing events relating to one or more medical samples; augmenting the historical medical sample event data with laboratory data; generating a historical feature set from the augmented historical medical sample event data, the historical feature set including time period data derived from the augmented historical medical sample event data; using the set of historical features as inputs to train a machine learning model to predict the volume of future medical sample events within a prediction time range; 11. A computer-implemented method comprising:

15. The medical sample event data comprises: Ordered medical samples, and / or a medical sample arriving at said healthcare laboratory A computer-implemented method or system according to any preceding claim, which describes one or more of:

Images