A clinical decision support system for estimating drug-related treatment optimization

Abstract

The present invention provides a clinical decision support system for estimating drug-related treatment optimization. The clinical decision support system includes: a computing unit designed to host multiple prediction models, the computing unit including an input interface designed to receive input data and an output interface designed to output results; multiple different trained prediction models, wherein each model is trained to predict the probability of treatment results for multiple different drug-related treatment options and for a specific patient group; a selection unit designed to automatically select one of the prediction models based on the input data according to a predefined selection scheme; wherein the clinical decision support system is designed to generate output results by processing the input data using the selected prediction model. The present invention also describes a prediction method, a method for manufacturing such a CDS system, and a data processing system.

Description

Technical Field

[0001] The present invention describes a clinical decision support ("CDS") system for estimating drug-related treatment optimization for inflammatory diseases as well as a predictive method for computational decision support and a method for making such a CDS system. Background Art

[0002] Inflammatory diseases - such as rheumatic diseases, such as rheumatoid arthritis, psoriatic arthritis, other musculoskeletal diseases, chronic obstructive pulmonary disease (COPD), asthma, multiple sclerosis or Crohn's disease - include a wide range of medical conditions that cause chronic pain and inflammation. For example, rheumatic diseases affect joints, tendons, ligaments, muscles and bones. Most of these conditions occur when the immune system begins to attack its own tissues for reasons that are not yet clear. Typically, inflammatory diseases are characterized by periods of disease inactivity (also known as "remission"), alternating periods of low and moderate disease activity, and periods of increased (high) disease activity (known as "flare-ups").

[0003] Although most diseases in such diseases are incurable, there are different types of drugs that can help keep disease activity at low levels. Suitable drugs and dosages are, for example, specified in specific disease guidelines (such as ACR and EULAR guidelines for rheumatoid arthritis). However, suitable drugs and dosages are derived based on clinical studies and statistical analysis at the population level.

[0004] Due to the large differences in patient characteristics such as demographics, diet and lifestyle, genetic predisposition, sensitivity to external factors like weather conditions and possibly other factors as well, it remains challenging for rheumatologists to find the right drug and / or the right dose for an individual patient.

[0005] Often, effective treatments need to be modified to accommodate the patient's current condition (eg, to pause immunosuppressive therapy due to planned surgery or acute infection), to reduce the risk of adverse events of a drug, and / or to reduce healthcare costs.

[0006] Furthermore, according to the "targeted therapy" strategy described in the guidelines for the treatment of rheumatoid arthritis, the dose of drugs, especially biologic drugs, should be gradually reduced once a stable remission is achieved. Rheumatologists are then again faced with the challenge of which patients are eligible for tapering and by how much the drug can be reduced in each individual case. This is often a process of trial and error, which is accompanied by a decrease in the patient's quality of life and an increase in healthcare costs, until the right treatment is found.

[0007] In the outpatient clinic routine of rheumatism patients, rheumatologists examine these rheumatism patients at fixed time intervals or irregular time intervals (for example, in the case of complications such as sudden illness). During a typical patient visit, inspection data is collected, and demographic data and lifestyle data previously collected are sometimes confirmed or updated. Based on this data, rheumatologists make ideal treatment decisions with patients. During a patient visit, blood samples are usually collected and sent to a laboratory for analysis, and what is most often concerned about is inflammatory biomarkers such as C-reactive protein (CRP). Later-usually a few days after the patient's visit and after having made a treatment decision-the result of this blood test can be obtained.

[0008] Based on the newly obtained laboratory data, the rheumatologist sometimes gained new insights and adjusted the patient's treatment over the phone. In all cases, treatment decisions were based on multiple relevant variables related to the patient's demographic characteristics (e.g., age, sex), examination data (e.g., patient questionnaire, number of tender and swollen joints), blood values ​​(e.g., ESR, CRP), and medications (e.g., substance, combination therapy, dosage, etc.).

[0009] Finally, clinical guidelines emphasize shared decision making between clinicians and patients regarding treatment, which is not an easy task given the large number of relevant variables involved.

[0010] The specific questions are:

[0011] 1. Physicians (eg, rheumatologists) often have difficulty finding an initial medication and / or dosage that may be effective for an individual patient.

[0012] 2. Doctors often need to taper a drug, but do not know whether the taper is safe and still effective for an individual patient and do not know how much to taper a drug to safely and still be effective for an individual patient.

[0013] 3. Data relevant to treatment decisions become available at different points in time.

[0014] 4. Due to the complex nature of making high-dimensional decisions in the field of inflammatory diseases (e.g., rheumatology), data-driven black-box decision support systems are often considered to lack transparency, which negatively affects the acceptance of both clinicians and patients. Summary of the invention

[0015] Therefore, it is an object of the present invention to improve known methods and to provide a clinical decision support system for estimating drug-related treatment optimization regarding inflammatory diseases, in particular for data-driven clinical decision support of therapy planning in rheumatic diseases.

[0016] This object is achieved by the clinical decision support system according to the technical solution of the present invention, the prediction method according to the technical solution of the present invention, the method for manufacturing a CDS system according to the technical solution of the present invention, and the data processing system according to the technical solution of the present invention.

[0017] In the following, the invention may be described using an example of predicting the probability of sudden onset of rheumatoid arthritis, but the invention is not limited to this application. The invention and its various aspects may be used in particular to predict the future state of patients with known inflammatory diseases like, for example, psoriatic arthritis, other musculoskeletal diseases, chronic obstructive pulmonary disease (COPD), asthma, multiple sclerosis or Crohn's disease.

[0018] A clinical decision support system for estimating drug-related treatment optimization for inflammatory diseases according to the present invention comprises the following components:

[0019] a computing unit designed to host a plurality of prediction models, the computing unit comprising an input interface designed to receive input data and an output interface designed to output results,

[0020] - a plurality of different trained prediction models, wherein each model is trained to predict the probability of a treatment outcome for a plurality of different drug-related treatment options and for a specific patient group based on input data,

[0021] a selection unit designed for automatically selecting one of the prediction models based on the input data according to a predefined selection scheme,

[0022] The CDS system is designed to generate output results by processing input data using the selected prediction model.

[0023] Typically, clinical decision support (CDS) systems are known in the art. However, the CDS system provides an estimate of the probability of drug-related treatment optimization risk in a future time period. The expression "drug-related" in this context may mean related to drug reactions and / or side effects regarding drug dosage and / or drug type. Alternatively, "drug-related" may mean a patient's (negative or positive) reaction to an applied (certain dose of) drug. For example, for a specific amount or type of drug, the drug reaction probability (whether the drug is helpful) can be estimated and / or the risk of side effects for a specific patient can be predicted.

[0024] A suitable computing unit should have enough memory and computing power to host multiple prediction models. This means that the computing unit must be able to process these prediction models in order to obtain results from the input data through the prediction models. However, unused prediction models do not necessarily need to be hosted by the computing unit. These unused prediction models may, for example, not appear in the memory until they are needed. Such computing units with input interfaces and output interfaces are known in the art. The output interface can be a data interface or a display. Any means for outputting the results are possible as long as they can provide the data format required by the user.

[0025] Regarding the prediction models, these prediction models are models trained for different purposes. Preferably, these prediction models have been trained using training data of different patient groups (each of these models has a training data set for a different patient group) and / or using training data related to different drugs (although a set of different drugs can also be used to train the model). Models can also be trained using different types of data (e.g., demographic data, drug data, examination data, or laboratory data), for example, using laboratory data to train one model and using demographic data to train another model. In general, the training of prediction models and the architecture of these prediction models are well known in the art (see, for example, EP 3 573 068A1).

[0026] What all prediction models have in common is that each model is trained to predict the probability of a treatment outcome based on input data. These treatment outcomes may involve a positive response (pain relief, improvement of condition), a negative response (side effects) or a neutral response (no response at all) to the drug by the patient. Therefore, "treatment outcome" can be understood as a side effect and / or disease outcome. This is done for a variety of (one or more) different drug-related treatment options, such as a specific drug group (or multiple such groups) or different doses of different drugs. In addition, this is done for a specific patient group. Preferably, a total of more than 10 different trained models are used, in particular more than 30 or even more than 50.

[0027] The model may reside in the computing unit itself or in a memory used by the computing unit. For example, information about the architecture and parameters of the prediction model resides in the memory, and the selected prediction model is downloaded from the memory to the (random access memory of the) computing unit.

[0028] The selection unit is designed to automatically select one of the prediction models. The selection is based on a predefined selection scheme and data input into the CDS system for processing by the model. The selection scheme can be a table stored in the memory of the computing system or a decision tree hardwired in the algorithm of the selection unit. According to the input data (e.g., diagnosis; laboratory data; data including information about certain drugs applied to the patient, the age or gender of the patient), the model in the multiple models is selected by the selection unit based on the selection scheme. Such a selection is beneficial because it is very complicated (or even impossible) to train a single prediction model for all possible cases and for all possible patients. In addition, sometimes it turns out that for special cases (special patient groups, special diseases or special use cases), prediction models of different architectures are better than models with another architecture. Therefore, the present invention can estimate which prediction model (architecture and training) will be the best for which situation when building the CDS system, and the prediction model is selected as part of the CDS system. Then, if a special situation occurs, the model is selected by the selection unit, and the model provides the best results for the special situation.

[0029] Preferably, the selection unit is designed to: search for a predefined data type and / or a value of a predefined data type in the input data; determine whether a predefined data type exists in the input data; and / or compare a value with a predefined threshold; and / or decide whether the value meets a predefined requirement. For example, the selection unit can be designed to: check whether the patient's gender is male, female or undefined, or check whether there is available laboratory data in the input data, or check whether the diagnosis is, for example, rheumatoid arthritis.

[0030] The CDS system is designed to generate output results by processing input data using a selected prediction model. It is known in the art how to process data using a trained model.

[0031] Thus, depending on the model used and the selected scenario, the CDS system is able to predict the patient's response to a certain drug (regarding side effects and / or relief), or to predict exacerbations during therapy de-escalation (particularly "flare" predictions for PsA and RA and "exacerbation" predictions for COPD and asthma). Regarding COPD, there is no de-escalation of, for example, biologics, as they are not approved to date for the treatment of these diseases, but there is de-escalation of other applicable drugs, especially corticosteroids.

[0032] For example, if a patient with rheumatoid arthritis is entering Phase 1 of treatment, the CDS system can estimate whether the patient will respond positively to drugs recommended by the corresponding guidelines, approved for use, and available at the treatment facility. If the model predicts that this particular patient will not respond to any Phase 1 drugs, Phase 2 can be started immediately, thus avoiding months of suffering due to drug ineffectiveness or side effects.

[0033] The prediction method for computing decision support according to the present invention comprises the following steps:

[0034] - providing a CDS system according to the present invention,

[0035] - providing input data to the CDS system, wherein the input data is preferably automatically selected and provided, in particular in case new data becomes available for predefined patients,

[0036] - using a CDS system to determine the result, wherein a prediction model is automatically selected by the CDS system based on the input data, and the result is automatically determined by the selected prediction model,

[0037] - Outputting results, in particular, where if a patient's results have changed substantially compared to earlier results for the same patient, the user is notified, e.g. in the form of a warning message or an icon in the patient list, so that the user knows that he has to open the case.

[0038] The input data is preferably data related to a single patient, and the input data preferably includes data from the group of demographic data (possibly including lifestyle data), medication data, examination data, and laboratory data. In addition, the input data includes information about expected changes in treatment, such as information about a reduction or increase in a drug or drug dosage to be applied to the patient. Furthermore, the input data may potentially include omics data (e.g., proteomics, genomics, metabolomics) and medical image data acquired using different imaging modalities (e.g., magnetic resonance imaging, ultrasound, computed tomography).

[0039] The method according to the invention for manufacturing a CDS system according to the invention comprises the following steps:

[0040] - providing at least a first model group and a second model group, each model group comprising a plurality of untrained machine learning models, in particular comprising a plurality of models with different internal architectures and / or different hyperparameters,

[0041] - providing at least a first training data set and a second training data set, each training data set comprising data with different distinguishing characteristics,

[0042] - training the first model group using the first training data set, and training the second model group using the second training data set,

[0043] - ranking each trained prediction model in the set of models using predefined quality criteria, preferably wherein the prediction models are developed and compared offline,

[0044] - Manually or automatically selecting the best ranked prediction model in each model group as the prediction model for the clinical decision support system.

[0045] It should be noted that the model groups mentioned here are not models of the CDS system. Only the "winner" (or winners) in the group will have a place in the final CDS system. In practice, there should be more than 1 group, such as more than 10, more than 30 or more than 50 groups.

[0046] The training data set should be selected for the purpose of a single trained model. If the model is applied to patients aged 60 or above, the training data set should only include data for patients aged 60 or above. If the model is applied to a certain drug, the corresponding training data set should include data on patient responses based on the drug.

[0047] The training criteria may depend on the status of the patient from which the training data was obtained. In the case where validation occurs and the predictive quality of the trained model is quantified, the criteria may assign a quality score to the predictive model. The training criteria may also be derived from the criteria of a nested cross-validation process.

[0048] For example, there are multiple prediction models (e.g., 52) trained for different diseases, drugs, actions (application of new drugs or drug reduction) and patient groups. These models can then be applied to: CDS systems for, for example, RA (rheumatoid arthritis); Phase II (EULAR guidelines); prediction of response to a certain drug or PsA (psoriatic arthritis); Phase IV, gradual reduction (also known as "dose reduction" or "therapy de-escalation"). For a given purpose, the best model is automatically selected by the selection unit.

[0049] With respect to patients with a certain disease, the selection can be accomplished by determining which of the prediction models is trained with the group of patients that is most similar to the actual patients, and accordingly, determining which of the prediction models is trained with the patients with the actual disease. With respect to information about (planned or applied) medications in the input data, it can be determined which of the prediction models is trained with the group of patients who received these medications. With respect to certain types of input data (e.g., laboratory data), it can be determined which of the prediction models is trained with such data.

[0050] The data processing system of the present invention, which is in particular a computer network system, comprises a data network, a plurality of client computers and a service computer system, wherein the service computer system comprises the clinical decision support system of the present invention.

[0051] The above-mentioned units or modules of the present invention can be implemented completely or partially as software modules running on a processor of a computing system. Mainly implemented in the form of software modules can have the following advantages: applications already installed on existing systems can be updated with relatively little effort to install and run the method of the present application. The purpose of the present invention is also achieved by a computer program product having a computer program, which can be directly loaded into the memory of a computing system and includes a program unit to perform the steps of the method of the present invention when the program is executed by the computing system. In addition to the computer program, such a computer program product may also include other parts, such as documents and / or additional components and hardware components that facilitate access to the software, such as hardware keys (dongles, etc.).

[0052] Computer readable media such as memory sticks, hard disks or other removable or permanently installed carriers can be used to transport and / or store executable portions of a computer program product so that these executable portions can be read from a processor unit of a computing system. The processor unit may include one or more microprocessors or their equivalents.

[0053] In particular, as disclosed in the following description, advantageous embodiments and features of the present invention are given by the technical solution of the present invention. Features of different claim categories can be appropriately combined to give other embodiments not described herein.

[0054] With respect to a trained prediction model or its training, there are some parameters (or "data") that refer to different types of data. Preferably, there are demographic data, medication data, examination data, or laboratory data (also including related scores and derived variables).

[0055] Demographic data may be data concerning the patient, in particular sex (male, female), height, weight, body mass index, age, smoking status (never smoked, smoker, former smoker), alcohol intake (yes, no, amount), list of comorbidities, time since diagnosis.

[0056] Preferred drug data refer to data for treatment with active agents, preferably as listed below: in particular biologics / biosimilars, methotrexate, other conventional disease-modifying antirheumatic drugs (cDMARDs), targeted synthetic disease-modifying antirheumatic drugs (tsDMARTs), nonsteroidal anti-inflammatory drugs (NSAIDs), glucocorticoids. For any of the active agents, data may refer to any member of the group treatment (yes / no), actual substance, mode of administration, prescribed dose, prescribed interval, start time and stop time.

[0057] Preferred examination data are data relating to tender joint count (e.g., 0 to 28), swollen joint count (e.g., 0 to 28), patient assessment of pain (e.g., 0 to 100), patient assessment of disease activity (e.g., 0 to 100), physician assessment of disease activity (e.g., 0 to 100), Health Assessment Questionnaire (e.g., 0 to 3), "Funktionsfragebogen Hannover" (e.g., 0 to 100), Clinical Disease Activity Index (e.g., 0 to 76).

[0058] Preferred laboratory data are data related to rheumatoid factor (positive, negative), anti-cyclic citrullinated peptide (positive, negative), seropositive rheumatoid arthritis (positive, negative), C-reactive protein (e.g., >0.01), erythrocyte sedimentation rate (e.g., 0 to 100), ESR-based disease activity score (e.g., 0 to 9.1), CRP-based disease activity score (e.g., 0 to 8), simple disease activity index (e.g., 0 to 86), duration of remission (e.g., >=0), count of previous flare-ups (e.g., >=0).

[0059] All of these possible data may be included in the input data and used by the selection unit.

[0060] According to a preferred embodiment of the CDS system, for a plurality of different prediction models, each prediction model has been trained for a different patient group, and each prediction model is selected based on patient-related information in the input data, preferably based on demographic data and / or examination data, and in particular based on information about distinguishing features from groups including sex, disease type (e.g., seropositive vs. seronegative), age, underlying health conditions, body mass index. This means that there are prediction models specifically trained for a particular patient group, which can be identified by a particular value of the patient-related data. For different patients whose respective values ​​are different, different prediction models are automatically selected.

[0061] According to a preferred embodiment of the CDS system, for multiple different prediction models, each prediction model has been trained for a different position in the clinical pathway, and each prediction model is selected based on input data related to the patient's position in the clinical pathway, preferably each prediction model is selected based on examination data.

[0062] According to the preferred embodiment of CDS system, for a plurality of different prediction models, each prediction model has been trained for different drugs, and each prediction model is selected based on the drug type given (indicated) in the input data, and the drug is particularly based on cDMARDs (lightweight and cheap conventional disease-modifying anti-rheumatic drugs), for example, based on methotrexate, sulfasalazine, hydroxychloroquine and leflunomide. However, there are many biological DMARDs (expensive, partially serious side effects, but usually much greater impact on disease activity). Recently, there are also biosimilars and targeted synthetic DMARDs on the market. And there are also NSAIDs (for example, aspirin, ibuprofen) for very mild symptoms, and there are also dangerous steroid drugs such as glucocorticoids or cortisone for short-term application in the case of acute severe sudden illness.

[0063] Suitable active agents (drugs) for specific inflammatory diseases are listed below:

[0064] Preferred drugs used in the treatment of rheumatoid arthritis are conventional disease-modifying antirheumatic drugs (cDMARDs) or biologics or other drugs that temporarily relieve pain and inflammation. Preferred cDMARDs for the treatment of RA include hydroxychloroquine, leflunomide, methotrexate, sulfasalazine or minocycline. Preferred biologics include abatacept, rituximab, tocilizumab, anakinra, adalimumab, etanercept, infliximab, certolizumab pegol or golimumab. Preferred tsDMARDs include Janus-associated kinase inhibitors such as tofacitinib or baricitinib. Preferred nonsteroidal anti-inflammatory drugs (NSAIDs) include ibuprofen / hydrocodone, ibuprofen / oxycodone, naproxen sodium, aspirin, celecoxib, nabumetone, naproxen (-sodium), piroxicam, diclofenac, diflunisal, indomethacin, ketoprofen, etodolac, fenoprofen, flurbiprofen, ketorolac, meclofenamic acid, mefenamic acid, meloxicam, oxaprozin, sulindac, salicylic acid, tolmetin, diclofenac / misoprostol, topical capsaicin or opioid analgesics such as codeine, acetaminophen / codeine, fentanyl, hydrocodone, hydromorphone, morphine, pethidine, oxycodone, tramadol. Preferred steroidal drugs include corticosteroids such as betamethasone, prednisone, dexamethasone, cortisone, hydrocortisone, methylprednisolone, prednisolone. Preferred immunosuppressants include cyclosporine, cyclophosphamide, azathioprine or hydroxychloroquine.

[0065] Preferred drugs used in the treatment of psoriatic arthritis (PsA) include disease-modifying antirheumatic drugs (DMARDs), immunosuppressants, and tumor necrosis factor-alpha (TNF-alpha) inhibitors. Preferred DMARDs for the treatment of PsA include methotrexate, sulfasalazine, cyclosporine, or leflunomide. Preferred nonsteroidal anti-inflammatory drugs (NSAIDs) include ibuprofen or naproxen. Preferred immunosuppressant drugs include azathioprine. Preferred TNF-alpha inhibitors include adalimumab, etanercept, golimumab, or infliximab.

[0066] Preferred medicines used in the treatment of chronic obstructive pulmonary disease (COPD) are, for example, short-acting bronchodilators, corticosteroids, methylxanthines, long-acting bronchodilators, combination medicines, roflumilast, mucus-active medicines, vaccines, antibiotics, cancer drugs or biological medicines. Examples of short-acting bronchodilators include salbutamol, levosalbutamol, ipratropium bromide or salbutamol / ipratropium bromide. Preferred corticosteroids include fluticasone or prednisolone. Preferred long-acting bronchodilators are aclidinium bromide, arformoterol, formoterol, glycopyrrolate, indacaterol, olodaterol, revefenacin, salmeterol, tiotropium bromide or umeclidinium. Recommended LABA / LAMA combined bronchodilator therapy includes aclidinium / formoterol, glycopyrrolate / formoterol, tiotropium / olodaterol, or umeclidinium / vilanterol. Combinations of inhaled corticosteroids with long-acting bronchodilators include budesonide / formoterol, fluticasone / salmeterol, or fluticasone / vilanterol.

[0067] Preferred drugs used in the treatment of asthma are bronchodilators or anti-inflammatory drugs, either quick-relief drugs or long-term asthma control drugs, respectively. Preferred are short-acting beta agonists, such as albuterol or levosalbutamol. Also preferred are anticholinergics, such as ipratropium bromide (Atrovent HFA). Preferred long-term asthma control drugs include: inhaled corticosteroids, such as beclomethasone, budesonide, flunisolide, fluticasone or mometasone; corticosteroids, such as prednisone, methylprednisolone or hydrocortisone; long-acting beta agonists, such as formoterol or salmeterol. Preferred combination inhalers include budesonide and formoterol or fluticasone and salmeterol. Preferred leukotriene modifiers include montelukast, zafirlukast or zileuton. Preferred methylxanthines include theophylline. Preferred immunomodulators include mepolizumab, omalizumab or reslizumab.

[0068] Preferred drugs used in the treatment of multiple sclerosis (MS) are interferon beta-1b, interferon beta-1a, glatiramer acetate, pegylated interferon beta 1-a, mitoxantrone, natalizumab, fingolimod or other sphingosine-1-phosphate receptor modulators, teriflunomide, pyrimidine, cladribine, ocrelizumab, siponimod, cladribine, dioxazolidinone fumarate, ozanimod, monomethyl fumarate.

[0069] Preferred drugs used in the treatment of Crohn's disease are drugs for treating any infection (usually antibiotics) and drugs for reducing inflammation (usually aminosalicylic acid anti-inflammatory drugs and corticosteroids). In particular, drugs for treating the symptoms of Crohn's disease include 5-aminosalicylic acid (5-ASA) preparations, prednisone, immunomodulators such as azathioprine (given as a prodrug of 6-mercaptopurine), methotrexate, infliximab, adalimumab, certolizumab pegol, vedolizumab, ustekinumab and natalizumab. Hydrocortisone should be used for severe attacks of Crohn's disease.

[0070] Some of the above mentioned active agents, such as TNF-α inhibitors or immunomodulators, belong to the group of expensive biological drugs.

[0071] A preferred embodiment of the CDS system is designed to select a prediction model based on the type of input data available, preferably wherein the prediction model is selected depending on whether laboratory data is part of the input data or not. This has the following advantages: in case of a preliminary interview or examination (no laboratory results), a prediction model can be selected that allows a first impression of the possible outcome, and after the laboratory data are available, another prediction model is automatically selected that allows an enhanced and optimized prediction. The availability of different data items may be independent of the path (for patients who have not received treatment, there may also be no recent available laboratory data). Selection based on different kinds of available data is advantageous. If laboratory data are not available, other models should be used instead of utilizing the available laboratory data.

[0072] Since there are many possible constellations of input data and inflammatory diseases, some illustrative examples for rheumatoid arthritis are listed below.

[0073] The EULAR guidelines list three stages for the treatment of RA using pharmacological non-local treatments, cDMARDs, biological disease-modifying antirheumatic drugs (bDMARDs) and targeted synthetic disease-modifying antirheumatic drugs (tsDMARDs).

[0074] In the first stage, usually between the methotrexate combined with short-term glucocorticoid and the sulfasalazine combined with short-term glucocorticoid, select. The CDS system can be used to predict the effectiveness of these medicines and provide help for the estimation of the applied medicines. The input data for prediction can be the demographic data and examination data of the corresponding patients. The prediction model can be trained for the reaction of the medicines and the patient about these medicines.

[0075] However, at this stage, if it is planned to taper off the medication already applied (i.e., if the patient is in sustained remission), the prediction model may also be selected. The input data for the prediction may then include demographic and examination data of the respective patient and information about the medication applied. The prediction model may be trained for dose reduction of the respective medication in a sustained remission scenario and the patient's response to the taper.

[0076] In the second phase, expensive activating agents are usually used. Usually, a choice is made between adding bDMARD or JAK inhibitor on the one hand and changing the cDMARD already applied or adding cDMARD on the other hand. The CDS system can be used to predict the effectiveness of these alternatives and provide help for selection. In addition to the data about the applied medicine, the input data for prediction can also be the demographic data and inspection data of the corresponding patient. The prediction model can be trained for the data and the patient's reaction about the corresponding medicine.

[0077] However, in this stage, the prediction model can also be selected in the case of a planned dose reduction or time interval increase under sustained remission. Then, the input data for prediction can also be the demographic data and examination data of the corresponding patient and information about the applied medicine. The prediction model can be trained for the dose reduction or time interval increase of the corresponding medicine under the sustained remission scenario and the patient's response.

[0078] In the third phase, it is usually decided whether the applied drug (e.g., another bDMARD or JAK inhibitor) should be changed. The CDS system can be used to predict the effectiveness of such changes. In addition to the applied medicines, the input data for prediction can also be the demographic data and examination data of the corresponding patients. The prediction model can be trained for the medicines and the patient's response to these medicines.

[0079] However, as in stage 2, in this stage, a prediction model can also be selected in the case of a planned dose reduction or time interval increase under sustained remission. Then, the input data for prediction can also be the demographic data and examination data of the corresponding patient and information about the applied drug. The prediction model can be trained for a dose reduction or time interval increase of the corresponding drug under sustained remission scenarios and the patient's response.

[0080] For example, one study showed that methotrexate was ineffective in approximately 43% of patients in phase 1 (see, e.g., https: / / arthritis-research.biomedcentral.com / articles / 10.1186 / s13075-018-1645-5). Therefore, it would be advantageous to identify those patients who respond positively to methotrexate and those who do not. During dose reduction or time interval increase, the probability of flare-up (within a certain time frame) can be calculated by a predictive model.

[0081] Regarding psoriatic arthritis, there are similar EULAR guidelines that include four stages with a similar process as described above. Here, the effect of applying a new drug or the risk of sudden disease after drug reduction can also be predicted by automatically selecting a prediction model.

[0082] Preferably, the first selection of the prediction model is based on the doctor's diagnosis (type of disease and stage of the EULAR guidelines) and the patient's demographic data. Then, the medicine used or the medicine planned to be used and the planned action (response prediction or dose reduction) can be part of the input data. Finally, the presence of certain data types (e.g., only examination data or also laboratory data) may also be a criterion for selecting the prediction model used. Finally, historical data (patient's medical records), comorbidities or lifestyle of the patient, potentially available omics data (e.g., metabolomics, proteomics, genomics) and imaging data (e.g., computed tomography images) may also be part of the input data and serve as the basis for selecting the prediction model.

[0083] According to a preferred embodiment of the CDS system, multiple different prediction models are trained to determine the probability that an individual patient will respond to a particular drug and / or the risk of a situational flare-up and / or the risk of an adverse drug event for tapering of different drugs.

[0084] Preferably, the prediction model is trained to:

[0085] - Determine the probability of response to first-line drugs such as methotrexate and sulfasalazine, and / or

[0086] - Determine the choice of second-line drugs, and / or

[0087] - Particularly for RA patients in stable remission receiving biologics, preferably for multiple dose regimens, tapering of the drug during any treatment phase (according to EULAR tapering recommendations).

[0088] According to a preferred embodiment of the CDS system, multiple prediction models are trained to determine a patient's drug response to multiple drugs, preferably, wherein a single model determines a patient's drug response to multiple drugs and / or a model in a group of multiple models determines a patient's drug response to a single drug.

[0089] Preferred embodiments of the CDS system are designed to output the probability of a flare (particularly in relation to the use and / or dosage of a predefined drug), the probability of an adverse event (e.g., a side effect of a drug), and / or the probability that the patient will not respond to the drug. For example, the CDS system is designed to predict the value of a relevant disease activity score, such as DAS28-ESR in RA patients (e.g., instead of or in addition to predicting a flare), which is defined as DAS28-ESR>2.6.

[0090] Preferably, the prediction model of the clinical decision support system is trained to determine and output a confidence score for a prediction, preferably where the prediction is a binary value involving a classification and the confidence is a probability value, and / or preferably where the prediction is a regression, the output comprises a prediction interval for a point prediction.

[0091] A preferred embodiment of the CDS system is designed to output information about which group of input parameters has the greatest impact on the output, preferably, a preferred embodiment of the CDS system is designed to generate values ​​for each parameter of the group to determine the extent of their impact on the output information. This provides a quantitative impression of the importance of these parameters. For example, if it is obvious for a certain patient that the results are strongly dependent on the body mass index, specific efforts can be made to actively change the body mass index. Therefore, an advantage of this embodiment is that the user can infer from the output that the high risk of sudden illness is due to a specific drug regimen and selectively change this specific drug regimen.

[0092] According to a preferred method for manufacturing a CDS system, a prediction method according to the present invention is performed using a clinical decision support system according to the present invention, and a feedback data set is provided for a plurality of patients, wherein the feedback data set is used to further train a prediction model. It should be noted that the prediction model is related to the distinguishing features of the feedback data. Preferably, the feedback data set is used for training, wherein the DAS28_ESR score of the patient's sudden illness is higher than 2.6.

[0093] In a preferred case of a CDS system for rheumatoid arthritis (but also for other diseases), there are multiple (especially more than 10) prediction models specifically trained for different input data.

[0094] There may be some such prediction model groups, wherein each group comprises a plurality (especially more than 10) prediction models, which are specially trained for different input data of different diseases (e.g. RA, PsA, spondyloarthropathies - SpA). Then, the selection unit is designed to parse the input data to obtain information about the diagnosis of the disease, thereby determining the actual prediction model group that should be used to select the corresponding prediction model, i.e., to filter all prediction models in the case where these prediction models were trained using data related to the disease.

[0095] Preferably, there are a plurality of prediction models which are specially trained for different treatment phases in a disease process. A preferred selection unit parses the input data for information about the treatment phase and filters the prediction models accordingly based on their training (particularly together with filters about a disease).

[0096] It can be seen that, preferably, the specific data for which the prediction model is trained is labeled on the prediction model, for example, the labels may include information about the disease, treatment stage, patient (e.g., gender, age, BMI), drug or use case (response to the drug or gradual reduction of the drug).

[0097] Preferably, there are multiple prediction models, which are specifically trained for different use cases (e.g., change of medication or tapering of medication). A preferred selection unit parses the input data for information about the use case and filters the prediction models accordingly based on their training (especially together with filters for a certain disease and / or treatment stage).

[0098] Preferably, there are a plurality of prediction models trained specifically for laboratory data and other prediction models trained for other examination data (e.g. examinations performed by a doctor). A preferred selection unit parses the input data by checking whether there is available laboratory data / examination data and filters the prediction models accordingly according to their training (especially together with filters about a certain disease and / or treatment stage and / or use case).

[0099] In the absence of explicit description, the various embodiments or their various aspects and features can be combined or exchanged with each other without limiting or expanding the scope of the invention described, as long as such combination or exchange is meaningful and within the meaning of the invention. In particular, some of the features described herein can be combined with other features of the present description to form a corresponding invention. The advantages described with respect to one embodiment of the invention (where applicable) are also advantageous for other embodiments of the invention.

[0100] Other objects and features of the present invention will become apparent from the following detailed description considered in conjunction with the accompanying drawings. It should be understood, however, that the drawings are designed only for the purpose of illustration and not as a definition of the limits of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] Figure 1 A preferred data processing system having a CDS system according to the present invention is shown.

[0102] Figure 2 A block diagram of a prediction method according to the present invention is shown.

[0103] Figure 3 A block diagram of a method for manufacturing a CDS system according to the present invention is shown.

[0104] Figure 4 The EULAR regimen for the treatment of rheumatoid arthritis is shown.

[0105] Figure 5 The EULAR regimen for the treatment of psoriatic arthritis is shown.

[0106] Figure 6 A possible decision tree for selecting a unit is shown. DETAILED DESCRIPTION

[0107] Figure 1 A preferred data processing system 7 with a CDS system 1 according to the invention is shown. The data processing system 7 comprises a client computer 8 connected to a service computer system 9 via a data network N. The service computer system 9 comprises the clinical decision support system 1 according to the invention.

[0108] A clinical decision support system 1 for estimating drug-related treatment optimization for inflammatory diseases comprises the following components:

[0109] A computing unit 2 comprising an input interface 3 designed to receive input data D (see the figure below) and an output interface 4 designed to output a result R. The computing unit 2 is designed to host a plurality of prediction models M, i.e. process these prediction models M in order to obtain a result R from the input data D by means of the prediction models M. However, unused prediction models do not need to be actively hosted by the computing unit 2.

[0110] The memory 5 is used to store and provide multiple prediction models M for use in the case where the calculation unit 2 needs the prediction model M.

[0111] A plurality of different trained prediction models M, which are stored in the memory 5. Each model M is trained to predict the probability of a treatment outcome for a plurality of different drug-related treatment options and for a specific patient group. For example, some prediction models M are trained for different patient groups, and some prediction models M should be selected based on the patient-related information in the input data D; some prediction models M are trained for different positions in the clinical pathway, and some prediction models M are selected based on the corresponding input data D; or some prediction models M are trained for different drugs, and some prediction models M are selected based on the drug type given in the input data D.

[0112] Preferably, the prediction model M is trained to determine the probability that a single patient will respond to a particular drug and / or the risk of flare-up for different drug tapering scenarios, particularly for determining the probability of response to a first-line drug or determining the selection of a second-line drug or determining tapering of a drug in a later treatment phase (i.e., not an actual phase or period). There may be a prediction model M trained for only a single drug or for multiple drugs.

[0113] A selection unit 6 designed for automatically selecting one of the prediction models M based on the input data D according to a predefined selection scheme. The prediction model M may be selected, for example, based on the diagnosis, the type of input data D available, preferably wherein the prediction model M is selected depending on whether laboratory data are part of the input data D or not.

[0114] The clinical decision support system 1 is designed to generate an output result R by processing input data D using a selected prediction model M. In particular, the clinical decision support system 1 is designed to output the probability of a sudden illness, the probability of an adverse event (e.g., a side effect of a drug), and / or the probability that a patient will not respond to a drug. To achieve this, the prediction model M of the clinical decision support system 1 can be trained to determine and output a confidence score for the prediction, preferably, where the prediction is a binary value involving a classification and the confidence is a probability value, and / or preferably, where the prediction is a regression, the output includes a prediction interval for a point prediction.

[0115] The clinical decision support system 1 can be designed to output information about which input parameter group has the greatest impact on the output (e.g., in the result), and the clinical decision support system 1 is preferably designed to generate, for each parameter of the group, a value of their degree of impact on the output result R. This can be achieved, for example, by using the SHAP interpretable AI framework (Shapley Additive exPlanations).

[0116] Figure 2A block diagram of a prediction method according to the present invention is shown.

[0117] In step I, providing, for example, Figure 1 The clinical decision support system 1 is shown.

[0118] In step II, input data D are provided to the clinical decision support system 1 , wherein the input data D may be automatically selected and provided, in particular when new data become available for predefined patients. However, the doctor may also upload selected data sets into the CDS system 1 .

[0119] In step III, the result R is determined using the clinical decision support system 1 , wherein the prediction model M is automatically selected by the clinical decision support system 1 based on the input data D. The result R is automatically determined by the selected prediction model M.

[0120] In step IV, the results R are output by the CDS system 1. If the patient's results have changed substantially compared to earlier results for the same patient, the user may be notified, for example, in the form of a warning message or an icon in the patient list, indicating that the case should be opened.

[0121] Figure 3 A method for manufacturing a CDS system 1 according to the present invention (see for example Figure 1 ) method.

[0122] It should be noted that although the method uses two or more (preferably multiple) model groups G and a corresponding number of training data sets T, only a single model group G is considered in this example. However, for each model group G, the process is similar.

[0123] In step TI, a model group G having a plurality of untrained machine learning models m is provided. Preferably, the untrained models m have different internal architectures and / or different hyperparameters, so that it is possible to evaluate which architecture / hyperparameters will be best used for a specific task.

[0124] Also in step TI, a training data set T is provided, which includes data with different distinguishing characteristics compared to other training data sets T. For example, all patients are female or the training data set T includes laboratory data.

[0125] In step TII, the training of the model group G is performed using the training data set T.

[0126] In step TIII, the trained prediction models M of the model group G are ranked using predefined quality criteria. It can be seen that the trained prediction model M is the "winner" of this ranking. The prediction model M can be developed and compared offline.

[0127] In step TIV, the best ranked prediction model M in the model group G is manually or automatically selected as the prediction model M of the clinical decision support system 1 .

[0128] In step TV, a feedback dataset F for a plurality of patients is provided and the (selected) prediction model M of the CDS system 1 is further trained using the feedback dataset F. The prediction model M is here related to the distinguishing features of the feedback dataset F. For example, the feedback dataset F can be used for training, wherein the DAS28-ESR score of the patient's sudden illness is higher than 2.6. The patient can also self-report sudden illness, which is also included in the feedback dataset F.

[0129] Figure 4 The EULAR schematic guidelines for the treatment of rheumatoid arthritis are shown. The EULAR guidelines list three stages for the treatment of RA using cDMARDs that may be combined with glucocorticoids, bDMARDs that may be combined with cDMARDs and tsDMARDs (such as JAK inhibitors). Dashed ellipses are added to treatment decisions that may be supported by a prediction model M specifically trained for the response to the drug, and dot-dash ellipses are added to treatment decisions that may be supported by a prediction model M specifically trained for tapering of the drug.

[0130] In the first stage, a choice is usually made between methotrexate combined with a short-term glucocorticoid and sulfasalazine combined with a short-term glucocorticoid. The CDS system 1 can be used to predict the effectiveness of these drugs and provide assistance for estimating the success of the drugs used. The input data D for prediction can be the demographic data and examination data of the corresponding patients. The prediction model M can be trained according to the drugs and the patient's reaction to these drugs.

[0131] However, at this stage, assuming that the patient is in sustained remission, the prediction model M can also be selected in the case of a plan to gradually reduce the already applied drugs. Then, the input data D for the prediction can also be the demographic data and examination data of the individual patients and the information about the applied drugs. The prediction model M can be trained for the dose reduction of the corresponding drug in the case of sustained remission and the patient's response to the gradual reduction.

[0132] In the second stage, more expensive activating agents are usually applied. Usually, a bDMARD or JAK inhibitor (tsDMARD) is added on the one hand and a bDMARD or cDMARD is added on the other hand. The CDS system 1 can be used to predict the effectiveness of these alternatives and provide help for selection. In addition to the medicine used, the input data D for prediction can also be the demographic data and inspection data of the corresponding patient. The prediction model M can be trained for the reaction of the medicine and the patient about these medicines.

[0133] However, in this stage, in the case of a planned dose reduction or time interval increase in sustained remission, the prediction model M can also be selected. Then, the input data D for prediction can also be the demographic data and examination data of the corresponding patient and the information about the applied drug. The prediction model M can be trained for the dose reduction or time interval increase of the corresponding drug in the sustained remission scenario and the patient's response.

[0134] In the third phase, it is usually decided whether the applied drug should be changed (e.g., another bDMARD or JAK inhibitor is used due to poor prognostic factors or ineffectiveness or adverse events observed in the second phase). The CDS system 1 can be used to predict the effectiveness of such drug changes. In addition to the applied drugs, the input data D for prediction can also be demographic data and examination data of the corresponding patients. The prediction model M can be trained for the drugs and the patient's response to these drugs.

[0135] However, as in stage 2, in this stage, the prediction model M can also be selected in the case of a planned dose reduction or time interval increase in sustained remission. Then, the input data D for prediction can also be the demographic data and examination data of the corresponding patient and information about the applied drug. The prediction model M can be trained for the dose reduction or time interval increase of the corresponding drug in the context of sustained remission and the patient's response.

[0136] Figure 5 The EULAR protocol for the treatment of psoriatic arthritis is shown. Four phases of treatment of PsA are listed in the EULAR guidelines, where the algorithm is similar to the treatment of RA. Again, dashed and dotted ellipses are added to these parts that can be predicted by predicting the model M specifically trained for the response to the drug.

[0137] Regarding psoriatic arthritis, there are similar EULAR guidelines that include four stages with a similar process as described above. Here, the effect of applying a new drug or the risk of sudden disease after drug tapering can also be predicted by automatically selecting the prediction model M.

[0138] Figure 6 The selection unit 6 (see above) is shown Figure 1 and Figure 2 ) possible decision trees.

[0139] First (upper part) a diagnosis is made to determine the actual disease the patient suffers from. This information is entered into the input data D, and the selection unit 6 is designed to determine the actual disease and select a prediction model M trained for this disease based on the input data D. However, there may be a large number of possible prediction models M, so the selection should be filtered.

[0140] Second (the next stage from top to bottom), the stage of treatment (see e.g. Figure 4 and Figure 5 ) is added to the input data D, and the selection unit 6 can be designed to: determine the actual stage according to the input data D and select the prediction model M trained for this stage.

[0141] Third (the next stage from top to bottom), a use case (changing medication or gradually reducing medication) can be added to the input data D, and the selection unit 6 can be designed to: select a prediction model M based on the input data D, wherein the prediction model M is trained to predict the effects of certain drugs on patients or the effects of gradually reducing the dose of drugs on patients.

[0142] Next (bottom stage), the input data D may be automatically checked by the selection unit 6 whether there are available examination and / or laboratory data, and a prediction model M trained to make predictions on such data may be selected by the selection unit 6 .

[0143] Although the present invention has been disclosed in the form of preferred embodiments and variations thereon, it should be understood that numerous additional modifications and variations could be made to the present invention without departing from the scope of the invention.

[0144] For the sake of clarity, it should be understood that the use of "a" or "an" throughout this application does not exclude a plurality, and "comprising" does not exclude other steps or elements. References to a "unit" or "module" do not exclude the use of more than one unit or module.

Claims

1. A clinical decision support system (1) for estimating drug-related treatment optimization for inflammatory diseases, the clinical decision support system (1) being designed to output the probability of sudden illness associated with the use and / or dosage of a predefined drug, the clinical decision support system (1) comprising: a computing unit (2) designed to host a plurality of prediction models (M), said computing unit (2) comprising an input interface (3) designed to receive input data (D) and an output interface (4) designed to output results (R), - a plurality of different trained prediction models (M), wherein each model (M) is trained to: predict the probability of a treatment outcome for a plurality of different drug-related treatment options and for a specific patient group based on input data, and wherein the plurality of different prediction models are trained to determine the risk of a flare-up for different drug tapering scenarios, a selection unit (6) designed to automatically select one of the prediction models (M) based on the input data (D) according to a predefined selection scheme, The clinical decision support system (1) is designed to: generate an output result (R) by processing the input data (D) using the selected prediction model (M), and Wherein, the selection unit is designed to: search the input data for a predefined data type and / or a value of a predefined data type; and determine whether the predefined data type exists in the input data; and / or compare the value with a predefined threshold; and / or determine whether the value meets predefined requirements.

2. The clinical decision support system according to claim 1, wherein: For a plurality of different prediction models (M), each prediction model (M) has been trained for a different group of patients and each prediction model (M) is selected based on patient-related information in the input data (D).

3. The clinical decision support system according to claim 2, wherein: For a plurality of different prediction models (M), each prediction model (M) has been trained for a different group of patients and each prediction model (M) has been selected based on demographic data and / or examination data.

4. The clinical decision support system according to claim 3, wherein: For multiple different prediction models (M), each prediction model (M) has been trained for a different patient group and each prediction model (M) has been selected based on information about distinguishing characteristics from the group including gender, disease type, age, underlying health conditions, body mass index.

5. The clinical decision support system according to claim 4, wherein: The disease types include seropositive rheumatoid arthritis and seronegative rheumatoid arthritis.

6. The clinical decision support system according to any one of claims 1 to 5, wherein: For the plurality of different prediction models (M), each prediction model (M) has been trained for a different position in the clinical pathway, and each prediction model (M) is selected based on input data (D) relating to the position of the patient in the clinical pathway.

7. The clinical decision support system according to claim 6, wherein: Each prediction model (M) is selected based on the inspection data.

8. The clinical decision support system according to any one of claims 1 to 5, wherein: For the plurality of different prediction models (M), each prediction model (M) has been trained for a different drug, and each prediction model (M) is selected based on the drug type given in the input data (D).

9. The clinical decision support system according to claim 8, wherein: The drugs are based on DMARDs or NSAIDs.

10. The clinical decision support system according to claim 8, wherein: The drugs are based on bDMARDs, cDMARDs or tsDMARDs.

11. A clinical decision support system according to any one of claims 1 to 5, designed to select a prediction model (M) based on the type of available input data (D).

12. The clinical decision support system according to claim 11, wherein: The prediction model (M) is selected depending on whether laboratory data is part of the input data (D).

13. A clinical decision support system according to any one of the preceding claims 1 to 5, wherein: A plurality of said different prediction models (M) are trained to determine the probability that an individual patient will respond to a particular drug and / or the risk of flare-up for different drug tapering scenarios.

14. The clinical decision support system according to claim 13, wherein: The prediction model (M) is trained to: - Determine the probability of response to first-line drugs, and / or - Determine the choice of second-line drugs, and / or - Scenarios where drug tapering is performed in the late treatment phase.

15. The clinical decision support system according to claim 13, wherein: The prediction model (M) was trained for the scenario of drug tapering in the late treatment phase for RA patients in stable remission receiving biologics for multiple dosing regimens.

16. The clinical decision support system according to claim 14, wherein: The first-line drugs include methotrexate and sulfasalazine.

17. The clinical decision support system according to any one of claims 1 to 5, wherein: A plurality of prediction models (M) are trained to determine a patient's drug response to a plurality of drugs.

18. The clinical decision support system according to claim 17, wherein: A single model determines the drug response of a patient to multiple drugs, and / or the models in the set of multiple models determine the drug response of a patient to a single drug.

19. The clinical decision support system according to any one of claims 1 to 5, wherein the clinical decision support system is designed to output the probability of an adverse event and / or the probability of a patient not responding to a drug.

20. The clinical decision support system according to claim 19, wherein: The prediction model (M) of the clinical decision support system (1) is trained to determine and output a confidence score for the prediction.

21. The clinical decision support system according to claim 20, wherein: The prediction is a binary value involving classification, and the confidence is a probability value.

22. The clinical decision support system according to claim 20, wherein: Forecasting is a regression and the output includes prediction intervals for point forecasts.

23. The clinical decision support system according to claim 19, wherein: The adverse events include side effects of the drug.

24. The clinical decision support system according to any one of claims 1 to 5, the clinical decision support system being designed to output information on which input parameter group has the greatest impact on the output.

25. The clinical decision support system according to claim 24, wherein the clinical decision support system is designed to generate, for each parameter of the group, a value of the degree of influence of each parameter on the output result (R).

26. A prediction method for computational decision support, comprising the following steps: - providing a clinical decision support system (1) according to any one of claims 1 to 25, - providing input data (D) to the clinical decision support system (1), - determining a result (R) using the clinical decision support system (1), wherein a prediction model (M) is automatically selected by the clinical decision support system (1) based on the input data (D), and the result (R) is automatically determined by the selected prediction model (M), and - output the result (R).

27. The prediction method according to claim 26, wherein: In case new data becomes available for predefined patients, said input data (D) is automatically selected and provided.

28. The prediction method according to claim 26, wherein: If a patient's results have changed substantially compared to earlier results for the same patient, the user is notified in the form of a warning message or an icon in the patient list so that he knows that he has to open the case.

29. A method for manufacturing a clinical decision support system (1) according to any one of claims 1 to 25, comprising the following steps: - providing at least a first model group (G) and a second model group (G), each model group (G) comprising a plurality of untrained machine learning models (m), - providing at least a first training data set (T) and a second training data set (T), each training data set (T) comprising data having different distinguishing characteristics, - using the first training data set (T) to train the first model group (G), and using the second training data set (T) to train the second model group (G), - ranking each trained prediction model (M) of the set of models (G) using predefined quality criteria, and - manually or automatically selecting the best ranked prediction model (M) in each model group (G) as the prediction model (M) of the clinical decision support system (1).

30. The method of claim 29, wherein: Each model group (G) includes multiple models (m) with different internal architectures and / or different hyperparameters.

31. The method of claim 29, wherein: Develop and compare prediction models offline (M).

32. The method of claim 29, wherein: The clinical decision support system (1) is used to perform the prediction method according to any one of claims 26 to 28, and a feedback data set is provided for a plurality of patients, wherein the feedback data set is used to further train the prediction model (M), and the prediction model (M) is related to the distinguishing features of the feedback data.

33. The method of claim 32, wherein: The feedback dataset was used for training in which patients had disease-emerging DAS28-ESR scores above 2.

6.

34. A data processing system (7), comprising a data network (N), a plurality of client computers (8) and a server computer system (9), wherein: The service computer system (9) comprises a clinical decision support system (1) according to one of claims 1 to 25.

35. The data processing system according to claim 34, wherein: The data processing system (7) is a computer network system.

36. A computer program product, comprising a computer program, which can be directly loaded into the memory of a control unit of a computer system, and the computer program comprises program elements, which are used to perform the steps of the method according to any one of claims 26 to 33 when the computer program is executed by the control unit of the computer system.

37. A computer-readable medium having stored thereon program elements which can be read and executed by a computer unit in order to perform the steps of the method according to any one of claims 26 to 33 when the program elements are executed by the computer unit.

Images