Systems and methods for computing risk of predicted medical outcomes in patients treated with multiple medications

1. A computer implemented method of training a machine learning model for assessing risk of at least one predicted medical outcome for a patient having a plurality of patient-specific parameters and future planned treatment with a plurality of patient-specific medications, comprising:

in a pre-processing step, for each of a plurality of sample patients:

mapping a plurality of patient-specific medications and a plurality of patient-specific parameters to a plurality of risk scores of a plurality of medical outcomes by a mapping data structure that includes a plurality of relationship-mappings that map between a plurality of medications, a plurality of predicted medical outcomes, and a plurality of patient parameters, wherein each relationship-mapping includes at least one respective medication of the plurality of medications, at least one respective predicted medical outcome of the plurality of medical outcomes, at least one respective patient parameter of the plurality of patient parameters, and at least one respective risk score for the at least one respective predicted medical outcome,

wherein at least one of the plurality of relationship-mappings includes a predicted medical outcome denoting a medication induced event of a medication of the plurality of medications, and at least one patient parameter selected from the group consisting of: a medication influencing factor directly affecting the medication which affects the medication induced event, and an event influencing factor directly affecting the medication induced event,

wherein the medication influencing factor is selected from the group consisting of the following primary parameters affecting the medication: blood level, absorption, distribution, metabolism, elimination, half life elimination, and pharmacodynamics,

wherein at least one patient-specific medication is mapped to a plurality of active ingredients, and wherein a subset of the plurality of relationship-mappings each include a respective active ingredient of the plurality of active ingredients, and the medication influencing factor is for the respective active ingredient;

computing an aggregated risk score for each respective predicted medical outcome by aggregating the risk scores of the plurality of relationship-mappings including each respective predicted medical outcome identified by the mapping data-structure; and

automatically designating the aggregated risk score as a ground truth label for a record corresponding to the sample patient for creating a training dataset for training a machine learning model;

in a training step, creating the training dataset comprising a plurality of records for a plurality of sample patients, wherein a record of a sample patient is computed by:

defining an input vector as a respective plurality of patient-specific medications and a respective plurality of patient-specific parameters of the sample patient mapped by the mapping data-structure, and

designating an output vector as the ground truth label, the output vector including the aggregated risk score computed for each respective predicted medical outcome; and

training a machine learning model on the training dataset using a supervised learning approach using the ground truth label, by iteratively teaching the machine learning model to infer a function that maps the input vector of each record of the training dataset to the corresponding ground truth label of the record,

wherein the machine learning model is iteratively trained to learn the function that predicts a target aggregated risk score for each respective target predicted medical outcome in response to being fed a plurality of target patient-specific medications and a plurality of target patient-specific parameters for a target patient,

wherein the machine learning model generates the predicted target aggregated risk score without accessing the mapping data-structure.

2. The method of claim 1 , wherein the medication influencing factor is selected from the group consisting of the following secondary parameters affecting the primary parameter: other medication, medication condition, and non-genetic parameters of the patient that are unrelated to known genetic sequences.

3. The method of claim 1 , wherein each aggregated risk score denotes a personalized predicted risk assessment of the patient being treated with the plurality of patient-specific medications and having the patient-specific parameters developing the respective predicted medical outcome, and each respective risk score of each respective relationship-mapping is indicative of a probability of a certain patient being treated with the one respective medication and having the one respective patient parameter, developing the one respective predicted medical outcome.

4. The method of claim 1 , wherein the identified plurality of relationship-mappings are clustered according to respective predicted medical outcomes, wherein relationship-mapping members of each cluster have a common predicted medical outcome, and wherein the computing the aggregated risk score comprises aggregating the risk scores of the relationship-mapping members of each respective cluster.

5. The method of claim 1 , wherein at least one of the plurality of patient-specific parameters comprises an influence of a certain patient-specific medication on another certain patient-specific medication, and the predicted medical outcome is a result of the influence.

6. The method of claim 1 , wherein the aggregated risk score for each respective predicted medical outcome is computed by multiplication of individual risk scores of the identified plurality of relationship-mapping including the respective predicted medical outcome.

7. The method of claim 1 , wherein the aggregated risk score denotes a positive increase or a negative decrease relative to a baseline risk score of the respective predicted medication outcome.

8. The method of claim 1 , further comprising receiving an indication signal denoting an adjusted at least one of the plurality of patient-specific medications created by an adjustment of the at least one of the plurality of patient-specific medications, wherein the adjustment is selected from the group consisting of: removal, replacement, adjustment of dose, and addition, and iteratively feeding the adjusted at least one of the plurality of patient-specific medications into the trained machine learning model, for obtaining respective predicted target aggregated risk scores.

9. The method of claim 8 , wherein the adjustment is iterated until the aggregated risk score for each respective predicted medical outcome is less than a risk threshold, and instructions are generated for treating the patient to prevent or reduce risk of each respective predicted medical outcome below the risk threshold.

10. The method of claim 1 , further comprising receiving an indication signal denoting an adjustment of at least one of the plurality of patient-specific parameters, wherein the adjustment is selected from the group consisting of: removal, replacement, adjustment of value, and addition, and iterating using the adjustment of the at least one of the plurality of patient-specific parameters, the identifying the plurality of relationship-mappings, the computing the aggregated risk factor, and instructions are generated for treatment of the patient using the adjustment.

11. The method of claim 10 , wherein the adjustment is iterated until the aggregated risk score for each respective predicted medical outcome is less than a risk threshold, and the instructions are generated for treating the patient to prevent or reduce risk of each respective predicted medical outcome below the risk threshold.

12. The method of claim 1 , further comprising creating and/or updating the mapping data-structure, by:

obtaining raw medical data;

extracting a plurality of features from the raw medical data corresponding to at least one medication, at least one predicted medical outcome, at least one patient parameter, and values for computing a corresponding risk value;

computing at least one relationship-mapping between the extracted features; and

storing the at least one relationship-mapping in the mapping data-structure.

13. The method of claim 12 , wherein the raw medical data is extracted using natural language processing (NLP) from one or more members of the group consisting of: formal prescribing information, pharmaceutical leaflets, drug information databases, medical database, medical literature, clinical trials, adverse drug reaction reports.

14. The method of claim 1 , wherein computing the aggregated risk score comprises determining a risk classification category of a plurality of risk classification categories each indicative of a range of values of risk scores.

15. The method of claim 14 , further comprising generating instructions for presenting a table of risk classification categories, and placing each respective predicted medical outcome into one of the risk classification categories of the table according to the respective determined risk classification category.

16. The method of claim 15 , wherein each predicted medical outcome stored in the mapping data-structure is associated with a certain severity category selected from a plurality of severity categories, and the presented table includes a plurality of cells each denoting a respective risk classification category and a respective severity category, wherein each respective predicted medical outcome is placed into one of the plurality of cells based on the corresponding aggregated risk score and corresponding certain severity category.

17. The method of claim 16 , wherein when at least one of the plurality of patient-specific parameters matches at least one predicted medical outcome, creating a set of cells in the table denoting current medical outcomes, and placing each of the plurality of patient-specific parameters into one of the cells according to respective severity category of the respective current medical outcome.

18. The method of claim 1 , wherein the plurality of patient-specific parameters are non-genetic parameters that are unrelated to known genetic sequences.

19. The method of claim 18 , wherein the non-genetic patient-specific parameters of the patient are selected from the group consisting of: smoking, alcohol, nutrition, amount of exercise, occupation, lifestyle data, past usage of medications, current usage of medications, current medical conditions, past medical condition, past medical treatments, current diagnosis, symptoms, lab test results, and imaging results.

20. The method of claim 1 , further comprising, analyzing for each respective predicted medical outcome, the identified plurality of relationship-mappings including the respective predicted medical outcome to determine at least one medication and/or at least one patient parameter contributing statistically significantly and disproportionally to the computed aggregated risk score relative to other medications and/or other patient parameters, and providing the determined at least one medication and/or at least one patient parameter as a significant risk factor for the patient for developing the respective medical outcome, and further comprising selecting an adjustment for the at least one medication and/or at least one patient parameter for reducing risk of the respective medical outcome to below a risk threshold, and instructions are generated for treatment of the patient based on the selected adjustment.

21. The method of claim 1 , further comprising generating instructions for generating a plurality of dynamic drug labels, each dynamic drug label computed for each one of the plurality of patient-specific medications, each dynamic drug label including a plurality of sub-aggregated risk scores, each sub-aggregated risk score computed for each respective predicted medical outcome of the respective patient-specific medication by aggregating the risk scores of the identified plurality of relationship-mappings including the respective predicted medical outcome and the respective patient-specific medication.

22. The method of claim 21 , further comprising dynamically updating each one of the plurality of dynamic labels based on an update of the mapping data-structure.

23. The method of claim 1 , wherein at least one of the plurality of predicted medical outcomes denote an efficacy of treating a target disease, and the aggregated risk score for the at least one of the plurality of predicted medical outcomes comprises a change in efficacy relative to a baseline efficacy.

24. The method of claim 1 , further comprising monitoring aggregated risk scores for the predicted medical outcomes, by: computing a plurality of aggregated risk scores for each respective predicted medical outcomes over a plurality of time intervals, computing a trend according to the plurality of aggregated risk scores, and analyzing the trend to generate an alert when at least one of the aggregated risk scores is predicted to cross a threshold.

25. A system for training a machine learning model for assessing risk of at least one predicted medical outcome for a patient having a plurality of patient-specific parameters and future planned treatment with a plurality of patient-specific medications, comprising:

at least one hardware processor executing a code for:

in a pre-processing step, for each of a plurality of sample patients:

mapping a plurality of patient-specific medications and a plurality of patient-specific parameters to a plurality of risk scores of a plurality of medical outcomes by a mapping data structure that includes a plurality of relationship-mappings that map between a plurality of medications, a plurality of predicted medical outcomes, and a plurality of patient parameters, wherein each relationship-mapping includes at least one respective medication of the plurality of medications, at least one respective predicted medical outcome of the plurality of medical outcomes, at least one respective patient parameter of the plurality of patient parameters, and at least one respective risk score for the one respective predicted medical outcome,

wherein at least one of the plurality of relationship-mappings includes a predicted medical outcome denoting a medication induced event of a medication of the plurality of medications, and at least one patient parameter selected from the group consisting of: a medication influencing factor directly affecting the medication which affects the medication induced event, and an event influencing factor directly affecting the medication induced event,

wherein the medication influencing factor is selected from the group consisting of the following primary parameters affecting the medication: blood level, absorption, distribution, metabolism, elimination, half life elimination, and pharmacodynamics,

wherein at least one patient-specific medication is mapped to a plurality of active ingredients, and wherein a subset of the plurality of relationship-mappings each include a respective active ingredient of the plurality of active ingredients, and the medication influencing factor is for the respective active ingredient;

computing an aggregated risk score for each respective predicted medical outcome by aggregating the risk scores of the plurality of relationship-mappings including each respective predicted medical outcome identified by the mapping data-structure; and

automatically designating the aggregated risk score as a ground truth label for a record corresponding to the sample patient for creating a training dataset for training a machine learning model;

in a training step, creating the training dataset comprising a plurality of records for a plurality of sample patients, wherein a record of a sample patient is computed by:

defining an input vector as a respective plurality of patient-specific medications and a respective plurality of patient-specific parameters of the sample patient mapped by the mapping data-structure, and

designating an output vector as the ground truth label, the output vector including the aggregated risk score computed for each respective predicted medical outcome; and

training a machine learning model on the training dataset using a supervised learning approach using the ground truth label, by iteratively teaching the machine learning model to infer a function that maps the input vector of each record of the training dataset to the corresponding ground truth label of the record,

wherein the machine learning model is iteratively trained to learn the function that predicts a target aggregated risk score for each respective target predicted medical outcome in response to being fed a plurality of target patient-specific medications and a plurality of target patient-specific parameters for a target patient,

wherein the machine learning model generates the predicted target aggregated risk score without accessing the mapping data-structure.

26. A method of assessing risk of at least one predicted medical outcome for a patient having a plurality of patient-specific parameters and future planned treatment with a plurality of patient-specific medications, comprising:

feeding a plurality of target patient-specific medications and a plurality of target patient-specific parameters for a target patient into a machine learning model; and

obtaining an outcome of a predicted target aggregated risk score for each one of a plurality of target predicted medical outcomes,

wherein the machine learning model generates the predicted target aggregated risk score without accessing the mapping data-structure,

wherein the machine learning model is trained on a training dataset using a supervised learning approach using ground truth labels, by iteratively teaching the machine learning model to iteratively infer a function that maps the input vector of each record of the training dataset to the corresponding ground truth label of the record,

wherein the training set comprises: in a pre-processing step, for each of a plurality of sample patients:

mapping a plurality of patient-specific medications and a plurality of patient-specific parameters to a plurality of risk scores of a plurality of medical outcomes by a mapping data structure that includes a plurality of relationship-mappings that map between a plurality of medications, a plurality of predicted medical outcomes, and a plurality of patient parameters, wherein each relationship-mapping includes at least one respective medication of the plurality of medications, at least one respective predicted medical outcome of the plurality of medical outcomes, at least one respective patient parameter of the plurality of patient parameters, and at least one respective risk score for the at least one respective predicted medical outcome,

wherein at least one of the plurality of relationship-mappings includes a predicted medical outcome denoting a medication induced event of a medication of the plurality of medications, and at least one patient parameter selected from the group consisting of: a medication influencing factor directly affecting the medication which affects the medication induced event, and an event influencing factor directly affecting the medication induced event,

wherein the medication influencing factor is selected from the group consisting of the following primary parameters affecting the medication: blood level, absorption, distribution, metabolism, elimination, half life elimination, and pharmacodynamics,

wherein at least one patient-specific medication is mapped to a plurality of active ingredients, and wherein a subset of the plurality of relationship-mappings each include a respective active ingredient of the plurality of active ingredients, and the medication influencing factor is for the respective active ingredient;

computing an aggregated risk score for each respective predicted medical outcome by aggregating the risk scores of the plurality of relationship-mappings including each respective predicted medical outcome identified by the mapping data-structure;

automatically designating the aggregated risk score as a ground truth label for a record corresponding to the sample patient for creating a training dataset for training a machine learning model;

in a training step, creating the training dataset comprising a plurality of records for a plurality of sample patients, wherein a record of a sample patient is computed by:

defining an input vector as a respective plurality of patient-specific medications and a respective plurality of patient-specific parameters of the sample patient mapped by the mapping data-structure, and

designating an output vector as the ground truth label, the output vector including the aggregated risk score computed for each respective predicted medical outcome.

27. The method of claim 1 , wherein the machine learning model is trained using a weak supervision approach.

28. The method of claim 1 , further comprising treating the patient in view of the aggregated risk score for each unique predicted medical outcome.

29. The method of claim 1 , wherein the mapping data-structure is implemented as a member selected from a group consisting of: a table, and a set of pointers, and wherein the machine learning model is implemented as a member selected from a group consisting of: neural network, Markov chains, support vector machine (SVM), logistic regression, k-nearest neighbor, and decision trees.