Constructing prediction targets from a clinically-defined hierarchy
View Patent ↗A method, a computing system and a computer program product are provided. A model is generated and trained. The model is based on clinical data with outcomes from clinically-defined hierarchical metadata in a selected level of clinically-defined hierarchical metadata serving as an initial set of prediction targets. A score is determined for each of the prediction targets based on the generated model and the set of evaluation factors. The set of prediction targets, the generated model, and the scores for the set of prediction targets are updated until the updated scores for the updated set of prediction targets satisfy acceptance criteria. The updated generated model, using the updated set of prediction targets, is applied to predict one of a set of updated prediction targets of mutually exclusive outcome categories.
1. A method for generating a set of prediction targets from clinically-defined hierarchical metadata represented by a hierarchical arrangement of nodes, the method comprising:
training, by a computer system, a model based on clinical data of patients with a given condition to determine outcome categories, wherein the outcome categories serve as a set of prediction targets for the model and correspond to selected nodes of the clinically-defined hierarchical metadata, wherein each node of the clinically-defined hierarchical metadata corresponds to an outcome category and includes one or more outcomes for the outcome category, and the outcome categories of child nodes provide a greater level of detail than outcome categories for corresponding parent nodes;
updating, by the computer system, the set of prediction targets and the model until scores for the set of prediction targets satisfy a threshold value, wherein the updating comprises:
determining, by the computer system, a score for each prediction target in the set of prediction targets based on the trained model, the score being calculated based on a number of correct predictions for the prediction target and existence of a corresponding intervention action for the prediction target;
identifying each prediction target having a score failing to satisfy the threshold value;
merging each node corresponding to an identified prediction target having a corresponding parent node into the corresponding parent node, wherein the corresponding parent node includes the one or more outcomes of the node corresponding to the identified prediction target;
removing each node corresponding to an identified prediction target at a highest level of the clinically-defined hierarchical metadata;
updating the set of prediction targets with the outcome categories of the merged nodes and the outcome categories of nodes corresponding to prediction targets with scores satisfying the threshold value; and
training the model based on the set of prediction targets to produce an updated trained model;
applying, by the computer system, the updated trained model to predict one of the set of prediction targets of outcome categories for a patient with the given condition based on clinical data of the patient; and
controlling, by the computer system, a health device used by the patient based on the predicted one of the set of prediction targets of the outcome categories to prevent the given condition.
2. The method of claim 1 , wherein the clinically-defined hierarchical metadata includes one or more from a categorization of medical diagnoses, procedures, medications, healthcare providers, geographical locations, and medical events.
3. The method of claim 1 , further comprising:
updating a particular prediction target of the set of prediction targets when the score of the particular prediction target is less than a score of a new prediction target at a higher level of the clinically-defined hierarchical metadata including the particular prediction target.
4. A computer system for generating a set of prediction targets from clinically-defined hierarchical metadata represented by a hierarchical arrangement of nodes, the computer system comprising:
at least one processor;
a memory; and
a communication bus connecting the at least one processor and the memory, wherein the at least one processor is configured to perform:
training a model based on clinical data of patients with a given condition to determine outcome categories, wherein the outcome categories serve as a set of prediction targets for the model and correspond to selected nodes of the clinically-defined hierarchical metadata, wherein each node of the clinically-defined hierarchical metadata corresponds to an outcome category and includes one or more outcomes for the outcome category, and the outcome categories of child nodes provide a greater level of detail than outcome categories for corresponding parent nodes;
updating the set of prediction targets and the model until scores for the set of prediction targets satisfy a threshold value, wherein the updating comprises:
determining a score for each prediction target in the set of prediction targets based on the trained model, the score being calculated based on a number of correct predictions for the prediction target and existence of a corresponding intervention action for the prediction target;
identifying each prediction target having a score failing to satisfy the threshold value;
merging each node corresponding to an identified prediction target having a corresponding parent node into the corresponding parent node, wherein the corresponding parent node includes the one or more outcomes of the node corresponding to the identified prediction target;
removing each node corresponding to an identified prediction target at a highest level of the clinically-defined hierarchical metadata;
updating the set of prediction targets with the outcome categories of the merged nodes and the outcome categories of nodes corresponding to prediction targets with scores satisfying the threshold value; and
training the model based on the set of prediction targets to produce an updated trained model;
applying the updated trained model to predict one of the set of prediction targets of outcome categories for a patient with the given condition based on clinical data of the patient; and
controlling a health device used by the patient based on the predicted one of the set of prediction targets of the outcome categories to prevent the given condition.
5. The computer system of claim 4 , wherein the clinically-defined hierarchical metadata includes one or more from a categorization of medical diagnoses, procedures, medications, healthcare providers, geographical locations, and medical events.
6. The computer system of claim 4 , wherein the at least one processor is further configured to perform:
updating a particular prediction target of the set of prediction targets when the score of the particular prediction target is less than a score of a new prediction target at a higher level of the clinically-defined hierarchical metadata including the particular prediction target.
7. A computer program product comprising:
at least one computer readable storage medium having computer readable program code embodied therewith for execution on at least one processor for generating a set of prediction targets from clinically-defined hierarchical metadata represented by a hierarchical arrangement of nodes, the computer readable program code being configured to be executed by the at least one processor to perform:
training a model based on clinical data of patients with a given condition to determine outcome categories, wherein the outcome categories serve as a set of prediction targets for the model and correspond to selected nodes of the clinically-defined hierarchical metadata, wherein each node of the clinically-defined hierarchical metadata corresponds to an outcome category and includes one or more outcomes for the outcome category, and the outcome categories of child nodes provide a greater level of detail than outcome categories for corresponding parent nodes;
updating the set of prediction targets and the model until scores for the set of prediction targets satisfy a threshold value, wherein the updating comprises:
determining a score for each prediction target in the set of prediction targets based on the trained model, the score being calculated based on a number of correct predictions for the prediction target and existence of a corresponding intervention action for the prediction target;
identifying each prediction target having a score failing to satisfy the threshold value;
merging each node corresponding to an identified prediction target having a corresponding parent node into the corresponding parent node, wherein the corresponding parent node includes the one or more outcomes of the node corresponding to the identified prediction target;
removing each node corresponding to an identified prediction target at a highest level of the clinically-defined hierarchical metadata;
updating the set of prediction targets with the outcome categories of the merged nodes and the outcome categories of nodes corresponding to prediction targets with scores satisfying the threshold value; and
training the model based on the and set of prediction targets to produce an updated trained model;
applying the updated trained model to predict one of the set of prediction targets of outcome categories for a patient with the given condition based on clinical data of the patient; and
controlling a health device used by the patient based on the predicted one of the set of prediction targets of the outcome categories to prevent the given condition.
8. The computer program product of claim 7 , wherein the clinically-defined hierarchical metadata includes one or more from a categorization of medical diagnoses, procedures, medications, healthcare providers, geographical locations, and medical events.
9. The computer program product of claim 7 , wherein the computer readable program code is further configured to be executed by the at least one processor to perform:
updating a particular prediction target of the set of prediction targets when the score of the particular prediction target is less than a score of a new prediction target at a higher level of the clinically-defined hierarchical metadata including the particular prediction target.