APPLICATION OF BAYESIAN NETWORKS TO PATIENT SCREENING AND TREATMENT

• US 20110082712A1
• Filed: 09/30/2010
• Published: 04/07/2011
• Est. Priority Date: 10/01/2009
• Status: Active Grant

• Current Assignee: Decisionq Corporation
• Original Assignee: Decisionq Corporation
• Inventors: Radano, Todd A., Kalina, Philip M., Eberhardt, John S. III
• Granted Patent: US 11,562,323 B2
• Time in Patent Office: Days
• US Class Current: 705/4
• CPC Class Codes: G06Q 10/10 Office automation; Time management
• CPC Class Codes: G06Q 40/08 Insurance
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: G16H 50/30 for calculating health indices; for individual health risk assessment
• CPC Class Codes: G16H 50/50 for simulation or modelling of medical disorders
• CPC Class Codes: G16Z 99/00 Subject matter not provided for in other main groups of this subclass

Abstract

According to one aspect of the invention, health insurance claim data for a first group of individuals is obtained to generate a training corpus, including a training set of claim data and a holdout set of claim data. The first group of individuals represents enrollees of one or more first health insurance plans and the health insurance claim data represents historic insurance claim information for each individual in the first group. A Bayesian belief network (BBN) model is created by training a BBN network based on the training set of claim data using predetermined machine learning algorithms. The BBN model is validated using the holdout set of claim data. The BBN model, when having been successfully validated, is configured to identify at least one of individuals with risk for a disorder and individuals with risk who are most likely to benefit from intervention and treatment for the disorder.

Claims

1. A computer-implemented method for evaluating enrollees of a health insurance plan, the method comprising:

obtaining health insurance claim data for a first group of individuals to generate a training corpus, including a training set of claim data and a holdout set of claim data, the first group of individuals representing enrollees of one or more first health insurance plans and the health insurance claim data representing historic insurance claim information for each individual in the first group;

creating a Bayesian belief network (BBN) model by training a BBN network based on the training set of claim data using a predetermined machine learning algorithm; and

validating the BBN model using the holdout set of claim data, wherein the BBN model, when having been successfully validated, is configured to identify at least one of individuals with risk for a disorder and individuals with risk who are most likely to benefit from intervention and treatment for the disorder; and

using the validated BBN model to develop enrollee-specific estimates of disease risk, enrollee-specific future estimates of utilization and cost, and enrollee-specific estimates of the change which would result from successful intervention and/or treatment.

2. The method of claim 1, wherein the BBN model comprises at least one screening model to identify a subset of the individuals in the first group with potential risk characteristics of the disorder and at least one cost model to calculate a cost estimate to produce an enrollee-specific cost estimate.

3. The method of claim 2, wherein creating a BBN model comprises:

building a list of a plurality of BBN model candidates based on the training set of claim data according to a plurality of categories using the predetermined machine learning algorithm;

scoring each of the BBN model candidates in the list using a predetermined scoring method; and

selecting the BBN model from the list of BBN model candidates, the selected BBN model having the highest score among the BBN model candidates to be validated.

4. The method of claim 3, wherein the predetermined scoring method comprises a minimum description length (MDL) or Bayesian Information Criterion (BIC) compatible method.

5. The method of claim 3, wherein building a list of BBN model candidates comprises:

calculating distributions of discrete states for variables of the BBN network;

performing a preliminary modeling operation on the variables to identify appropriate machine learning parameters, optionally generating a first BBN model;

performing a global modeling operation to set the appropriate machine learning parameters and to observe global data structures, optionally generating a second BBN model;

performing a naï

ve modeling operation to identify a contribution of each of the variables, optionally generating a third BBN model;

performing a focused modeling operation on a subset of the variables identified in at least one of the preliminary modeling operation, the global modeling operation, and the naï

ve model operation, optionally generating a fourth BBN model;

using k-fold cross-validation to assist in attribute selection; and

scoring, using the predetermined scoring method, all BBN models to select the best BBN model.

6. The method of claim 2, wherein validating the BBN model comprises:

generating a receiver operating characteristic (ROC) curve based on an outcome of the BBN model operated on the holdout set of claim data; and

calculating an area under the curve (AUC) based on the ROC curve, wherein the AUC is used to evaluate predictive correctness of the BBN model given the holdout set of claim data.

7. The method of claim 2, further comprising:

receiving a second set of claim data from a client, the second set of claim data being associated with a second group of individuals representing enrollees of one or more second health insurance plans;

performing a screening operation using at least one screening BBN model based on the second set of claim data to identify a second subset of individuals in the second group having risk characteristics associated with the disorder;

performing a cost estimation on the second subset of individuals using at least one cost BBN model to produce enrollee specific cost estimates; and

transmitting the enrollee-specific disease risk and cost estimates to the client.

8. The method of claim 7, further comprising retraining at least one screening BBN model and at least one cost BBN model based on the second set of claim data and the enrollee specific cost estimates for future usages.

9. A computer-readable storage medium having computer instructions stored therein, which when executed by a computer, cause the computer to perform a method for evaluating enrollees of a health insurance plan, the method comprising:

obtaining health insurance claim data for a first group of individuals to generate a training corpus, including a training set of claim data and a holdout set of claim data, the first group of individuals representing enrollees of one or more first health insurance plans and the health insurance claim data representing historic insurance claim information for each individual in the first group;

creating a Bayesian belief network (BBN) model by training a BBN network based on the training set of claim data using a predetermined machine learning algorithms; and

validating the BBN model using the holdout set of claim data, wherein the BBN model, when having been successfully validated, is configured to identify at least one of individuals with risk for a disorder and individuals with risk who are most likely to benefit from intervention and treatment for the disorder.

10. The computer-readable storage medium of claim 9, wherein the BBN model comprises at least one screening model to identify a subset of the individuals in the first group with potential risk characteristics of the disorder and at least one cost model to predict a cost estimate to produce patient specific cost estimate.

11. The computer-readable storage medium of claim 10, wherein creating a BBN model comprises:

building a list of a plurality of BBN model candidates based on the training set of claim data according to a plurality of categories using the predetermined machine learning algorithm;

scoring each of the BBN model candidates in the list using a predetermined scoring method; and

selecting the BBN model from the list of BBN model candidates, the selected BBN model having the highest score among the BBN model candidates to be validated.

12. The computer-readable storage medium of claim 11, wherein the predetermined scoring method comprises a minimum description length (MDL) compatible method.

13. The computer-readable storage medium of claim 11, wherein building a list of BBN model candidates comprises:

calculating distributions of discrete states for variables of the BBN network;

performing a preliminary modeling operation on the variables to identify appropriate machine learning parameters, optionally generating a first BBN model;

performing a global modeling operation to set the appropriate machine learning parameters and to observe global data structures, optionally generating a second BBN model;

performing a naï

ve modeling operation to identify a contribution of each of the variables, optionally generating a third BBN model;

performing a focused modeling operation on a subset of the variables identified in at least one of the preliminary modeling operation, the global modeling operation, and the naï

ve model operation, optionally generating a fourth BBN model;

using k-fold cross-validation to assist in attribute selection; and

scoring, using the predetermined scoring method, all BBN models to select the best BBN model.

14. The computer-readable storage medium of claim 10, wherein validating the BBN model comprises:

generating a receiver operating characteristic (ROC) curve based on an outcome of the BBN model operated on the holdout set of claim data; and

calculating an area under the curve (AUC) based on the ROC curve, wherein the AUC is used to evaluate predictive correctness of the BBN model given the holdout set of claim data.

15. The computer-readable storage medium of claim 10, wherein the method further comprises:

receiving a second set of claim data from a client, the second set of claim data being associated with a second group of individuals representing enrollees of one or more second health insurance plans;

performing a screening operation using the at least one screening BBN model based on the second set of claim data to identify a second subset of individuals in the second group having risk characteristics associated with the disorder;

performing a cost estimation on the second subset of individuals using the at least one cost BBN model to produce enrollee specific cost estimates; and

transmitting the enrollee specific cost estimates to the client.

16. The computer-readable storage medium of claim 15, wherein the method further comprises retraining the at least one screening BBN model and the at least one cost BBN model based on the second set of claim data and the enrollee specific cost estimates for future usages.

17. A data processing system, comprising:

a processor; and

a memory coupled to the processor to store instructions, which when executed from the memory, cause the processor to

obtain health insurance claim data for a first group of individuals to generate a training corpus, including a training set of claim data and a holdout set of claim data, the first group of individuals representing enrollees of one or more first health insurance plans and the health insurance claim data representing historic insurance claim information for each individual in the first group,

create a Bayesian belief network (BBN) model by training a BBN network based on the training set of claim data using a predetermined machine learning algorithms, and

validate the BBN model using the holdout set of claim data, wherein the BBN model, when having been successfully validated, is configured to identify at least one of individuals with risk for a disorder and individuals with risk who are most likely to benefit from intervention and treatment for the disorder.

18. A computer-implemented method for evaluating enrollees of a health insurance plan, the method comprising:

receiving a first set of claim data from a client, the first set of claim data being associated with a first group of individuals representing enrollees of one or more first health insurance plans; and

performing a screening operation using at least one screening Bayesian belief network (BBN) model based on the first set of claim data to identify a subset of individuals in the first group having risk characteristics associated with a disorder.

19. The method of claim 18, further comprising:

performing a cost estimation on the subset of individuals using at least one cost BBN model to produce enrollee specific cost estimates, wherein the at least one screening BBN model and the at least one cost BBN model were trained using a predetermined machine learning algorithm based on a second set of claim data associated with a second group of individuals of one or more second health insurance plans, and wherein the second set of claim data representing historic insurance claim information for each individual in the second group; and

transmitting the enrollee specific cost estimates to the client.

20. The method of claim 19, further comprising retraining the at least one screening BBN model and the at least one cost BBN model based on the first set of claim data and the enrollee specific cost estimates for future usages.

21. The method of claim 19, further comprising training the at least one screening BBN model and the at least one cost BBN model prior to receiving the first set of claim data, including

partitioning the second set of claim data into a training set and a holdout set;

building a list of a plurality of BBN model candidates based on the training set of claim data according to a plurality of categories using the predetermined machine learning algorithm;

scoring each of the BBN model candidates in the list using a predetermined scoring method; and

selecting a BBN model as a final candidate from the list of BBN model candidates, the selected BBN model having the highest score among the BBN model candidates.

22. The method of claim 21, wherein the predetermined scoring method comprises a minimum description length (MDL) or Bayesian Information Criterion (BIC) compatible method.

23. The method of claim 21, wherein building a list of BBN model candidates comprises:

calculating distributions of discrete states for variables of an initial BBN network;

performing a preliminary modeling operation on the variables to identify appropriate machine learning parameters, optionally generating a first BBN model;

performing a global modeling operation to set the appropriate machine learning parameters and to observe global data structures, optionally generating a second BBN model;

performing a naï

ve modeling operation to identify a contribution of each of the variables, optionally generating a third BBN model;

performing a focused modeling operation on a subset of the variables identified in at least one of the preliminary modeling operation, the global modeling operation, and the naï

ve model operation, optionally generating a fourth BBN model;

using k-fold cross-validation to assist in attribute selection; and

scoring, using the predetermined scoring method, all BBN models to select the best BBN model.

24. The method of claim 21, further comprising validating the selected BBN model, including

generating a receiver operating characteristic (ROC) curve based on an outcome of the selected BBN model operated on the holdout set of claim data;

and

calculating an area under the curve (AUC) based on the ROC curve, wherein the AUC is used to evaluate predictive correctness of the selected BBN model given the holdout set of claim data.

25. A computer-readable medium having computer instructions stored therein, which when executed by a computer, cause the computer to perform a method for evaluating enrollees of a health insurance plan, the method comprising:

receiving a first set of claim data from a client, the first set of claim data being associated with a first group of individuals representing enrollees of one or more first health insurance plans; and

performing a screening operation using at least one screening Bayesian belief network (BBN) model based on the first set of claim data to identify a subset of individuals in the first group having risk characteristics associated with a disorder.

26. The computer-readable medium of claim 25, wherein the method further comprises:

performing a cost estimation on the subset of individuals using at least one cost BBN model to produce enrollee specific cost estimates, wherein the at least one screening BBN model and the at least one cost BBN model were trained using a predetermined machine learning algorithm based on a second set of claim data associated with a second group of individuals of one or more second health insurance plans, and wherein the second set of claim data representing historic insurance claim information for each individual in the second group; and

transmitting the enrollee specific cost estimates to the client.

27. The computer-readable medium of claim 26, wherein the method further comprises retraining at least one screening BBN model and at least one cost BBN model based on the first set of claim data and the enrollee specific cost estimates for future usages.

28. The computer-readable medium of claim 26, wherein the method further comprises training at least one screening BBN model and at least one cost BBN model prior to receiving the first set of claim data, including

partitioning the second set of claim data into a training set and a holdout set;

building a list of a plurality of BBN model candidates based on the training set of claim data according to a plurality of categories using the predetermined machine learning algorithm;

scoring each of the BBN model candidates in the list using a predetermined scoring method; and

selecting a BBN model as a final candidate from the list of BBN model candidates, the selected BBN model having the highest score among the BBN model candidates.

29. The computer-readable medium of claim 28, wherein the predetermined scoring method comprises a minimum description length (MDL) or Bayesian Information Criterion (BIC) compatible method.

30. The computer-readable medium of claim 28, wherein building a list of BBN model candidates comprises:

calculating distributions of discrete states for variables of an initial BBN network;

performing a preliminary modeling operation on the variables to identify appropriate machine learning parameters, optionally generating a first BBN model;

performing a global modeling operation to set the appropriate machine learning parameters and to observe global data structures, optionally generating a second BBN model;

performing a naï

ve modeling operation to identify a contribution of each of the variables, optionally generating a third BBN model;

performing a focused modeling operation on a subset of the variables identified in at least one of the preliminary modeling operation, the global modeling operation, and the naï

ve model operation, optionally generating a fourth BBN model;

using k-fold cross-validation to assist in attribute selection; and

scoring, using the predetermined scoring method, all BBN models to select the best BBN model.

31. The computer-readable medium of claim 28, wherein the method further comprises validating the selected BBN model, including

generating a receiver operating characteristic (ROC) curve based on an outcome of the selected BBN model operated on the holdout set of claim data;

and

calculating an area under the curve (AUC) based on the ROC curve, wherein the AUC is used to evaluate predictive correctness of the selected BBN model given the holdout set of claim data.