Context-aware prediction in medical systems

• US 9,754,081 B2
• Filed: 05/13/2014
• Issued: 09/05/2017
• Est. Priority Date: 05/14/2013
• Status: Active Grant

• Current Assignee: Regents of the University of California (University of California)
• Original Assignee: Regents of the University of California (University of California)
• Inventors: Ghasemzadeh, Hassan, Suh, Myung-Kyung, Lan, Mars, Sarrafzadeh, Majid, Alshurafa, Nabil
• Primary Examiner(s): Misir, Dave
• Application Number:
  US14/276,660
• Publication Number:
  US 20140344208A1
• Time in Patent Office: 1,211 Days
• Field of Search: 706 52
• 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

A method includes receiving contextual data related to at least one of environmental, physiological, behavioral, and historical context, and receiving outcome data related to at least one outcome. The method further includes creating a feature set from the contextual data, selecting a subset of features from the feature set, assigning a score to each feature in the subset of features according to the probability that the feature is a predictor of the at least one outcome, and generating a characteristic curve for the at least one outcome from the subset of features, the characteristic curve being based on the scoring. The method further includes calculating the area under the characteristic curve, and using, the area under the characteristic curve, identifying whether the subset of features is a suitable predictor for the at least one outcome.

Claims

1. A method of context-aware prediction of medical conditions, comprising:

by a first computing device executing instructions from a non-transitory computer-readable medium for;

receiving contextual data related to at least one of environmental, physiological, behavioral, and historical context;

receiving outcome data related to at least one outcome;

creating a feature set from the contextual data;

selecting a subset of features from the feature set;

assigning a score to each feature in the subset of features according to a probability that the feature is a predictor of the at least one outcome;

generating a characteristic curve for the at least one outcome from the subset of features, the characteristic curve based on the scores of the features in the subset of features;

calculating an area under the characteristic curve;

identifying, using the area under the characteristic curve, whether the subset of features is a suitable predictor for the at least one outcome;

responsive to identifying that the subset of features is a suitable predictor for the at least one outcome, generating a prediction model relating the subset of features to the at least one outcome; and

providing to a display device for display a prediction of a likelihood of occurrence of the at least one outcome by applying the prediction model to data received from a monitoring device,

wherein generating the characteristic curve includes iteratively;

setting a probability threshold;

selecting a feature group from the subset of features, wherein, for each feature in the feature group, the assigned score is greater than the probability threshold;

determining for the contextual data of features in the feature group a true positive rate and a false positive rate of prediction of the outcome; and

plotting the true positive rate and the false positive rate for the probability threshold.

2. The method of claim 1, wherein the characteristic curve is one of a plurality of characteristic curves, further comprising comparing the area under the characteristic curve to areas under other characteristic curves in the plurality of characteristic curves and identifying a preferred predictor based on the comparing.

3. The method of claim 1, wherein the characteristic curve is one of a plurality of characteristic curves, further comprising comparing the area under the characteristic curve to areas under other characteristic curves in the plurality of characteristic curves and identifying a preferred scoring technique based on the comparing.

4. The method of claim 1, further comprising:

receiving additional contextual data; and

predicting an outcome based on the additional contextual data using the prediction model.

5. The method of claim 4, wherein the additional contextual data includes data regarding a group of individuals, and the predicted outcome is a prediction of the outcome for the group of individuals.

6. The method of claim 1, wherein the contextual data includes data from a plurality of sources.

7. The method of claim 1, wherein the contextual data includes data relating to a plurality of individuals.

8. The method of claim 1, wherein the contextual data includes data taken at a plurality of times.

9. The method of claim 1, wherein the contextual data comprises a weight, a blood pressure, a heart rate, and a symptom response, and wherein the outcome data comprises a use of a monitoring device;

further comprising determining whether a subject is using the monitoring device by applying the prediction model to the contextual data.

10. The method of claim 1, wherein the contextual data comprises a body mass index, a waist circumference, a high density lipoprotein, a low density lipoprotein, and a questionnaire associated with a subject;

and wherein the outcome data comprises an effectiveness of a health-improvement program, a body mass index loss, a waist circumference loss, a high density lipoprotein increase, or a low density lipoprotein decrease of the subject.

11. The method of claim 1, wherein the outcome data comprises a medical condition of a subject, the medical condition including a chronic health failure or diabetes.

12. The method of claim 1, wherein receiving the contextual data further comprises receiving the contextual data at a predefined interval.

13. A system for context-aware prediction of medical conditions, comprising:

a first computing device comprising a memory including processor-executable instructions and a processor configured to execute instructions from the memory;

wherein the instructions include instructions for the processor to;

receive contextual data and outcome data;

create a feature set from the contextual data;

select a plurality of feature subsets from the feature set;

apply each of the plurality of feature subsets and the outcome data to a classifier, and determine a score for each of the plurality of feature subsets;

select a preferred feature subset based on the score for each of the plurality of feature subsets; and

generate a prediction model of an outcome using the preferred feature subset; and

an interface providing access to the prediction model by a second computing device;

wherein the prediction model is configured to input subject data received from a monitoring device, make a prediction based on the received subject data regarding a likelihood of the outcome occurring, and output the prediction to the second computing device,

wherein the outcome data represents at least one outcome, and wherein the instructions to determine the score for each of the plurality of feature subsets includes instructions for the processor to, for each feature subset;

use the classifier to determine a probability measure for each feature in the feature subset, the probability measure being an indication of how predictive the feature is for an outcome of the at least one outcome;

create a characteristic curve of true positive rate versus false positive rate, wherein each point of the characteristic curve represents a portion of the feature set, and each portion of the feature set is selected based on a probability measure threshold; and

calculate the score as an area under the characteristic curve.

14. The system of claim 13, wherein the instructions further include instructions for the processor to cluster the contextual data, and create the feature set from at least a portion of the clustered contextual data.

15. The system of claim 14, wherein the clustered contextual data includes multiple data clusters, and wherein the instructions to create the feature set include instructions to create a plurality of features sets from respective ones of the multiple data clusters.

16. The system of claim 13, wherein the outcome data includes data representing a plurality of different outcomes.

17. The system of claim 13, wherein the instructions further include instructions for the processor to receive additional contextual data, and predict an outcome from the additional contextual data using the prediction model.