Missing data approaches in medical decision support systems

• US 20060184475A1
• Filed: 02/15/2006
• Published: 08/17/2006
• Est. Priority Date: 02/16/2005
• Status: Active Grant

• Current Assignee: Cerner Innovation, Inc. (Oracle Corporation)
• Original Assignee: Siemens Medical Solutions USA Incorporated (Siemens AG)
• Inventors: Krishnan, Sriram, Rao, R. Bharat
• Granted Patent: US 7,650,321 B2
• Time in Patent Office: Days
• US Class Current: 706/20
• CPC Class Codes: G16H 10/60 for patient-specific data, e.g. for electronic patient records
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

Missing data is addressed in a medical decision support system. The classifier applied to the patient record with missing data is obtained as a function of the available data. For example, one of a plurality of different classifiers is selected based on the features available in the patient record to be classified. The different classifiers are developed using different feature sets. The classifier developed using a feature set closest to or a sub-set of the features available in the patient record is selected for classifying the patient record. As another example, features in a training set corresponding to features available in the patient record are used to build a classifier. The classifier is applied to the patient record by inputting the available features of the patient record.

Claims

1. A method for addressing missing data in a medical decision support system, the method comprising:

identifying a set of M features in a patient record;

assigning, with a processor, a classifier as a function of the M features; and

classifying the patient record with the classifier as a function of the M features.

2. The method of claim 1 wherein identifying comprises identifying each of the M features.

3. The method of claim 1 wherein identifying comprises determining missing features from a set of N possible features, the set of M features being a sub-set of the set of N possible features.

4. The method of claim 1 wherein assigning comprises selecting, with the processor, from a collection of at least two classifiers, each of the at least two classifiers operable with different sets of features.

5. The method of claim 4 wherein selecting, with the processor, comprises selecting the classifier for the set of M features or a sub-set of the M features and not selecting any classifier with at least one feature not in the set of M features.

6. The method of claim 1 wherein assigning comprises:

selecting, with the processor, the set of M features or a sub-set of the M features from training set patient records; and

constructing, with the processor, the classifier from the selected set of M features from the training set patient records.

7. The method of claim 6 wherein constructing is performed separately for different patient records to be analyzed, the patient record being one of the different patient records to be analyzed, and wherein the constructing is performed with the training set for the different patient records to be analyzed, the training set patient records free of the different patient records to be analyzed.

8. The method of claim 6 wherein selecting comprises selecting a sub-set of the training set patient records where at least one of the unselected training set patient records is free of information for one of the M features.

9. The method of claim 1 wherein assigning comprises assigning from different types of classifiers.

10. The method of claim 1 further comprising:

determining, with the processor, an estimate of performance of the classifier; and

outputting the estimate.

11. In a computer readable storage media having stored therein data representing instructions executable by a programmed processor for addressing missing data in a medical decision support system, the storage media comprising instructions for:

determining a list of features available from data for a current patient;

obtaining a classifier as a function of the list; and

applying the classifier to the data.

12. The instructions of claim 11 wherein obtaining comprises selecting from a plurality of prior developed classifiers, each of the prior developed classifiers operable with different lists of features.

13. The instructions of claim 11 wherein obtaining comprises constructing the classifier from training data values for the features or a sub-set of features in the list, the training data being previously collected.

14. The instructions of claim 11 further comprising:

outputting an estimate of performance of the classifier.

15. A system for addressing missing data in a medical decision support, the system comprising:

a memory operable to store variables available for a patient record;

a processor operable to determine a classifier as a function of the variables available for the patient record and operable to determine a diagnosis with the classifier and the variables.

16. The system of claim 15 wherein the memory is operable to store a plurality of classifiers, each classifier corresponding to different sets of variables, and wherein the processor is operable to determine the classifier by matching the variables available for the patient record to one of the plurality of classifiers.

17. The system of claim 15 wherein the memory is operable to store training data, and wherein the processor is operable to build the classifier from a sub-set of the training data, the sub-set of the training data being training variables corresponding to the variables available for the patient record.

18. In a computer readable storage media having stored therein data representing instructions executable by a programmed processor for addressing missing data in a medical decision support system, the storage media comprising instructions for:

obtaining a patient record for medical decision support analysis;

determining that information for at least one feature is missing from a patient record;

selecting a classifier based on the missing information; and

analyzing the patient record with the classifier.

19. The instructions of claim 18 wherein the classifier is a previously trained classifier;

further comprising repeating obtaining, determining, selecting and analyzing for a different patient record.

20. In a computer readable storage media having stored therein data representing instructions executable by a programmed processor for addressing missing data in a medical decision support system, the storage media comprising instructions for:

obtaining a patient record for medical decision support analysis;

determining that information for at least one feature is missing from a patient record;

training a classifier from a sub-set of training data, the sub-set free of information from the training data corresponding to the missing feature; and

analyzing the patient record with the classifier.

21. The instructions of claim 20 further comprising repeating obtaining, determining, training and analyzing for a different patient record with the same training data.