Systems and methods for providing health care search recommendations
Disclosed embodiments include computer-implemented methods and systems that can efficiently generate highly relevant recommended search terms to clinicians. A hybrid collaborative filtering model recommends search terms for a specific patient to the clinician. The model draws on information from patients' clinical encounters and the searches that were performed during the clinical encounters. To generate recommendations, the model uses search terms which are (1) frequently co-occurring with the ICD codes recorded for the patient and (2) highly relevant to the most recent search terms. One variation of the model uses only the most recent ICD codes assigned to the patient. Another variation uses all ICD codes. Comprehensive experiments of embodiments of the methods and systems have demonstrate high levels of performance.
1. A computer-implemented method for providing search term recommendations in connection with a current encounter for a subject, the method implemented by one or more processors, comprising:
receiving previous search term information associated with search terms searched during one or more previous encounters for the subject;
receiving previous ICD code information associated with one or more previous encounters for the subject;
accessing a co-occurrence information source based on the previous search term information and the previous ICD code information, wherein the co-occurrence information source includes co-occurrence information associated each of a plurality of reference search terms and a plurality of reference ICD codes;
receiving the co-occurrence information from the co-occurrence information source;
generating, for each reference search term in the co-occurrence information source, a search term component score based on the previous search term information and the received co-occurrence information;
generating, for each reference search term in the co-occurrence information source, an encounter component score based on the previous ICD code information and the received co-occurrence information;
generating, for each reference search term in the co-occurrence information source, a recommendation score based on the search term component score and the encounter component score; and
generating a set of one or more recommended search terms based upon the recommendation scores; and
presenting the set of one or more recommended search terms on an interface.
2. The method of claim 1 , wherein generating the set of recommended search terms comprises:
ranking the reference search terms based upon the associated recommendation scores; and
selecting a predetermined number of the reference search terms based upon the rankings.
3. The method of claim 1 , wherein receiving previous search term information includes receiving one or more of (1) information about search terms within a predetermined period of time prior to the current patient encounter, (2) information about search terms within a predetermined number of patient encounters prior to the current patient encounter, or (3) information about search terms during the current patient encounter.
4. The method of claim 1 , wherein receiving previous encounter information includes receiving one or more of (1) information about previous encounters within a predetermined period of time prior to the current patient encounter, (2) information about patient encounters within a predetermined number of patient encounters prior to the current patient encounter, or (3) information about ICD codes during the current patient encounter.
5. The method of claim 1 , wherein receiving previous encounter information includes receiving information about all patient encounters prior to the current patient encounter.
6. The method of claim 1 , wherein:
the method further includes generating weighted previous encounter information based upon all the patient encounters prior to the current patient encounter; and
generating the encounter component score includes generating the encounter component score based upon the weighted previous encounter information.
7. The method of claim 1 , wherein:
the method further comprises receiving information representative of search terms used by a clinician during the patient encounter; and
generating the set of one or more recommended search terms includes excluding the search terms used by the clinician during the current patient encounter.
8. The method of claim 1 , wherein:
generating a search term component score includes generating a null score when the search term information reflects no previously searched terms; and
generating the set of one or more search terms includes generating the set of search terms based solely on the encounter component score.
9. The method of claim 1 wherein accessing the co-occurrence information source includes accessing a co-occurrence information source constructed using representation learning.
10. The method of claim 9 wherein accessing the co-occurrence information source includes accessing a co-occurrence information source constructed using one or both of matrix factorization or an optimization problem.
11. The method of claim 1 , further comprising a A method for generating the co-occurrence information source, comprising:
receiving encounter information from a plurality of electronic health records, wherein the plurality of electronic health records includes electronic health records of a plurality of subjects;
identifying, for each subject, one or more subject encounters in the encounter information;
identifying, for each subject, search terms in the encounter information;
identifying, for each subject, ICD codes in the encounter information;
associating, for each subject, the identified search terms with one of the subject encounters;
associating, for each subject, the identified ICD codes with one of the subject encounters;
determining, for each subject encounter, associated search terms and ICD codes, based upon the identified search terms associated with the subject encounters and the identified ICD codes associated with the subject encounters; and
determining the co-occurrence information, for each of the identified search terms with respect to each of the identified ICD codes, based upon the determined associated search terms and ICD codes, wherein the co-occurrence information defines the co-occurrence information source.
12. The method of claim 11 , wherein determining the co-occurrence information for each of the identified search terms with respect to each of the identified ICD codes includes determining a number of the associated search terms and ICD codes corresponding to the identified search terms and the identified ICD codes.
13. The method of claim 11 , wherein associating the search terms and the subject encounters includes associating search terms and subject encounters based on temporal proximity between the search terms and the subject encounters.
14. The method of claim 11 , further comprising representation learning.
15. The method of claim 14 wherein the representation learning includes one or both of matrix factorization or an optimization problem.