Generating personalized treatment options using precision cohorts and data driven models

Patent metadata

Field	Value
Publication number	US-11610688-B2
Application number	US-201815967635-A
Country	US
Kind code	B2
Filing date	May 1, 2018
Priority date	May 1, 2018
Publication date	Mar 21, 2023
Grant date	Mar 21, 2023

Abstract

A plurality of attributes are extracted from a plurality of electronic health records, where each electronic health record is associated with a patient in a plurality of patients. Additionally, a training data set and a scoring data set are generated based on the plurality of attributes, and a patient similarity model is trained based on the training data set. A precision cohort is identified, where the precision cohort includes patients in the plurality of patients from the scoring data set that are similar to a first patient based on an electronic health record of the first patient and the similarity model. At least one result statistic for each of a plurality of treatments given to patients in the precision cohort is determined, and a first treatment of the plurality of treatments is selected for the first patient based at least in part on the determined result statistics.

First claim

What is claimed is:

1. A method, comprising:
extracting a plurality of attributes from a plurality of electronic health records, wherein each electronic health record is associated with a patient in a plurality of patients, comprising:
identifying a first decision point in a first electronic health record of the plurality of electronic health records, wherein the first decision point corresponds to a patient attending an appointment with a healthcare provider; and
extracting a treatment selected by the healthcare provider at the first decision point, wherein the selected treatment comprises a decision to continue a current treatment plan;

generating a training data set based on the plurality of attributes without reference to an index patient;
generating a scoring data set based on the plurality of attributes, wherein:
the scoring data set and the training data set differs by at least one attribute;
the scoring data set is updated upon new information becoming available, wherein the new information corresponds to a new patient; and
the training data set is updated only periodically, such that the scoring data set is updated more frequently than the training data set, wherein the training data set includes:
a first row with information corresponding to a first healthcare encounter by a first patient,
a second row with information corresponding to a second healthcare encounter by the first patient, and
a third row with information corresponding to a first healthcare encounter by a second patient, and
wherein the scoring data set includes at least a first variable that is excluded from the training data set;

training a patient similarity model based on the training data set using locally supervised metric learning (LSML) to learn a weight matrix that aligns similar patients more closely together, as compared to disparate patients, in a multidimensional vector space, wherein:
the patient similarity model is specific to a first disorder, of a plurality of disorders, and
the patient similarity model is updated periodically alongside the training data set;

identifying a precision cohort of patients in the plurality of patients in the scoring data set that are similar to the index patient based on an electronic health record of the index patient and the similarity model, comprising:
prior to generating similarity scores, identifying a subset of the plurality of patients by filtering the plurality of patients based on one or more guideline variables and without use of the similarity model, wherein at least one of the one or more guideline variables specifies a numerical value for a biometric reading, and wherein a value of the biometric reading for each of the patients in the subset of the plurality of patients matches a value of the biometric reading for the index patient; and
subsequent to identifying the subset of the plurality of patients, generating similarity scores for the patients in the subset of the plurality of patients using the similarity model;

determining at least one result statistic for each of a plurality of treatments given to patients in the precision cohort; and
selecting a first treatment of the plurality of treatments for the first disorder and the index patient based at least in part on the determined result statistics, comprising displaying the first variable to a healthcare provider, wherein the first variable is not used to affect selection, scoring, or filtering of the precision cohort.

2. The method of claim 1 , wherein extracting the plurality of attributes comprises:
identifying a plurality of decision points in each of the plurality of electronic health records;
extracting one or more treatments selected at least one identified decision point;
extracting one or more guideline variables associated with the one or more treatments; and
extracting one or more results variables associated with at least one identified decision point.

3. The method of claim 2 , wherein generating the training data set comprises identifying and selecting one or more salient variables from the plurality of attributes, the one or more guideline variables, and the one or more results variables using a feature selection model.

4. The method of claim 1 , wherein training the patient similarity model comprises processing the training data set using one or more metric learning methods.

5. The method of claim 1 , wherein training the patient similarity model further comprises processing the training data set using one or more clustering methods.

6. The method of claim 2 , wherein identifying the precision cohort further comprises:
identifying a first set of attributes that correspond to the one or more guideline variables and are exhibited by the index patient, based on the electronic health record of the index patient;
identifying a group of patients in the plurality of patients with attributes matching the first set of attributes;
generating a similarity score for each patient in the identified group of patients by processing each of the identified group of patients with the patient similarity model; and
selecting one or more of the identified group of patients for inclusion in the precision cohort based on the generated similarity scores.

7. The method of claim 2 , wherein determining at least one result statistic for each of a plurality of treatments comprises:
generating one or more clusters of patients in the precision cohort based on a respective treatment in the plurality of treatments that each patient received;
computing the at least one result statistic for each of the plurality of treatments based on the extracted one or more results variables for each patient in each of the one or more clusters.

8. A computer program product comprising a computer-readable storage medium having computer-readable program code embodied therewith, the computer-readable program code executable by one or more computer processors to perform an operation comprising:
extracting a plurality of attributes from a plurality of electronic health records, wherein each electronic health record is associated with a patient in a plurality of patients, comprising:
identifying a first decision point in a first electronic health record of the plurality of electronic health records, wherein the first decision point corresponds to a patient attending an appointment with a healthcare provider; and
extracting a treatment selected by the healthcare provider at the first decision point, wherein the selected treatment comprises a decision to continue a current treatment plan;

generating a training data set based on the plurality of attributes without reference to an index patient;
generating a scoring data set based on the plurality of attributes, wherein:
the scoring data set and the training data set differs by at least one attribute;
the scoring data set is updated upon new information becoming available, wherein the new information corresponds to a new patient; and
the training data set is updated only periodically, such that the scoring data set is updated more frequently than the training data set, wherein the training data set includes:
a first row with information corresponding to a first healthcare encounter by a first patient,
a second row with information corresponding to a second healthcare encounter by the first patient, and
a third row with information corresponding to a first healthcare encounter by a second patient, and
wherein the scoring data set includes at least a first variable that is excluded from the training data set;

training a patient similarity model based on the training data set using locally supervised metric learning (LSML) to learn a weight matrix that aligns similar patients more closely together, as com

Assignees

• IBM
• Merative Us L P

Inventors

• Ng Kenney
• Kartoun Uri
• Tang Paul C
• Stavropoulos Charalambos
• Park Yoonyoung
• Chiu Amy
• Das Amarendra

Classifications

• G16H20/00

  ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance · CPC title

• G16H10/60 Primary

  for patient-specific data, e.g. for electronic patient records · CPC title

• G16H50/50

  for simulation or modelling of medical disorders · CPC title

• G16H50/70 Primary

  for mining of medical data, e.g. analysing previous cases of other patients · CPC title