Electronic clinical decision support device based on hospital demographics

Patent metadata

Field	Value
Publication number	US-2019164648-A1
Application number	US-201716323802-A
Country	US
Kind code	A1
Filing date	Aug 1, 2017
Priority date	Aug 8, 2016
Publication date	May 30, 2019
Grant date	—

Abstract

An electronic clinical decision support (CDS) device (10) employs a trained CDS algorithm (30) that operates on values of a set of covariates to output a prediction of a medical condition. The CDS algorithm was trained on a training data set (22). The CDS device includes a computer (12) that is programmed to provide a user interface (62) for completing clinical survey questions using the display and the one or more user input devices. Marginal probability distributions (42) for the covariates of the set of covariates are generated from the completed clinical survey questions. The trained CDS algorithm is adjusted for covariate shift using the marginal probability distributions. A prediction of the medical condition is generated for a medical subject using the trained CDS algorithm adjusted for covariate shift (50) operating on values for the medical subject of the covariates of the set of covariates.

First claim

1 . An electronic clinical decision support (CDS) device employing a trained CDS algorithm that operates on values of covariates of a set of covariates to output a prediction of a medical condition, the trained CDS algorithm having been trained on a training data set of training samples, the CDS device comprising:
a computer including a display and one or more user input devices, the computer programmed to:
adjust the trained CDS algorithm for covariate shift by computing covariate shift adjustment weights for the training samples of the training data set using marginal probability distributions for the covariates of the set of covariates and performing update training on the training data set with the training samples weighted by the covariate shift adjustment weights;
generate a prediction of the medical condition for a medical subject by applying the trained CDS algorithm adjusted for covariate shift to values for the medical subject of the covariates of the set of covariates; and
display the generated prediction of the medical condition for the medical subject on the display.

2 . The electronic CDS device of claim 1 wherein the computer is further programmed to:
provide a user interface for completing clinical survey questions using the display and the one or more user input devices; and
generate the marginal probability distributions for the covariates of the set of covariates from the completed clinical survey questions.

3 . The electronic CDS device of claim 1 wherein:
the trained CDS algorithm adjusted for covariate shift comprises the trained CDS algorithm and a covariate shift predictor that operates on the prediction of the medical condition output by the trained CDS algorithm and the values of covariates on which the trained CDS algorithm operated to output the prediction; and
performing update training comprises training the covariate shift predictor on the training data set with the training samples weighted by the covariate shift adjustment weights wherein the training of the covariate shift predictor does not modify the trained CDS algorithm.

4 . The electronic CDS device of claim 3 wherein the covariate shift predictor comprises a logistic regression predictor.

5 . The electronic CDS device of claim 1 wherein the adjustment of the trained CDS algorithm for covariate shift does not use any training samples other than the training data set of training samples.

6 . The electronic CDS device of claim 1 wherein computing covariate shift adjustment weights for the training samples of the training data set using the marginal probability distributions includes:
optimizing joint probability distributions over the set of covariates with respect to the training data set constrained by the marginal probability distributions for the covariates of the set of covariates; and
computing the covariate shift adjustment weights for the training samples of the training data set from the optimized joint probability distributions.

7 . The electronic CDS device of claim 6 wherein optimizing the joint probability distributions includes maximizing the effective sample size of the training data set.

8 . (canceled)

9 . (canceled)

10 . (canceled)

11 . (canceled)

12 . (canceled)

13 . The electronic CDS device of claim 1 wherein the trained CDS algorithm is adjusted for covariate shift without using any training samples other than the training data set of training samples.

14 . An electronic clinical decision support (CDS) method employing a CDS algorithm that operates on values of covariates of a set of covariates to output a prediction of a medical condition, the CDS method comprising:
training the CDS algorithm on a training data set of training samples using a first computer; after the training, performing CDS operations using a second computer different from the first computer, the CDS operations including:
adjusting the trained CDS algorithm for covariate shift using marginal probability distributions for the covariates of the set of covariates;
generating a prediction of the medical condition for a medical subject by applying the trained CDS algorithm adjusted for covariate shift to values for the medical subject of the covariates of the set of covariates; and
displaying the generated prediction of the medical condition for the medical subject on a display.

15 . The electronic CDS method of claim 14 wherein the adjusting of the trained CDS algorithm for covariate shift includes:
computing covariate shift adjustment weights for the training samples of the training data set using the marginal probability distributions; and
performing update training on the training data set with the training samples weighted by the covariate shift adjustment weights.

16 . The electronic CDS method of claim 15 wherein computing covariate shift adjustment weights for the training samples of the training data set includes:
optimizing joint probability distributions over the set of covariates with respect to the training data set constrained by the marginal probability distributions for the covariates of the set of covariates; and
computing the covariate shift adjustment weights for the training samples of the training data set from the optimized joint probability distributions.

17 . The electronic CDS method of claim 16 wherein optimizing the joint probability distributions includes maximizing the effective sample size of the training data set.

18 . The electronic CDS method of claim 15 wherein:
the trained CDS algorithm adjusted for covariate shift comprises the trained CDS algorithm and a covariate shift predictor that operates on the prediction of the medical condition output by the trained CDS algorithm and the values of covariates on which the trained CDS algorithm operated to output the prediction; and
performing update training comprises training the covariate shift predictor on the training data set with the training samples weighted by the covariate shift adjustment weights wherein the training of the covariate shift predictor does not modify the trained CDS algorithm.

19 . The electronic CDS method of claim 14 wherein the trained CDS algorithm is adjusted for covariate shift without using any training samples other than the training data set of training samples.

20 . The electronic CDS method of claim 14 wherein the training includes selecting the covariates of the set of covariates using the training data set.

Assignees

• Koninklijke Philips Nv

Inventors

• Conroy Bryan
• Potes Blandon Cristhian Mauricio
• Xu Minnan

Classifications

• G06N7/01 Primary

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G16H10/60

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

• G16H50/70

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

• G06N20/00

  Machine learning · CPC title

• G16H50/20 Primary

  for computer-aided diagnosis, e.g. based on medical expert systems · CPC title