System and method for predicting and summarizing medical events from electronic health records

Patent metadata

Field	Value
Publication number	US-11410756-B2
Application number	US-201715690714-A
Country	US
Kind code	B2
Filing date	Aug 30, 2017
Priority date	Jul 28, 2017
Publication date	Aug 9, 2022
Grant date	Aug 9, 2022

Abstract

A system for predicting and summarizing medical events from electronic health records includes a computer memory storing aggregated electronic health records from a multitude of patients of diverse age, health conditions, and demographics including medications, laboratory values, diagnoses, vital signs, and medical notes. The aggregated electronic health records are converted into a single standardized data structure format and ordered arrangement per patient, e.g., into a chronological order. A computer (or computer system) executes one or more deep learning models trained on the aggregated health records to predict one or more future clinical events and summarize pertinent past medical events related to the predicted events on an input electronic health record of a patient having the standardized data structure format and ordered into a chronological order. An electronic device configured with a healthcare provider-facing interface displays the predicted one or more future clinical events and the pertinent past medical events of the patient.

First claim

We claim:

1. A system, comprising in combination:
a) computer memory storing aggregated electronic health records from a multitude of patients of diverse age, health conditions, and demographics including as elements thereof at least medications, laboratory values, diagnoses, vital signs, and free text medical notes, and obtained in different formats, wherein the aggregated electronic health records are converted into a single standardized data structure format and ordered per patient into an ordered arrangement; and
b) a computer executing one or more deep learning models trained on the aggregated health records converted into the single standardized data structure format and in the ordered arrangement to predict one or more future clinical events based on an input electronic health record of a patient, and wherein a particular one of the one or more deep learning models includes an attention mechanism indicating how much attention the particular model gave to elements in the input electronic health record to predict the one or more future clinical events, wherein the particular model generates an output comprising both the predicted one or more future clinical events and the result of the attention mechanism, and wherein the computer executing the particular model of the one or more deep learning models comprises (1) generating, for a plurality of the elements of the input electronic health record, a respective data embedding vector and a respective time embedding vector, wherein generating the time embedding vector for a particular element of the input electronic health record comprises determining, based on a timing of the particular element, a plurality of single-valued functions of the timing of the particular element, (2) concatenating the data embedding vectors in an order to generate a data embedding matrix and concatenating the time embedding vectors in the same order to generate a time embedding matrix, (3) combining the time embedding matrix with a product of a projection matrix and the data embedding matrix to generate an attention vector as the result of the attention mechanism, and (4) predicting the one or more future clinical events based on the data embedding matrix and the attention vector.

2. The system of claim 1 , wherein the aggregated health records comprise health records arranged in different data formats.

3. The system of claim 1 , wherein the standardized data structure format comprises Fast Health Interoperability Resources (FHIR).

4. The system of claim 1 , wherein the aggregated health records contain hospitalization diagnoses, and wherein the diagnoses are mapped to single-level Clinical Classification Software (CCS) codes.

5. The system of claim 1 , wherein combining the time embedding matrix with the product of the projection matrix and the data embedding matrix to generate the attention vector comprises using a column dot product operator to multiply the product of the projection matrix and the data embedding matrix with the time embedding matrix.

6. A method of generating training data for machine learning from a set of raw electronic health records of a multitude of patients from diverse sources in diverse data formats, comprising the steps of:
a) obtaining the set of raw electronic health records;
b) converting the set of raw electronic health records into a single standardized data structure format;
c) ordering the electronic health records converted into the single standardized data structure format into a time-sequenced order per patient; and
d) storing the time-sequenced ordered electronic health records in the standardized data structure format in a data storage device; and
using the time-sequenced ordered electronic health records in the standardized data structure format to train one or more deep learning models to predict one or more future clinical events based on an input electronic health record of a patient, wherein a particular one of the one or more deep learning models includes an attention mechanism indicating how much attention the particular model gave to elements in the input electronic health record to predict the one or more future clinical events, wherein the particular model generates an output comprising both the predicted one or more future clinical events and the result of the attention mechanism, and wherein executing the particular model of the one or more deep learning models comprises (1) generating for a plurality of the elements of the input electronic health record, a respective data embedding vector and a respective time embedding vector, wherein generating the time embedding vector for a particular element of the input electronic health record comprises determining, based on a timing of the particular element, a plurality of single-valued functions of the timing of the particular element, (2) concatenating the data embedding vectors in an order to generate a data embedding matrix and concatenating the time embedding vectors in the same order to generate a time embedding matrix, (3) combining the time embedding matrix with a product of a projection matrix and the data embedding matrix to generate an attention vector as the result of the attention mechanism, and (4) predicting the one or more future clinical events based on the data embedding matrix and the attention vector.

7. The method of claim 6 , wherein the standardized data structure format comprises Fast Health Interoperability Resources (FHIR).

8. The method of claim 7 , wherein the set of electronic health records includes both structured data and unstructured data including free text notes.

Assignees

• Google Llc

Inventors

• Oren Eyal
• Cui Yingwei
• Flores Gerardo
• Duggan Gavin
• Zhang Kun
• Litsch Kurt
• Sundberg Patrik
• Zhang Yi

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/08

  Learning methods · CPC title

• G06N3/09

  Supervised learning · CPC title