Granted Patent B2

US 12,190,263 · App. 17/218,111 · Granted Jan 7, 2025

Inventors: Brian Denton (Ann Arbor, MI); Erkin Ötles (Ann Arbor, MI)

Assignee: REGENTS OF THE UNIVERSITY OF MICHIGAN

G06Q10/063114G06N3/08G16H50/30

Abstract

Claims (68)

1. A computer-implemented method for predicting a work-status of an individual patient, the method comprising, via one or more processors:

receiving, by the one or more processors, observation data and historical work-statuses of the individual patient, wherein the observation data is represented by an input observation matrix and the historical work-statuses are represented by a work-statuses matrix; and

training, by the one or more processors, a machine learning algorithm to predict work status based on (i) observation data and (ii) historical work-statuses, wherein training the machine learning algorithm includes:

transforming the input observation matrix to lower dimensional vectors;

inputting, into the machine learning algorithm, the lower dimensional vectors;

predicting, with the machine learning algorithm, a work-status of the individual patient following a predetermined time period based on the observation data of the individual patient;

splitting the work-status into a training dataset and a validation dataset;

filtering and normalizing the training dataset to minimize information leakage;

validating the machine learning algorithm by comparing the predicted work-status with the validation dataset; and

dynamically updating the machine learning algorithm using the training dataset.

2. The computer-implemented method of claim 1 , wherein the observation data includes data of diagnoses, treatments, patient outcomes, and medications.

3. The computer-implemented method of claim 1 , wherein:

the observation data, and the historical work-statuses each include a plurality of timesteps; and

the prediction of the work-status of the individual patient is made at each timestep of the plurality of timesteps.

4. The computer-implemented method of claim 1 , wherein the prediction of the work-status of the individual patient is made as a probability.

5. The computer-implemented method of claim 1 , wherein:

the observation data includes: (i) a plurality of timesteps, and (ii) feature data; and

the feature data includes: (i) dynamic data that changes between two timesteps of the plurality of timesteps, and (ii) static data that remains constant across all timesteps of the plurality of timesteps.

6. The computer-implemented method of claim 1 , wherein the observation data includes feature data, and wherein the feature data includes:

an age of an input patient;

a gender of the input patient;

an occupation of the input patient;

a diagnosis of the input patient;

a treatment of the input patient;

a diagnosis accompanying the treatment;

a medication of the input patient, and dispense amount of the medication; and

a health status of the input patient.

7. The computer-implemented method of claim 1 , wherein:

the observation data includes a low dimension category feature including one of gender data or health status data;

the observation data further includes a high dimension category feature including one of occupation data, diagnosis data, treatment data, or medication data; and

the computer-implemented method further includes:

converting the low dimension category feature to a one-hot-encoding vector, and aggregating values of the one-hot-encoding vector; and

mapping the high dimension category feature to a real-space vector with a dimension proportional to a number of category values of the high dimension category feature.

8. The computer-implemented method of claim 1 , wherein training the machine learning algorithm further includes:

using an objective function to minimize binary-cross entropy between a vector of the observation data and a vector of the historical work-statuses; and

using a special generator function to pad variable lengths of the observation data.

9. The computer-implemented method of claim 1 , wherein:

input observations of the observation data are represented by X n,t ; and

the historical work-statuses are represented as binary health-state variables represented by Y n,t .

10. A computer system for predicting a work-status of an individual patient, the computer system comprising one or more processors configured to:

receive observation data and historical work-statuses of the individual patient, wherein the observation data is represented by an input observation matrix and the historical work-statuses are represented by a work-statuses matrix; and

train a machine learning algorithm to predict work status based on (i) observation data and (ii) historical work-statuses, wherein training the machine learning algorithm includes:

transform the input observation matrix to lower dimensional vectors;

input, into the machine learning algorithm, the lower dimensional vectors;

predict, with the machine learning algorithm, the work-status of the individual patient following a predetermined time period based on the observation data of the individual patient;

split the work-status into a training dataset and a validation dataset;

filter and normalize the training dataset to minimize information leakage;

validate the machine learning algorithm by comparing the predicted work-status with the validation dataset; and

dynamically update the machine learning algorithm using the training dataset.

11. The computer system of claim 10 , wherein the observation data includes data of diagnoses, treatments, and medications.

12. The computer system of claim 10 , wherein:

the observation data includes: (i) a plurality of timesteps, and (ii) feature data; and

the feature data includes: (i) dynamic data that changes between two timesteps of the plurality of timesteps, and (ii) static data that remains constant across all timesteps of the plurality of timesteps.

13. A computer device for predicting a work-status of an individual patient, the computer device comprising:

one or more processors; and

one or more memories coupled to the one or more processors;

the one or more memories including computer executable instructions stored therein that, when executed by the one or more processors, cause the one or more processors to:

receive observation data and historical work-statuses of the individual patient, wherein the observation data is represented by an input observation matrix and the historical work-statuses are represented by a work-statuses matrix; and

train a machine learning algorithm to predict work status based on (i) observation data and (ii) historical work-statuses, wherein training the machine learning algorithm includes:

transform the input observation matrix to lower dimensional vectors;

input, into the machine learning algorithm, the lower dimensional vectors;

predict, with the machine learning algorithm, a work-status of the individual patient following a predetermined time period based on the observation data of the individual patient;

split the work-status into a training dataset and a validation dataset;

filter and normalize the training dataset to minimize information leakage;

validate the machine learning algorithm by comparing the predicted work-status with the validation dataset; and

dynamically update the machine learning algorithm using the training dataset.

14. The computer device of claim 13 , wherein the observation data includes data of diagnoses, treatments, and medications.

15. The computer device of claim 13 , wherein the machine learning algorithm is a deep learning algorithm.

Assignments (1)

Continuity (2)

References Cited (116)