# US 2023/0274845 A1 · Neural network artificial intelligence system for class-based modeling

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2023/0274845 A1
- Document ID: US20230274845A1
- Status: Published application
- Kind: A1
- Application number: 18/144,330
- Filed: 2023-05-08
- Published: 2023-08-31
- Earliest claimed: 2019-07-15
- Assignees: Express Scripts Strategic Development, Inc.
- Applicants: Express Scripts Strategic Development, Inc.
- Inventors: Christopher G. Lehmuth; Alexi E. Makarkin
- Web page: https://patentgrep.com/patent/US20230274845A1

## CPC

- G16H 70/20: relating to practices or guidelines
- G16H 70/40: relating to drugs, e.g. their side effects or intended usage
- G16H 20/10: relating to drugs or medications, e.g. for ensuring correct administration to patients
- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems
- G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients
- G16H 15/00: ICT specially adapted for medical reports, e.g. generation or transmission thereof
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA

## Abstract

A method includes receiving historical data collected from a client associated with members. The historical data includes per-member metrics for the client and demographic information for the members. The method includes identifying therapeutic classes for the client based on the per-member metrics and the demographic information. The method includes segmenting the historical data into a data set for each therapeutic class. The method includes, for each therapeutic class of the set of therapeutic classes, determining a pattern for the per-member metrics corresponding to the respective therapeutic class, generating a respective predictive model based on the pattern, and training a neural network of the respective predictive model using a two-stage training process. The predictive model is configured to generate, as output for the therapeutic class, a per-member metric prediction for an input period of future time. The method includes generating predictions for the therapeutic classes using the predictive models.

## Claims

### Claim 1

A method comprising:
receiving historical data collected from a client, wherein the client is associated with a plurality of members, and wherein the historical data includes per member metrics for the client and demographic information for the plurality of members;
identifying a set of therapeutic classes for the client based on the per member metrics and the demographic information;
segmenting the historical data into a respective data set for each therapeutic class, wherein the respective data set for the respective therapeutic class includes per member metrics corresponding to the respective therapeutic class;
for each therapeutic class of the set of therapeutic classes:
determining a pattern for the per member metrics corresponding to the respective therapeutic class;
generating a respective predictive model based on the pattern, wherein the respective predictive model is configured to generate, as output for the respective therapeutic class, a per member metric prediction for an input period of future time; and
training a neural network of the respective predictive model using a two-stage training process, wherein a first stage of the two-stage training process trains the respective predictive model with the historical data collected from the client, and wherein a second stage of the two-stage training process trains the respective predictive model with drug market event data; and
generating predictions for the set of therapeutic classes by using the predictive models of each therapeutic class.

### Claim 2 (depends on 1)

The method of claim 1 wherein the training of the neural network includes:
obtaining the drug market event data and historical market data characterizing an impact of the historical drug market event data on per member metrics;
determining a predicted impact for the obtained drug market event data based on the historical market data;
generating training data for the second stage of the two-stage training process that includes the obtained drug market event data associated with the predicted impact; and
training the respective predictive model during the second stage of the two-stage training using the training data generated for the second stage training process.

### Claim 3 (depends on 2)

The method of claim 2 wherein:
the drug market event data and the historical market data characterize an impact of the historical drug market event data on drug spending trends; and
for each therapeutic class, the respective predictive model is configured to generate, as output, a drug spending trend prediction for the input period of future time.

### Claim 4 (depends on 1)

The method of claim 1 further comprising providing the predictions via a web portal in at least one of a displayable graphical format and a downloadable data structure.

### Claim 5 (depends on 4)

The method of claim 4 wherein the web portal is configured to facilitate selection of a training data set for each respective predictive model.

### Claim 6 (depends on 1)

The method of claim 1 wherein the per member metric corresponds to a specific length of time.

### Claim 7 (depends on 6)

The method of claim 6 wherein the per member metric is based on total spending within the specific length of time.

### Claim 8 (depends on 1)

The method of claim 1 wherein the predictions are further categorized for brand name, generic, and specialty drugs for each of the therapeutic classes.

### Claim 9 (depends on 1)

The method of claim 1 wherein:
the method further comprises validating the trained models using first data collected over a first time period;
the historical data used to train the predictive models during the first stage corresponds to second data from a second time period; and
the first time period is shorter in duration and more recent than the second time period.

### Claim 10 (depends on 9)

The method of claim 9 wherein the drug market event data includes data about new drugs to be released and about past impacts of release of other new drugs on year-over-year change in drug spending and on per member metrics.

### Claim 11 (depends on 9)

The method of claim 9 further comprising:
training the validated models using the first data and the second data; and
generating the predictions using the trained validated models.

### Claim 12 (depends on 9)

The method of claim 9 further comprising:
reconfiguring one of the predictive models in response to the one of the predictive models failing validation;
retraining all of the predictive models including the reconfigured model using the second data; and
revalidating, subsequent to the retraining, all of the predictive models including the reconfigured model using the first data.

### Claim 13 (depends on 9)

The method of claim 9 further comprising:
generating a new predictive model to replace one of the predictive models in response to the one of the predictive models failing validation due to an anomaly; and
using the new predictive model to generate a portion of the predictions.

### Claim 14

A non-transitory computer-readable medium comprising processor-executable instructions, the instructions including:
receiving historical data collected from a client, wherein the client is associated with a plurality of members, and wherein the historical data includes per member metrics for the client and demographic information for the plurality of members;
identifying a set of therapeutic classes for the client based on the per member metrics and the demographic information;
segmenting the historical data into a respective data set for each therapeutic class, wherein the respective data set for the respective therapeutic class includes per member metrics corresponding to the respective therapeutic class;
for each therapeutic class of the set of therapeutic classes:
determining a pattern for the per member metrics corresponding to the respective therapeutic class;
generating a respective predictive model based on the pattern, wherein the respective predictive model is configured to generate, as output for the respective therapeutic class, a per member metric prediction for an input period of future time; and
training a neural network of the respective predictive model using a two-stage training process, wherein a first stage of the two-stage training process trains the respective predictive model with the historical data collected from the client, and wherein a second stage of the two-stage training process trains the respective predictive model with drug market event data; and
generating predictions for the set of therapeutic classes by using the predictive models of each therapeutic class.

### Claim 15 (depends on 14)

The computer-readable medium of claim 14 wherein the training of the neural network includes:
obtaining the drug market event data and historical market data characterizing an impact of the historical drug market event data on per member metrics;
determining a predicted impact for the obtained drug market event data based on the historical market data;
generating training data for the second stage of the two-stage training process that includes the obtained drug market event data associated with the predicted impact; and
training the respective predictive model during the second stage of the two-stage training using the training data generated for the second stage training process.

### Claim 16 (depends on 15)

The computer-readable medium of claim 15 wherein:
the drug market event data and the historical market data characterize an impact of the historical drug market event data on drug spending trends; and
for each therapeutic class, the respective predictive model is configured to generate, as output, a drug spending trend prediction for the input period of future time.

### Claim 17 (depends on 14)

The computer-readable medium of claim 14 wherein the instructions include providing the predictions via a web portal in at least one of a displayable graphical format and a downloadable data structure.

### Claim 18 (depends on 14)

The computer-readable medium of claim 14 wherein:
the per member metric corresponds to a specific length of time; and
the per member metric is based on total spending within the specific length of time.

### Claim 19 (depends on 14)

The computer-readable medium of claim 14 wherein:
the instructions further include validating the trained models using first data collected over a first time period;
the historical data used to train the predictive models during the first stage corresponds to second data from a second time period; and
the first time period is shorter in duration and more recent than the second time period.

### Claim 20 (depends on 19)

The computer-readable medium of claim 19 wherein the instructions include:
reconfiguring one of the predictive models in response to the one of the predictive models failing validation;
retraining all of the predictive models including the reconfigured model using the second data; and
revalidating, subsequent to the retraining, all of the predictive models including the reconfigured model using the first data.
