Machine learning systems for predictive modeling and related methods

Patent metadata

Field	Value
Publication number	US-11545248-B2
Application number	US-202117367253-A
Country	US
Kind code	B2
Filing date	Jul 2, 2021
Priority date	Apr 18, 2019
Publication date	Jan 3, 2023
Grant date	Jan 3, 2023

Abstract

A machine learning system for training a data model to predict data states in medical orders is described. The machine learning system is configured to train a data model to predict whether a medical order requires prior authorization (“PA”) for medical orders within a medical order data set so that related systems may process incoming medical orders with PA determinations predicted by the data model. The machine learning system includes a first data warehouse system. The first prescription processing system generates a data model of historical orders and payer responses, apply a predictive machine learning model to the data model to generate a trained predictor of whether a medical order requires PA, associated with order data, apply the trained predictor to a plurality of production orders to determine PA for each of the plurality of production orders, and process the plurality of production orders with each associated PA determination.

First claim

What is claimed is:

1. A system for determining prior authorization (PA) requirements for prescriptions, the system comprising:
at least one processor; and
at least one memory component having instructions stored thereon, which, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
receiving a first portion of a plurality of historical orders and a first portion of a plurality of payer responses, the plurality of historical orders being associated with respective ones of the plurality of payer responses;
applying at least one data balancing operation to the first portion of the plurality of historical orders and the first portion of the payer responses;
generating a data model of the first portion of the plurality of historical orders and the first portion of the payer responses, the data model being represented by a tree-structure including a plurality of leaves, nodes, and edges;
generating a predictive machine learning model;
applying the predictive machine learning model to the data model to generate a trained predictor of whether a medical order requires prior authorization (PA), associated with order data;
receiving a plurality of prescription production orders;
applying the trained predictor to the plurality of prescription production orders to determine whether the medical order requires a PA for respective ones of the plurality of prescription production orders; and
processing the plurality of prescription production orders with associated PA requirement determinations for the respective ones.

2. The system of claim 1 , wherein the operations further comprise:
updating the trained predictor, by:
applying the predictive machine learning model to the data model; and

training the predictive machine learning model using at least one of a) a k-nearest neighbor algorithm, b) a logistic regression algorithm, c) a random forest algorithm, and d) a naive Bayesian algorithm.

3. The system of claim 2 , wherein training the predictive machine learning model further comprises:
applying the k-nearest neighbor algorithm, including considering a set of neighbors comprising the nearest twenty neighbors identified for the plurality of leaves of the tree-structure.

4. The system of claim 1 , wherein the operations further comprise:
receiving a second portion of the plurality of historical orders and a second portion of the plurality of payer responses;
applying the trained predictor to the second portion of the plurality of historical orders to determine whether a medical order requires PA for the second portion of the plurality of historical orders;
receiving a pre-defined threshold for error;
determining an error rate for the trained predictor by comparing the determined PA requirement for the second portion of the plurality of historical orders to the associated second portion of the plurality of payer responses; and
determining whether the trained predictor requires re-training.

5. The system of claim 4 , wherein the operations further comprise:
upon determining that the trained predictor requires re-training,
re-training the trained predictor by generating a second data model of the second portion of the plurality of historical orders and the second portion of the payer responses, the second data model being represented by a tree-structure including a plurality of leaves, and
applying the predictive machine learning model to the second data model to generate a re-trained predictor of whether a medical order requires PA, associated with order data;
receiving a third portion of the plurality of historical orders and a third portion of the plurality of payer responses;
applying the re-trained predictor to the third portion of the plurality of historical orders to determine whether a medical order requires PA for respective ones of the third portion of the plurality of historical orders;
receiving the pre-defined threshold for error; and
determining a second error rate for the re-trained predictor by comparing the determined PA requirement for the respective ones of the third portion of the plurality of historical orders to the associated third portion of the plurality of payer responses.

6. The system of claim 1 , wherein the operations further comprise:
re-balancing the first portion of the plurality of historical orders and the first portion of the plurality of payer responses, to create balanced data; and
generating the data model based on the balanced data.

7. The system of claim 1 , wherein the operations further comprise:
receiving additional pluralities of historical orders and additional pluralities of historical payer responses; and
iteratively re-training the trained predictor on the additional pluralities of historical orders and the additional pluralities of historical payer responses in parallel with applying the trained predictor to the plurality of prescription production orders.

8. A method for determining prior authorization (PA) requirements for prescriptions, the method comprising:
receiving a first portion of a plurality of historical orders and a first portion of a plurality of payer responses, the plurality of historical orders being associated with respective ones of the plurality of payer responses;
applying at least one data balancing operation to the first portion of the plurality of historical orders and the first portion of the payer responses;
generating a data model of the first portion of the plurality of historical orders and the first portion of the payer responses, the data model being represented by a tree-structure including a plurality of leaves, nodes, and edges;
generating a predictive machine learning model;
applying the predictive machine learning model to the data model to generate a trained predictor of whether a medical order requires prior authorization (“PA”) associated with order data;
receiving a plurality of prescription production orders;
applying the trained predictor to the plurality of prescription production orders to determine whether the medical order requires PA for respective ones of the plurality of prescription production orders; and
processing the plurality of prescription production orders with associated PA requirement determinations for the respective ones.

9. The method of claim 8 , further comprising:
applying the predictive machine learning model to the data model; and
training the predictive machine learning model using at least one of a) a k-nearest neighbor algorithm, b) a logistic regression algorithm, c) a random forest algorithm, and d) a naive Bayesian algorithm.

10. The method of claim 9 , further comprising:
using the k-nearest neighbor algorithm that is configured to consider a set of neighbors comprising a nearest twenty neighbors identified for respective ones of the plurality of leaves of the tree-structure.

11. The method of claim 8 , further comprising:
receiving a second portion of the plurality of historical orders and a second portion of the plurality of payer responses;
applying the trained predictor to the second portion of the plurality of historical orders to determine whether a medical order requires PA for respective ones of the second portion of the plurality of historical orders;
receiving a pre-defined threshold for error;
determining an error rate for the trained predictor by comparing the determined PA requirement for the respective ones of the second portion of the plurality of historical orders to the associated second portion of the plurality of payer responses; and
determining whether the trained predictor requires re-training.

12. The method of claim 11 , fu

Assignees

• Express Scripts Strategic Dev Inc
• Express Scripts Strategie Dev Inc

Inventors

• Nambirajan Balakrishnan
• Anderson Garret L
• Durosko Heather L
• Gorbett Angela

Classifications

• G06N7/01

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

• G16H20/10 Primary

  relating to drugs or medications, e.g. for ensuring correct administration to patients · CPC title

• G16H50/70

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

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N20/20

  Ensemble learning · CPC title