A Methodology for Using Large Language Models to Create User-Friendly Applications for Medicaid Redetermination and Other Social Services

Background

Following the unwinding of Medicaid’s continuous enrollment provision, states must redetermine Medicaid eligibility, creating uncertainty about coverage [1 (opens in new tab)] and the widespread administrative removal of beneficiaries from rolls [2 (opens in new tab)].

Existing research demonstrates that Large Language Models (LLMs) can automate clinical trial eligibility query extraction [3 (opens in new tab)], generation [4 (opens in new tab)], and classification [5 (opens in new tab)]. Given that Medicaid redetermination follows eligibility rules similar to those in clinical trials, we thought LLMs might help with Medicaid redetermination, as well.

Therefore, using the State of Washington, South Carolina, and North Dakota as examples, we applied LLMs to extract Medicaid rules from publicly available documents and transform those rules into a web application that could allow users to determine whether they are eligible for Medicaid. This paper describes the methodology we used.