From General to Specific: Tailoring Large Language Models for Personalized Healthcare

TL;DR

The paper addresses the lack of true individual-level personalization in medical LLMs by introducing Personalized Medical Language Model (PMLM), which constructs per-patient hard prompts through self-informed predictions and peer-informed similarities, then refines them with reinforcement learning. The method preserves privacy by avoiding internal LLM parameter changes and enables use with proprietary models, while enabling cross-disease applicability. Through an obstetrics/gynecology real-world dataset, PMLM achieves state-of-the-art personalization, outperforming both disease-focused and some fine-tuned baselines, with GPT-4-based PMLM delivering the best results. Overall, PMLM provides a practical framework for highly personalized, privacy-preserving medical AI that can adapt across LLMs and disease contexts, advancing personalized healthcare delivery.

Abstract

The rapid development of large language models (LLMs) has transformed many industries, including healthcare. However, previous medical LLMs have largely focused on leveraging general medical knowledge to provide responses, without accounting for patient variability and lacking true personalization at the individual level. To address this, we propose a novel method called personalized medical language model (PMLM), which explores and optimizes personalized LLMs through recommendation systems and reinforcement learning (RL). Specifically, by utilizing self-informed and peer-informed personalization, PMLM captures changes in behaviors and preferences to design initial personalized prompts tailored to individual needs. We further refine these initial personalized prompts through RL, ultimately enhancing the precision of LLM guidance. Notably, the personalized prompt are hard prompt, which grants PMLM high adaptability and reusability, allowing it to directly leverage high-quality proprietary LLMs. We evaluate PMLM using real-world obstetrics and gynecology data, and the experimental results demonstrate that PMLM achieves personalized responses, and it provides more refined and individualized services, offering a potential way for personalized medical LLMs.

Figures (13)

• Figure 1: Previous so-called personalized medical LLMs only contain knowledge of a single disease, with personalization limited to variations in the input data.
• Figure 2: Architecture of the proposed PMLM.
• Figure 3: Example prompt for the predictor.
• Figure 4: Example of the coarse-grained personalized prompt. The labels for similar patients are ground truth values, as these records come from past cases.
• Figure 5: Example prompt for evaluation, where the <Output> is omitted when evaluating proprietary LLMs.