Identifying clusters of people with Multiple Long-Term Conditions using Large Language Models: a population-based...

Area of Science:

• Computational linguistics
• Health informatics
• Population health

Background:

• Identifying patterns in Multiple Long-Term Conditions (MLTC) is crucial for personalized healthcare.
• Existing methods may not fully capture the complexity of co-occurring chronic diseases.

Purpose of the Study:

• To develop and apply a language model for identifying gender-specific clusters of patients with MLTC.
• To leverage electronic health record (EHR) data for uncovering distinct disease patterns.

Main Methods:

• A pipeline incorporating a DeBERTa language model (EHR-DeBERTa) was developed.
• The model was pre-trained on longitudinal EHR data from 5.8 million UK patients.
• Gender-specific patient embeddings were generated and clusters identified using K-Means analysis.

Main Results:

• Fifteen clusters were identified in females and seventeen in males.
• Clusters were categorized into low disease burden, mental health, cardiometabolic, respiratory, and mixed disease groups.
• Cardiometabolic and mental health conditions showed significant separation across clusters, with older patients prevalent in cardiometabolic groups.

Conclusions:

• Large language models (LLMs) can offer interpretable insights into complex disease patterns.
• This approach supports precision medicine for individuals with MLTC.
• Future research incorporating clinical outcomes can enhance risk prediction.