Large Medical Model (LMM)

AI model · Forecasting / time series · Time series → Time series by GenHealth.ai

Predict costs, risks, and care paths with a transformer model trained on 140M patient records.

The GenHealth Large Medical Model (LMM) is a transformer-based generative AI model specifically trained on longitudinal records of 140 million patients. It is designed for healthcare organizations, health plans, and developers to predict medical costs, risks, and future care paths with high precision. By treating patient histories as sequences of events, the model can simulate possible medical futures and surface early interventions for chronic conditions.

Languages
English
Training Data Size
140 million patients
Architecture
Generative Pre-trained Transformer (GPT)
Primary Modality
Patient event sequences to predictions
Cost Prediction Improvement
14.1% over top models

Overview

The GenHealth Large Medical Model (LMM) is a transformer-based generative AI model specifically trained on longitudinal records of 140 million patients. It is designed for healthcare organizations, health plans, and developers to predict medical costs, risks, and future care paths with high precision. By treating patient histories as sequences of events, the model can simulate possible medical futures and surface early interventions for chronic conditions.

Predictive Analytics

  • Cost Prediction: Predicts future medical costs based on longitudinal patient histories and event sequences.
  • Risk Assessment: Identifies potential health risks and conditions before they manifest in clinical records.
  • Care Path Simulation: Simulates possible care paths and future medical events for individual patients.
  • MLR Forecasting: Predicts medical loss ratio (MLR) for health plans using massive healthcare datasets.

Model Architecture and Training

  • 140M Patient Dataset: Trained on a massive dataset covering 140 million patients across the United States.
  • Transformer Architecture: Utilizes transformer-based architectures optimized for patient event sequences.
  • Benchmark Performance: Improves cost prediction accuracy by 14.1% over previous top-performing models.
  • Synthetic Data Generation: Generates accurate synthetic patient histories for any combination of conditions and demographics.

Specifications

Training Data Size
140 million patients
Architecture
Generative Pre-trained Transformer (GPT)
Primary Modality
Patient event sequences to predictions
Cost Prediction Improvement
14.1% over top models

Added by AntSpace from public sources (genhealth.ai).

More AI models by GenHealth.ai