Large language models forecast patient health trajectories enabling digital twins
Bibliographic record
Abstract
Abstract Background Generative artificial intelligence (AI) accelerates the development of digital twins, which enable virtual representations of real patients to explore, predict and simulate patient health trajectories, ultimately aiding treatment selection and clinical trial design. Recent advances in forecasting utilizing generative AI, in particular large language models (LLMs), highlights untapped potential to overcome real-world data (RWD) challenges such as missingness, noise and limited sample sizes, thus empowering the next generation of AI algorithms in healthcare. Methods We developed the Digital Twin - Generative Pretrained Transformer (DT-GPT) model, which utilizes biomedical LLMs using rich electronic health record (EHR) data. Our method eliminates the need for data imputation and normalization, enables forecasting of clinical variables, and preliminary explainability through a human-interpretable interface. We benchmarked DT-GPT on RWD including long-term US nationwide non-small cell lung cancer (NSCLC) and short-term Intensive Care Unit (ICU) datasets. Findings DT-GPT surpassed state-of-the-art machine learning methods in patient trajectory forecasting on mean absolute error (MAE) for both the long-term (3.4% MAE improvement) and the short-term (1.3% MAE improvement) dataset. Additionally, DT-GPT was capable of preserving cross-correlations of clinical variables (average R 2 of 0.98), handling data missingness and noise. Finally, we discovered the ability of DT-GPT to provide insights into a forecast’s rationale and to perform zero-shot forecasting on variables not used during fine-tuning, outperforming even fully trained task-specific machine learning models on 13 clinical variables. Interpretation DT-GPT demonstrates that LLMs can serve as a robust medical forecasting platform, empowering digital twins which virtually replicate patient characteristics beyond their training data. We envision that LLM-based digital twins will enable a variety of use cases, including clinical trial simulations, treatment selection and adverse event mitigation.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.013 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".