Answers are grounded in this paper and cite their sources.
OBJECTIVE: To develop and validate InterGenEHR+, an interactive framework for natural language-conditioned generation of synthetic tabular data and longitudinal EHR trajectories for critically ill patients requiring mechanical ventilation and venovenous extracorporeal membrane oxygenation (VV-ECMO), enabling scenario-based evaluation of machine learning models in high-risk cohorts. METHODS: InterGenEHR+ uses an optional large language model (LLM) parser for text-driven input. For single-time-point generation, we integrate variational autoencoders (VAEs) and diffusion models conditioned on observed variables. Multiple diffusion architectures support conditional longitudinal generation. The framework is evaluated on a multi-institutional VV-ECMO cohort of 874 patients using distributional fidelity metrics, imputation accuracy, adherence to clinical constraints, downstream mortality prediction tasks, and clinician evaluation. RESULTS: Gemma 3 4B with few-shot prompting achieves the best performance in parsing clinician prompts while maintaining a low spurious slot rate. The experimental results demonstrate strengths across IGE variants: IGE-TabDDPM and IGE-PMVAE show strong performance in single-time-point generation, whereas IGE-CSDI, IGE-FGTI, and IGE-MTSCI achieve the best performance on different longitudinal fidelity metrics. In downstream evaluations across five random seeds and three classifiers, models trained on synthetic cohorts retain predictive utility for mortality prediction on real test data. Clinician ratings further indicate that the generated trajectories are clinically plausible and consistent with the requested scenarios. CONCLUSION: InterGenEHR+ enables clinician-specified conditional generation of clinically coherent EHR trajectories with strong statistical fidelity and downstream predictive utility in the evaluated VV-ECMO cohort and could be extended to other clinical settings. SIGNIFICANCE: .
No takes yet. Share an insight, caveat, or question.
Yang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: