pyhealth

You can train clinical ML models on electronic health records by loading datasets like MIMIC-III or MIMIC-IV, defining the mortality prediction task, selecting an EHR model, and using the built-in training and evaluation pipeline. This standardizes the healthcare modeling workflow end-to-end.

Supported healthcare datasets include MIMIC-III, MIMIC-IV, eICU, and OMOP. These datasets provide electronic health records, clinical codes, and physiological data required for clinical prediction and drug recommendation tasks.

Yes, medical coding interoperability is supported to translate and standardize diagnosis and medication codes across ICD and drug vocabularies. This creates consistent model inputs for clinical machine learning workflows.

Build a drug recommendation model by loading a supported clinical dataset, setting the drug recommendation task, and selecting a model through the unified healthcare pipeline. The toolkit handles patient-level splitting and evaluation compatible with clinical metrics.

Yes, multimodal healthcare modeling is supported alongside EHR-based tasks. You can apply clinical prediction across modalities including sleep, EEG, image, and text-based clinical tasks using the standardized training and evaluation pipeline.

Patient-level splitting utilities are provided to correctly partition clinical health data for model training. This prevents data leakage across patient records and ensures evaluation metrics accurately reflect model performance on unseen patients.