DeepPatientLevelPrediction R package
This R package helps researchers build and test deep learning models to predict individual patient outcomes using real-world health data stored in a standard format called OMOP CDM.
At a glance
Use when
When predicting complex patient outcomes using high-dimensional observational data and when leveraging deep learning methods within a standardized, generalizable framework is desired.
Avoid when
When working with small datasets, limited computational resources, or when simpler models (e.g., logistic regression) are sufficient for the task.
Inputs
Patient-level observational data in OMOP Common Data Model format, including demographics, diagnoses, procedures, and medications; defined prediction problem (e.g., outcome, time windows); model configuration parameters.
Outputs
Trained deep learning models for patient-level prediction, performance metrics (e.g., AUC, calibration), and tools for model interpretation and validation.
How it works
An R package that enables patient-level prediction modeling using deep learning (via PyTorch through reticulate) on observational healthcare data structured in the OMOP Common Data Model. It integrates with the OHDSI PatientLevelPrediction framework and supports models such as MLP, ResNet, Transformer, and RealMLP. The package requires R ≥ 4.0.0 and Python (recommended via uv), with GPU support recommended for training.
- Project
- EHDEN
- Funding
- IMI
- Project status
- Completed 2024
- HTA domains
- Clinical Effectiveness
- Categories
- RWEPredictive Modelling
- Technology
- Non-specific
- Assumptions
- Data is standardized to the OMOP CDM; sufficient sample size and event rates for deep learning training; access to computational resources (preferably GPU) for model training.
- Strengths
- Integrates deep learning into the OHDSI ecosystem; supports multiple neural network architectures; enables model customization; leverages standardized data for reproducibility and portability.
- Limitations
- Requires technical expertise in R, Python, and deep learning; GPU often needed for practical use; steep learning curve for new users; limited interpretability of deep learning models.
- Also known as
- DeepPatientLevelPrediction, DeepPatientLevelPrediction R package
Questions this answers
- › Can I use deep learning to predict patient outcomes from real-world data?
- › How can I build and validate predictive models using observational health databases?
- › Is it possible to use neural networks within the OHDSI modeling framework?
- › What types of deep learning models are supported for patient-level prediction?
- › Can I customize or add my own deep learning architecture in this tool?
- › Does this package work with large-scale electronic health record data?
Related methods
Similar by meaning
Beta record. Generated from the primary source via AI extraction and independent audit, pending final human review.