PatientLevelPrediction

PatientLevelPrediction is an R package designed for building and validating patient-level predictive models using data in the OMOP Common Data Model (CDM) format.

It addresses the prediction problem where, for a population at risk, the goal is to predict which patients at a specific moment in time ($t=0$) will experience a defined outcome during a specified time-at-risk (TAR). The prediction is based solely on patient information collected during an observation window prior to $t=0$.

Key capabilities include:

• Developing models for multiple combinations of Target Cohorts (T) and Outcome Cohorts (O).
• Extracting data for multiple covariate settings from OMOP CDM databases.
• Utilizing a wide range of covariates (drugs, diagnoses, procedures, age, comorbidity indexes, etc.).
• Supporting various machine learning algorithms (Regularized logistic regression, Random forest, GBM, etc.) and allowing for custom algorithms, feature engineering, and sampling.
• Providing tools for external validation, performance plotting (ROC and Calibration), and learning curves.
• Offering a Shiny app for interactive exploration of results and report generation.

Follow these resources to begin using the package:

1. Installation: Refer to the Package Installation guide.
2. Core Workflow: Read the main vignette: Building Single Patient-Level Predictive Models.
3. Advanced Usage:
   • Building Multiple Patient-Level Predictive Models
   • Adding Custom Machine Learning Algorithms
   • Creating Learning Curves
4. Specialized Modeling:
   • For Deep Learning, use DeepPatientLevelPrediction.
   • For Ensemble Models, use EnsemblePatientLevelPrediction.
5. Reference: For a full list of functions, see the Package Reference.