|
Abstract:
|
We present both a conceptual framework for developing patient-level prediction models using observational data and the corresponding open-source software developed to implement the framework across computational environments and observational healthcare databases. We propose a five step framework based on a review of best practices for model development, validation and reporting. The framework includes standardized steps for i) transparently defining the problem, ii) selecting suitable dataset/s, iii) constructing variables from the observational data, iv) learning the predictive model and v) fairly validating the model performance. We also develop open-source software implementing the framework that utilizes the Observational Medical Outcomes Partnership common data model (CDM) for convenient sharing of models and reproducing model evaluation across multiple observational datasets. We develop 21 prediction models for 21 different outcomes within a target population of pharmaceutically-treated depressed people across 4 different US insurance claims databases. All 84 models are available in an accessible online repository and can be used by any user with an observational database
|
