Predictive modeling of ART outcomes with the CORGData framework: a design and pilot protocol

The CORGData framework has so far been demonstrated on synthetic genetic data. Its stated near-term application is assisted reproductive technology (ART) outcome modeling: predicting cycle and cumulative outcomes from clinical and embryological predictors to support counseling and laboratory decision-making. This paper is a registered study protocol for the framework's first empirical pilot. It specifies a retrospective cohort design on de-identified IVF/ICSI cycles, with cumulative live birth per oocyte retrieval as the primary outcome; a pre-declared candidate predictor set spanning patient, stimulation, and embryo-level variables; a penalized-regression modeling strategy with a pre-specified validation scheme (temporal split plus bootstrap optimism correction); and governance safeguards for clinical data including de-identification, access control, and a prohibition on person-level reconstruction. Reporting will follow the TRIPOD statement for prediction models. Methods are fully specified; no data have been analyzed and results are marked accordingly. This protocol exists so that the eventual model, whatever its performance, can be judged against a fixed plan rather than against the flexibility of retrospective analysis.