Skip to content

Synthetic Transport Case

The synthetic transport case study in cases/synthetic_transport_example/ provides an end-to-end evaluation of risk prediction transportability under controlled covariate shift and selection bias.


Overview

In this scenario: - Target population: \(N_{\text{target}} = 5{,}000\) - Source clinic population: \(N_{\text{source}} = 2{,}000\) (with high event enrichment) - External reference registry: \(N_{\text{reference}} = 5{,}000\)

The analysis demonstrates that: 1. Unconstrained ML fitted on the source cohort exhibits severe calibration-in-the-large drift on the target cohort. 2. Propensity Score Matching alone reduces covariate discrepancy but does not eliminate calibration intercept drift. 3. Constrained MLE successfully anchors risk predictions back to reference benchmarks, restoring calibration.


Execution

Run the case study runner:

python -m cases.synthetic_transport_example.run_case

The script writes summary tables and diagnostics to data/synthetic_transport/.