Calibration under matched case-control sampling

When the goal is to predict a binary disease outcome, but both the training and testing data are matched case-control sampling data. In this situation, can I just focus on discrimination accuracy and ignore calibration as it will be difficult to calibrate the model under this complex sampling design? Is calibration necessary in this situation?

You can assess relative calibration, i e. Calibration slope = 1.0.

1 Like