Thread by @lamelaspablo

#WeeklyCardioReview New CTO success score CASTLE (from @EuroCtoClub over 20k procedures) shows good discrimination (validation AUC 0.68) and good visual calibration, by @jcspratt in @JACCJournals interventions.onlinejacc.org/content/12/4/3… Great to have contemporary scores to inform patients

Have a methods question: derivation/validation datasets not selected randomly, but derived in an older cohort (A) and validated in a newer cohort (B). Not surprisingly, discrimination properties (AUC) remained similar, but calibration poorer in validation cohort (overestimation)

Likely due to better success rates over time as technology and expertise improve. But calibration is key for talking to patients: they don't want to know if they are more likely to have success vs other patients (discrimination), they want to know their % success (calibration)

My question: should they re-calibrate the original score using a calibration factor for contemporary procedures, or just use the absolute numbers of success they got in the validation analysis? @rwyeh @f2harrell @AlesqQuadros @ADAlthousePhD @PedroMDMSc @hvanspall @GreggWStone

@mmamas1973 @rahatheart1 @g_giustinoMD @DocSavageTJU @djc795 @DavidLBrownMD @SanjitSJolly @RogueRad @willsuh76 @ShariqShamimMD @venkmurthy @mirvatalasnag