Temporal External Validation of Successive RCRI (Revised Cardiac Risk Index) Updates across Observational Healthcare Databases: a Multinational Retrospective Cohort Study

Background We aim to evaluate whether the Revised Cardiac Risk Index (RCRI) recalibration methods worked. Methods We performed temporal external validation and external validation of different recalibrated versions of the RCRI, an implemented prediction model, on observational health databases. The three versions included the original RCRI model (oRCRI), as well as the Duceppe and MDCalc recalibrations. (I) Temporal external validation was performed in 6-month intervals from 2010 to 2025, while (II) external validation overall covered the entire period from 2010 to 2025. Results We performed temporal external validation on eight databases. The Area Under the Receiver Operating Characteristic Curve showed a statistically significant (p = 0.0099) decrease. Calibration metrics and Net-Benefit were relatively consistent. Based on calibration metrics and Net-Benefit values, the RCRI versions can be ranked from best performing to worst performing as follows: MDCalc, oRCRI and Duceppe. In addition, we performed external validation on nine (including the eight used for temporal external validation) databases with an AUROC range of 0.62–0.86 and calibration plots that indicate an overestimation of risk. For an RCRI score of 1, the Net-Benefit range was the most substantial for the MDCalc version (-0.0013–0.022), followed by the oRCRI version (-0.00056–0.023) and the Duceppe version ( -0.025 – (-0.012)). Conclusions While the RCRI recalibrations show consistent calibration and Net‑Benefit, the significant decline in AUROC highlights challenges that extend beyond the RCRI itself. As patient populations and care patterns evolve, even well‑established clinical prediction models require ongoing reassessment, illustrating both the value and limitations of recalibration alone.