Validation of the PADIT (Prevention of Arrhythmia Device Infection Trial) risk score for infection and infection subtypes

Abstract Background Cardiac implantable electronic device (CIED) infection carries a substantial burden of morbidity, mortality, and cost. The Prevention of Arrhythmia Device Infection Trial (PADIT) risk score improves identification of high-risk patients and may guide targeted strategies to reduce infection. Recent work has categorized CIED infection into localized pocket versus systemic infection, with early reports suggesting different risk factors for each. However, no current risk score has been validated for infection subtypes. Objectives (i)Independently validate the PADIT infection risk score.(ii)Compare risk factors for infection subtypes.(iii)Assess PADIT performance in predicting subtype-specific infection. Methods A prospective registry was initiated at the University of Ottawa Heart Institute in 2007 to capture all CIED procedures and prospectively identify infections in collaboration with the infection prevention team. PADIT risk score components were documented for each procedure. All suspected infections were adjudicated independently by two physicians (with a third if required), blinded to PADIT score and baseline variables, and subclassified as pocket or systemic infection. Logistic regression models were generated to validate PADIT performance for each subtype, with evaluation using Akaike and Bayesian information criteria (AIC/BIC), C-statistics, and calibration slope. Results Between 2007 and 2020, 14,225 procedures were performed (mean age 72 ± 14 years, 35% female, 70% new implants, 18% generator changes, 11% upgrades). A total of 103 infections (0.73%) were adjudicated, of which 71 (69%) were pocket and 32 (31%) systemic. The PADIT score showed good predictive performance with a C-statistic of 0.687 (95% CI 0.655–0.743), similar to the derivation cohort (0.702, 95% CI 0.661–0.741). Notably, the number of prior procedures was strongly associated with pocket infection but not systemic infection. PADIT discrimination was consistent across subtypes: pocket infection C-statistic 0.691 (95% CI 0.649–0.761) and systemic infection 0.746 (95% CI 0.707–0.848). Calibration slopes demonstrated good agreement between predicted and observed events, with the best fit for systemic infection. Conclusion The PADIT score was independently validated with discrimination and calibration similar to the original derivation cohort. Importantly, prior procedures predicted pocket but not systemic infection. Overall, PADIT performed well in predicting both subtypes, with the strongest model fit observed for systemic infection.