Abstract

Objectives There is a pressing need for workable risk models for patients undergoing PCI. A few risk models have been proposed, but to be applicable to a national audit process, they must be validated not only on the local populations from which they were derived but also on other geographical populations (″internal″± vs ″external″± validation). We sought to validate two proposed risk adjustment models (Mayo Clinic, US & NWQIP, UK models) for in-hospital percutaneous coronary intervention (PCI) complications on an independent data set of relatively high-risk patients undergoing PCI.

Setting Tertiary centre in North England

Methods Between September 2002 and August 2006, 5034 consecutive PCI procedures (Validation set) were performed on a patient group characterized by a high incidence of AMI (16.1%) and cardiogenic shock (1.7%). Two external models, one developed by the North West Quality Improvement Programme (NWQIP model, UK) and the other by the Mayo Clinic Risk Score (MC model, USA) were externally validated

Main outcome measure Major adverse cardiovascular and cerebrovascular events (MACCE), which were in-hospital mortality, Q wave MI, emergent CABG, and cerebrovascular accidents.

Results In this patient group, an overall in-hospital complication rate of 2% was observed. Multivariate regression analysis identified risk factors for in-hospital complications that were similar to the risk factors identified by the two external models. When fitted to the data set, both external models had an area under the ROC curve ≥0.85 (c index (95% CI), NWQIP; 0.86 (0.82 to 0.9), MC; 0.87(0.84 to 0.9)) indicating overall excellent model discrimination and calibration (Hosmer-Lemeshow test, p > 0.05). The NWQIP model was accurate in predicting in-hospital complications in different patient subgroups (age groups, cardiogenic shock, diabetes, women, previous MI, previous CABG)

Conclusions We have externally validated both models. Despite differences in variable selection, both these predictive models yield comparable results that provided excellent model discrimination and calibration when applied to patient groups in a different geographic population other than the one on which the original model was developed.