Risk scoring is a useful tool for prognostication, decision-making and, increasingly, assessing individual and institutional performance. At the South Yorkshire Cardiothoracic Centre, we have been using the New York risk score as a clinical routine which was developed in 2002 on a large population from the state of New York. We developed a “Sheffield risk score”, originally created in 2004, validated internally in a second Sheffield population in 2007 and externally in another UK centre 3 years ago. This study aimed to assess the ability of both scores to predict 30-day mortality.Methods
The data from all patients undergoing PCI at the South Yorkshire Cardiothoracic Centre from January 2009 to December 2011 were collected prospectively. The variables for the New York PCI risk score (age, sex, haemodynamic status, previous MI, congestive heart failure, left ventricular function, renal function, peripheral arterial disease and left main stem disease) and the Sheffield score (age, extent of coronary disease, urgency of the PCI and cardiogenic shock) were extracted from this database. 30-day all-cause mortality was collected from the Office of National Statistics (ONS). The predictive power of the two scores was assessed using ROC curve analysis and Hosmer-Lemeshow (HL) tests.Results
4,607 patients underwent PCIs during the study period. The age of the patients was 63 ± 12 years, 72% were males, 71% had PCI for ACS and 1.0% had cardiogenic shock. The 30-day mortality in this population was 2.5%. Both the New York score and the Sheffield score discriminated well between patients who did and did not die within 30 days (area under curve 0.84 and 0.81, respectively). The New York score did not show superior predictive power compared with the Sheffield score. However, both scores showed poor calibration and a sub-optimal fit for this population (HL statistic 54.4, p < 0.01 for the New York score and 36.6, p < 0.01 for the Sheffield score).Conclusion
Risk models developed from a distinct population group need not only geographical validation, but also periodic update as their value may diminish with changing patient profile and treatment options. Our findings highlight their use as a guide; clinical judgement, informed consent and multidisciplinary review remain indispensable.