Abstract 3865: Validity of the TIMI risk scores for UA/NSTEMI and STEMI using a UK dataset
Objective: To determine the validity of the TIMI risk score for unstable angina/NSTEMI for 14-day death and TIMI risk score for STEMI for 30-day death in a UK dataset of community-based cohort of patients.
Methods: We used the MINAP database, a UK registry of acute coronary syndromes and to undertake basic statistical summaries and classical logistic modelling of clinical end points appropriate to each risk model. The performance of the fitted risk models were evaluated using the area under the receiver operating curve.
Results: We had data for 187,069 patients.
TIMI risk score for UA/NSTEMI: Of the 49,951 eligible patients, the mean risk score = 2.553 (SD = 1.177), the 14-day mortality = 6.8% and there was a death rate gradient by risk score (0/1 = 3.0%, 2 = 6.6%, 3 = 8.0%, 4 = 8.8%, 5 = 9.1%, 6 = 9.6%). Logistic regression of 14-day mortality on the TIMI risk score entered as a continuous covariate showed a trend to increased risk with increasing score OR = 1.288, P < 0.001. However, the model did not discriminate well between survival and death within 14-days (C index = 0.584, 95% CI = 0.574 to 0.594).
TIMI risk score for STEMI: Of the 34,722 eligible the mean risk score = 2.5559 (SD = 2.0782, minimum = 0, maximum = 12), the 30-day mortality showed an increasing trend with increasing risk score (OR = 1.604, Wald score 2503.24, P < 0.001) and the fitted model was a good discriminator of survival and death at 30-days (C index = 0.763, 95% CI = 0.755 to 0.770).
Conclusion: Risk models help medical decision-making for patients with acute coronary syndromes but many are developed from randomised controlled trials and are only validated in the derivation cohort. In comparison to the TIMI risk score for STEMI, the commonly used TIMI unstable angina /NSTEMI risk score was not a good predictor of 14-day death in a UK community-based population. Risk scores should be validated in the population to which they are to be applied, so that accurate predictions of clinical end points may be made.