Forecasting risk in acute myocardial infarction

Coronary heart disease is the most common cause of death worldwide with an estimated 7 million deaths per year. The majority of these deaths are due to acute myocardial infarction (AMI) so the burden of illness and mortality from AMI worldwide is immense. Existing short-term risk assessment strategies in AMI are limited to Western patient populations. In this thesis we have proposed risk models for prediction of mortality after AMI based on the geographically diverse Hirulog and Early Reperfusion or Occlusion (HERO-2) trial. The HERO-2 trial randomised 17 073 patients to either unfractionated heparin or bivalirudin in conjunction with fibrinolytic therapy with streptokinase, for the treatment of ST-segment Elevation MI. Patients were recruited from 46 countries from Europe, North and Latin America and Asia, including Australia, New Zealand and Russia. We have developed a comprehensive risk model to identify significant predictors of 30-day mortality. This model was subsequently simplified to a basic risk index and predictive accuracy was compared. We have also proposed two new methods for directly comparing the calibration and ranking performance of two risk strategies. -- The geographical diversity of the HERO-2 trial also provided a unique opportunity to examine international differences in clinical outcomes following AMI. We have undertaken a comprehensive comparison of patient characteristics, treatment and outcomes across 5 pre-specified regions: Western countries, Latin America, Eastern Europe, Russia and Asia. We found that mortality rates were lower in Western countries and that these differences could not be attributed to patient case-mix, treatments or national health and economic statistics. -- An important issue in applying findings from randomised clinical trials is the procedure to estimate risk among members of other patient populations. Using the HERO-2 trial we compared methods for updating risk models for AMI. A variety of re-calibration and model revision strategies were compared with a global modeling strategy having a built-in region effect. The relative performance of these methods in the different geographical regions, which vary in sample size, was of primary interest. Model revision was found to only provide a slight improvement in predictive performance over the global model. We concluded that a global model with regional re-calibration is adequate. We also studied data from 5 additional multinational trials: GUSTO-1, GUSTO-2b, GUSTO-3, ASSENT-2 and ASSENT-3. We further explored the adequacy of applying simple re-calibration to update a model for the context of applying a previously developed model to a new trial. We found that new models do not need to be developed for risk assessment in new trials; prior models with recalibration will suffice.