Research Projects


Forecasting risk in acute myocardial infarction

Impact of the censoring distribution on time-to-event problems in the presence of competing risks

A model to predict risk of recurrent events in the LIPID cardiovascular trial

A method to adjust for differential background treatments in long-term trials 

Models to predict breast cancer metastasis to internal mammary nodes

Forecasting risk in acute myocardial infarction

Existing short-term risk assessment strategies in acute myocardial infarction are limited to Western populations. We have proposed risk models for prediction of mortality after acute myocardial infarction based on the geographically diverse HERO-2 trial. HERO-2 randomised 17 073 patients to either unfractionated heparin or bivalirudin in conjunction with streptokinase to treat ST-segment-elevation myocardial infarction. Patients were recruited from 46 countries from Europe, Russia, North America, Latin America and Asia, including Australia and New Zealand. We also examined variations in outcomes across geographical regions and propose new methods for comparing the calibration and ranking performance of risk strategies.

Back to top

Impact of the censoring distribution on time-to-event problems in the presence of competing risks   

Methods accounting for competing risks in time-to-event problems are becoming common in mainstream statistical analyses. Standard approaches include those based on log-rank type tests (Gray) and cumulative incidence regression (Fine and Gray). These approaches are based on weighting competing events by the censoring distribution. The usual cumulative incidence regression uses weights based on the pooled censoring distribution. However, the effects of the pattern of events and censoring in these approaches are still unclear. We are examining two aspects of this problem: the amount of competing risk present (by using a proportional-hazards model) and the pattern of censoring between groups in the presence of competing risks.

Back to top

A model to predict risk of recurrent events in the LIPID cardiovascular trial

Traditional methods for analysing clinical and epidemiological data have focused on the first occurrence of the outcome or event being measured. These methods can be unsuitable for analysing recurring events because a first event may signal another one; that is, recurrent events are not independent of each other.

A new study focused on recurring events in the CTC's multicentre trial, LIPID, which had shown that lipid-lowering with a statin prevented a coronary event. We investigated recurrent events and whether risk factors were different for first and recurrent events. A semiparametric proportional-hazards model and a parametric conditional model were both found to be useful tools for exploring the biological cardiovascular process. The analysis also showed that the study drug, pravastatin, prevented first and second cardiovasacular events to a similar degree.

Back to top

A method to adjust for differential background treatments in long-term trials 

An advance in trial methods arose from difficulties in the statistical analysis of the FIELD diabetes trial. In this large international trial, 9795 patients were randomly assigned to fenofibrate or placebo and followed up for an average of 5 years. Cardiovascular outcomes were measured.

Over the 5 years of the trial, many patients started taking newly approved cholesterol-lowering drugs, confounding the effect of the study drug. FIELD investigators and CTC statisticians devised a novel method using the results of other clinical trials to adjust the estimates of efficacy of the study drug - a method with potential for wide application in long-term trials.

Back to top

Models to predict breast cancer metastasis to internal mammary nodes

An important prognostic factor in breast cancer is the status of the internal mammary lymph nodes, that is, whether there is tumour in the nodes near the middle of the chest. These nodes are less accessible than axillary lymph nodes and less likely to be visualised with radioisotope mapping or to be biopsied. Models to predict metastasis in these lymph nodes have been developed on the basis of anatomy and tumour biology. These will assist cancer clinicians to make decisions about treatment when the status of these lymph nodes is not known.

Back to top