From 1981 to 1984, I spent 3 great years at Harvard School of Public Health (HSPH), completing a Masters of Science in Biostatistics and undertaking research for a subsequent doctorate in medicine. I was fortunate to have Professor Marvin Zelen as my supervisor and mentor, and to work with many other inspiring teachers and friends. The excellent research opportunities at HSPH and the Dana Farber Cancer Institute (DFCI) led to important initiatives that have both launched my research career and influenced many others.

The first of these was the problem of publication bias in clinical trials. Early in my time at Harvard I compared the benefits and harms of single-agent or combination chemotherapy in ovarian cancer as a decision analysis assignment. In reviewing the trial evidence, I discovered that the combined results and conclusions were quite different depending on whether trials were selected from the published literature or from a register of trials (CLINPROT): the published trials contained more results in favour of combination therapy than the unpublished trials. The example subsequently led me to call for the prospective registration of all clinical trials as a solution to publication bias. Twenty years later, prospective registration of all clinical trials is becoming a reality.

This same decision-analysis problem was the beginning of research interests in assessing trade-offs in outcomes in trials: using statistical decision theory and other approaches such as quality-adjusted survival analysis to combine trial outcomes of quality and duration of life and using utility assessment to more formally assess trade-offs in these outcomes. Today, we are using these approaches to routinely incorporate health economic assessments into current clinical trials.

Another problem I encountered at Harvard related to interpreting multiple comparisons in clinical trial outcomes. I introduced a less-conservative extension of the Bonferroni procedure – the Simes test, which became a basis for other tests, including the Hochberg and Hommel procedures. The issue of multiplicity of trial data is just as important today. An alternative approach to using these statistical tests to interpret multiple comparisons is to seek confirmatory evidence for specific hypotheses from similar studies– based on systematic reviews of prospectively registered trials.

A few years after returning to Australia I became the founding director of the NHMRC Clinical Trials Centre, established from a model inspired by times and experiences at Harvard. Over the past 20 years our Centre has grown to about 140 staff and we have collaborated with many investigators in Australia and internationally in undertaking clinical trials research, particularly in cancer and cardiovascular disease. This has seen over 60,000 patients recruited to more than 100 clinical trials. Examples of major milestones from trial results over this period have included: better survival for patients with coronary heart disease and cancers (breast, bowel, testicular); improvements in quality of life for patients in cancer trials; and reduced cardiovascular disease events in patients with CHD and with diabetes. There have also been many examples of research into how we can design or undertake better clinical trials that are more likely to change practice.

We have a growing team of researchers at the NHMRC Clinical Trials Centre supporting clinical trials research and undertaking methodological research in relation to these studies. In addition to the earlier examples, I will also illustrate some more recent examples involving prospective meta-analysis and the problem of dealing with non-adherence or contamination on long-term trials. A central theme in each case has been to see each problem we face when undertaking or interpreting clinical trials as an opportunity for further research: an approach born out of beginnings at Harvard.