At the American Association for Cancer Research annual meeting in Washington DC last week, a recurrent theme was complexity. The deeper scientists have delved into the fundamental nature of cancer, the more they have come to recognize its vast genetic diversity, which can make even tumours of the same cancer type seem unrelated.

It is encouraging to see researchers embracing new methods to deal with that complexity. One especially promising technique highlighted in several talks was the 'adaptive' clinical trial, which allows researchers to avoid being locked into a single, static protocol for the duration of the trial. Instead, investigators can evaluate data as they come in, and use that information to change a trial's structure (see page 1258).

Such flexibility is particularly important in cancer research. Investigators have struggled to plan clinical trials that can incorporate an ever-proliferating list of molecular biomarkers — features such as mutations or gene-expression patterns that can be used to distinguish one person's tumour from the next person's. The idea is to find markers that can identify a subset of people for which a given therapy is working, even when it seems to have little effect on the patient population as a whole. Adaptive trials thus allow investigators to analyse the data midstream, correlate those results with known biomarkers, and then alter the course of the trial in light of that information — perhaps by enrolling additional people with cancer on the basis of their biomarker status. The trial then continues, targeting those people most likely to benefit.

Researchers have dabbled with such experiments for years. But it was only in 2007 that the European Medicines Agency (EMA) published a paper outlining appropriate adaptive-trial conduct, and only in February 2010 that the US Food and Drug Administration (FDA) followed suit with its draft guidance to industry. The FDA draft is open for public comment until 1 June.

Undertaking adaptive clinical trials is an experiment in itself: there is much still to be learned about the unforeseen pitfalls, and what the best practices are.

Guidance from these key agencies is likely to spur interest in the trials, but it will take more than that to fully exploit the technique's potential. Because adaptive trials require more statistical sophistication than conventional ones, for example, medical researchers will have to work with statisticians on trial design from the outset. And because the trials require extra months of planning time, pharmaceutical companies will have to revise the common practice of financially rewarding the speed with which a clinical-trial team enrols its first patient, to avoid tempting researchers to shortcut the planning.

And there are potential pitfalls with adaptive trials. Perhaps most importantly, researchers who use adaptive techniques will have to take care that they do not fool themselves. By changing the patient population to one that is more likely to benefit from the treatment, they can inflate the risk of reaching a false-positive conclusion. They also run the risk that the interim data required to take that step might compromise a trial's double-blind safeguards, influence patient and investigator behaviour, and colour the results even further. For these reasons, the FDA and the EMA advise against using an adaptive trial when a standard trial will do, and urge extra caution in designing the late-stage clinical trials that are crucial to determining drug approvals.

The two documents also outline the statistical methods that can be used to control the false-positive rate, and call for appointing an independent body to handle the interim review. But these guidelines cannot be taken as the final word. Undertaking adaptive clinical trials is an experiment in itself: there is much still to be learned about the unforeseen pitfalls, and what the best practices are. As always in science, investigators must be ready to adapt their approaches as the data come in.