Changing what you can measure: endpoints and stuff

The Avandia (rosiglitazone)story covered at theheart.org and elsewhere, has been portrayed as a big bad pharma versus heroic meta-analysis contest. While GSK has not covered themselves in glory, the FDA conclusion that there is a defensible case for Avandia remaining on the market - but with higher profile warnings and caveats - seems reasonable.
One major casualty is likely to be the reliance in clinical trials of easy to measure surrogate outcomes, in Avandia's case blood glucose. The glitazones are good at controlling blood glucose levels, and that is in part how they captured such a large part of the diabetes market. But it turned out that their effect on clinical outcomes heart attacks and strokes was more equivocal, and in the case of heart failure had an adverse effect.
The use of surrogate endpoints has, in the past, come up trumps. Treatments to reduce high blood pressure and high blood cholesterol, were licensed, marketed and widely used before there was evidence that they had effects on hard clinical outcomes. Lowering blood LDL as a rationale for using statins was controversial in its day. But large scale clinical trials such as the Heart Protection Study later found important clinical benefits too.
It is at least questionable whether the statins would now have been licensed without showing a reduction in hard endpoints such as heart attacks and strokes. That would have witheld their benefits from thousands of people. So a more cautious approach to the evidence on which drugs are approved for use has drawbacks.
There was criticism of the speed at which glitazones began to be prescribed, particularly in north America, without evidence of benefits in clinical outcomes rather than biochemical outcomes. The speed of uptake of statins was much slower, and really only picked up after the first large scale clinical trials showed benefits, initially in higher risk patients and then more widely.
This is perhaps one lesson. We can choose to accept evidence from surrogate outcome studies in licensing drugs if we accept that it is not then appropriate for treatments to be heavily marketed before they have proved themselves in terms of harder clinical endpoints. Or we can decide not to accept surrogate outcomes, and delay the introduction of treatments possibly by many years (and may prevent the development of some treatments altogether).
It's not an easy choice to make. Either way there are risks - from unexpected adverse events, or from the unavailability of possible treatments.

What's inside the box.: the future of clinical trials

Last week I reported for a newspaper on a huge trial on different approaches to diabetes screening published in the Lancet. It looked at 8 different ways of screening for type 2 diabetes in a population of 325,000 people over 50 years.
Except it didn’t really. It used a computer model, the Archimedes model, built up over many years. The model has had in the past, some fantastic successes, predicting the results of the CARDS trial with great accuracy, for example.
But there are times when for all the hundreds of thousands of data points, and the most sophisticated algorithms in the model, it gets things wrong. The model did not predict the results from the Illuminate trial that resulted in Pfizer stopping development on torcetrapib, its drug to reduce HDL levels, because of unexpected adverse results.
There is still a lot about the complex interaction of treatments and human physiology that we don’t understand and where we have to use empirical evidence.
Developing computer models of human physiology and the likely effects of medicines (or screening strategies) is likely only to increase as computing power grows. The Lancet trial would have been impossible to do in real life. It would have had to have begun before I was born and would have cost tens of millions of pounds to carry out. It came up with some results which should inform public policy on managing the diabetes epidemic.
But there will be fewer and fewer people who can understand what is going on inside the box. What are the assumptions made, what is the integrity of the data? Today a reasonably informed person can, with a little work, read published clinical trials and come to a sensible conclusion about their quality. If the findings come as the result of the workings of arcane formulae, and manipulation of complex datasets, there is much less opportunity for a non-specialist to be able to make any kind of judgement. I have seen the future of clinical trials, and it worries me.