Big Pharma vs. personalized medicine: Big is not necessarily better

I was at a local biotech conference today and there was a panel discussion on personalized medicine. The speakers, biotech execs and doctors, noted that while there was much interest and even current business from doctors, there has been very little interest in partnerships or acquisitions from Big Pharma. One of the speakers noted that Lipitor has some of the best evidence for clinical efficacy of drugs widely used in medicine, and yet its benefit of reducing risk of cardiovascular events from 3% to 2% still meant that 97% of patients experienced no actual benefit. Personalized medicine in its various forms aims to be more efficient than that. And surprise surprise, big pharma is not interested.

That got me thinking about trial sizes. I think in recent years there has been some expectation inflation, or expectation confusion, when it comes to trial sizes. We have gotten used to trial sizes of 1000-10000 patients run by big pharma companies testing drugs like Lipitor that are meant to be given to lots of people to provide benefits in low-frequency outcomes. When the effect is real, the p value can be very small. When there are few or no side effects in this large population, then everybody is happy, including the biostatisticians, and the drug gets approved. However, a large trial is not an inherent good in and of itself. In fact large trials should be seen as necessary evils, if they end up taking up more time and money than a smaller one. If you were throwing a Japanese-themed dinner party, for example, would you spend the time and money to make twice the amount of sushi you think you need to be 95% sure nobody is left out, or would you make 10 times the amount, just in case the Japanese national sumo team happens to show up?

Much has been made of the fact that trials for Provenge and other immunotherapies have been small. However, not one of the critics who have raised this issue in public (e.g. oncologist Howard Scher and biostatistician Thomas Fleming), seem to have thought about if the benefit of more trials outweigh their harms. They seem to be misapplying anachronistic guidelines suitable for non-customized drugs with small therapeutic windows, e.g. chemotherapy, to a different new situation of personalized medicine. But large trials and biostatistical purity are not requirements for FDA approval. Demonstration of substantial evidence of efficacy is the requirement, and there are no written rules about how that is defined. Lately, large trials have been accepted as useful tools for most small molecule chemicals to arrive at that demonstration. However, if the indication is urgent and your drug is so good that you can see effects in a smaller number of patients, then that satisfies all the clinical and regulatory requirements for approval. That happened with rituximab, which was approved in 1997 on the basis of a 166-patient noncontrolled unblinded Phase II trial. It is happening now with Provenge, where the 3-year survival rate is tripled over placebo. Now larger trials may have gotten smaller p-values, and that's nice, but one has to ask if the intellectual satisfaction of being 99.9% certain vs. being 99% certain is worth delaying drug availability and putting more patients on placebos, which are real and irreversible harms inflicted to patients. Also, one can make the case that continued testing may reveal unexpected side effects, and this is a valid point for non-life-threatening conditions, but again you could take this argument to infinite extremes, and it's perhaps unnecessary to ask for multi-year safety data in a disease with a 1-year life expectancy.