CANCER PATIENT AND COMPLETING NEW DRUG TIRALS
If an agent shows encouraging activity in phase 2 with acceptable toxicity, it will then proceed to phase 3 trials in which the agent is compared to the current standard of care. Where the agent is a new drug, this will generally involve the drug company discussing the trial with the regulatory organizations such as the UK Medicines and Healthcare Regulatory Agency (MHRA), European Medicines Agency (EMA), and the US Food and Drug Administration (FDA). These bodies will have an opinion as to the appropriate comparator treatment and also the outcome required to obtain a licence. The comparator may be an existing drug or combination of drugs, or it may be what is termed ‘best supportive care’. This latter option is chosen when there is no clear-cut standard therapy – patients receive whichever palliative measures the clinician thinks appropriate.
The hallmark feature of phase 3 trials is that the patients are randomly assigned between the treatment options. This ensures that patients will be evenly distributed between the various arms of the trial and minimizes the risk of differences in outcomes arising due to patients with a better or worse prognosis being concentrated in one arm of the trial. Whilst the design makes good scientific sense and is regarded as the ‘gold standard’ method of assessment, as always there are limitations.
Firstly and most obviously, where the control arm is best supportive care or worse still a placebo medication, there is understandable reluctance on the part of patients. Careful explanation and support are clearly required, particularly to make the point that if there is no other proven alternative, then treatment outside the trial will be no different to the control arm. Often, however, a phase 3 trial is not comparing the new drug with placebo but with the current standard therapy. This is generally a much easier discussion in the clinic as everyone receives treatment and the new medicine may be less good than the old one – we don’t know until we do the trial. Even if the control is placebo, it is by no means a given that the new drug will turn out better – there are plenty of examples of trials in which the drug was no better than placebo, and even examples when the drug was worse – the drug was both toxic and ineffective.
Secondly, most new medicines will be only a little better than the existing ones, hence the likely differences between the trial arms will be small. In order to detect small differences, large sample sizes are necessary to ensure statistical confidence in the outcomes. Statistics is a much mocked, maligned, and misunderstood science, so it is helpful to illustrate why sample sizes need to be big with a simple example. Suppose we want to assess whether a coin used for a coin toss is evenly balanced or biased to either heads or tails. If we toss once, then we get either heads or tails (ignoring the possibility that the coin balances on its edge!). If we toss again and get the same, we have (say) 100% heads, 0% tails. No one would say the coin was based on this size of the sample, though. Suppose we carry on and get to 10 tosses – 6 heads, 4 tails – would we be confident that the coin was biased? Probably not. However, if we get to 100 tosses with 60 heads and 40 tails, or 1,000 tosses with 600 heads and 400 tails, we would have increasing confidence that the coin was indeed biased. The reverse of the problem is more difficult: if we got 501 versus 499, would we say the coin was biased? Again, probably not, but how about 510 versus 490? 520 versus 480? How similar can the numbers be in order that the difference is probably by chance rather than due to a biased coin? Even a big difference like 600 versus 400 can occur by chance with an unbiased coin, but would be very unlikely. The statistics plan for a trial is therefore a key and will specify how many patients will be needed to reliably detect the minimum difference deemed to be clinically important in advance of the trial starting. For a trial testing a new drug in advanced cancer, this will be along the lines of an average improvement in survival of at least three months. As with our coin flip, this could arise by chance so the trial statistician will calculate how many patients are needed to show (or exclude) this difference reliably – usually defined (largely arbitrarily) as the chance result occurring fewer than 1 in 20 times.