Sir Michael Rawlins: I believe we need clinical trials analysed
by Bayesian approaches alongside frequentist approaches |
We should be looking at different approaches to designing and
analysing clinical trials, Sir Michael Rawlins, chairman of the National
Institute
for Clinical Excellence, told participants during his keynote address
at the British Pharmaceutical Conference.
“The classical approach to looking at clinical efficacy is the
randomised controlled parallel group trial,” said Professor Rawlins.
The analysis of clinical trial data is conducted according to frequentist
principles,
ie, the probability of a random event occurring according to its relative
frequency, using a null hypothesis and generating a P-value. Professor
Rawlins explained that this approach is based on the work of Ronald Fisher
and two of his enemies, Jerzy Neyman and E. Pearson, in the early part
of this century. “Fisher was a genius but he was also a bully ...
and he had many vile disagreements with both Neyman and Pearson,” he
said. As a result of Fisher belittling other forms of statistical analysis,
few statisticians over the next 50 or 60 years even considered arguing
against the basis of the frequentist approach. Alternative statistical approaches
However there are alternative approaches to the design and analysis
of clinical trials, said Professor Rawlins. He explained that historical
controlled trials have almost fallen into disrepute. However many treatments
that we use are based entirely on data from these trials. He cited
insulin for diabetic ketoacidosis and thyroxine for myxoedema as examples.
Historical trials can be analysed following Bayesian principles, ie,
that probability applies to degrees of plausibility given incomplete
knowledge. The Bayesian theory is named after Thomas Bayes, a non-conformist
minister who lived in the 18th century, explained Professor Rawlins. “He
produced a book describing his approach to probability and the calculation
of probability densities.” Probability densities are another
way of looking at the effectiveness of a drug.
In 1992, Byar et al promoted criteria for accepting data from historical
trials. These are:
· There is a biological basis for the trial
· The condition is life-threatening or permanently disabling
· The condition has a known and predictable natural history
· There is no effective treatment
Professor Rawlins said that he would add “effect size is large” to
this list.
The Bayesian approach has many advantages over the frequentist approach,
he explained. It does not require a null hypothesis, it does not require
predetermined power calculations or type I and type II errors, you do
not have to have prespecified subgroup analysis and there are no P-values.
Bayes’s hypothesis (the way of calculating the probability density)
requires that you have a “prior”, that is what you believe
about the state of knowledge before the experiment. “In the case
of a historical trial the prior is the data from the historical control,” said
Professor Rawlins. But in some circumstances it can be difficult and
has led some Bayesian statisticians to resort to getting the opinions
of experts about how good they think a treatment might be before it is
given. “As you can imagine this has produced some degree of sneering
among frequentist statisticians,” said Professor Rawlins.
He explained that historical controlled trials could be more widely used
than in the past, particularly if the natural history of diseases can
be established in advance of their treatment. “It would be a great
advantage if we could do this as it would cut down the rate and the duration
of drug development substantially.” He concluded: “I believe
we need clinical trials analysed by Bayesian approaches alongside frequentist
approaches”.
A participant at the conference asked Professor Rawlins who will carry
out these studies since it will not be a cheap investment. Professor
Rawlins replied that he envisages that a company running a conventional
trial via a frequentist approach could run a Bayesian statistical analysis
in parallel. “That would be funded by public funds. I do not think
it is reasonable to expect the company to do that,” he said. He
added that the Government, through the Medical Research Council and the
Department of Health, is interested in principle in joining forces with
the pharmaceutical industry in this way.
From market to NHS
Professor Rawlins then went on to talk about getting the drug from
the marketplace to the NHS. “This is where NICE comes in.”
He summarised the roles of NICE and set out his thoughts for the future. “We
need to develop methodology better. There are some areas, particularly
looking at cost-effectiveness and guidelines, where methodological developments
need to be undertaken,” said Professor Rawlins. He told participants
that although implementation was not part of NICE’s role when it
was set up, it will now be taking on responsibility for encouraging it. “Somebody
else was going to do it, but unless someone else does it, we are a waste
of space,” said Professor Rawlins.
Another important area for NICE is public health. “We have been
asked to take over responsibility for giving advice on public health
that formerly rested with the Health Development Agency,” he said.
He predicted that this will almost double the staff and the budget of
NICE. “It offers a real opportunity for us to ensure that public
health issues come into our guidance and that public health benefits
from the experience that we have gained in the past few years, particularly
in the area of cost-effectiveness.” |
The
Pharmaceutical Journal