PARIS, France--A treatment that even modestly improves survival may be important, but most cancer trials are too small to detect such differences, Professor Richard Peto, of Oxford's ICRF Clinical Trial Service Unit, said at the Fifth International
PARIS, France--A treatment that even modestly improves survivalmay be important, but most cancer trials are too small to detectsuch differences, Professor Richard Peto, of Oxford's ICRF ClinicalTrial Service Unit, said at the Fifth International Congress onAnti-Cancer Chemotherapy.
"Cancer trials need to be 10 times larger than their currentsize," Prof. Peto urged, noting that the average cancer trialincludes no more than several hundred patients. Although the problemof false-positive results is well controlled by randomization,he said, it takes either a very large "megatrial" ora systematic overview of many different trials to address theproblem of false-negatives.
Prof. Peto cautioned against undue emphasis on subgroups, particularlyin small trials. "If you have a treatment that works, subgroupanalysis is like a machine for producing false-negative results,and, if you have a treatment that doesn't work, subgroup analysiswill likely produce false-positive results," he warned.
To illustrate, he reanalyzed the ISIS-2 trial (which proved thataspirin saves lives in an acute heart attack) according to patients'zodiac signs. Even though the ISIS-2 trial is large and its overallresults are highly significant, this subgroup analysis "revealed"that aspirin apparently failed to benefit patients born underLibra or Gemini.
Often, the ideal target for a cancer trial, Prof. Peto said, isseveral thousand patients. "This isn't going to be cutting-edgecancer research, but it is going to give useful information aboutwidespread cancer treatments," he said. "First, aska good, relevant question, and then radically simplify every aspectof the trial," he advised. "Minimize work, maximizesize."
To get these large numbers, entry criteria cannot be excessivelystringent. Prof. Peto suggested use of the "uncertainty principle"--if,and only if, the physician and patient are both substantiallyuncertain about the appropriateness of each of the trial treatments,then the patient should be considered eligible.
He described a formula that has proved successful in large-scalecardiovascular trials, which includes telephone randomizationwith no entry forms and simple 1-page forms completed just fromclinical notes.
"There's a tendency today for trials to be made more complicatedin terms of procedures, documentation, and audits," Prof.Peto observed. He decried the new massive audit rules imposedlast year by Congress and the National Cancer Institute in responseto the widely publicized auditing problems uncovered at the NSABP(National Surgical Adjuvant Breast and Bowel Project).
"In my view, this is entirely inappropriate," he said."The amount of auditing that needs to be done for a trialis a scientific question that should be discussed by those designingthe trial at the design stage."
Prof. Peto went on to say that Congress does not dictate to scientistshow big their trials ought to be--that is part of the scientificprocess--and, similarly, Congress should not be telling researchershow much auditing there should be. "I think these kinds ofexternal impositions on the design of trials are destructive,"he said.