Falsification of clinical trial data can hinder drug development, damage reputations and, potentially, harm patients. Many efforts focus on detecting fraud but biopharm and CROs (contract research organisation) can take steps to decrease the likelihood of data falsification before a study even starts.
“Two potentially successful preventive strategies are simplifying the eligibility criteria for the trial and reducing the numbers of variables being recorded”, Scotland-based statisticians wrote in the journal Significance.
In general the statisticians recommend focusing “on the data items that are most easily affected by fraud”. Suggested items to focus on include adverse events, compliance data, patient diaries and eligibility criteria, such as the age, sex and medical history of patients.
Location can also help discourage data falsification. “In large clinical trials, embedding assessments as far as possible within routine clinical practice is one way to encourage adherence to the protocol by triallists”, the statisticians wrote.
Although industry could do more to prevent fraud reported cases of data fabrication are rare. Of the 22 trials recorded in the UK ethics review database from 1997 to 2011, three relate to potential data fabrication.
Also, the impact of fraud is minimised by randomisation and blinding, meaning the conclusions of a trial can still be valid. The exception is when the “fraud subverts the randomisation process or the blinded assessment of patient outcomes”, the authors wrote.
Detecting fraud can be as simple as spotting different handwriting in patient diaries but statistical approaches are also used. It is very hard “to make a set of several interrelated measurements look genuine”, the authors write, and this makes it possible to detect data fabrication.
Conventional statistical analysis may focus on outlying data that can have excessive influence over conclusions. However, when detecting data fraud data points that are too close to the mean are also of interest as those fabricating findings are likely to choose values that appear feasible.