Handle with care: changing endpoints in clinical trials

With the US Food and Drug Administration (FDA) showing more than a passing interest in ‘adaptive’ clinical trials, where study parameters such as dose selection, treatment duration or sample size are open to adjustment in the light of emerging safety or efficacy data, an article in PLoS Clinical Trials has tackled the thorny question of whether clinical endpoints should be changed once a study is underway.

For Scott Evans of the Center for Biostatistics in AIDS Research at the Harvard School of Public Health in Boston, US, the answer is a qualified ‘yes’ – although revisions to primary endpoints in particular “should be uncommon”, he says. If not appropriately evaluated, changes to endpoints “lead to misguided research and suboptimal patient care”, he warns.

Setting out endpoints in advance is a fundamental principle in the design of randomised trials, Evans points out. Failure to do so can introduce bias into a trial and lays it open to manipulation. Sometimes, though, new information – for example, results from other trials or identification of better biomarkers – may come to light that merits adjusting endpoints in the course of the trial.

Nor, indeed, is this unusual. Evans cites a study published in 2004 by Chan et al in the Journal of the American Medical Association that compared published articles with protocols for 102 randomised trials approved by the scientific-ethical committees for Copenhagen and Frederiksberg in Denmark during 1994-1995. They reported that 62% of the studies had at least one primary endpoint that had been changed, introduced or omitted.

Given this frequency, Evans writes, it is important to determine when changes to endpoints are appropriate and how they should be reported. The main consideration, he says, is whether the decision to modify an endpoint is independent of the data obtained from the trial to date. If this is so, he argues, the revisions may have some merit or even be worth encouraging. If not, then there is a danger of cherrypicking – for example, by selecting new endpoints because they show a trend towards significance and ignoring other candidate endpoints because the trend is not desirable.

Investigators and reviewers should ask three key questions to gauge whether a change in endpoint really is independent of the trial data, Evans suggests: what is the source of the new information that prompts consideration of the change (e.g., results from another trial, which may make the revision credible); whether interim data on the endpoint in question (or related data) have already been reviewed; and, “most importantly”, who is making the decision to revise the endpoint (e.g., the trial sponsors or an independent external advisory committee).

Appropriate decision-makers “should have no knowledge of the endpoint (or related trial data) results”, he stresses. Even if there has been no formal interim analysis, investigators may have a ‘sense’ of the endpoint result or a related variable, perhaps by noticing changes in certain patients at their site that could be attributed to the study drug.

Another important consideration, Evans notes, is the scientific relevance of the endpoints at issue. If the current state of knowledge makes the trial uninformative or inefficient, then changing endpoints “may be constructive, and perhaps even ethically necessary, to ensure that the study generates a scientific contribution”. Nonetheless, he cautions, there is a risk of introducing operational bias. Participating clinicians and patients may believe the change was due to lack of treatment efficacy, which can affect their willingness to carry on.

If trial endpoints are modified, appropriate documentation is crucial, Evans says. The changes should be described in amendments to the protocol and analysis plan, and declared when trial results are submitted to a medical journal. Reporting of a clinical trial with any modified endpoint should include: a clear statement that information obtained after the trial started led to the change in endpoint; a description of the reasons and decisions procedure; a discussion of any potential biases introduced by the change in endpoint; and, if the decision was not independent of the trial data, a disclaimer stating that the results should be interpreted with caution.

In some cases it may be appropriate to change or identify endpoints once a trial is underway, even when this decision is based on data from the trial, Evan adds. For example, with a very large study of long duration, the investigators may split the trial into a hypothesis-generating stage, in which endpoints are identified, and a subsequent hypothesis-testing stage. In this scenario, he notes, statistical testing would be based only on data collected after completion of the first stage.