ACR 2011: Dealing with Loss of Data in Clinical Trials

There is a lot of work involved in producing a successful clinical trial. Even the planning stage alone can take a significant amount of time. But when data is incomplete-especially in longitudinal clinical trials-it can be a real struggle for organizers.

There is a lot of work involved in producing a successful clinical trial. Even the planning stage alone can take a significant amount of time. But when data is incomplete—especially in longitudinal clinical trials—it can be a real struggle for organizers. In this morning’s session “Are You Losing It? How to Deal with Loss to Follow Up in Research Studies and Clinical Trials,” presenters Alyssa B. Dufour, MA and Robert R. McLean, DSc, MPH, spoke a bit about what to do in these situations.

First, McLean noted the common reasons for data going missing. Here are some that he mentioned:

  • Death
  • Illness
  • Disability
  • Migration

Death and illness, he said, are extremely common in studies that involve an older population, and, to a lesser extent, in longitudinal studies altogether. Disability and migration also happen with frequency mostly because many of these studies rely on volunteers. “We see a lot of times, a volunteer will get involved and then miss one session and then decide not to be a part of a study anymore, and that is their right.” He also added that, often times, people who remain in a study are generally healthier.

In the best case scenario, incomplete data is a nuisance, but, in the worst case, it reduces sample size, compromises generalizability, and can lead to bias. Durfour explained to the audience that there were three types of missing data: Completely Missing at Random (CMAR; very infrequently the case), Missing at Random (MAR), and Missing Not at Random (MNAR; the most common situation). She gave a brief overview of all three and then touched on some of the ways in which researchers try to cope with the incompleteness of their data without trying to skew the results. Single amputation was mentioned, but it should be noted that this technique was frowned upon by some of the attendees who waited for the Q&A portion at the end. This prompted McLean to stress that this method was indeed “falling out of favor.”

The best way to avoid having to even worry about which method should be taken to finish a study without having all the data would be to make sure to collect all of the data; however, this is unrealistic considering many of the variables outside of researchers’ control. But Dufour did have some tips and advice for the audience:

  • Define variables in study planning
  • Anticipate data collection needs of participants with varying health and functional abilities
  • Have a plan for alternative data collection strategies in case primary strategy fails
  • Prioritize data collection sequence
  • Measure adherence to treatment protocol
  • Develop codes for missing data
  • Be flexible with schedules, sites, and protocols
  • Be prepared to modify protocol if necessary
  • Quantify any amount of missing data (>5% is usually negligible) and then examine reasons and mechanisms

This activity is not sanctioned by, nor a part of, the American College of Rheumatology.