Extent of missing outcome data for participants in trials included in five Cochrane systematic reviews: a descriptive analysis

Tags: Poster
Ramly E1, Kahale L1, Barba M2, Sperati F2, Terrenato I2, Ballout R1, Guyatt G3, Schunemann H3, Akl E1
1American University of Beirut, Lebanon, 2National Cancer Institute Regina Elena, Italy, 3McMaster University, Canada

Background:

Missing outcome data for participants in randomized clinical trials (RCTs) may bias results of meta-analyses. Moreover, it is not always clear whether the outcomes of certain categories of participants (e.g., non-compliant) are missing.

Objectives:

To describe the extent and categories of missing participant data (MPD) in RCTs included in five Cochrane systematic reviews.

Methods:

We conducted this study during the process of updating a series of five Cochrane systematic reviews on the topic of anticoagulation in patients with cancer. In duplicate and independently, pairs of reviewers abstracted the number of randomized participants belonging to the following categories:

1. excluded from follow-up pre-treatment (e.g. never received intervention; deemed ineligible);

2. likely or definite MPD (e.g. withdrew consent, lost to follow-up);

3. possible MPD (e.g. non-compliant).

For each trial, we calculated the percentage of randomized participants belonging to each of the MPD categories. Then, we calculated the mean, standard deviation (SD), and interquartile range (IQR) of the percentages across all included trials.

Results:

We included a total of 58 trials in this study. The table shows the distribution across included trials, of the percentage of randomized participants belonging to each of the MPD categories. On average, about 1% of participants were excluded pre-treatment; 5% had likely or definitely MPD; and 7% had possible MPD. The total percentage was 13.6%. The mean percentage of participants with likely or definitely MPD was statistically significantly higher in the intervention arm (5.39%) compared with the control arm (4.40%).

Conclusions:

In this sample of RCTs, the percentage of participants excluded pre-treatment was relatively small. However, the percentage of those with “likely or definitely” MPD is substantive and slightly higher in intervention arm, which may increase the risk of associated bias.