Premature trial discontinuation often not accurately reflected in registries: comparison of registry records with publications

abstract:
Premature trial discontinuation often not accurately reflected in registries: comparison of registry records with publications
 

Background

One quarter of randomized clinical trials (RCTs) are prematurely discontinued and frequently remain unpublished. Trial registries can document whether a trial is ongoing, suspended, discontinued, or completed and therefore represent an important source for trial status information. The accuracy of this information is unclear.

Objective

To examine the accuracy of completion status and reasons for discontinuation documented in trial registries as compared to corresponding publications of discontinued RCTs (reference standard), and to investigate potential predictors for accurate trial status information in registries.

Methods

We conducted a cross-sectional study comparing information provided in publications (reference standard) to corresponding registry entries. First, we reviewed publications of RCTs providing information on both discontinuation and registration. We identified eligible RCT publications through systematic searches of MEDLINE and EMBASE (2010-2014) and an international cohort of 1017 RCTs initiated between 2000 and 2003. Second, pairs of investigators independently and in duplicate extracted data from publications and corresponding registry records. Third, for each discontinued RCT, we compared publication information to registry information. We used multivariable regression to examine whether accurate labeling of trials as discontinued (vs. other status) in the registry was associated with recent initiation of RCT, industry sponsorship, multicenter design, or larger sample size.

Results

We identified 173 publications of RCTs that were discontinued due to slow recruitment (55%), harm (16%), futility (11%), benefit (5%), other reasons (3%), or multiple reasons (9%). Trials were registered with clinicaltrials.gov (77%), isrctn.com (14%), or other registries (8%). Of the 173 corresponding registry records, 77 (45%) trials were labeled as discontinued and 57 (33%) provided a reason for discontinuation (of which 53, 93%, provided the same reason as in the publication). Labeling of discontinued trials as “discontinued” (vs. other label) in corresponding trial registry records improved over time (adjusted odds ratio 1.16 per year, confidence interval 1.04-1.30) and was possibly associated with industry sponsorship (2.01, 0.99-4.07) but unlikely with multicenter status (0.81, 0.32-2.04) or sample size (1.07, 0.89-1.29).

Conclusions

Trial status information matched in less than half of registry records of RCTs in our sample with corresponding publications that explicitly mentioned trial discontinuation; one third of registry records provided a reason for discontinuation. Current trial status information in registries should be viewed with caution.