Data-sharing and trustworthiness of trials evaluating cervical ripening in induction of labour: a meta-epidemiological study of randomised controlled trials

Abstract

Background: Induction of labour is a common medical intervention in obstetrics, but its uptake is highly variable in well-resourced and low-resourced settings. Cervical ripening is the initial phase of prepping the cervix when it is not 'ready'. Approximately 30% of women are induced in high-income countries, while this proportion is generally lower in low-resource settings. One way to assess the trustworthiness of randomised controlled trials (RCTs) is by examining individual participant data (IPD), but data sharing remains a significant challenge. It is still unclear whether there are substantial differences in trial characteristics, trustworthiness, effect sizes, and the certainty of evidence between RCTs that share data and those that do not.

Methods: In this meta-epidemiological study of RCTs comparing different methods of induction of labour, we included nine IPD meta-analyses. RCTs in the shared and non-shared groups were assessed on the following criteria: trial characteristics, trial registration, trustworthiness (independently evaluated via the Trustworthiness in RAndomised Controlled Trials [TRACT] checklist by two investigators), excessive similarity or difference in baseline characteristics beyond what would be expected by chance, and statistical analysis results that cannot be reproduced with summary data. The TRACT checklist aims to identify and triage RCTs at risk of trustworthiness issues, and includes seven domains: governance, author group, plausibility of intervention usage, timeframe, drop-out rates, baseline characteristics, and outcomes. We performed random-effects meta-analyses separately for the two groups, followed by the GRADE approach, and compared their effect estimates using ratio of odds ratios (ROR). This study was prospectively registered at the Center for Open Science OSF registries network (WU93A) before data extraction.

Findings: Of 265 eligible RCTs (65,115 women), 64 (24.2%) trials shared data, while 201 (75.8%) did not. Adequate trial registration (after 2010) was found in 44.0% (22/50) of shared RCTs vs 14.1% of (14/99) non-shared RCTs (p < 0.001). In the shared group, 84.4% (54/64) were considered to have no trustworthiness concerns, compared to 49.1% (87/177) in the non-shared RCTs (p < 0.001). Risk of bias assessment revealed that 20.3% (13/64) of shared RCTs had an overall high risk, compared to 28.4% (57/201) of non-shared RCTs (p = 0.20). In the shared RCTs, simulation-generated p-value distributions of baseline characteristics were likely consistent with the expected uniform distribution (p = 0.50), but not in the non-shared RCTs (p = 0.006). Proportions of RCTs with at least one inconsistently reported p-value that cannot be reproduced with summary data were not significantly different between shared and non-shared groups (for baseline characteristics: 17.2% vs 25.4%; for caesarean delivery: 42.2% vs 30.8%; for uterine hyperstimulation: 17.2% vs 12.9%, respectively). Non-shared RCTs showed exaggerated effect estimates compared to the shared group, with statistically significant RORs ranging from 2.36 to 1.29 in three out of nine induction of labour comparisons. The GRADE assessment showed higher certainty of evidence for nine effect estimates and equal certainty for seven when comparing the shared groups to non-shared groups across all IPD meta-analysis projects on caesarean section and hyperstimulation. This was more pronounced among RCTs without trustworthiness concerns (than in RCTs with trustworthiness concerns), demonstrating greater certainty of evidence of RCTs without such trustworthiness issues.

Interpretation: RCTs on labour induction without IPD-sharing are more likely to have lower quality, more trustworthiness concerns, and exaggerated effect estimates than those with shared IPD. IPD-sharing and IPD meta-analyses should be encouraged. Trustworthiness and quality assessment should be prioritised whenever using RCTs for evidence synthesis and clinical guidelines development to ensure better evidence informs clinical practice.

Funding: This study is supported by three NHMRC Investigator Grants for BWM, WenL, and DLR.

Keywords: Cervical ripening; Data sharing; Induction of labour; Quality of research; Trustworthiness.

Fig. 1

Risk of bias among randomised controlled trials evaluated in this study. a. Shared group; b. Non-shared group; c. RCTs meeting trustworthiness criteria (shared + non-shared); d. RCTs not meeting trustworthiness criteria (shared + non-shared). RCT: Randomised controlled trials. The Cochrane risk of bias-2 tool was used.

Fig. 2

The cumulative distribution of Monte Carlo simulations generated p-values for continuous baseline characteristics. a. Shared group: 166 variables from 50 RCTs, p = 0.50∗; b. Non-shared group: 466 variables from 138 RCTs, p = 0.006∗; c. RCTs meeting trustworthiness criteria (shared + non-shared): 137 variables from 41 RCTs in the shared group, p = 0.34∗ (blue line) and 288 variables from 79 RCTs in the non-shared group, p = 0.37∗ (light green line); d. RCTs not-meeting trustworthiness criteria (shared + non-shared): 29 variables from nine RCTs in the shared group, p = 0.92∗ (dark green line) and 175 variables from 58 RCTs in the non-shared group, p = 0.003∗ (red line). RCT: randomised controlled trials. ∗According to the Kolmogorov–Smirnov test.

Fig. 3

The cumulative distribution of the calculated p-value/reported p-value ratio for inconsistently reported baseline characteristics and outcomes. a. Shared group; b. Non-shared group; c. RCTs meeting trustworthiness criteria (shared + non-shared); d. RCTs not-meeting trustworthiness criteria (shared + non-shared). RCT: Randomised controlled trials.