No consistent evidence of data availability bias existed in recent individual participant data meta-analyses: a meta-epidemiological study.

OBJECTIVES

The objective of the study was to assess trial-level factors associated with the contribution of individual participant data (IPD) to IPD meta-analyses, and to quantify the data availability bias, namely the difference between the effect estimates of trials contributing IPD and those not contributing IPD in the same systematic reviews (SRs).

STUDY DESIGN AND SETTING

We included SRs of randomized controlled trials (RCTs) with IPD meta-analyses since 2011. We extracted trial-level characteristics and examined their association with IPD contribution. To assess the data availability bias, we retrieved odds ratios from the original RCT articles, calculated the ratio of odds ratios (RORs) between aggregate data (AD) meta-analyses of RCTs contributing IPD and those of RCTs not contributing IPD for each SR, and meta-analytically synthesized RORs.

RESULTS

Of 728 eligible RCTs included in 31 SRs, 321 (44%) contributed IPD, whereas 407 (56%) did not. A recent publication year, larger number of participants, adequate allocation concealment, and impact factor ≥10 were associated with IPD contribution. We found the SRs yielded widely different estimates of RORs. Overall, there was no significant difference in the pooled effect estimates of AD meta-analyses between RCTs contributing and not contributing IPD (ROR 1.01, 95% confidence interval, 0.86-1.19).

CONCLUSIONS

There was no consistent evidence of a data availability bias in recent IPD meta-analyses of RCTs with dichotomous outcomes. Higher methodological qualities of trials were associated with IPD contribution.