Heterogeneity of design features in studies included in systematic reviews with meta-analysis of cognitive outcomes in children born very preterm.

BACKGROUND

Meta-analyses of the voluminous scientific literature on the impact of very preterm (VPT, <32 weeks' gestation) birth on cognition find a marked deficit in intelligence quotient (IQ) among children born VPT relative to term-born peers, but with unexplained between-study heterogeneity in effect size.

OBJECTIVES

To conduct an umbrella review to describe the design and methodology of primary studies and to assess whether methodological heterogeneity affects the results of meta-analyses.

DATA SOURCES

Primary studies from five systematic reviews with meta-analysis on VPT birth and childhood IQ.

STUDY SELECTION AND DATA EXTRACTION

Information on study design, sample characteristics and results was extracted from studies. Study features covered study type, sample size, follow-up rates, adjustment for social context, management of severe impairments and test type.

SYNTHESIS

We used random-effects subgroup meta-analyses and meta-regressions to investigate the contribution of study features to between-study variance in standardised mean differences (SMD) in IQ between groups.

RESULTS

In 58 cohorts (56%), children with severe impairments were excluded, while 23 (22%) cohorts accounted for social factors. The least reported feature was the follow-up rate (missing in 38 cohorts). The largest difference in SMDs was between studies using full scale IQ tests (61 cohorts, SMD -0.89, 95% CI -0.96, -0.82) versus short-form tests (27 cohorts, SMD -0.68, 95% CI -0.79, -0.57). The proportion of between-study variance explained by the type of test was 14%; the other features explained less than 1% of the variance.

CONCLUSIONS

Study design and methodology varied across studies, but most of them did not affect the variance in effect size, except the type of cognitive test. Key features, such as the follow-up rate, were not consistently reported limiting the evaluation of their potential contribution. Incomplete reporting limited the evaluation of the full impact of this methodological diversity.