Objective: To use meta-analytic techniques to formally examine (1) publication bias, and (2) between-study heterogeneity in the reported regression coefficient of systolic blood pressure (SBP) on birth weight, adjusting for current weight.
Data identification: A total of 55 separate analyses reporting on 382 514 male and female subjects ranging from 0 to 75 years, summarized in a recent systematic review.
Results: Some 52 of 55 analyses reported a negative regression coefficient (beta) of SBP on birth weight, adjusting for current size, and in 27 of 55 (49%) the 95% confidence interval (CI) did not include zero. A pooled estimate of the regression coefficient across studies under a random-effects model was -1.38 mmHg/kg (95% CI: -1.66 to -1.10). The magnitude of the regression coefficient was negatively correlated with the age of the study population (r = -0.44, P = 0.0007). However, there was strong evidence of publication bias with smaller studies tending to report more negative betas - the Pearson correlation between beta and its standard error (SE(beta)) was -0.55 (P < 0.0001). To examine independent effects, meta analysis regressions of beta on age and SE(beta) were run. After controlling for age, there remained strong evidence of publication bias (P = 0.002), whereas after controlling for publication bias as measured by SE(beta), the relationship between beta and age was not significant (P = 0.08).
Conclusions: This analysis strongly suggests, but does not prove, that publication bias is present when systematically examining the relationship between blood pressure and birth weight. Controlling for publication bias lessens the apparent relationship between the regression coefficient and age of the study population.