Fig 2.

Aggregation bias when using meta-regression of study level results rather than individual participant data meta-analysis of treatment-covariate interactions. The research question was whether blood pressure lowering treatment is more effective among women than men. Evidence is shown from a meta-analysis of 10 trials of antihypertensive treatment, comparing the across trial association of treatment effect and proportion men (solid line)—which is steep and statistically significant—with participant level interactions of sex and treatment effect in each trial (dashed lines) —which are flat and neither clinically nor statistically important. This case study is based on previous work.^{47 58 60} Each block represents one trial, with block size proportional to trial size. Across trial association is denoted by gradient of solid line, derived from a meta-regression of the trial treatment effects against proportion of men, which suggests a large effect of a 15 mm Hg (95% confidence interval 8.8 to 21 mm Hg) greater reduction in systolic blood pressure in trials with only women compared with only men. However, the treatment-sex interaction based on participant level data is denoted by gradient of dashed lines within each trial, and on average these suggest only a 0.8 mm Hg (−0.5 to 2.1 mm Hg) greater treatment effect for women than for men, which is neither clinically nor statistically significant