Systematically missing data in causally interpretable meta-analysis.