Introduction Control selection is a crucial step at the study designing phase. Although, the concept of the different types of matching for control selection has been discussed in the context of case-control studies, here we targeted analytical cross-sectional studies to explore the effects of each type of control selection on the amount of bias and precision of the OR.
Methods 41 coronary atherosclerotic patients and 92 disease-free hospital controls were recruited to assess the relationship between opium consumption (OpiumHx) and coronary atherosclerosis (Outcome). Considering the OpiumHx as the main independent factor and age as the confounder, we calculated point estimate and the CI for OR in different scenarios of matching for control selection, namely exact, stratified, frequency and propensity matching. Syntaxes were developed by STATA 10.
Results The crude OR was 3.4 (95% CI 1.5 to 7.9). By exact matching on age, 21 pairs remained for the analysis and the OR was equal to 3.3 (0.8 to 18.8). Stratified matching on age group kept 41 pairs and gave us the OR of 0.9 (0.5 to 1.7). Frequency matching kept 88 subjects for the analysis and led to the OR of 3.0 (1.2 to 7.4). By propensity matching, 27 pairs remained which gave the OR of 3.5 (1.1 to 14.6).
Conclusion Matching techniques influence effect size and precision, seriously. Although the most bias reduction happened in pair matched techniques, a large reservoir of controls would be needed to prohibit immense decrease in precision. These findings should be considered at both protocol development and analysis phases of observational studies with caution.