Selection bias and the use of controls with malformations in case-control studies of birth defects

We compared four methods of control selection to assess the effect of using infants with malformations as controls in case-control studies of birth defects. We identified cases and controls using data from the Slone Epidemiology Unit Births Defect Study for the years 1976-1992. Cases were defined as infants with cleft lip and palate and no other malformations (N = 494). Controls (N = 8356) were chosen from infants with other malformations, excluding other oral cleft conditions or syndromes associated with clefts. Maternal smoking during the first 13 weeks of pregnancy was the exposure of interest. We then assessed the measures of association resulting from using controls with varying restrictions. When we excluded all defects potentially associated with maternal smoking (based on reports in the literature), the crude odds ratio for smoking and oral cleft risk was 1.6 (1.3-1.9). When we eliminated all defect groups with a smoking prevalence that was one or more standard deviations above or below the total control group mean, the odds ratio was 1.5 (1.2-1.8); with controls restricted to infants with Mendelian-inherited disorders (with presumably no causal effect of smoking), the odds ratio was 1.6 (1.1-2.7); and when selection was unrestricted, the crude odds ratio was 1.5 (1.2-1.8). When used selectively, infants with malformations other than the anomaly of interest can be a suitable source of controls.