Reducing Participation Bias in Case-Control Studies: Type 1 Diabetes in Children and Stroke in Adults

Background: Case-control studies have been used extensively in determining the aetiology of rare diseases. However, case-control studies often suffer from participation bias in the control group, resulting in biased odds ratios that cause problems with interpretation. Participation bias can be hard to detect and is often ignored. Methods: Population data can be used in place of the possibly biased control group, to investigate whether participation bias may have affected the results in previous studies, or in place of controls in future studies. We demonstrate this approach by reanalysing and comparing the results of two case-control studies: Type 1 diabetes in Yorkshire children and stroke in Indian adults. Findings: Using population data to represent the control groups reduced the width of the confidence intervals given in the original studies and confirmed the findings for the two diabetes risk factors used; caesarean birth (odds ratio (OR) = 2.12 (1.53, 2.95) compared with 1.84 (1.09, 3.10)) and amniocentesis (OR = 3.38 (2.09, 5.47) compared with 3.85 (1.34, 11.04)). The three stroke risk factors investigated were found to have increased odds ratios when using population data; hypertension (OR = 5.645 (5.639, 5.650) compared with 3.807 (2.114, 6.856)), diabetes (OR = 12.212 (12.200, 12.224) compared with 3.473 (1.757, 6.866)) and smoking (OR = 5.701 (5.696, 5.707) compared with 2.242 (1.255, 4.005)). Interpretation: Participation bias can greatly affect the results of a study and cause some potential risk factors to be over-or underestimated. This approach allows previous studies to be investigated for participation bias and presents an alternative to a control group in future studies, while improving precision.