Assessing bias in a prospective study of diabetes that implemented substitution sampling as a recruitment strategy

Abstract

Objective: Strategies such as reminders are frequently used to maximize baseline recruitment and for this reason are collectively termed “usual practice.” The objective of this study was to investigate substitution sampling as an alternative recruitment strategy. Study Design and Settings: Data are from the Living with Diabetes Study, which is a prospective cohort study providing a comprehensive examination of health care utilization. Baseline information was collected for 3,197 of 11,470 eligible individuals between November 2008 and October 2009. Follow-up occurred 12 months after recruitment, with outcome of interest being emergency department attendance. Biases resulting from the two recruitment programs were investigated through the comparison of adjusted logistic regression coefficients and absolute relative biases (ARBs). Results: Corresponding estimates resulting from both programs were similar except for age (75+ years). This effect was significant (β: −0.59; 95% confidence interval [CI]: −1.04, −0.13) under substitution sampling, but not under “usual practice” (β: −0.36; 95% CI: −0.78, 0.07). Analysis using the ARB metric reinforced similarity, with the Wilcoxon signed-rank test failing to detect significant difference between programs (median difference: −1.01; 95% CI: −5.88, 2.02). Conclusion: Substitution sampling deserves consideration as a recruitment option alongside “usual practice,” as concerns about additional bias may be unwarranted.