Koen Bernardus Pouwels1, Bianca Mulder2, Eelko Hak3. 1. Unit of PharmacoEpidemiology & PharmacoEconomics, Department of Pharmacy, University of Groningen, Antonius Deusinglaan 1, 9713 AV Groningen, The Netherlands. Electronic address: k.b.pouwels@gmail.com. 2. Unit of PharmacoEpidemiology & PharmacoEconomics, Department of Pharmacy, University of Groningen, Antonius Deusinglaan 1, 9713 AV Groningen, The Netherlands. 3. Unit of PharmacoEpidemiology & PharmacoEconomics, Department of Pharmacy, University of Groningen, Antonius Deusinglaan 1, 9713 AV Groningen, The Netherlands; Department of Epidemiology, University Medical Center Groningen, University of Groningen, Hanzeplein 1, Postbus 30 001, 9700 RB Groningen, The Netherlands.
Abstract
OBJECTIVES: To empirically evaluate the concordance of effect estimates between case-only and parallel group designs and to identify predictors of discrepancies. STUDY DESIGN AND SETTING: MEDLINE and EMBASE databases were searched through June 31, 2013. Studies that used both a case-only (case crossover or self-controlled case series) and a parallel group design (cohort or case-control) were identified. Spearman correlation coefficient was used to evaluate the concordance between designs. Z-scores were used to assess whether differences in the effect estimates were common, using an absolute threshold value of 1.96. A prediction model was built to identify predictors of discrepancies. RESULTS: The search identified 1,367 articles of which 53 were included for analysis. In total, 519 comparisons were made. The correlation coefficient between case-only vs. parallel group studies was 0.64 (P < .001). In 221 of the 519 comparisons (43%), the difference between both study designs was larger than the predetermined threshold. The following predictors of discrepancy were found: intermittent exposure, rare event, acute outcome, length of hazard period, type of case-only design, and sample size (C statistic of 0.783). CONCLUSION: The concordance between effect estimates of case-only and parallel group designs is moderate. Such discrepancies could be predicted by failure to meet assumptions of case-only designs.
OBJECTIVES: To empirically evaluate the concordance of effect estimates between case-only and parallel group designs and to identify predictors of discrepancies. STUDY DESIGN AND SETTING: MEDLINE and EMBASE databases were searched through June 31, 2013. Studies that used both a case-only (case crossover or self-controlled case series) and a parallel group design (cohort or case-control) were identified. Spearman correlation coefficient was used to evaluate the concordance between designs. Z-scores were used to assess whether differences in the effect estimates were common, using an absolute threshold value of 1.96. A prediction model was built to identify predictors of discrepancies. RESULTS: The search identified 1,367 articles of which 53 were included for analysis. In total, 519 comparisons were made. The correlation coefficient between case-only vs. parallel group studies was 0.64 (P < .001). In 221 of the 519 comparisons (43%), the difference between both study designs was larger than the predetermined threshold. The following predictors of discrepancy were found: intermittent exposure, rare event, acute outcome, length of hazard period, type of case-only design, and sample size (C statistic of 0.783). CONCLUSION: The concordance between effect estimates of case-only and parallel group designs is moderate. Such discrepancies could be predicted by failure to meet assumptions of case-only designs.