Bias from historical control groups used in orthodontic research: a meta-epidemiological study

Abstract

The validity of meta-analysis is dependent upon the quality of included studies. Here, we investigated whether the design of untreated control groups (i.e. source and timing of data collection) influences the results of clinical trials in orthodontic research. This meta-epidemiological study used unrestricted literature searching for meta-analyses in orthodontics including clinical trials with untreated control groups. Differences in standardized mean differences (ΔSMD) and their 95% confidence intervals (CIs) were calculated according to the untreated control group through multivariable random-effects meta-regression controlling for nature of the interventional group and study sample size. Effects were pooled with random-effects synthesis, followed by mixed-effect subgroup and sensitivity analyses. Studies with historical control groups reported deflated treatment effects compared to studies with concurrent control groups (13 meta-analyses; ΔSMD = -0.31; 95% CI = -0.53, -0.10; P = 0.004). Significant differences were found according to the type of historical control group (based either on growth study or clinical archive; 11 meta-analyses; ΔSMD = 0.40; 95% CI = 0.21, 0.59; P < 0.001). The use of historical control groups in orthodontic clinical research was associated with deflation of treatment effects, which was independent from whether the interventional group was prospective or retrospective and from the study's sample size. Caution is warranted when interpreting clinical studies with historical untreated control groups or when interpreting systematic reviews that include such studies. PROSPERO (CRD42015024179). None.