Abstract
Objectives
Clinical trial registration is widely recommended, since it allows tracking of trials that helps ensure full and unbiased reporting of their results. Aim of the present overview was to provide empirical evidence on bias associated with trial registration via a meta-epidemiological approach.
Study Design and Settings
Six databases were searched in September 2017 for randomized clinical trials and systematic reviews thereof assessing the effects of orthodontic clinical interventions. After duplicate study selection and data extraction, statistical analysis included a two-step meta-epidemiological approach within- and across-included meta-analyses with a Paule-Mandel random-effects model to calculate Differences in Standardized Mean Differences (ΔSMD) between registered and unregistered trials and their 95% Confidence Intervals (CI), followed by subgroup and sensitivity analyses.
Results
A total of 16 meta-analyses with 83 trials and 4988 patients collectively were finally included, which indicated that registered trials reported less beneficial treatment effects than unregistered trials (ΔSMD=-0.36; 95% CI=-0.60,-0.12). Although some small-study effects were identified, sensitivity analyses according to precision and risk of bias indicated robustness.
Conclusions
Signs of bias from lack of trial protocol registration were found with non-registered trials reporting more beneficial intervention effects than registered ones. Caution is warranted by the interpretation of non-registered randomized trials or systematic reviews thereof.
Clinical trial registration is widely recommended, since it allows tracking of trials that helps ensure full and unbiased reporting of their results. Aim of the present overview was to provide empirical evidence on bias associated with trial registration via a meta-epidemiological approach.
Study Design and Settings
Six databases were searched in September 2017 for randomized clinical trials and systematic reviews thereof assessing the effects of orthodontic clinical interventions. After duplicate study selection and data extraction, statistical analysis included a two-step meta-epidemiological approach within- and across-included meta-analyses with a Paule-Mandel random-effects model to calculate Differences in Standardized Mean Differences (ΔSMD) between registered and unregistered trials and their 95% Confidence Intervals (CI), followed by subgroup and sensitivity analyses.
Results
A total of 16 meta-analyses with 83 trials and 4988 patients collectively were finally included, which indicated that registered trials reported less beneficial treatment effects than unregistered trials (ΔSMD=-0.36; 95% CI=-0.60,-0.12). Although some small-study effects were identified, sensitivity analyses according to precision and risk of bias indicated robustness.
Conclusions
Signs of bias from lack of trial protocol registration were found with non-registered trials reporting more beneficial intervention effects than registered ones. Caution is warranted by the interpretation of non-registered randomized trials or systematic reviews thereof.
| Original language | English |
|---|---|
| Journal | Journal of Clinical Epidemiology |
| Early online date | 26 Apr 2018 |
| DOIs | |
| Publication status | E-pub ahead of print - 26 Apr 2018 |
Keywords
- Randomized clinical trials
- Metaepidemiology
- Meta-analysis
- Protocol registration
- Empirical bias