Prediction intervals should be included in meta-analyses published in dentistry

Abstract

This focus article aims to highlight the value of reporting prediction intervals (PIs) in random effects meta-analysis and to assess the prevalence of PI-reporting in periodontology and implant dentistry meta-analyses. We searched in the PubMed database for meta-analyses published in the fields of periodontology and implant dentistry. We selected meta-analyses related to primary outcomes with at least three trials. Additionally, we extracted information on the type of the meta-analysis model (fixed or random) and whether the random effects meta-analyses included PIs in addition to the 95% confidence intervals. Three-hundred and forty-nine meta-analyses were found in 94 systematic reviews. Two-hundred and sixty-three (75.4%) subgroup and full meta-analyses used the random-effects model, 81 (23.2%) used fixed-effect methods, and 5 (1.4%) did not specify the model used. In 75 systematic reviews, we found 231 meta-analyses with three or more trials (173 full meta-analyses and 58 subgroup meta-analyses). Only one systematic review reported PIs. Interpretation of the results of random effects meta-analyses which ignore heterogeneity can be misleading. Heterogeneity should be explored, and two common approaches include subgroup analyses and meta-regression. Random effects meta-analyses should include PIs because they convey the extent of heterogeneity in treatment effects across studies in a clinically relevant context.