Multi-center randomized clinical trials in oral and maxillofacial surgery: modeling of fixed and random effects

The multi-center randomized clinical trial (MCCT) is an important tool to evaluate treatment of rare diseases. An important and challenging analytic consideration is how to model the variability of the set of clinical centers composing an MCCT. The purpose of this paper was to demonstrate how changing the assumptions regarding the variability (fixed effect versus random effect) of the set of clinical centers may alter the results. The data for this paper were derived from a recently completed MCCT. The MCCT was designed as a prospective, randomized clinical trial comparing the stability of two techniques, i.e., wire versus rigid internal fixation (RIF), for stabilizing the mandible after bilateral sagittal split ramus osteotomy (BSSO) for patients requiring mandibular advancement. Three treatment centers were involved. The key outcome variable was change in mandibular position (B-point) over time. We developed two different analytic models by varying the underlying statistical assumptions regarding the variability of the clinical treatment centers, i.e., random or fixed effects. Analyses based on the random-effects model demonstrated no significant difference between treatment groups in terms of relapse (P = 0.13). With the fixed-effects model, however, wire fixation had significantly more relapse at B-point over time than RIF (P = 0.02). The results from these two sets of analyses demonstrate how changing assumptions regarding the variability of the set of clinical centers can alter the interpretation of the treatment effect. The choice of statistical modeling of the set of clinical centers is an important consideration when performing analyses of MCCTs and it is a decision that should be made prior to initiating the study.