Diederik Hentenaar

87 Erythritol air-polishing in surgical peri-implantitis treatment in combination with implants nested in patients. With a mean group size of 2 infected implants per patient and a minimum amount of 50 patients, it was estimated to detect a medium effect size with 80% power at a significance level of α=0.05. Since our study focused on clinical relevant effects, small effect sizes were less important and detection of medium effect sizes were supposed to be sufficient for our study. According to the non-surgical peri-implantitis literature at the time of the study design, we estimated a 20% success rate for our non-surgical patient treatment phase (Muthukuru, et al., 2012). Therefore, it was assumed that 80% of the patients would need surgical follow-up. To compensate for patient withdrawal and losses to follow up (10%), a sample size of 80 patients was used at baseline. This was an intentional slight overestimation in order to assure enough available participants for the surgical phase of our design. Randomization Patients were randomly assigned to one of the two groups (test and control) following stratified randomization, taking into account the preceding non-surgically performed treatment (air-polishing/ultrasonic). Predefined generated notes with either ‘air- polishing’ or ‘conventional’ were equally divided over coded (AA, AB, etc…), identically sealed envelopes. On the day of the intervention, an operator assistant opened a coded envelop to decide which therapy to apply. Accordingly, all included implants per patients were treated with the randomized therapy. The code was written down and a decoding list saying which code belongs to which procedure was kept sealed until data analysis. This way the investigator performing the clinical assessments and data analysis (DH), which was not present at the surgical procedure, did not know which therapy was applied. Statistical analysis To analyse the difference in clinical and radiographical effects between both treatments, generalized linear mixed models (GLMMs) were used (IBM SPSS Statistical software, version 23.0. for Windows, Armonk, NY: IBM Corp). A three-level structure was chosen with patient, implant and time as level 1, 2 and 3, respectively. The patient was considered unit of analysis, whereas the implant unit of observation. First, the clinical and radiographical outcomes were analysed while controlling for the corresponding baseline parameters BoP, SoP, Plq, PPD and MBL ( i.e., crude analysis). Then, the primary and secondary outcomes were analysed while controlling for the baseline values and confounding effects ( i.e., adjusted analysis). The following a priori defined confounders were used in the adjusted mixed model: history of periodontitis (dichotome), smoking, implant surface modification (nominal), mean periodontal plaque level at T12 and mean marginal bone loss at baseline (linear). For skewed data (SoP and Plq) a gamma distribution was used. Within- group differences of the peri-implant clinical parameters 4