This initial electronic search produced 1045 articles and 33 were identified in the manual search. After eliminating duplication, examining, and applying inclusion criteria, 78 articles were included for data extraction and full-text evaluation. However, 22 articles were excluded because they did not meet the objectives of the review or they did not have a clear methodology. Therefore, a total of 56 articles [20-75], were selected as relevant to the objectives of the review. The PRISMA flowchart in Figure 1 synthesizes the screening and selection processes. The study designs were as follow: 18 cross-sectional studies, 18 longitudinal studies, 1 case-control study, 16 cohort studies, and 3 randomized controlled trials.
In order to reduce the heterogeneity of the results and to facilitate their interpretation, the studies were grouped according to diagnostic criteria into four groups (Tables 1-4). One study  was included in groups 1 and 2.
The peri-implantitis mean prevalence obtained was 19.6% (95% CI, 15.25 to 24.75%) at the patient-level and 12.39% (9.46 to 15.43%) at the implant-level. By diagnostic criteria groups, for group 1 (BOP + loss of supporting bone ³ 3mm), 16.4% (0.9 to 31.89%) at the patient-level and 12.12% (2.96 to 21.29%) at the implant-level, while for group 2 (BOP + loss of supporting bone ³ 2mm), 20.67% (15.89 to 25.44%) at the patient-level and 12.65% (8.98 to 16.31%) at the implant-level. Table 5 shows all the results of the study. Given the high specific weight of group 2 compared to the other groups (54.39%), the totals were calculated by weighted average.
In addition, an analysis of the influence of the load time or time variable was carried out based on implants on the registered peri-implantitis prevalence at the patient level and at the implant level, that is displayed in Table 6. No significant differences were observed in prevalence among studies with follow-up period of 5 to 9 years and studies with greater longevity, both at patient-level (17.19% and 17.75% respectively) as at implant-level (11.11% versus 9.43%).
Considering the Consensus report of the 2017 World Workshop on the classification of periodontal and peri-implant diseases and conditions that recommended that probing depth should not be included as a diagnostic criterion, studies have been divided according to this variable to study its impact on peri-implantitis prevalence (Table 7). Prevalence in studies that used probing depth, as one more diagnostic criterion was higher than those that do not used it, both at patient-level (25.26% and 17.17% respectively) and at implant-level (15.66% and 11.07% respectively). However, no significant differences were observed (p = 0.24 and p= 0.26 respectively).
A meta-analysis of group 2 was performed due to its large specific weight (31 articles). Heterogeneity was evaluated using the Q and I2 test . The prevalence of peri-implantitis at patient level obtained was 19.6% (CI-95%, 18.4-20.8) for the fixed effects model and 20% (CI-95%, 16.6-23.7) for the random effects model. Heterogeneity analysis in both models was high (Q = 187.04; p < 0.05; I2 = 87.169%). To prevent the presence of publication bias, those studies farther from the mean, one at a time, were eliminated, controlling with the Egger´s regression test its absence (p ³ 0.1) . After performing this sensitivity analysis excluding studies of Gatti , Koldsland , Marrone , Romandini , and Tey , the prevalence at the patient level was 18.1% (CI-95%, 16.2-19.9%). Meta-analysis found heterogeneity (Q = 34.13; p = 0.018; I2 = 44.3%). The Egger’s intercept test was 0.17 (CI-95%, 0.12-0.23; t = 6.70; df = 19; p = 0.846), indicating no small-study effects (Figure 2a, Figure 2b).
The prevalence of peri-implantitis at implant-level obtained was 12.3% (CI-95%, 11.7-12.9%) for the fixed effects model and 11.5% (CI-95%, 8-15.4%) for the random effects model. High heterogeneity in both models was also observed (Q =968.47; p < 0.05; I2 = 97,21%). After performing the sensitivity analysis excluding studies of Fransson , Gatti , González-González , Koldsland , Lee , Marrone , Ravald , Rodrigo , Romandini  and Vandeweghe , the prevalence at the implant level was 9.1% (95% CI, 8.1-10.2%). Meta-analysis found heterogeneity among the studies (Q = 31.59; p = 0.017; I2 = 46.2%). The Egger’s test was 0.06 (CI-95%, 0.00-0.39; t= 5.96; df= 17; p = 0.01) indicating the presence of small-study effects (Figure 3a, Figure 3b).