Estimating the Impact of Co-localized Risk Variables on Peri-implant Diseases

Aim & Hypothesis: This study hypothesizes that peri-implant diseases, including peri-implant mucositis and peri- implantitis, result from the combined influence of site-level anatomical factors and patient-level behavioral or systemic characteristics. Specifically, features such as inadequate gingival attachment, shallow vestibular depth, and poor interproximal hygiene access significantly increase the risk of disease onset and progression. Peri-implant diseased states are associated with distinct shifts in the peri-implant microbiome, marked by a transition from health-associated commensals to pathogenic, dysbiotic microbial communities. The aim of this study is to quantify the contribution of site-level and patient-level variables to peri-implant disease risk using multivariate modeling and machine learning techniques. Materials & Methods: In this cross-sectional analysis, 240 participants with single, non-splinted implants in function for at least one year were enrolled. Clinical and radiographic assessments categorized implants as healthy (n = 72), peri-implant mucositis (n = 135), or peri-implantitis (n = 33). Site-level factors (gingival attachment, keratinized mucosa, vestibular depth, plaque and inflammation indices, implant position) and patient-level variables (age, hygiene behavior) were analyzed using multinomial logistic regression and Random Forest classification. Linear Discriminant Analysis (LDA) of Bray-Curtis distances was applied to examine clustering based on combined variables. Results: Of the 240 implants analyzed, 47.1% were healthy, 40.4% had mucositis, and 12.5% had peri- implantitis. Multinomial logistic regression identified gingival attachment (GA <1 mm) and vestibular depth (VD <11 mm) as the strongest predictors of disease, with odds ratios of 4.2 and 7.8. The covaiate risk of the Interaction effects between shallow VD and inadequate keratinized mucosa (KM) significantly elevated mucositis risk. Implant adjacency without tooth support, shorter implant length, and posterior positioning were associated with peri-implantitis. The Modified Plaque Index (mPI) consistently predicted both disease states, with its impact intensified in sites with reduced GA, shallow VD, or limited KM. Random Forest models further highlighted age, interproximal hygiene access, and anatomical constraints as key features, achieving over 95% accuracy in disease classification. Conclusions: In conclusion, this study reinforces the multifactorial and interdependent nature of peri-implant disease risk. By integrating site-level and patient-level variables into a unified framework, we provide novel insights into how co-localized factors synergize to influence disease onset and progression. These findings advocate for early identification of site-level disease predictors, and preventive approach in implant therapy, highlighting the importance of individualized treatment planning and maintenance strategies for optimal long-term outcomes.