Aim: To cluster peri-implantitis patients and explore non-linear patterns in peri-implant bone levels.
Materials and methods: Clinical and radiographic variables were retrieved from 94 implant-treated patients (340 implants, mean 7.1 ± 4.1 years in function). Kernel probability density estimations on patient mean peri-implant bone levels were used to identify patient clusters. Inter-relationships of all variables were evaluated by principal component analysis; a k-nearest neighbours method was performed for supervised prediction of implant bone levels at the patient level. Self-similar patterns of mean bone level per implant from different jaw bone sites were examined and their associated fractal dimensions were estimated.
Results: Two clusters of implant-treated patients were identified, one at patient mean bone levels of 1.7 mm and another at 4.0 mm. Five of thirteen available variables (number of teeth, age, gender, periodontitis severity, years of implant service), were predictive for peri-implant bone levels. A high jaw bone fractal dimension was associated with less severe peri-implantitis.
Conclusions: Non-linearity of peri-implantitis was evidenced by finding different peri-implant bone levels between two main clusters of implant-treated patients and among six different jaw bone sites. The patient mean peri-implant bone levels were predicted from five variables and confirmed complexity for peri-implantitis.
Keywords: complex disease; computational biology; data mining; dental implant; non-linearity; peri-implantitis; statistical learning theory.
© 2015 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd.