Logistic Regression Analysis of the Factors Involved in the Failure of Osseointegration and Survival of Dental Implants with an Internal Connection and Machined Collar: A 6-Year Retrospective Cohort Study

Abstract

Background. Although the long-term success rate of dental implants is currently close to 95%, it is necessary to provide more evidence on the factors related to the failure of osseointegration and survival. Purpose. To establish the risk factors associated with the failure of osseointegration and survival of dental implants with an internal connection and machined collar and to establish a predictive statistical model. Materials and Methods. An analytical, retrospective, and observational clinical study of a sample of 297 implants with a follow-up of up to 76 months. Independent variables related to the implant, patient, and surgical and rehabilitative procedures were identified. The dependent variables were failure of osseointegration and failure of implant survival after prosthetic loading. A survival analysis was carried out by applying the Kaplan-Meier model (significance for $p<0.05$ ). The log-rank test and the Cox regression analysis were applied to the factors that presented differences. Finally, the regression logit function was used to determine whether it is possible to predict the risk of implant failure according to the analyzed variables with the data obtained in this study. Results. The percentages of osseointegration and survival were 97.6 and 97.2%, respectively. For osseointegration, there were significant differences according to gender ( $p=0.048$ ), and the risk of nonosseointegration was 85% lower in women. Regarding survival, the Cox analysis converged on only two factors, which were smoking and treatment with anticoagulant drugs. The risk of loss was multiplied by 18.3 for patients smoking more than 10 cigarettes per day and by 28.2 for patients treated with anticoagulants. Conclusions. The indicated risk factors should be considered, but the analysis of the results is not sufficient to create a predictive model.