Predicting Dental Implant Survival Using Frailty Methods for Clustered Observations

Material and Methods: To address the research purpose we conducted a retrospective cohort study. The cohort consisted of subjects having at least one Bicon implant placed. Covariates used in the survival predictions included smoking status, implant staging, and timing of placement. These covariates were previously identified using a Cox frailty regression model that adjusted for clustered observations and heterogeneity of implant failure risks. We applied the flexible, semi-parametric Cox proportional hazards frailty model to predict survival. Specifically, we use this model to obtain point and interval estimates for the survival probabilities of a dental implant conditioning on the covariates and as well as information about the failure times of other implant(s) of the same subject.

Results: The cohort composed of 677 patients having 2349 implants placed. The 12- and 60-month survival estimates for the first implant in a non-smoker, one stage implant placement, and immediate implant placement after extraction were 85.6% (95%CI: 76.1%,81.8%) and 70.4% (95%CI: 51.1, 83.3%) respectively. For the joint distribution predictions for two implants surviving for 12- and 60- months given the same clinical parameters were 74.3% (95%CI: 60.5% to 84.7%) and 53.0% (95%CI: 31.4%, 70.9%) respectively.

Conclusions: Survival predictions of clustered datasets are exceedingly challenging proposition commonly encountered in dental research. The use of this innovative flexible Cox proportional hazards frailty model for prediction allows unique predictions about implant failure based on prior information about the survival status of other implants within the same subject.

Supported partially by NIH grants R37AI024643 (TC), K24DE000448 (TBD), and MGH OMFS Research Fund (SKC, TBD)