Objectives: A large number of multivariable models which associate independent variables with the outcome tooth loss exist. Directly or indirectly, these make predictions as to the relative risk of tooth loss. We aimed to validate six of these prediction models.
Methods: We applied each model, if needed after adaptions, in a cohort of 301 compliant periodontitis patients who had been under supportive periodontal treatment (SPT) in a university setting over 21.7 ± 5.6 years. The models employed a range of tooth-level and patient-level parameters. Model accuracy, that is, the ability to rightly predict tooth loss during SPT using baseline parameters, was investigated by the area under the receiver-operating-characteristics curve (AUC).
Results: Most models showed low accuracy (AUC ranged between 0.52 and 0.67). The classification model from Avila et al. (2009) Journal of Periodontology, 80, 476-491, expressing the risk of tooth loss in five grades, was most accurate (mean AUC: 0.67, 95%CI: 0.65/0.69). When applying this model, the risk of false-positively predicting tooth loss was high, except when the highest grade (i.e. a tooth being considered as having a hopeless prognosis) was used. In this case, the specificity was 84% and the sensitivity 46%.
Conclusions: Predicting tooth loss in this specific cohort of periodontitis patients was only limitedly possible.
Keywords: periodontal therapy; periodontitis; prediction; risk model; tooth loss.
© 2018 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd.