Validation of a multivariate prediction rule for history of periodontitis in a separate population.

AIM

Validation of a previously derived prediction rule for periodontitis in an external population sample.

MATERIALS AND METHODS

Age, smoking and self-reported tooth mobility were used in logistic models to predict moderate and severe periodontitis as diagnosed from panoramic radiographs of 246 patients attending private practices in Germany. Coefficients derived from these models were used to predict periodontitis in a representative population-based sample of 3297 residents of the region of Pomerania, Germany.

RESULTS

In the full derivation sample, the predictive power of the logistic model as measured by the c-statistic was 0.82 and 0.84 for moderate and severe periodontitis, respectively. In the validation set, these models yielded c-statistics of 0.82 for both moderate and severe periodontitis. Lower c-statistics were obtained among subjects aged 40 years and older in the derivation set (c=0.74 and 0.77), and the performance was poorer in the validation set with c-statistics of 0.69 and 0.72, respectively.

CONCLUSIONS

A prediction rule based on age, smoking and self-reported tooth mobility can yield a moderately useful external validity. Validity may be dependent on specific population characteristics, and derivation of a prediction rule based on a clinical subsample of the target population with a larger set of predictors may provide better results in an application.