Scoring algorithm for predicting periodontitis in dentate adults using self‐report measures — National Health and Nutrition Examination Survey 2009–2012

Abstract Aim Our goal was to develop and externally validate oral health self‐report measures for predicting periodontitis in a representative U.S. adult population (30–79 years old) and to evaluate a predictive scoring tool for periodontitis constructed from the best performing model parameter estimates. Methods The predictive models for periodontitis using demographic characteristics and self‐reported oral health measures were developed and tested with the National Health and Nutrition Examination Survey (NHANES) 2009–2012 data (development 2009–2010, validation 2011–2012). The best performing model was externally validated against clinical periodontitis cases defined by measurements from a full‐mouth periodontal examination at six sites around all teeth excluding third molars. A predictive scoring tool derived from the transformed sum of the model coefficient estimates was also externally validated. Model performances were evaluated by their sensitivity, specificity, predictive accuracy, and area under the receiver‐operating characteristic curve (AUROC). Results Our best model used self‐reported oral health, smoking, and demographics. Predictive Risk Scores (PRS) of ≥65 captured about 98% of the true periodontitis cases. Three forms of the model (1—individual risk factor variables, 2—continuous PRS, and 3—PRS categories) were applied to the development and validation data sets. Overall, all three forms had high sensitivity (>84%) in both the development and validation data sets and had similar AUROC (around 80%). Specificity was low to moderate. When externally validated, the model incorporating PRS as a continuous measure had high sensitivity (84.0%) and low specificity (57.5%), with AUROC of 79.5% and predictive accuracy of 71.6%. Similarly, when PRS as a categorical variable was externally validated, the model had a high sensitivity (82.8%) and low specificity (59.9%), with an AUROC of 79.3% and predictive accuracy of 72.0%. Conclusion Overall, modeling of four self‐report oral health measures, combined with smoking and demographic characteristics, performs well in predicting clinical periodontitis in a nationally representative sample of the adult dentate US adult population. Compared with clinical periodontal examination, this approach is promising as a viable, non‐clinical, and much less resource‐intensive alternative method for estimating the burden of periodontitis.