Periodontal disease (PD) screening models based on a self-reported questionnaire were previously established and externally validated. The aim of the present study is to explore whether the screening models could be modified to improve prediction performance; this methodology is called 'updating'. Updatingthe models for 'total' and 'severe' PD was performed using two datasets. One dataset from a previous study (n = 155) was used to explore the updating, and a second (n = 187, built for the current study) was used to validate whether updating improved performance. Updating was based on different statistical approaches, including model recalibration and revision. Discrimination and calibration were assessed after updating. For 'total' PD, the update based on model revision improved its performance. However, still low AUCs were found: 0.64 (0.56-0.73) and 0.61 (0.53-0.69) with corresponding O:E ratios 1.00 (0.80-1.23) and 0.92 (0.75-1.13) in the update and validation cohorts, respectively. For 'severe' PD, performance of the original model without update performed still the best; AUCs were 0.72 (0.61-0.83) and 0.75 (0.66-0.84) in the update and validation cohorts, respectively, with corresponding O:E ratios 0.60 (0.38-0.84) and 0.62 (0.42-0.87). The updating methodology did not further improve the performance of the original 'severe' PD screening model; it performed satisfactorily in the medical care setting. Despite updating attempts, the screening model for 'total' PD remained sub-optimal. Screening for 'severe' PD can now be implemented in the medical care setting.
Read full abstract