IADR Abstract Archives
Diabetes Risk Assessment in Dental Settings: Model Development and Validation
Objectives: To develop and externally validate a risk assessment tool for early diabetes detection in dental settings using data routinely available to dental teams, including dental risk factors.
Methods: The prediction tool was developed and validated using population-based cohort studies conducted in Northeast Germany, “Studies of Health in Pomerania” (SHIP). SHIP-TREND (n=3339) had 329 events and relevant medical and dental data for model development. An HbA1c level ≥6.0% was employed as the dependent variable, with no prior diabetes diagnosis. Missing data were multiply imputed. Variables were selected using backward elimination. Internal validation was undertaken to allow adjustment for optimism. A second independent dataset was used for external validation (SHIP-0, n=2381, 403 events).
Results: The final model included: age, sex, body mass index, smoking status, parent or sibling with diabetes, dental prosthesis, mobile teeth, history of periodontal treatment within the last 5 years and probing pocket depths ≥5mm. Pre-specified interaction terms were included in the model. After adjusting for optimism (shrinkage factor, 0.91) the final model had an AUC of 0.72 (95%CI 0.69-0.75) and calibration of 0.91. In the validation set AUC was 0.69 (95%CI 0.67-0.72), calibration was 0.91. A paper-based score was created without interaction terms for ease of use in the clinical setting with sensitivity of 0.76 and specificity of 0.53 at a threshold ≥ 22.
Conclusions: Dental variables may be utilised within a prediction tool to aid dental teams identify those at high risk of type 2 diabetes. The model’s performance is comparable to the current recommended tool (Leicester risk score: AUC 0.72, sensitivity 81%, specificity 45%). However, the newly developed model is potentially more clinically acceptable as it uses data routinely available to dental teams. Further validation is required to determine the viability of using such a model on a UK population.
Methods: The prediction tool was developed and validated using population-based cohort studies conducted in Northeast Germany, “Studies of Health in Pomerania” (SHIP). SHIP-TREND (n=3339) had 329 events and relevant medical and dental data for model development. An HbA1c level ≥6.0% was employed as the dependent variable, with no prior diabetes diagnosis. Missing data were multiply imputed. Variables were selected using backward elimination. Internal validation was undertaken to allow adjustment for optimism. A second independent dataset was used for external validation (SHIP-0, n=2381, 403 events).
Results: The final model included: age, sex, body mass index, smoking status, parent or sibling with diabetes, dental prosthesis, mobile teeth, history of periodontal treatment within the last 5 years and probing pocket depths ≥5mm. Pre-specified interaction terms were included in the model. After adjusting for optimism (shrinkage factor, 0.91) the final model had an AUC of 0.72 (95%CI 0.69-0.75) and calibration of 0.91. In the validation set AUC was 0.69 (95%CI 0.67-0.72), calibration was 0.91. A paper-based score was created without interaction terms for ease of use in the clinical setting with sensitivity of 0.76 and specificity of 0.53 at a threshold ≥ 22.
Conclusions: Dental variables may be utilised within a prediction tool to aid dental teams identify those at high risk of type 2 diabetes. The model’s performance is comparable to the current recommended tool (Leicester risk score: AUC 0.72, sensitivity 81%, specificity 45%). However, the newly developed model is potentially more clinically acceptable as it uses data routinely available to dental teams. Further validation is required to determine the viability of using such a model on a UK population.