From No-Show Prediction to Operational Action: A Retrospective Secondary Analysis of Calibrated, Duration-Weighted Dental Appointment Prioritization

Background/Objectives: Patient non-attendance disrupts dental scheduling and may leave reserved appointment time unused when missed slots cannot be reassigned. Most no-show prediction studies focus on discrimination or the number of missed appointments identified, without considering appointment duration, probability calibration, temporal robustness, or limited intervention capacity. This study developed and evaluated a calibrated, duration-weighted framework for operational prioritization of dental appointments. Methods: A retrospective secondary analysis was conducted using 196,018 appointments recorded in 2019 at a dental clinic in Riyadh, Saudi Arabia. L2-regularized logistic regression was used as the primary model, with CatBoost as a secondary nonlinear comparator. Model development, probability calibration, and testing were performed in separate chronological periods. Performance was assessed using ROC-AUC, average precision, Brier score, calibration measures, rolling-origin temporal validation, dentist-level transportability analysis, and sensitivity analyses. Four prioritization strategies were compared at intervention budgets from 5% to 30%: random, probability-only, duration-only, and calibrated probability multiplied by scheduled duration. Results: In the independent November–December test set, the calibrated logistic model achieved a ROC-AUC of 0.6886 (95% CI: 0.6827–0.6941), average precision of 0.6087, and Brier score of 0.2238. At the prespecified 10% intervention budget, probability-only targeting captured 15.88% of scheduled no-show minutes, duration-only targeting captured 18.49%, and duration-weighted targeting captured 23.04%. Duration-weighted targeting exceeded probability-only targeting by 7.16 percentage points (95% CI: 6.47–7.79) and duration-only targeting by 4.54 percentage points (95% CI: 4.00–5.39). The advantage persisted across rolling-origin validation and sensitivity analyses. Conclusions: Combining calibrated no-show probability with scheduled appointment duration identified more scheduled time associated with non-attendance than either probability or duration alone under the same intervention budget. The framework demonstrated a retrospective ranking advantage for identifying scheduled minutes associated with non-attendance but did not establish recoverable clinical capacity, reductions in missed appointments, or improvements in clinical efficiency. Independent external validation and prospective intervention studies are required before routine clinical implementation.