Risk-Informed Complaint Management in Neurosurgery: Development and Temporal Validation of an Early-Warning Model for Dispute Escalation

Background: Neurosurgery is a high-risk surgical specialty with substantial clinical uncertainty, procedural complexity, and potential for severe adverse outcomes. Although timely risk stratification after an initial complaint may facilitate structured communication, senior review, and coordinated dispute management, practical tools for estimating escalation risk at the first complaint remain limited. Methods: We conducted a single-center retrospective study of neurosurgical complaint events recorded at the Second Xiangya Hospital from January 2016 to December 2025. Time zero was the first documented complaint date. The endpoint was a final institutional severity grade of 7–10, denoting hospital-level handling or beyond, versus grades 1–6. Candidate predictors were restricted to information available on or before time zero and were selected by univariable screening followed by multivariable logistic regression. The sociodemographic support-needs profile was excluded from the model to avoid stigmatization and was used in fairness analyses. Results: Among 975 eligible complaint events, 161 (16.5%) reached high-grade escalation. Four predictors were retained: ICU admission, invasive manipulation failure, a pre-complaint communication-context factor, and anger at the first complaint. The model achieved moderate discrimination and calibration in the apparent cohort (AUC, 0.781; PR-AUC, 0.417; Brier score, 0.115) and similar performance in the 2023–2025 temporal test cohort (AUC, 0.759 [95% CI, 0.687–0.826]; PR-AUC, 0.364; Brier score, 0.119). At a threshold of 0.20, the positive predictive value in the test cohort was 0.333. Exploratory risk strata separated high-grade escalation rates from 5.8% in the low-risk group to 16.6% and 37.3% in the intermediate- and high-risk groups. Conclusions: Routinely available clinical and communication indicators, recorded by the time of the first complaint, can support risk-informed complaint management in neurosurgery. Because two of every three high-risk flags did not escalate, the model should support senior review and early communication rather than serve as a stand-alone classifier; external validation and local recalibration are required before use elsewhere.