Head and neck cancer risk predictive models: a solution to streamline urgent suspected cancer referrals?

Most head and neck cancers (HNC) are diagnosed at advanced stages, highlighting the need for strategies to enhance early recognition. Since 2009, urgent suspected cancer (USC) referrals have increased by 377%, but HNC conversion rates have declined to 2.2%, straining resources and reducing efficiency. Predictive tools aim to support clinicians to identify high-risk patients. This study evaluates the predictive capability of HaNC-RC-v.2 compared with a restructured Wakefield Model and assesses the impact of HNC risk-based stratification on service organisation. Retrospective analysis of 1183 USC referrals collected data surrounding demographics, social history, 16 signs/symptoms, and diagnoses. Univariate and multivariate regression identified significant HNC predictors. Performance of the Wakefield Model was evaluated through sensitivity, specificity, area under the receiver operator curve (AUC), Akaike information criterion, and Hosmer-Lemeshow tests. DeLong tests compared model predictive capabilities. The Wakefield Model incorporated age, gender, smoking, alcohol, and seven HNC predictors, achieving an AUC of 82.70%. HNC-risk thresholds defined high-risk (≥7.35%; sensitivity: 75.71%; specificity: 80.41%) and low-risk (≤1.70%; sensitivity: 76.47%; specificity: 58.21%) groups. The model's predictive capability was significantly greater than that of the HaNC-RC-v.2 (AUC: 77.70%). Stratifying patients by HNC-risk thresholds ensured that >95% of HNC cases were assessed within 14 days, whilst reallocating 26.20% - 32.12% of USC assessments, improving service efficiency. HNC predictive models help clinicians to identify high-risk individuals, prioritise assessments, and optimise service provision. Implementing models at the point of referral could potentially stratify referrals into suitable assessment pathways.