Evaluating the Performance of Melanoma Risk Prediction Models

Importance Evaluation of melanoma risk prediction models in independent datasets is crucial for determining their potential utility for preventive strategies. Objective To assess the discrimination and calibration of 6 published melanoma risk prediction models in a large prospective longitudinal cohort in Australia. Design, Setting, and Participants This population-based study uses data from the Sax Institute’s 45 and Up Study cohort, which includes 267 357 residents of New South Wales (NSW), Australia, aged 45 years or older, recruited through the Services Australia Medicare enrollment database in 2005 to 2009, with linked cancer data to December 31, 2019. The study sample included participants who completed the 2020 follow-up survey. Incident melanoma diagnoses were obtained through data linkage to the NSW Cancer Registry, separately for outcomes of invasive melanoma only (all models), or invasive and in situ melanoma (1 model). The data analysis occurred between May 2025 and February 2026. Exposures Risk factors for each risk model were assessed from baseline and 2020 follow-up questionnaires. Main Outcomes and Measures Main outcomes were discrimination and calibration measures for predicting 10-year risk of a first primary melanoma. Results A total of 45 061 participants completed the 2020 follow-up survey (52.8% response rate). The analysis sample included up to 43 235 participants (median [IQR] age, 57 [51-64] years; 24 159 [55.9%] female; 19 076 [44.1%] male), and 946 invasive and 1406 in situ incident melanomas. Discrimination assessed using the area under the receiver operating characteristic curve (AUC) at 10 years follow-up ranged from 0.58 (95% CI, 0.56-0.60) to 0.70 (95% CI, 0.68-0.72). Most risk models underestimated the number of melanomas diagnosed in the cohort; the 2 models with the best calibration had calibration in the large values of −0.16 (95% CI, −0.20 to −0.11]; slope, 0.74 [95% CI, 0.69-0.78]) and 0.59 (95% CI, 0.52-0.66]; slope, 0.72 [95% CI, 0.66-0.78]). Conclusions and Relevance In this cohort study, several prediction models demonstrated moderate discriminatory performance comparable to their original studies, supporting their potential use for risk stratification to guide prevention and early detection strategies. Some recalibration may be needed to improve estimation of absolute risks at the population level.