Poster Presentation Australasian Melanoma Conference 2026

Risk prediction models for first primary melanoma: a systematic review (#426)

Ellie B Medcalf 1 , Xu Huang 1 , Deonna M Ackermann 1 , Shannon Baker 2 , Rehana A Salam 2 , Katy JL Bell 1
  1. Sydney School of Public Health, The University of Sydney, Camperdown, NSW, Australia
  2. The Daffodil Centre, The University of Sydney, and Cancer Council NSW, Sydney, NSW, Australia

Background

Melanoma incidence continues to rise globally. Targeting screening to higher risk individuals may retain benefits of early detection (identifying and treating clinically progressive lesions at an early stage) while minimising harms (false positiveresults and overdiagnosis of indolent lesions). Validated risk assessment tools that identify individuals at the highest risk for first primary melanoma could assist targeted screening programs. This systematic review systematically searched for and assessed published risk prediction models for first primary melanoma in adults at population average risk.

Methods

We searched PubMed, CENTRAL, and Embase from inception to 31st January 2026 for studies reporting risk prediction models in adults with no prior history of melanoma and that included phenotypic characteristics. Two reviewers independently screened titles, abstracts, assessed full-text articles, extracted data using the CHARMS checklist and evaluated risk of bias and applicability using PROBAST + AI.

Results

 From 3540 records screened, 23 eligible studies were included in the review. Most studies used case-control study design (20, 87%). Of the 23 studies, 27 models were developed and/or evaluated. In these models, the most common included risk factors were hair colour (17, 63%), naevi count (16, 59%), age (15, 56%), sex (14, 52%), sunburns (13, 48%), genetic risk factors (11, 41%), and nonmelanoma skin cancer (11, 41%). While all models were assessed for discrimination, just over half assessed calibration (15, 56%) and only a few assessed clinical utility/net benefit (3, 11%). One study was rated low concern for quality/risk of bias, however, most were rated high concern due to case-control design, self-reported predictor measurement, not reporting predictive performance metrics, or not undertaking internal or external validation. No studies were rated low concern for applicability. While 12 studies described models predicting invasive melanoma, none reported results separately for those that are unambiguously potentially lethal (i.e., unlikely to be overdiagnosed - invasive melanoma stage IB or higher), contributing to unclear/high concern applicability ratings.

Conclusion

No validated model was found that reported risk prediction for first primary melanoma that are unambiguously potentially lethal. Development and external validation of models addressing this outcome would better support targeted risk screening for melanoma.