New Melanoma Risk Prediction Tool Superior in Accuracy to Previous Tools

An improved risk-prediction tool has been developed, providing enhanced accuracy for predicting patients’ future risk of invasive melanoma, new findings suggest.1

This tool, comprised of 7 predictors and referred to as MP7, was evaluated by a series of investigators in an analysis authored by such researchers as David C. Whiteman, MBBS, PhD, of the Department of Population Health at QIMR Berghofer Medical Research Institute in Australia.

In their analysis’s introduction, Whiteman and colleagues highlighted that target approaches are increasingly being used to systematically detect melanomas, and those at highest risk are prioritized for screenings.2

“Moreover, the number of newly diagnosed invasive melanomas in the QSkin cohort has almost tripled, providing substantially greater statistical power than previously,” Whiteman and coauthors wrote.1 “Herein, we describe the development and validation of a new prediction model for invasive melanoma in the QSkin cohort over 10 years to inform risk-stratification strategies.”

Cohort Study Design

The investigative team implemented data for their analysis drawn from the QSkin Sun and Health Study. QSkin was a large population-based prospective cohort based in Queensland, Australia. QSkin had recruited individuals of adult age, specifically aged 40–69 years, between November 2010 - December 2011 using a random sampling of their state’s population.

Those who were shown to have a registry-confirmed history of melanoma (either in situ or invasive) before their enrollment were not included by Whiteman et al. Ethical approval was granted by the Human Research Ethics Committee of the QIMR Berghofer Medical Research Institute, with all subjects providing their written informed consent.

For the investigation, participants had been followed for a decade after the 2011 baseline survey. Only individuals who had been listed as melanoma-free at recruitment and who completed the comprehensive baseline risk factor questionnaire were included in this analysis. Whiteman and colleagues' nalyses were carried out in the period between October 2024 - April 2025.

At the study's entry, 31 baseline variables were prespecified as potential predictors of subsequent occurrences of invasive melanoma. Incident cases of invasive melanoma among participants, determined histologically, were found by the investigators via record linkage with the Queensland Cancer Register for diagnoses. Censoring of participants at the time of melanoma took place in situ diagnosis or death. To build the prediction model, Whiteman and collagues used Cox proportional hazards regression with both forward and backward variable selection strategies.

Melanoma Prediction Accuracy Findings

Among the 41,919 participants deemed eligible for inclusion, Whiteman and coauthors noted that 55% were female and that the mean age at baseline was 55.4 years. Across 401,356 person-years of observation, there were 706 individuals who developed a new invasive melanoma.

In the final model, there were 14 baseline predictors: sex, age, patient ancestry, nevus density, hair color, freckling density, tanning response, history of sunburns in adulthood, family melanoma history, previous diagnosis of another cancer, skin cancer excision history, smoking status, previous actinic keratoses, and height. There were also 2 additional terms: age squared and an age-by-sex interaction.

The investigative team concluded that the prediction tool attained a discrimination index (C statistic) of 0.74 (95% CI, 0.73–0.76).1 Through the use of the Youden index, the optimal screening threshold was noted as having corresponded to targeting the top 40% at predicted risk. This would identify 74% of future melanoma cases, with a number needed to screen of 32.

Overall, the team's cohort study identified the tool as an option allowing for enhanced accuracy for the prediction of one's future risk of invasive melanoma compared with other current tools.

“In future research, we will assess the benefit of adding genetic information to the tool, and we have recently recruited a second, independent cohort in which to validate the tool externally in the years ahead,” Whiteman and colleagues concluded.1 “In the meantime, we see strong merit in assessing the performance of this tool independently in other settings.”

References

1. Whiteman DC, Olsen CM, Pandeya N, et al. A Risk Prediction Tool for Invasive Melanoma. JAMA Dermatol. Published online September 10, 2025. doi:10.1001/jamadermatol.2025.3028.

2. Collins GS, Dhiman P, Ma J, et al. Evaluation of clinical prediction models (part 1): from development to external validation. BMJ. 2024;384:e074819. doi:10.1136/bmj-2023-074819.