Design: External validation of six risk models in a population-based prospective cohort
Population: Up to 43,235 adults aged 45 and over in New South Wales, Australia
Primary outcome: Discrimination and calibration for 10-year risk of first primary melanoma
Effect: AUC 0.58 (0.56 to 0.60) to 0.70 (0.68 to 0.72); most models underestimated risk

Six published melanoma risk prediction models were tested in up to 43,235 adults aged 45 and over from the Australian 45 and Up Study, followed for first primary melanoma; 946 invasive and 1406 in situ melanomas occurred.

Discrimination at 10 years ranged from an AUC of 0.58 to 0.70. Most models underestimated how many melanomas occurred, and even the best-calibrated two needed recalibration.

These models perform about as well in new data as in the original studies, which supports using them to sort people into broad risk groups for prevention. They are not precise enough to decide individual screening on their own, and Australian risk levels differ greatly from those in India.

Use melanoma risk models to identify broad high-risk groups, not to make individual screening decisions alone.
Keep asking about naevus count, prior melanoma, family history and sunburn — the core inputs of these models.
Recognise that models built in high-incidence, fair-skinned populations may not transfer to Indian patients.
Expect most models to underestimate absolute risk unless recalibrated for the local population.

Why it matters

Risk-stratified screening depends on models whose real-world accuracy has rarely been checked independently.

The statistics, in plain English

An AUC of 0.5 is a coin toss and 1.0 is perfect; 0.58 to 0.70 is modest. Calibration describes whether predicted numbers match observed numbers; a slope below 1 means the model's predictions were too spread out, and most models predicted fewer melanomas than occurred.