This article is more than 5 years old. Information may no longer be current.

Comprehensive melanoma risk prediction models limited

Comprehensively developed and assessed melanoma risk prediction models are limited, according to recent study results.

Researchers in Sydney, Australia, conducted a systematic search using Embase, Medline, Premedline and the Cochrane databases from inception through April 30, 2013, for published studies reporting multiple risk prediction models for incident primary cutaneous melanoma for adults. Two independent reviewers searched the data, using terms “melanoma, risk prediction,” and “statistical model.”

From 3,322 titles screened, 19 studies describing 28 melanoma risk prediction tools were included in the study. The number of predictors ranged from two to 13, with nevi, skin type, freckle density, age, hair color and sunburn history being the most common.

“There was limited reporting and substantial variation among the studies in model development and performance,” the researchers reported.

Nine studies reported discrimination — defined as how well the model separates patients with melanoma from those without disease — that ranged from fair (AUC=0.62) to very good (AUC=0.86). The models’ internal or external validity or use in clinical and public health practice was assessed in limited studies.

The risk prediction tool developed by Fears and colleagues in 2006 for US populations “appears to be the most clinically useful and may also assist in identifying high-risk groups for melanoma prevention strategies,” the researchers said.

“More external validation and prospective evaluation will help translate melanoma risk prediction models into useful tools for clinical and public health practice,” the researchers concluded.

Disclosure: The researchers report no relevant financial disclosures.