Model proves suspected melanoma risk factors

August 1, 2005

Boston — A recent analysis supports what has long been believed about risk factors thought to contribute to melanoma development, including age, sex, family history of melanoma, number of moles, history of sunburn and hair color.

Boston - A recent analysis supports what has long been believed about risk factors thought to contribute to melanoma development, including age, sex, family history of melanoma, number of moles, history of sunburn and hair color.

Eunyoung Cho, Sc.D., an instructor in medicine at the Channing Laboratory of Brigham and Women's Hospital and Harvard Medical School here is lead author of the study (J Clin Oncol. 23:2669-2675) establishing an association between these risk factors and melanoma development in large prospective studies.

"We built a statistical model incorporating these risk factors so that individuals can assess their risk of developing melanoma," he says.

In total, researchers combined three large prospective studies that together followed more than 150,000 women and 25,000 men who were cancer-free at baseline for up to 14 years. During the period studied, researchers documented 535 incident melanoma cases (444 in women and 91 in men).

Initially, researchers analyzed the three patient cohorts (from the Nurses' Health Study/NHS, the NHS II and the Health Professionals Follow-up Study/HPFS) separately. But since strong risk factors yielded similar effect estimates in the studies, researchers combined these cohorts to achieve more powerful analysis.

Using the combined data set, researchers then built models with predictors of melanoma risk in individual studies. One by one, they added the following factors: age, family history of melanoma, number of nevi, hair color and history of severe and painful sunburn. All factors strongly predicted melanoma risk (p <0.05).

As part of their analysis, researchers also evaluated other potential risk factors one at a time. Among men, these included skin reaction to sun, latitude of residence at birth and other ages, and body mass index. Among women, factors included oral contraceptive use and menopausal status. However, none of these variables exerted a statistically significant effect within the model. Nor did researchers find sunscreen use to be related to melanoma risk.

In the final model, researchers could calculate an individual's risk score by adding a regression coefficient for various risk factors and for the level of each risk factor (except age) the individual reported, plus the regression coefficient for age multiplied by age in years. As an example, a 50-year-old woman with no family history of melanoma, one to two episodes of severe sunburn, three to five moles and blonde hair would have an estimated relative risk of 4.54 compared with a woman of the same age without the same risk factors.

To check the generalizability of the formula in the United States and its accuracy, Dr. Cho and her colleagues calculated subjects' 10-year risk of being diagnosed with melanoma and compared them with SEER cancer statistics and found generally similar risk levels. Likewise, the model yielded a concordance statistic (adjusted for age and sex) of 0.62, which researchers say is comparable to those of predictive models for other cancers sites.

Therefore, Dr. Cho says, "our model is as good as other statistical models for other cancers sites."

Presently, few statistical models for estimating melanoma risk exist.

"This is a rapidly developing area of research," she says. "So far, there has been a statistical model for breast cancer, called the Gail model, which has been frequently used in the clinical setting (Gail MH et al. J Natl Cancer Inst 81: 1879-186, 1989.)."

As for the melanoma model, researchers believe it will be useful to estimate individuals' risk of developing melanoma and to identify high-risk populations that might be included in prevention trials or targeted for screening and prevention efforts.

Whether it's sufficiently accurate at the individual level to ultimately reduce the volume of excisions for premalignant lesions, however, has yet to be established.

Limitations

Since researchers had no direct measure of cumulative sun exposure during a subject's childhood, adolescence or lifetime, the model is limited. Furthermore, although they checked their model for goodness of fit, the model has yet to be verified independently in a distinct population.