Variation in Incidence and Fatality of Melanoma by Season of Diagnosis in New South Wales, Australia

Seasonal variation in cutaneous melanoma incidence with a summer peak is an intriguing phenomenon. It has been observed in different parts of the world and in both northern and southern hemispheres, but is poorly understood (1). Some authors suggest that it could be due to greater awareness of skin lesions in summer, leading to an increase in excision of pigmented skin lesions and thus greater and probably earlier diagnosis of melanoma. Others have suggested that there might be a true increase in melanoma occurrence in summer resulting from a late promotional effect of sun exposure (1).

One report showed that seasonal variation in incidence was confined to invasive melanoma; it was not found in in situ melanoma (2). Increased awareness of skin lesions would be expected to increase diagnosis of both in situ and invasive melanomas. In addition, Braun et al. (3) showed that there is seasonal variation in melanoma incidence among members of melanoma-prone families whose skin is screened regularly for melanoma. This observation also seems inconsistent with a summer increase in awareness of skin lesions being the main or only cause of the summer incidence peak.

We hypothesized that if seasonal variation in incidence of melanoma were due to a seasonal increase in awareness, we would observe a parallel seasonal increase in thinner melanomas and better survival from melanoma diagnosed in summer depended on it. On the other hand, if late-stage promotion of melanoma were responsible for the summer peak, we might expect to see an increase in more aggressive melanomas in summer and a worse prognosis independent of lesion thickness.

We tested this hypothesis in the population of the state of New South Wales, Australia, because of its high incidence of melanoma, known seasonal variation in incidence, well-known programs for prevention of sun exposure and early diagnosis of melanoma, and the reasonably complete data held by the New South Wales Central Cancer Registry for all relevant characteristics of melanoma.

New South Wales occupies most of the southeastern corner of Australia and has a population of >6.6 million. The New South Wales Central Cancer Registry first reported on cancer incidence in 1972 and has done so for each year since. We used the period 1989 to 1998 to study seasonal variation of cutaneous melanoma. With the approval of The Cancer Council New South Wales Ethics Committee, the Registry supplied a data file containing deidentified unit records of all people registered with a newly diagnosed cutaneous melanoma during this period. Each record contained the following variables: sex; age at and month and year of diagnosis of melanoma; anatomic site; histopathologic type (classified according to SNOMED International) and Breslow thickness of melanoma; month and year of death; and cause of death if known to be dead. In addition to regularly recording all deaths from cancer in New South Wales, the registry periodically links its records to all deaths registered in New South Wales and the Australian Death Index (which includes all deaths registered in Australia) to ascertain missed deaths from cancer and deaths from other causes; this had been last done for all deaths up to the end of 2000 for the data file supplied.

Five histopathologic categories were created based on the morphology code from the International Classification of Diseases-O: superficial spreading melanoma, nodular melanoma, lentigo maligna melanoma, other, and melanoma not further specified. Acral lentiginous melanomas were only 122 cases in the series and this number is probably underestimated. These melanomas were in the melanoma not further specified category.

To characterize seasonal variation, we used the ratio of the number of melanoma cases in summer to that in winter; we used Nam's method to test the statistical significance of variation from a ratio of 1.0 and to calculate the 95% confidence interval (95% CI) of the ratio (4). We considered summer to be December, January, and February; autumn March, April, and May; winter June, July, and August; and spring September, October, and November.

We tested for variation in Breslow thickness by month of diagnosis with a Kruskal-Wallis test. We used Poisson regression, with a scale variable (square root of Pearson's χ²/degree of freedom) to control for extra Poisson variation, to adjust for possible confounding of any association of Breslow thickness with season by year of diagnosis, age (5-year age groups entered as a continuous variable), sex, anatomic site, and histologic type. Results were expressed as percentage change in Breslow thickness across categories of each variable. We also conducted stratified analysis by age, sex, anatomic site, and histologic type.

We used a Cox proportional hazards model to study the effect of season of diagnosis on fatality from melanoma (5). Because we only had year and month of diagnosis and year and month of death, we used “1” as the day of diagnosis and day of death. Cases were censored at the date of death from causes other than melanoma or December 31, 2000, whichever was earlier.

Both univariate and multivariate analyses were done. The independent variables included in the multivariate model were year of diagnosis, age (5-year age groups entered as a continuous variable), sex, anatomic location, histologic type, Breslow thickness, and season of diagnosis of melanoma. In both multivariate Poisson and Cox regression models, 2,736 cases were excluded because of missing values of age and Breslow thickness. Thus, these analyses were done on 23,109 melanoma cases, of which 1,729 had died from melanoma.

All analyses were done with SAS software (version 8.2, SAS Institute, Cary, NC), using PROC GENMOD for Poisson regression and PROC PHREG for Cox model.

During the period 1989 to 1998, 25,845 newly diagnosed cases of melanoma were reported to the Cancer Registry (10,869 females and 14,976 males). They were more frequent in men than women, and the median age at diagnosis was 59.5 years. The trunk was the main site, superficial spreading melanoma the main histologic type, and summer was the commonest season of diagnosis (Table 1). Among 23,116 cases, the mean Breslow thickness was 1.36 mm with a median of 0.70 and interquartile range of 0.45-1.50 (Breslow thickness was missing for 2,729). There was a significant decrease in thickness with time from a mean of 1.45 mm in 1989 to 1.31 mm in 1998 (P < 0.0001, Kruskal Wallis test on all years). The proportion with missing thickness also fell, from 11% in 1989 to 6% in 1998 (P < 0.0001, χ² test on all years).

There was significant seasonal variation in incidence of melanoma (P < 0.0001, Nam test; Fig. 1). The lower incidence in January than in December and February is probably due to the high proportion of people who take holidays in January. The amplitude of the seasonal variation, measured as the summer to winter ratio, was greater for women, younger people, lesions on the limbs, and superficial spreading melanoma (Table 2). We also found a small difference between first primary melanoma and later primaries. Amplitude was 1.36 (95% CI, 1.32-1.42) for first, 1.40 (95% CI, 1.19-1.65) for second, and 1.77 (95% CI, 1.16-2.70) for third and later melanomas. This heterogeneity in amplitude of seasonal variation was statistically significant (P = 0.03).

In a simple bivariate analysis, melanomas were thickest in winter and thinnest in summer (Table 3). This seasonal variation was highly statistically significant (P < 0.0001, Kruskal-Wallis test) and the curve of Breslow thickness according to month of diagnosis was close to a second-order polynomial (R² = 0.84) with a minimum in mid-January. It was mainly due to an increase number of thick melanomas diagnosed in winter; there was much more variation by season in the third quartile of thickness than in the median or in first quartile and 25.5% of melanomas were very thick (>1.5 mm) in winter compared with 21.6% in summer (Table 3). Breslow thickness was also significantly correlated (P < 0.0001) with other variables measured: thicker for males than females and for older than younger people, thicker on the head and neck than elsewhere, and thicker for nodular than for other types of melanoma.

To test the independence of the effect of season of diagnosis on Breslow thickness, we built a Poisson regression model with a scale variable (see Materials and Methods); 2,736 observations (10.6% of all observations) for which age or Breslow thickness were missing were excluded from the analysis. All variables remained significantly associated with thickness in this model and melanomas diagnosed in summer were on average 7.9% thinner than those diagnosed in winter (Table 4). The relationship between season of diagnosis and Breslow thickness in the model was stronger for males, for melanoma patients older than 50 years old, for the limbs, and for melanoma of the superficial spreading type, as was the case when these variables were considered individually.

The mean period of follow-up of cases (interval between date of diagnosis and date of censoring or death) was 68 months with a median of 63 months and interquartile range of 37 to 97 months; 2,710 (10.5%) died from their melanoma. Fatality from melanoma was greater in men than women and in older than younger people, and for melanomas of the head and neck, thicker melanomas, and nodular melanomas (Table 5). We also found a protective effect of being diagnosed in summer compared with being diagnosed in winter, with a relative fatality of 0.72 (95% CI, 0.65-0.81) in a univariate analysis; the cause-specific 5-year survival rate was 92.1% for diagnosis in summer and 89% for diagnosis in winter. This difference remained significant, although somewhat less, after adjustment for possible confounding effects of year of diagnosis, age, sex, Breslow thickness, anatomic site, and histologic type. Relative fatality for diagnosis in spring was now marginally lower (0.80) than for diagnosis in summer (0.82). The relative fatality for melanomas diagnosed in September was lower than for any other month and thereafter relatively constant for each month from October to February (Fig. 2); however, the CIs about monthly estimates of relative fatality were wide. Stratified analysis showed that the relationship observed between season of diagnosis and survival was stronger for melanoma in younger people, women, melanomas of the superficial spreading type, and melanomas of the upper and lower limbs (data not shown).

As expected, we observed seasonal variation in incidence of cutaneous melanoma with a summer peak. This seasonal increase in melanoma incidence was greater in women and those <50 years of age, for melanomas on the limbs, and for superficial spreading and lentigo maligna melanomas, and was associated with significantly thinner lesions. Melanoma diagnosed in summer had a lower fatality than melanoma diagnosed in winter, and this difference was substantially independent of Breslow thickness, the best predictor of melanoma survival, and other prognostic factors in our data set.

That Breslow thickness decreased as melanoma incidence increased in the summer months suggests that there is increased recognition and earlier diagnosis of melanoma in summer, possibly because more skin is exposed to view as temperatures increase. This view is supported by the greater seasonal variation in women and younger people, who are more likely to express the intention to do and to report recent skin self-examination than men and older people (6, 7). That this may not be a complete explanation, however, is suggested by the significant 18% greater incidence of melanoma of the more continually exposed head and neck in summer than in winter, as also reported from the United States (8). This summer peak in incidence of melanomas on the head and neck was not much less than the increase on the trunk (25%), although substantially less than that on the limbs (60% +). The limbs are probably the first body sites to become more exposed as temperatures increase and may thus have both greater detection and greater sun exposure effects on their seasonal incidence. Thus, a short-term, tumor-promoting effect of sun exposure, simply due, perhaps, to stimulation of melanocyte proliferation from exposure of skin to UV radiation (9), remains a possibility, as is also suggested by pathologic changes and increased DNA synthesis in nevi leading to increased diagnosis of melanoma in summer (10, 11).

A decrease in Breslow thickness due to greater examination of the skin in summer might also decrease fatality from melanomas diagnosed in summer, as we observed. However, although adjustment for seasonal variation in thickness, and in sex, age, anatomic site, and histologic type of melanoma reduced the fatality difference between melanomas diagnosed in winter and summer, it remained at ∼20% (fatality ratio 0.82; 95% CI, 0.72-0.94). Whereas an adjusted relative fatality in spring that is a little less than that in summer and substantially less than that in autumn might seem inconsistent with a protective effect of sun exposure against melanoma fatality, monthly average ambient UV radiation in Sydney, the capital city of New South Wales, is on average nearly 2 UV index points higher in the spring months of September, October, and November than the autumn months of March, April, and May (http://www.arpansa.gov.au/uvindex/models/syduvmodel.htm). In addition, it is possible that behavioral factors might increase exposure of the skin in spring relative to that in summer and autumn because of relatively less concern about effects of sun exposure on the skin in the milder spring months.

This independent seasonal variation in melanoma fatality, with lower fatality in spring and summer, is consistent with a recent report that people who have had high sun exposure during their life have lower fatality after a diagnosis of melanoma than those who had lower sun exposure (12). This lower fatality was also independent of melanoma thickness and sex, age, and site of melanoma, as well as histologic evidence of ulceration and mitotic index of the melanomas, which had also been measured in that study. It was, moreover, independent of history of physician skin examination, skin self-examination, and “skin awareness,” thus giving further evidence that the effect could not be explained by behaviors promoting early diagnosis of melanoma when sun exposure is high. In addition to confirming these earlier results, our seasonal variation data further advance knowledge by strongly suggesting that sun exposure around the time of diagnosis has an important influence on survival. No time relationship between exposure and effect could be inferred from the previous results; equally, our results do not exclude effects of exposure at other times.

Up to 30% lower fatality from cancers of the breast, colon, and prostate in Norway diagnosed in summer and autumn than in winter has also been reported recently. These results add plausibility to our findings and suggest that the underlying mechanism for melanoma is not one mediated through a direct effect of sun exposure on behavior of the lesion. Antineoplastic effects of vitamin D produced by exposure of skin to solar UV radiation are thought to be the most likely cause (13, 14). There is direct and persuasive evidence for protective effects of vitamin D against occurrence of colorectal cancer in humans and, to a much less extent, cancers of the breast and prostate (15). In addition, limited clinical studies suggest that administration of vitamin D3 or a vitamin D analogue may limit the progression of advanced prostate cancer; these are the only direct studies of vitamin D effects on cancer survival to date (16, 17). Alternatively to vitamin D, it has been suggested that melanization and induction of DNA repair capacity by sun exposure might reduce the accumulation of mutational change in melanomas and thus increase survival (12).

Direct investigations of the relationship between measures of vitamin D sufficiency and its genetic determinants or modifiers in people with melanoma and their survival from melanoma are clearly needed. It would not be wise, however, at this early stage in understanding of the relationship between sun exposure and melanoma survival to limit attention solely to vitamin D. As noted above, the possibility of late-stage effects of sun exposure on melanoma development, which may contribute to seasonal variation in melanoma incidence, cannot be rejected completely. However, as we hypothesized above, such effects might be expected to promote more aggressive disease in addition to promoting tumor growth, thus leading to higher, not lower, fatality. Thus far, most available evidence on effects of UV radiation on melanoma cell lines or melanomas themselves point in this direction (18). If more aggressive disease proves to be the dominant direct consequence of sun exposure in the late stages of melanomagenesis, protective effects of vitamin D could be more important in limiting fatality of melanoma than they might be for other cancers.