Fiessler et al.1 stated that there is no association between month of birth and risk of melanoma in adult age. The authors analysed a large database from the cancer registry of Bavaria and claimed that their results contradict available studies, including ours from Italy.2–4 The arguments presented were: (i) the Bavarian study was larger than the other positive studies; (ii) the study presented methodological improvements over previous available evidence; and (iii) methodological weakness of previous studies was responsible for positive results. We believe that the association between date of birth and adult skin cancer risk should not be dismissed so quickly.
The hypothesis of a link between neonatal exposure to UV radiation and adult risk of skin cancer is in line with a wealth of available evidence on UV exposure in the young5–8 and with biological/immunological studies on tolerance.9–10 Thus, available evidence is sound with respect to biological plausibility and epidemiological data available for older age groups.8
Methodological criticism by Fiessler et al.1 of published studies is inappropriate and thus previous evidence is not weakened. In particular, commenting on our finding of an excess skin cancer risk for people born in the spring, they stated that no adjustment was made in our paper for birth month of the reference population.1 A seasonal effect, similar to our study, is apparent in the Bavarian study. However, seasonality was ascribed to bias on our part since it disappeared with the inclusion of an offset accounting for uneven distribution of births by month. Thus Fiessler et al.1 indirectly affirmed that our results were biased. However, all our results were ‘adjusted’ using the study population by month or even week of birth (Figure 1).
Seasonal distribution of Umbria population by week of birth for the period 1994–2010.
Indeed, seasonality of births was extensively described11 and we were aware of this phenomenon when writing our paper. The population by date of birth was used as denominator to calculate rates and as an offset in negative binomial regression models presented in our study. People born in the spring were more frequent in our population. Indeed, the inclusion of populations by date of birth in our analyses reduced (but did not cancel) the risk estimates with respect to using a single average population (Figure 2). More precisely, our study dealt with skin cancers and results were significant for other skin cancers and non-significant for melanoma of the skin. Moreover, we used skin cancer cases and regional population by week of birth and spline modelling to provide accurate and flexible estimates of the correlation between date of birth and skin cancer risk.12 The risk function for melanoma was strikingly similar to other skin cancers, and we speculated that a common risk function for all skin cancers was possible.
Spline modelling of the risk of skin melanoma by week of birth obtained using two different offsets: (a) uniform average population (unadjusted), and (b) population by week of birth (adjusted for seasonal distribution of population by date of birth).
Of course, it would be entirely inappropriate to include distribution of births as an additional offset variable in our study. Thus, our findings were not biased because of birth distribution and consequently the paper from Bavaria did not represent a methodological improvement on our own study. Instead, it is reasonable to question whether the offset based on the month of birth used in the German paper is equivalent to the use of actual population by date of birth adopted in our paper. Using the distribution of births by period did not reflect population at risk since change due to deaths and migration were not considered. This strategy could therefore lead to bias, including bias toward the null as acknowledged by Fiessler et al.1 Moreover, the distribution of cases in the German study was similar to our own and a consistent, although very small, excess risk for people born in the spring was still present in the model including the offset based on distribution of births. This point is particularly relevant in the present case because all analyses based on cancer registry data lacked information on exposure. Month of birth was assumed as a proxy of early neonatal sun exposure. Consequently, published investigations based on cancer registry data were probably affected by a differential misclassification bias since it is unlikely that all people born in the spring shared neonatal exposure to UV. Small excess risks observed in our descriptive analyses were likely to be underestimated.
Another point not considered in the paper by Fiessler et al.1 was that the resulting pattern of risk was remarkably similar in all three studies2–4 and this evidence supports the hypothesis of a common factor yet to be identified. In addition, we used splines to investigate pattern of risk by date of birth.12 The criticism of Fiessler et al.1 of the method used in the Swedish cohort study does not apply to our more flexible technique. Thus, it is unlikely that the Swedish result was an artefact due to the sinusoidal model used.
In conclusion, the large Bavarian study1 provides only inconclusive evidence on the relationship between season of birth and adult risk of melanoma, with respect to previously published studies.2–4 Lack of information on neonatal exposure and the use of historical birth distributions in the analyses may have contributed to their negative results. Moreover, positive evidence2–4 is not weakened by their methodological criticism. The pattern of early exposures among newborns deserves further investigation and analytical studies including data on individual exposure. Available evidence cannot be considered proof of the role of early UV exposure in development of adult melanoma, but discarding such evidence on the basis of a single ‘negative’ study could result in avoidable health burden and could delay further understanding of the aetiology of melanoma.
