Dear Editor, Tobacco smoking poses a great global health burden to > 1 billion smokers and those subject to passive smoking, and is well known to be a causal factor in at least 12 types of cancer.1 In terms of cutaneous malignant melanoma (CMM), which was reported by the World Health Organization to account for 232 130 new cases and 55 489 deaths in 2012, its association with smoking has so far not been established. To date, only a few epidemiological studies have indicated a positive association,2 3 4 5 and, interestingly, the remaining majority reported statistically significant or nonsignificant inverse correlations.6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 To shed light on this relationship, we report herein the results of our meta‐analysis on the effects of smoking on CMM mortality and incidence. A systematic search of articles was conducted in MEDLINE, Embase and the Cochrane library, from their start dates to January 2015, with ‘melanoma’ OR ‘skin neoplasm’ OR ‘skin cancer’ in the title or abstract, AND ‘smok*’ OR ‘tobacco’ OR ‘cigar*’ OR ‘waterpipe’ OR ‘snuff’ in the full text in order to locate studies directly focused, or indirectly reporting, on smoking related to CMM. Studies were considered if data on smoking status were collected and could be tabulated with CMM, or indexes such as odds ratio (OR), relative risk (RR) and hazard ratio, with corresponding 95% confidence intervals (CIs), were reported. The largest or most recent reports were adopted in terms of repetition. Quality assessment of the studies by the Newcastle–Ottawa scale and data extraction procedures were performed by two independent reviewers; any discrepancy was resolved by discussion. As CMM is rare, we combined and also separately analysed data from case–control and cohort designs. A fixed‐effect model was applied if the heterogeneity index I2 was < 50%; otherwise, a random effects model was adopted in pooling summary estimates. All statistical analyses were performed using RevMan Version 5.3 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark). The systematic search yielded 451 citations. By screening the titles and abstracts, the full text of 88 articles was deemed suitable for assessment. After the exclusion of publications containing irrelevant study types, duplicate reports and studies failing to obtain risk estimates (after contacting the corresponding authors three times), 23 studies were included (Table 1). The quality of the included studies was generally high or moderate, except for five studies,3 12 13 18 22 which had total scores < 6. Basic characteristics of included studies NA, not available; BMI, body mass index. aNumber of datasets separately used for analysis in parentheses. Basic characteristics of included studies NA, not available; BMI, body mass index. aNumber of datasets separately used for analysis in parentheses. Based on information from one study,16 which took CMM mortality as the outcome, and one that studied both incidence and mortality,4 pooled estimated RRs for smoking on CMM mortality were 0·79 (95% CI 0·67–0·92) and 0·91 (95% CI 0·80–1·04) for current and former smokers, respectively (no information on ever smoking was available). Figure 1 presents pooled estimates of smoking on CMM incidence. RR estimates were 0·70 (95% CI 0·63–0·78), 0·90 (95% CI 0·85–0·95) and 0·92 (95% CI 0·87–0·94) for current, former and ever smokers, respectively. Forest plot of smoking and incidence of cutaneous malignant melanoma. Data are presented as odds ratio (95% confidence interval). NA, not applicable; HPFS, Health Professionals Follow‐up Study; NHS, Nurses’ Health Study. Based on these result, an inverse association of smoking with CMM may be seen. This is further strengthened by exploring the dose–response relationship of some of the included studies. In the Health Professionals Follow‐up Study cohort,23 the adjusted RR was 0·77 for men who smoked < 15 cigarettes daily and fell sharply to 0·32 for men who smoked more; the same pattern was also shown in women, although the adjusted estimates were weaker (1·0 and 0·94, respectively). Similar results were obtained in the cohort of Klatsky et al. and the case–control study of Westerdahl et al.10 24 Classification of the degree of smoking was set at none and 20 cigarettes daily. The unadjusted RRs fell from 0·9 to 0·6, and the adjusted ORs from 0·7 to 0·6, respectively. The dose–response studies of Østerlind et al. and Freedman et al. resulted in irregular trends but still showed a general inverse relationship for most subgroups.5 8 As a group I carcinogen,1 the traditional harmful impression of tobacco smoking gives rise to some explanations as to the observed ‘protective’ effect on CMM. In one simulation study, Thompson et al. demonstrated that competing risks from smoking‐related noncommunicable diseases and cancers may result in an inverse association when smoking has no effect on melanoma.25 Cohort study results may suffer from this type of bias. Therefore, we separately analysed the results by study type. In cohort studies, the pooled RRs were 0·69 (95% CI 0·62–0·77), 0·88 (95% CI 0·77–1·01) and 0·92 (95% CI 0·83–1·01) for current, former and ever smokers, respectively. However, a similar trend was observed in case–control studies (not affected by competing risk): ORs were 0·70 (95% CI 0·57–0·86), 0·84 (95% CI 0·75–0·95) and 0·83 (95% CI 0·76–0·91) for the three smoking statuses, respectively. In addition, the observed inverse relations were also not influenced by design details, such as matching method and study base (all RR < 1·0). Sensitivity analysis of study quality by excluding low‐quality studies increased the RR estimates nonsignificantly but did not affect the inverse relation: current smoking from 0·71 to 0·72, ever smoking from 0·92 to 0·93 and former smoking from 0·90 to 0·94. Another widely adopted concern in explaining the intriguing results is insufficiency in the adjustment for confounders between the exposure of smoking and outcome of CMM. By obtaining and analysing all the raw and different levels of adjusted estimates, no known factor (i.e. age, sun exposure, ethnicity and number of naevi) demonstrated the ability to reverse a harmful effect by confounding. One concern in inferring this association is selective reporting in the included studies (moderate bias was detected in the ever smoking analysis; P < 0·05). Subgroup analysis of study purpose was also carried out, which showed that the ‘protective’ effect may be over‐reported: RR 0·68 (95% CI 0·61–0·76) for direct report vs. 0·78 (95% CI 0·57–1·05) for indirect report. We found 38 CMM publications mentioning a measurement of smoking status with no relevant data shown; however, these efforts were in vain because of the inability to contact the authors of the earlier study and the unwillingness to provide data for recent studies. Thus, publication bias was not negligible. In summary, the results, based on > 8000 cases of CMM, of this analysis indicate that smoking is associated with a reduced risk of CMM. However, based on publication bias and the lack of adjustment for unknown confounders, the conclusion should be adopted with caution. More epidemiology reports and mechanistic studies are therefore warranted. Needless to say, the current results should definitely not be used as a prompt to start smoking. Funding sources: none. Conflicts of interest: none declared.
No takes yet. Share an insight, caveat, or question.
Li et al. (2015) studied this question.
Synapse has enriched 5 related papers. Consider them for comparative context: