To update the meta-analysis of 2006, we used the same methodological approach as previously described.11
Briefly, MB searched the literature published up to May 2012 using the databases PubMed, ISI Web of Science (Science Citation Index Expanded), Embase, Pascal, Cochrane Library, LILACS, and MedCarib. We used the following keywords for diseases: “skin cancer”, “squamous cell carcinoma”, “SCC”, “basal cell carcinoma”, “BCC”, and “melanoma”. To define exposure, we used the following keywords: “sunbed”, “sunlamp”, “artificial UV”, “artificial light”, “solaria”, “solarium”, “indoor tanning”, “tanning bed”, “tanning parlour”, “tanning salon”, and “tanning booth”. No language restriction was applied. We reviewed the titles and abstracts to identify potentially eligible studies and carried out a manual search of studies identified from references cited in reviews on skin cancer.
From the initial search we selected case-control, cohort, and cross sectional studies published as original articles. Non-eligible trials included ecological studies, case reports, reviews, and editorials.
PA and SG reviewed the selected articles and SG and MB abstracted the data using a standardised data collection protocol. The minimal common information on use of indoor tanning appliances for all studies was “ever used.” For those studies that did not strictly assess ever users of indoor tanning appliances compared with never users,13
we used the information closest to this category.
We also extracted the highest category of sunbed use reported in each study—that is, the greater duration (defined as “high use”) along with estimates of risk for the association with first use of sunbeds at a young age—before age 35 years.
We transformed every measure of association, adjusted for the maximum number of confounding variables, and 95% confidence intervals, into logarithms of relative risk and calculated the corresponding variance.15
When no estimates were reported, we used tabular data to calculate the crude estimates and 95% confidence intervals.
The meta-analysis was calculated from a random effect model as described previously16
—that is, a mixed effects model with summary relative risk obtained from maximum likelihood estimation. We calculated confidence intervals assuming an underlying t
distribution. Heterogeneity was assessed by Higgins and Thompson’s I2
statistic ranges from zero to 100%, zero indicating that the relative risks of the different studies included in the meta-analysis are homogeneous—that is, that the relative risks are consistent with each other.
We used a two step procedure to obtain summary risk estimates for dose-response. Firstly, we fitted a linear model within each study to estimate the relative risk per session of sunbed use. When sufficient information was published (the number of participants in usage category), we fitted the model according to a previously proposed method.18
This method provides the natural logarithm of the relative risk and an estimator of its standard error, taking into account that the estimates for separate categories depend on the same reference group. When the numbers of participants in each serum level category were not available from the publications, we calculated coefficients ignoring the correlation between the estimates of risk at the separate exposure levels. Secondly, we estimated the summary relative risk by pooling the study specific estimates with the mixed effects models.
All analyses were done with SAS Windows version 9.2. We used PROC MIXED in SAS to calculate the random effects models.
Heterogeneity and sensitivity analyses
We carried out several sensitivity analyses to evaluate the stability of the pooled estimates. Firstly we examined the pooled relative risks for case-control and prospective (cohort and nested case-control) studies separately. Then we examined changes to the results after the exclusion of specific studies.
To investigate heterogeneity between the studies we carried out metaregressions and subgroup analyses. Heterogeneity was investigated by looking at factors that could influence the quality of the studies and that could be responsible for heterogeneity, such as the study design, adjustment for confounding factors, features of the population, and publication year. As an additional analysis for heterogeneity, we compared risk estimates according to the average latitude of countries or areas where studies were done.
To investigate whether publication bias may have affected the validity of the estimates, we constructed funnel plots of the regression of log relative risk on the sample size, weighted by the inverse of the pooled variance. We evaluated publication bias using the Macaskill test.19
Sunbed use and burden of melanoma
To translate the estimation of risk in the current study to the burden in the general population, we provided a broad estimation of the burden of sunbed use in Europe. We gathered data on the prevalence of sunbed use from recent surveys carried out in Europe. As no survey was available for central European countries, we limited our estimation to the original 15 countries of the European Community (Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Spain, Sweden, Portugal, the Netherlands, and the United Kingdom) plus the three countries that are part of the European Free Trade Association (Iceland, Norway, and Switzerland). For these 18 countries, we extracted data on the incidence of melanoma from GLOBOCAN 2008.20
We identified seven surveys carried out in the 18 countries from which we extracted prevalence of ever having used a sunbed during lifetime.21
We also extracted the prevalence of sunbed use in the control group included in the Swedish cohort.14
Data were available for Denmark, France, Germany, Iceland, Spain, Sweden, Switzerland, and the United Kingdom. These countries represent 70% of all melanoma cases occurring in the 18 countries studied. Prevalence for the other 10 countries was determined from estimates for neighbouring countries.
We estimated the attributable fraction with Levin’s formula28
by using prevalence of ever use of sunbeds from surveys and the summary relative risk for ever use of sunbeds.