The incidence of melanoma has increased dramatically in countries with predominantly White populations in recent decades and is projected to continue to increase for several countries [1–4].
Two key potential drivers of increasing incidence are increased population exposure to ultraviolet radiation (the principal known environmental risk factor for melanoma) , and overdiagnosis of tumours that would not spread during that patient’s lifetime [6,7]. The former hypothesis (actual increase in the incidence of consequential illness) concords with increasing recreational exposure to sunlight since the 1950s [8,9]. However, stable trends in melanoma mortality despite increasing incidence have been observed, which may reflect either earlier diagnosis and more effective treatment of increasing consequential illness, or increasing overdiagnosis of relatively indolent lesions [10–14].
Analyses of stage-specific time trends of melanoma incidence could help to provide insights into mechanisms underlying increasing incidence: genuine increases in ultraviolet radiation exposure could be assumed to result in increased incidence of each stage equally, while overdiagnosis may lead to a disproportionately greater increase in early stage disease. Levell et al.  reported increasing incidence trends for stage I–III melanoma in 1991–2004, with a steeper increase for stage I disease. However, it is important to examine more recent patient cohorts, and potential differences in stage-specific variation over time by the sociodemographic group .
Patients and methods
We analysed anonymized data on cases of primary invasive malignant melanoma (International Classification of Diseases for Oncology-3 code C43) diagnosed during 1996–2015 in East Anglia, UK, for patients aged 15 years and older. These data were collected initially by the population-based cancer registry for the Anglia subregion of the East of England, and successor organizations who took over responsibility (currently the Public Health England National Cancer Registration and Analysis Service).
Patients who were first diagnosed with melanoma in the 15 years before the study period (1981–1995) (n = 309/10 199) were a priori excluded from analyses, while for those with multiple tumours diagnosed during the study period (1996–2015), the earliest diagnosis was selected [16,17]. For patients with more than one tumour in a given year, the tumour with the more advanced stage at diagnosis was included.
Information was available on the year of diagnosis, stage at diagnosis, sex, age group (15–49 years and 5-year age groups thereafter), deprivation group, the method of diagnosis (e.g. histological or other basis) and survival status to death or censoring. Tumour stage was assigned by registry staff on the basis of clinical, imaging, and pathological information according to the 4th edition of the TNM classification (i.e. unified AJCC/UICC classifications) . We combined stage III and IV cases into a single ‘advanced stage’ category, because of the small number of stage IV cases (n = 155/9890). The deprivation measure used was the Index of Multiple Deprivation (IMD) quintiles (fifths) of the income domain scores for the Lower Super Output Area of the patient’s residence at diagnosis: IMD 2004 for diagnoses in 1996–2002; IMD 2007 for 2003–2006; IMD 2010 for 2007–2011; IMD 2015 for 2012–2015. As we used all available data for the period 1996–2015, a sample size calculation was not carried out.
Our general approach was to calculate incidence rates by treating the number of melanoma cases as the numerator and the estimated mid-year population residing within the boundaries of East Anglia as the denominator. Population estimates were stratified by the year of diagnosis, sex, age group and deprivation . All analyses were performed in Stata v.13 (StataCorp., College Station, Texas, USA).
We first summarized the numbers of melanoma cases by sex, age group, deprivation, and stage, overall and by 5-year diagnosis periods. As 5.0% of patients (493/9890) had no recorded stage at diagnosis, we used multiple imputation before the main analyses – see below [20,21]. In a sensitivity analysis, we carried out a ‘complete case analysis’ (i.e. on those with nonmissing stage, n = 9397).
Incidence trends, by stage
We presented overall and stage-specific incidence trends, by plotting sex-specific age-standardized rates (using the European Standard Population 2013) by year , for all cases and by stage at diagnosis. We plotted these trends on the log-scale, to allow a fair representation of relative changes over time among stage groups with different baseline incidence rates.
We initially considered a Poisson model for assessing temporal trends while adjusting for stage, sex, age group, deprivation and the interaction term stage×year, but variability in incidence rates was larger than assumed by the Poisson model (P < 0.0001). Hence, we adopted negative binomial models that include an additional parameter for variance overdispersion (‘Model 1’ in Box S1, Supplemental digital content 1, http://links.lww.com/MR/A60) .
Negative binomial models provided estimates of adjusted incidence rate ratios (IRRs). The IRR for the independent variable year (hereafter referred to as the ‘annual IRR’) represents the percentage increase in incidence per year as: (IRR−1) × 100 (e.g. an annual IRR of 1.05 is equivalent to a 5% increase per year).
Stage-specific incidence trends, by patient group
We explored whether stage-specific incidence trends varied by patient group, by investigating whether including three-way and four-way interaction terms between the two demographic variables (age and sex) and the stage × year interaction improved the fit of the original negative binomial model (which included main effects variables for stage, sex, age group, deprivation, year and the interaction stage × year). The investigation of interaction terms was performed using a step-wise process. We included interactions with demographic variables in descending order of the strength of association for the main effects in the original model and used the likelihood ratio test at each step to assess the improvement of fit by including the variable at that step.
The final model included the same main effect variables (i.e. stage, sex, age group, deprivation and year), as well as a four-way interaction term sex × age group × stage × year (and its two-way and three-way components; ‘Model 2’ in Box S1, Supplemental digital content 1, http://links.lww.com/MR/A60). As this updated model required a substantially greater number of degrees of freedom, these variables were more broadly categorized as two stage groups (I, II–IV); three age groups (15–49, 50–69 and ≥70 years); and three deprivation groups (least deprived, 2nd to 4th quintiles, most deprived). Finally, we calculated the adjusted annual IRRs for each patient group (all combinations of stage, sex and age group), predicted from this updated model.
We created five datasets (following the generic rule that the number of imputations should correspond to the percentage of missing data) , where missing stage was imputed. We used multinomial logistic regression, on the basis of sex, age group, deprivation group, survival status, and the Nelson–Aalen estimator of the cumulative hazard , censoring survival information at 1 year post-diagnosis, to minimize bias from deaths unrelated to the melanoma. We carried out analyses as described above on each of these five datasets, using Rubin’s rules to combine resulting estimates . In sensitivity analysis, we carried out a ‘complete case analysis’ by restricting the analysis sample to only cases with observed stage, and compared the two approaches as the percentage change in the model coefficients (equivalent to the percentage change in IRRs on the log-scale).
There were 9890 cases of melanoma diagnosed in East Anglia during 1996–2015 in an approximate population of 2.5 million during the study period. Table 1 shows the patients’ characteristics by 5-year periods. Just over half (51%) of the cases were women, and just over half (53%) of the cases were under 65 years of age. Stage I disease accounted for the majority of cases (62%) during the entire study period. The proportion of younger patients (15–49 year-olds) decreased over time (from 31% in 1996–2000 to 20% in 2011–2015), while the proportion of stage I cases increased (from 57 to 64%).
Incidence trends, by stage
During the study period, the age-standardized incidence of melanoma increased substantially from 14.0 per 100 000 in 1996 to 29.4 in 2015 (i.e. by + 4% annually) (Fig. 1 and Table S1, Supplemental digital content 1, http://links.lww.com/MR/A60). Overall (all-stage) incidence rates increased with time with a slight acceleration around 2005 (women: 15.0 per 100 000 in 1996 to 27.6, men: 13.3–32.5, Fig. 1). This overall time trend was mirrored by that for stage I cases, where the age-standardized incidence increased from 8.2 per 100 000 in 1996 to 20.3 in 2015 ( + 5% annually). Age-standardized incidence of stage II and stage III/IV cases also increased, but less steeply than for stage I cases ( + 2.8 and + 1.9%, annually).
After adjustment for sex, age group and deprivation, there was strong statistical evidence for increasing incidence in stage I, II, and III/IV disease (P < 0.0001 for all adjusted annual IRRs), but a steeper increase for stage I disease [adjusted annual IRR for stage I disease: 1.05, i.e., a 5% increase per year, 95% confidence interval (CI): 1.05–1.06; stage II: 1.03, i.e., 3% increase per year, 95% CI: 1.02–1.04; stage III/IV: 1.02, i.e., 2% increase per year, 95% CI: 1.01–1.03; Table 2].
Stage-specific incidence trends, by sex–age subgroups
There was an increasing incidence of stage I disease for all age-sex strata, with adjusted annual IRR estimates significantly greater than 1 for all groups [lower bounds of 95% CIs for IRRs ranged from 1.01 to 1.10 (Fig. 2 and Table 3)]. Larger increases in the incidence of stage 1 disease over time were seen with increasing age [e.g. adjusted annual IRR for women, ≥ 70 years: 1.08 (95% CI: 1.08–1.09), women overall: 1.04 (95% CI: 1.04–1.05)]. Within each age group, adjusted annual IRRs were larger for men than women.
Considering the respective adjusted annual IRRs for stage II–IV melanoma, increasing temporal trends in incidence were observed both in older women [e.g. IRR for ≥ 70 years: 1.03 (95% CI: 1.02–1.03)] and, more so, in older men [e.g. ≥ 70 years: 1.05 (95% CI: 1.04–1.06)] (Fig. 2 and Table 3). In younger people (age: 15–49 years) the increase in the incidence of stage II–IV melanoma was minimal, and not significant for women [adjusted annual IRR for women: 1.00 (95% CI: 0.99–1.01), P = 0.38; men: 1.01 (95% CI: 1.00–1.02), P = 0.01]. Increases in rates of stage II–IV disease were greater in older age, but this gradient was not as steep as that for stage I disease.
The complete-case sensitivity analyses produced similar results to those obtained in the main analyses, where missing stage was imputed (also see Table 2 and Fig. S1 and Table S2, Supplemental digital content 1, http://links.lww.com/MR/A60).
Summary of main findings
The age-standardized incidence of melanoma increased substantially in the East Anglia region during 1996–2015. This increase predominantly reflected the increasing incidence of stage I cases, while the incidence of stage II–IV disease increased less steeply. In women aged under 50, though there was an increased incidence of stage I cases, this was not accompanied by an increase in the incidence of late-stage disease. For either sex, the increase in incidence was steeper in the oldest patients (≥70 years), and across age groups, increases in incidence were steeper in men than women.
Strengths and limitations
The principal strengths of this analysis are the use of population-based data covering an extensive period, using high quality and highly complete information on stage at diagnosis. To account for potential biases in estimates arising from missing stage data, we used multiple imputation. Nonetheless, as shown in sensitivity analyses, most estimates obtained from complete-case analysis were similar to those obtained using multiple imputed data.
The main limitation is that we cannot confidently infer the causes of the observed stage-specific incidence trends. Further analyses that consider changes in case-mix over time of other disease factors, for example, in tumour subsite (trunk, limbs, face, etc.) or diagnostic route (e.g. identification through healthcare encounters triggered by unrelated reasons), or in the rate of investigations for suspected melanoma in the general population, could help elucidate potential mechanisms. We examined data from the former East Anglia cancer registry, a geographically-defined population of South England whose residents are relatively older and more affluent, and more likely to be of White ethnic origin, compared with the rest of the English population . These considerations might limit the generalizability of the findings, though not their internal validity. Analyses of England-wide data for the same 20-year era (to overcome these generalizability limitations) are not possible for our study period, because of the historically poor nationwide completeness of stage information for melanoma, until recent years. We examined the stratification of stage-specific incidence by sex and age group, but larger studies could additionally enable stratification by socioeconomic status.
Comparisons with the literature
Our study indicates a continuation (to 2015) of previously reported trends in our population (1991–2004) . The observed steeper increase in stage I incidence mirrors upwards incidence rates of thin melanoma tumours ( < 1.5 mm), and incidence rates of melanoma in certain subsites (e.g. trunk, head and neck) in studies covering similar time periods in Denmark [28,29], Finland , Scotland  and Northern England .
Implications for policy, practice and research
There are two main implications arising from the findings. First, the observed trends are compatible with a hypothesis of potential overdiagnosis of indolent cases, across age groups but would be wrong to assign all of the increase in the incidence of melanoma to overdiagnosis. Had this been the case, no increase in advanced stage at diagnosis would have been observed in men (all age groups) and women (age ≥ 50 years). Second, the observed trends are also compatible with a hypothesis of increasing rates of genuinely consequential illness for older patients, particularly older men. Such a hypothesis is supported by evidence from South-Eastern Europe, where the increasing incidence of melanoma has been accompanied by increasing mortality for men and women aged 50 and older , and the USA, where increasing incidence was not accompanied by any decrease in tumour thickness for stage III and IV patients .
Quantifying the extent of overdiagnosis or increasing rates of consequential disease in a population is challenging. Nevertheless, our data suggest that both appear to be plausible, partial, explanations. These realizations emphasize the importance of shared decision making about investigating skin lesions and communication of issues around overdiagnosis in conversations with patients .
The 1–5% annual increase in advanced stage melanoma (II–IV) in women aged 50 and older and men in all age groups (Fig. 2 and Table 3), indicates a genuine increase in late-stage melanoma, which accounts for a 50% increase in incidence in older women and a 94% increase in older men over the 20-year study period (Table 3). This may reflect different risk behaviours with regard to sun exposure or different biological behaviours of melanoma in different age groups. The observation of different patterns in the incidence of stage II–IV melanoma women in the under 50s and those aged 50 and older warrants further investigation.
The increase in melanoma incidence puts pressure on limited health service budgets, particularly with regard to the use of novel, effective, but expensive biological treatments for late-stage disease, which have become available in recent years, offering life-extending management options. Our findings indicate an ongoing need for public health education campaigns, to increase awareness of melanoma risk factors and symptoms and achieve earlier diagnosis, particularly in the elderly, and in men, who experienced the steepest increases in stage II–IV disease.
The findings suggest that both a genuine increase in the incidence of consequential illness and a degree of overdiagnosis are likely to be responsible for the observed increasing incidence trends in our population during the study period. They also suggest that targeting of men and older people in future public health awareness campaigns may be justified.
G.L. and A.H. are supported by a Cancer Research UK Advanced Clinician Scientist Fellowship Award to G.L. (C18081/A18180). G.L. is an associate director (co-investigator) of the multi-institutional CanTest Research Collaborative funded by a Cancer Research UK Population Research Catalyst award (C8640/A23385). This project involves data derived from patient-level information collected by the NHS, as part of the care and support of cancer patients. The data are collated, maintained and quality assured by the National Cancer Registration and Analysis Service, which is part of Public Health England.
Conflicts of interest
There are no conflicts of interest.
1. De Vries E, Bray FI, Eggermont AM, Coebergh JW; European Network of Cancer
R. Monitoring stage
-specific trends in melanoma
incidence across Europe reveals the need for more complete information on diagnostic characteristics. Eur J Cancer
Prev. 2004; 13:387–395
2. Garbe C, Leiter U. Melanoma
epidemiology and trends. Clin Dermatol. 2009; 27:3–9
4. Whiteman DC, Green AC, Olsen CM. The growing burden of invasive melanoma
: projections of incidence rates and numbers of new cases in six susceptible populations through 2031. J Invest Dermatol. 2016; 136:1161–1171
5. Parkin DM, Mesher D, Sasieni P. 13. Cancers attributable to solar (ultraviolet) radiation exposure in the UK in 2010. Br J Cancer
. 2011; 105(Suppl 2)S66–S69
6. Welch HG, Black WC. Overdiagnosis
. J Natl Cancer
Inst. 2010; 102:605–613
7. Marcus PM, Prorok PC, Miller AB, DeVoto EJ, Kramer BS. Conceptualizing overdiagnosis
screening. J Natl Cancer
Inst. 2015; 107:djv014
8. Leiter U, Garbe C. Epidemiology of melanoma
and nonmelanoma skin cancer
: the role of sunlight. Adv Exp Med Biol. 2008; 624:89–103
9. Wallingford SC, Alston RD, Birch JM, Green AC. Regional melanoma
incidence in England, 1996–2006: reversal of north-south latitude trends among the young female population. Br J Dermatol. 2013; 169:880–888
10. Weyers W. The ‘epidemic’ of melanoma
between under- and overdiagnosis
. J Cutan Pathol
. 2012; 39:9–16
11. Glusac EJ. The melanoma
‘epidemic’: lessons from prostate cancer
. J Cutan Pathol. 2012; 39:17–20
12. De Giorgi V, Gori A, Grazzini M, Rossari S, Oranges T, Longo AS, et al. Epidemiology of melanoma
: is it still epidemic? What is the role of the sun, sunbeds, Vit D, betablocks, and others? Dermatol Ther. 2012; 25:392–396
13. Levell NJ, Beattie CC, Shuster S, Greenberg DC. Melanoma
epidemic: a midsummer night’s dream? Br J Dermatol. 2009; 161:630–634
14. Welch HG, Woloshin S, Schwartz LM. Skin biopsy rates and incidence of melanoma
: population based ecological study. BMJ. 2005; 331:481
15. NHS Digital.. Indicator Portal: Record of Stage
at Diagnosis. NHS Digital; 2015. Available at: https://indicators.hscic.gov.uk/webview
. [Accessed 10 May 2018]
16. Rowe CJ, Law MH, Palmer JM, MacGregor S, Hayward NK, Khosrotehrani K. Survival outcomes in patients with multiple primary melanomas. J Eur Acad Dermatol Venereol. 2015; 29:2120–2127
17. Hwa C, Price LS, Belitskaya-Levy I, Ma MW, Shapiro RL, Berman RS, et al. Single versus multiple primary melanomas: old questions and new answers. Cancer
. 2012; 118:4184–4192
18. Hermanek P, Sobin LH. TNM classification of malignant tumours. 19874th fully rev ed, Berlin, London: Springer
20. Howlader N, Noone AM, Yu MD, Cronin KA. Use of imputed population-based cancer
registry data as a method of accounting for missing information: application to estrogen receptor status for breast cancer
. Am J Epidemiol. 2012; 176:347–356
21. Falcaro M, Carpenter JR. Correcting bias due to missing stage
data in the non-parametric estimation of stage
-specific net survival for colorectal cancer
using multiple imputation. Cancer
Epidemiol. 2017; 48:16–21
22. Pace M, Lanzieri G, Glickman M, Grande E, Zupanic T, Wojtyniak B, et al. Revision of the European Standard Population: report of Eurostat’s task force;. 2013. Available at: http://ec.europa.eu/eurostat/documents/3859598/5926869/KS-RA-13-028-EN.PDF/e713fa79-1add-44e8-b23d-5e8fa09b3f8f
. [Accessed 2 March 2017]
23. Cameron AC, Trivedi PK. Regression analysis of count data. 1998, Cambridge: Cambridge University Press
24. von Hippel PT. Regression with missing Y’s: an improved strategy for analyzing multiply imputed data. Sociol Methodol. 2007; 37:83–117
25. White IR, Royston P. Imputing missing covariate values for the Cox model. Stat Med. 2009; 28:1982–1998
26. Rubin DB. Multiple imputation for nonresponse in surveys. 1987, Chichester, NY: Wiley
28. Fuglede NB, Brinck-Claussen UO, Deltour I, Boesen EH, Dalton SO, Johansen C. Incidence of cutaneous malignant melanoma
in Denmark, 1978–2007. Br J Dermatol. 2011; 165:349–353
29. Bay C, Kejs AM, Storm HH, Engholm G. Incidence and survival in patients with cutaneous melanoma
by morphology, anatomical site and TNM stage
: a Danish Population-based Register Study 1989–2011. Cancer
Epidemiol. 2015; 39:1–7
30. Stang A, Pukkala E, Sankila R, Soderman B, Hakulinen T. Time trend analysis of the skin melanoma
incidence of Finland from 1953 through 2003 including 16 414 cases. Int J Cancer
. 2006; 119:380–384
31. MacKie RM, Bray C, Vestey J, Doherty V, Evans A, Thomson D, et al. Melanoma
incidence and mortality in Scotland 1979–2003. Brit J Cancer
. 2007; 96:1772–1777
32. Downing A, Newton-Bishop JA, Forman D. Recent trends in cutaneous malignant melanoma
in the Yorkshire region of England; incidence, mortality and survival in relation to stage
of disease, 1993–2003. Br J Cancer
. 2006; 95:91–95
33. Barbaric J, Sekerija M, Agius D, Coza D, Dimitrova N, Demetriou A, et al. Disparities in melanoma
incidence and mortality in South-Eastern Europe: Increasing incidence and divergent mortality patterns. Is progress around the corner? Eur J Cancer
. 2016; 55:47–55
34. Shaikh WR, Dusza SW, Weinstock MA, Oliveria SA, Geller AC, Halpern AC. Melanoma
thickness and survival trends in the United States, 1989 to 2009. J Natl Cancer
Inst. 2016; 108::djv294
35. Ghanouni A, Renzi C, McBride E, Waller J. Comparing perceived clarity of information on overdiagnosis
used for breast and prostate cancer
screening in England: an experimental survey. BMJ Open. 2017; 7:e015955