Abstract
Background:
Guidelines for follow-up of melanoma patients are based on limited evidence.
Objectives:
To guide skin surveillance, we developed a risk prediction model for subsequent primary melanoma based on demographic, phenotypic, histopathologic, sun exposure, and genomic risk factors.
Methods:
Using Cox regression frailty models, we analysed data for 2,613 melanomas from 1,266 patients recruited to the population-based Genes, Environment and Melanoma (GEM) study in New South Wales, Australia, with a median of 14 years follow-up via the cancer registry. Discrimination and calibration were assessed.
Results:
The median time to diagnosis of a subsequent primary melanoma decreased with each new primary. The final model included 12 risk factors. Harrell’s C-statistic was 0.73 (95% confidence interval [CI] 0.68–0.77), 0.65 (95% CI 0.62–0.68) and 0.65 (95% CI 0.61–0.69) for predicting second, third and fourth primary melanomas, respectively. The risk of a subsequent melanoma was 4.75 times higher (95% CI 3.87–5.82) for the highest versus lowest quintile of the risk score. The mean absolute risk of subsequent primary melanoma within 5 years was 8.0% (standard deviation [SD] 4.1%) after the first melanoma and 46.8% (SD 15.0%) after the second, but varied substantially by risk score.
Conclusions:
The risk of developing a subsequent primary melanoma varies considerably between individuals and is particularly high for those with two or more previous melanomas. This risk prediction model and nomograms enables estimation of absolute risk of subsequent melanoma based on an individual’s risk factors and can be used to tailor surveillance intensity, communicate risk and provide patient education.
INTRODUCTION
People with a previous primary melanoma have a higher risk of developing a second primary invasive melanoma compared with the general population.1 Secondary prevention through routine surveillance is important to ensure that subsequent melanomas are detected and treated at an early stage when prognosis is better.2 However, skin surveillance protocols for melanoma patients, including intervals between clinic visits and duration of follow-up, differ considerably by country because the clinical practice guidelines are underpinned by low levels of evidence.3 In some countries, follow-up intervals are tailored to individual risk factors, particularly number of previous primary melanomas, dysplastic naevi, family history of melanoma and CDKN2A mutations.1,4 However, there are currently no detailed risk prediction models that combine multiple risk factors to estimate an individual’s absolute risk of developing a subsequent primary melanoma.
Compared to using a one-size-fits-all approach for skin surveillance, a tailored approach, whereby higher-risk individuals have more regular and/or more specialised skin surveillance than lower-risk individuals, could improve patient outcomes through earlier detection of new melanomas, and have psychological benefits such as reduced anxiety in lower-risk individuals. It may also lead to more efficient use of health care resources.3
To facilitate implementation of individual patient risk assessment and tailored skin surveillance for melanoma patients, a comprehensive prediction model for subsequent primary cutaneous melanoma was developed using data from a large population-based sample of melanoma patients with a median 14 years of follow-up. Demographic, phenotypic, histopathologic, sun exposure, and genomic risk factors were included, and nomograms were produced to facilitate clinical use.
METHODS
Study design, sample and data collection
Between 2000 and 2003, 1,282 participants with an incident single or subsequent primary cutaneous melanoma living in New South Wales (NSW), Australia, were identified from the NSW Cancer Registry and recruited to the international Genes, Environment and Melanoma (GEM) study.5 The GEM study was designed as a case-control study in which controls had single primary melanomas and cases had multiple primary melanomas, thus the sample was enriched for individuals with multiple primary melanomas.5 Physician approval was obtained and participants gave written informed consent; participation was 54%. Pathology reports and slides were centrally reviewed for the index melanomas. For this analysis, the NSW GEM study was converted from a case-control study to a cohort study design by obtaining data for participants’ other invasive or in situ primary melanoma diagnoses (for the period 1960–2011) and deaths, through linkage to the NSW Cancer Registry. Entry to the cohort was on the date of the participant’s first melanoma diagnosis recorded by the cancer registry (which may have been before GEM study enrolment) and follow-up ended on the date of last complete search of the registry for further primary melanomas or at date of death; the median length of follow-up was 14.3 years (range 0.4–51.5 years). Ethics approval was obtained from The University of Sydney and the NSW Population and Health Services Research Ethics Committees.
Data were collected by self-administered questionnaire, telephone interview, and buccal brushes or blood samples; more details are provided in the Supplementary online material.
Statistical methods
Twenty-one well-known demographic, sun exposure3,4, phenotypic, genomic and histopathologic risk factors that could be easily ascertained in clinical settings were assessed as candidate predictors for diagnosis of new primary melanoma (see Supplementary online material for a description of variables, the model selection process and internal validation steps). After excluding 16 participants (30 melanomas) with missing values, the analysis included 1,266 participants and 2,613 melanomas.
Data were analysed using a proportional hazard conditional frailty model,6 an extension of Cox regression. Frailty models maximise statistical power and account for the fact that some events (such as development of multiple melanomas) may not be independent, by incorporating within-cluster correlation as normally distributed random effects in the model. Patients were censored in the period from their last known melanoma diagnosis date to their last follow-up or death. This model allows multiple records for each patient; thus, a patient who developed two new melanomas contributed three time-intervals in the model.
Estimation of relative and absolute risks
Hazard ratios and 95% confidence intervals (CIs) were estimated for each predictor in the final model, analyzing all melanomas simultaneously. Absolute risks of subsequent melanoma and model performance were estimated using standard Cox proportional hazards modeling stratified by the number of previous primary melanomas. The 1, 5 and 10-year absolute risks for each individual were calculated using the risk score from the frailty model, and the baseline risks for new primary melanoma calculated from the survival function for each stratum.7 The baseline risks for development of a second primary melanoma were calculated from original GEM controls (recruited at their first primary melanoma) to avoid bias from oversampling of multiple primary melanomas in the cohort; whereas baseline risks for third and higher-order primaries were calculated from the whole sample. Nomograms were created for estimating absolute risks.8,9
Performance measures
Discrimination was assessed using Harrell’s C-statistic, a measure of the area under the Receiver Operating Characteristics curve (AUC) suitable for censored survival data, and by plotting Kaplan-Meier curves.9 Here, the C-statistic is the probability that the predicted risk is higher for an individual who develops another melanoma than for an individual who does not, and ranges from 0.5 (equivalent to a coin toss) to 1.0 (perfect discrimination).9 Calibration was assessed by comparing the predicted probability of a subsequent primary melanoma for each decile of risk using the risk prediction model, against the observed frequency of melanomas obtained from the Kaplan-Meier curves for the same decile subgroup, and stratified by the number of previous primary melanomas. Confidence intervals for the C-statistic were derived from 10 bootstrap samples. The fit of the Cox model was assessed by plotting Cox-Snell residuals against the cumulative hazard.8 We repeated the main results excluding the polygenic risk score from the prediction model, as these data are not currently routinely available for melanoma patients. Analyses were performed with SAS software (version 9.4). We followed the TRIPOD statement10 for reporting the study.
RESULTS
Of 1,266 participants, 62.5% were men. The median age at diagnosis of first melanoma was 58.9 years (standard deviation [SD] 14.8 years), and 17.1% were aged < 45 years. The majority (53.9%) of participants had high school as their highest level of education, and 31.8% had tertiary education. Of 2,613 melanomas, 16.5% were in situ, and 83.5% were invasive (of which 68% ≤ 1.0mm Breslow). The number of primary melanomas per individual ranged from 1 to 16, with over half the sample having more than one primary melanoma (Table 1). The median time between melanomas decreased with each new primary, from 3.7 years between the first and second to 0.7 years between sixth and subsequent primaries (Table 1).
Table 1.
Number of primary melanomas, time between primaries and follow-up time since participants’ first primary melanoma
| Number of primary melanomas1 per individual |
Subjects (N=1266) N (%) |
Follow-up time (years) since first primary melanoma Median (25th–75th centile) |
Timing | Time between melanoma primaries (years) Median (25th–75th centile) |
|---|---|---|---|---|
| 1 primary only | 590 (46.6%) | 14.3 (14.1–14.4) | 1st to 2nd primary | 3.7 (0.8–8.8) |
| 2 primaries only | 360 (28.4%) | 14.5 (12.7–19.7) | 2nd to 3rd primary | 3.0 (0.6–5.8) |
| 3 primaries only | 152 (12.0%) | 15.5 (12.8–22.3) | 3rd to 4th primary | 1.9 (0.5–4.0) |
| 4 primaries only | 93 (7.3%) | 15.1 (12.9–20.1) | 4th to 5th primaries | 1.8 (0.7–3.9) |
| 5 primaries only | 31 (2.4%) | 19.6 (14.2–21.9) | 5th to 6th primaries | 0.8 (0.1–3.0) |
| 6 + primaries (max 16) | 40 (3.2%) | 20.8 (16.0–28.8) | 6th to subsequent | 0.7 (0.1–2.1) |
| Overall | 14.3 (13.4–15.8) | |||
Of 2,613 melanomas, 16.5% were in situ, and 83.5% were invasive; of invasive melanomas: 68% were ≤ 1.0mm Breslow thickness, 17% were 1.1–2.0mm, 9% were 2.1–4.0mm, 4% were > 4mm, and 1% unknown Breslow thickness.
Twelve factors were retained in the final reduced model (Table 2). More details of the model selection results are provided in the Supplementary online material. The risk of a subsequent melanoma was 4.75 times higher (95% CI 3.87–5.82) for participants in the highest versus lowest quintile of risk. The 1-year, 5-year and 10-year estimated absolute risks derived from the risk prediction model for developing a second melanoma or third melanoma are shown in Figure 1, and for developing a fourth melanoma in eFigure 1. Over 5-years, the mean risks were 8.0% (SD 4.1%), 46.8% (SD 15.0%), and 52.5% (SD 15.3%) for people with one, two and three melanomas, respectively.
Table 2.
Distribution of risk factor characteristics in the sample overall and according to number of previous primary melanomas, and hazard ratios (HRs) for their association with subsequent melanoma derived from a multivariable Cox proportional hazards frailty model
| Predictor | HR1 (95% CI) | P-value | Total (N=1,266) N (%) |
Number of previous primary melanomas | ||
|---|---|---|---|---|---|---|
| 1 only (N=590) N (%) |
2 only (N=360) N (%) |
≥ 3 (N=316) N (%) |
||||
| Sex | ||||||
| Female | 1.00 | <0.0001 | 475 (37.5%) | 273 (46.3%) | 133 (36.9%) | 69 (21.8%) |
| Male | 1.46 (1.26, 1.70) | 791 (62.5%) | 317 (53.7%) | 227 (63.1%) | 247 (78.2%) | |
| Age category (years) at first melanoma | ||||||
| <45 | 1.00 | <0.0001 | 217 (17.1%) | 133 (22.5%) | 48 (13.3%) | 36 (11.4%) |
| 45 to 54 | 1.25 (0.99, 1.57) | 241 (19.0%) | 119 (20.2%) | 74 (20.6%) | 48 (15.2%) | |
| 55 to 64 | 1.79 (1.44, 2.24) | 301 (23.8%) | 112 (19.0%) | 99 (27.5%) | 90 (28.5%) | |
| 65 to 74 | 1.87 (1.49, 2.34) | 318 (25.1%) | 124 (21.0%) | 92 (25.6%) | 102 (32.3%) | |
| ≥ 75 | 1.78 (1.37, 2.32) | 189 (14.9%) | 102 (17.3%) | 47 (13.1%) | 40 (12.7%) | |
| Previous keratinocyte cancer | ||||||
| No | 1.00 | <0.0001 | 769 (60.7%) | 435 (73.7%) | 209 (58.1%) | 125 (39.6%) |
| Yes | 1.47 (1.30, 1.66) | 486 (38.4%) | 155 (26.3%) | 144 (40.0%) | 187 (59.2%) | |
| Don’t know | 1.66 (1.10, 2.49) | 11 (0.9%) | 0 (0%) | 7 (1.9%) | 4 (1.3%) | |
| Family history of melanoma in first degree relatives | ||||||
| No/unknown | 1.00 | 0.0006 | 1021 (80.6%) | 506 (85.7%) | 273 (75.8%) | 242 (76.6%) |
| Yes | 1.27 (1.11, 1.45) | 245 (19.4%) | 84 (14.2%) | 87 (24.2%) | 74 (23.4%) | |
| Skin colour | ||||||
| Olive/brown/dark | 1.00 | 0.05 | 153 (12.1%) | 91 (15.4%) | 41 (11.4%) | 21 (6.6%) |
| Fair | 1.33 (1.05, 1.68) | 849 (67.1%) | 381 (64.6%) | 240 (66.7%) | 228 (72.2%) | |
| Very fair | 1.37 (1.05, 1.78) | 264 (20.9%) | 118 (20.0%) | 79 (21.9%) | 67 (21.2%) | |
| Naevus (mole) density | ||||||
| None | 1.00 | 0.05 | 349 (27.6%) | 166 (28.1%) | 101 (28.1%) | 82 (25.9%) |
| Few | 1.08 (0.93, 1.24) | 664 (52.4%) | 331 (56.1%) | 170 (47.2%) | 163 (51.6%) | |
| Some | 1.23 (1.04, 1.46) | 200 (15.8%) | 72 (12.2%) | 74 (20.6%) | 54 (17.1%) | |
| Many | 1.31 (1.00, 1.71) | 53 (4.2%) | 21 (3.6%) | 15 (4.2%) | 17 (5.4%) | |
| Ability to tan | ||||||
| Deep/moderately tan | 1.00 | 0.01 | 663 (52.4%) | 321 (54.4%) | 188 (52.2%) | 154 (48.7%) |
| Freckle only/occasionally tan | 1.18 (1.04, 1.34) | 603 (47.6%) | 269 (45.6%) | 172 (47.8%) | 162 (51.3%) | |
| Polygenic risk score | ||||||
| <0.15 | 1.00 | 0.0005 | 413 (32.6%) | 234 (39.7%) | 93 (25.8%) | 86 (27.2%) |
| 0.15 to <0.7 | 1.21 (1.04, 1.41) | 431 (34.0%) | 195 (33.1%) | 124 (34.4%) | 112 (35.4%) | |
| 0.7+ | 1.34 (1.16, 1.54) | 422 (33.3%) | 161 (27.3%) | 143 (39.7%) | 118 (37.3%) | |
| CDKN2A functional mutation | ||||||
| No | 1.00 | 0.009 | 1202 (94.9%) | 560 (94.9%) | 347 (96.4%) | 295 (93.4%) |
| Yes | 1.46 (1.14, 1.86) | 18 (1.4%) | 5 (0.8%) | 5 (1.4%) | 8 (2.5%) | |
| Unknown | 1.09 (0.79, 1.51) | 46 (3.6%) | 25 (4.2%) | 8 (2.2%) | 13 (4.1%) | |
| Recreational sun exposure in beach and water activities from age 15 (average h/d) | ||||||
| <0.5 | 1.00 | 0.002 | 351 (27.7%) | 178 (30.2%) | 101 (28.1%) | 72 (22.8%) |
| 0.5 to <1 | 1.08 (0.92, 1.26) | 348 (27.5%) | 165 (28.0%) | 106 (29.4%) | 77 (24.4%) | |
| 1 to <1.5 | 1.25 (1.07, 1.47) | 257 (20.3%) | 109 (18.5%) | 70 (19.4%) | 78 (24.7%) | |
| 1.5+ | 1.30 (1.11, 1.51) | 310 (24.5%) | 138 (23.4%) | 83 (23.1%) | 89 (28.2%) | |
| Anatomical site of previous primary melanoma2 | ||||||
| Arms/other3 | 1.00 | 0.003 | 212 (16.7%) | 92 (15.6%) | 64 (17.8%) | 56 (17.7%) |
| Head/neck | 1.32 (1.08, 1.63) | 569 (44.9%) | 243 (41.2%) | 164 (45.6%) | 162 (51.3%) | |
| Legs | 1.09 (0.89, 1.32) | 208 (16.4%) | 118 (20.0%) | 58 (16.1%) | 32 (10.1%) | |
| Trunk | 1.32 (1.11, 1.57) | 277 (21.9%) | 137 (23.2%) | 74 (20.6%) | 66 (20.9%) | |
| Histologic subtype of previous primary melanoma2 | ||||||
| Superficial spreading | 1.00 | 0.006 | 776 (61.3%) | 403 (68.3%) | 214 (59.4%) | 159 (50.3%) |
| Nodular | 1.04 (0.85, 1.28) | 127 (10.0%) | 61 (10.3%) | 36 (10.0%) | 30 (9.5%) | |
| Lentigo maligna melanoma | 1.18 (0.97, 1.44) | 158 (12.5%) | 72 (12.2%) | 47 (13.1%) | 39 (12.3%) | |
| Lentigo maligna (in-situ) | 1.07 (0.75, 1.52) | 9 (0.7%) | 0 (0%) | 2 (0.6%) | 7 (2.2%) | |
| Superficial spreading in-situ | 1.24 (0.98, 1.56) | 32 (2.5%) | 0 (0%) | 18 (5.0%) | 14 (4.4%) | |
| Other | 1.05 (0.69, 1.60) | 40 (3.2%) | 24 (4.1%) | 9 (2.5%) | 7 (2.2%) | |
| Not otherwise specified | 1.42 (1.19, 1.69) | 124 (9.8%) | 30 (5.1%) | 34 (9.4%) | 60 (19.0%) | |
| Composite risk score (quintiles)4 | ||||||
| <1.53 | 1.00 | <0.0001 | 363 (28.7%) | 252 (42.7%) | 79 (21.9%) | 32 (10.1%) |
| 1.53 to <1.89 | 2.00 (1.62, 2.45) | 289 (22.8%) | 149 (25.3%) | 83 (23.1%) | 57 (18.0%) | |
| 1.89 to <2.16 | 2.81 (2.29, 3.43) | 253 (20.0%) | 98 (16.6%) | 95 (26.4%) | 60 (19.0%) | |
| 2.16 to <2.49 | 3.45 (2.82, 4.22) | 215 (17.0%) | 63 (10.7%) | 59 (16.4%) | 93 (29.4%) | |
| ≥ 2.49 | 4.75 (3.87, 5.82) | 146 (11.5%) | 28 (4.8%) | 44 (12.2%) | 74 (23.4%) | |
Adjusted for all other individual risk factors in the table.
The model uses data from the most recent diagnosis of primary melanoma. The N (%) shows data from the first primary melanoma.
There were 9 melanomas classified as ‘Other’, 4 of which were first primary melanomas, 4 were fourth primary and 1 was second primary.
The composite risk score was comprised of all twelve risk factors in this table for each melanoma and classified in quintiles.
Figure 1.
Distribution of 1-year, 5-year and 10-year estimated absolute risks derived from the risk prediction model, for development of a second primary melanoma for people who have one previous melanoma (top row), and for development of a third primary melanoma for people who have two previous melanomas (bottom row).
eFigure 2 shows the mean estimated 10-year absolute risks of developing a subsequent melanoma according to deciles of risk score from the prediction model and number of previous melanomas. The absolute risks for developing a second primary melanoma in 10 years ranged from 4.7% (in decile 1) to 26.5% (in decile 10); and for developing a third primary melanoma (for people with two) ranged from 36.7% (in decile 1) to 84.4% (in decile 10). The absolute risks for developing a fourth primary melanoma were similar to those for a third melanoma. Nomograms are shown in Figure 2 for second primary melanoma and eFigure 3 for third and fourth primary melanoma.
Figure 2.
Nomogram for estimating predicted 1-year, 5-year and 10-year risk of developing a second primary melanoma, for people who have one previous primary melanoma. To calculate an individual’s risk, sum the points (top row) that correspond to each of the individual’s risk factors, then find the 1-year, 5-year and 10-year absolute risks that correspond with the total points (bottom rows). The 12 factors are: age at first melanoma, sex, previous keratinocyte cancer, family history of melanoma in first degree relatives, skin colour, ability to tan, naevus density (none, few, some, many), polygenic risk score, CDKN2A functional mutation status, recreational sun exposure in beach and water activities from age 15 (average hours/day), anatomical site and histological subtype of the previous primary melanoma. Refer to Table 2 of the main paper, and the Supplementary online material for more details of the variables.
Harrell’s C-statistic was 0.73 (95% CI 0.68–0.77), 0.65 (95% CI 0.62–0.68) and 0.65 (95% CI 0.61–0.69) for predicting second, third and fourth primary melanomas, respectively. Kaplan-Meier curves, which showed the observed proportion of participants who developed a subsequent melanoma within each quintile of risk, indicated that there was good discrimination between the quintiles overall, particularly for quintiles 1 and 5 (Figure 3).
Figure 3:
Kaplan-Meier curves showing the proportion of patients within each quintile of risk from the prediction model who developed a subsequent melanoma during the follow-up period, for people with one previous melanoma (left), two previous melanomas (middle) or three previous melanomas (right).
Assessment of calibration comparing observed versus predicted number of melanomas developed during follow-up by decile of risk score showed that the risk prediction model tended to underestimate absolute risk of a second melanoma but overestimate risk of a third melanoma (eFigure 4). Plotting Cox-Snell residuals against the cumulative hazard indicated a reasonable model fit. Results for hazard ratios, discrimination and calibration estimates were similar when the polygenic risk score was excluded from the prediction model (see Supplementary online material).
DISCUSSION
This risk prediction model enables estimation of an individual’s relative and absolute risks of new primary melanoma based on a comprehensive set of risk factors including demographic, phenotypic, histopathologic and genomic factors, and average hours/day in recreational sun exposure in beach and water activities from age 15. It demonstrated good discrimination for predicting a second melanoma (Harrell’s C-statistic 0.73) and moderate discrimination (0.65) for predicting a third or fourth melanoma. Average C-statistics (AUCs) for first primary cancer risk prediction models range from around 0.63 for breast cancer,11 0.70 for colorectal cancer,12 0.70 for prostate cancer,13 0.75 for melanoma,14,15,16 and 0.78 for lung cancer.17,18 There was a strong upward linear trend in risk of subsequent melanoma across risk-score quintiles. The risk of a subsequent primary melanoma was also substantially higher for people with two or more melanomas than for people with one.
This risk prediction model can be used to tailor skin surveillance intensity and patient education to the individual’s level of risk. Nomograms (with and without the polygenic risk score) were created to facilitate clinical use of the risk prediction model if an online platform is not available. Relative risks are likely to be similar across different population settings, but caution should be used when applying the model-derived absolute risks to other settings as the baseline incidence rates of multiple primary melanoma are likely to differ.1 In addition, calibration of the model on internal validation, stratified by number of previous primaries, suggested underestimation of absolute risk of a second primary melanoma and overestimation of absolute risk of a third and fourth melanoma. External validation in independent datasets and other geographic regions is required to further evaluate calibration and the generalizability to different populations.
Current recommendations for the frequency and duration of follow-up after a melanoma diagnosis are usually based on stage of disease and focused on detecting loco-regional and distant metastatic spread of melanoma.19 Current Australian guidelines recommend annual follow-up for 10 years for patients with stage I disease, every 6 months for 2 years then annually for 8 years for patients with stage IIA disease, and more frequently (for the first 2–3 years) for patients with more advanced disease.19 The US National Comprehensive Cancer Network (NCCN) guidelines for early-stage (Ia-IIa) melanomas, representing the vast majority of cases, recommend follow-up every 3–12 months for 5 years, then annually.20 The importance of long-term surveillance for new primary melanomas should be recognised in melanoma clinical practice guidelines, and follow-up should extend at least 10 years and, perhaps, indefinitely depending on risk level.
We observed decreasing time between diagnoses with each new primary melanoma, from a median of 3.7 years for a second melanoma to 3.0, 1.9, 1.8, 0.8 and 0.7 years for each subsequent melanoma. This observation might be related to higher risks with age and with multiple primaries. A similar pattern was observed in studies in Sweden21 and Austria22 although the intervals were longer. Nosrati et al23 observed a median of 1.4 years between the first and second melanoma. Other studies have shown that the risk of subsequent melanoma tends to be higher within the first few years of diagnosis24–26 but remains elevated for at least 20 years.26
Tailoring follow-up schedules for skin surveillance according to individual risk would help ensure that those at higher risk receive more frequent surveillance to detect new primary melanomas at an early stage when they are most amenable to treatment, whilst reducing the intensity of surveillance for those at lower risk and thus minimising potential harms and reducing costs. Previous research has shown that it is cost-effective for patients at very high risk of melanoma to be managed in a program of specialised surveillance that includes six-monthly skin checks, total body photography, and sequential digital dermoscopy imaging for suspicious lesions.24,27 As an example, this program of specialised surveillance could be applied to individuals whose 10-year risk of subsequent melanoma is 20% or higher; this would include a small proportion of patients with one previous melanoma and nearly all patients with two or more primary melanomas. Patients with one melanoma whose 10-year risk is at the lower end of the curve, say <5%, might benefit psychologically and financially from fewer routine visits, for example every 2 years, without compromising other outcomes.28 These lower-risk patients might also be more appropriately managed in primary care or in a shared primary care/specialist care arrangement,29 thus helping to ensure that patients at higher risk have timely access to specialist clinics and more advanced diagnostic equipment. Further research is needed to determine the risk cut-points that represent the best balance of benefits, harms, resources and costs for different surveillance schedules, in different populations.
Few studies have built a multivariable risk prediction model.22,23,30,31 Commonly cited individual risk factors are family history of melanoma,22,30,32 high naevus count,22,30,33 and male sex.23,30,32,34 Others include older age,23,33,34 inability to tan,30 fair skin,30,32 and CDKN2A mutation status.22,35 Several different markers of genetic susceptibility were captured in our model, including family history of melanoma, CDKN2A mutation status (a rare, high-penetrance mutation) and polygenic risk score (common polymorphisms). Their joint inclusion might explain why a young age at diagnosis was not an independent predictor of risk, as observed by some others.21,25,26
This is the first study of multiple primary melanoma to include a polygenic risk score, which in the future is likely to become more readily available in everyday clinical practice. The risk estimates for individual variants in the polygenic risk score were taken from published genome-wide association studies rather than from the GEM dataset to avoid model overfitting,36 but the magnitude of risk estimates may differ for studies of first versus subsequent primary melanoma.5 Although polygenic risk score was a strong risk factor, discrimination of the model changed only minimally after removing the polygenic risk score.
We found that markers of intermittent sun exposure, such as melanoma on the trunk and recreational sun exposure in beach and water activities from age 15, as well as markers of more cumulative sun exposure, such as melanomas occurring on the head and neck, history of keratinocyte cancer and lentigo maligna melanoma subtype, were associated with increased risk of subsequent melanoma. Total sun exposure, sunburns and ambient UV were assessed but not retained in the multivariable model. Other studies have found increased risk associated with working outside for >10 years,33 melanomas occurring on the head and neck,25,26 and lentigo maligna melanoma subtype.30 Unlike Siskind et al,30 prior nodular melanoma was not associated with risk of subsequent primary melanoma in our study. We did not have data for dysplastic naevi, which in addition to common naevi, have been shown to be a risk factor for subsequent melanoma.31,37
We found mean estimates of absolute risks for a second primary melanoma at 5 and 10 years to be 8.0% and 13.1% respectively, and for a third melanoma 46.8% and 66.0% respectively. A meta-analysis of eight studies reported lower mean absolute risks of 3.5% at 5-years and 5.3% at 10-years;1 however, risks varied substantially and were higher for studies that were Australian, hospital-based, included melanoma in situ or people with multiple primaries. A registry-based Australian study in 1995 estimated 4.5% risk of second invasive primary melanoma within 5 years.34 A population-based study in 2006 from New Hampshire, US, found that an additional invasive or in situ melanoma occurred in 6% within one year of the initial diagnosis, and in 8% within 2 years.31 More recently, in a population-based cohort from Queensland, Australia, 11% of participants developed ≥ 2 invasive or in situ melanomas over three years.38 Single-centre studies from Sydney39 and New York37 reported similar mean absolute risk estimates to our study for a second melanoma and slightly lower risks for a third primary melanoma. In a high-risk group of patients in Sydney, 12.7% developed a new primary melanoma within two years, and previous multiple primaries were the strongest predictor of a new primary.24
This study has several strengths, including its population-based design, comprehensive risk factors, large sample size, and long follow-up. Frailty models included multiple melanomas simultaneously in the model, maximizing statistical power and simplifying clinical use by generating a single risk prediction model. However, this assumes that the hazard ratio for each risk factor in the model remains constant for each subsequent melanoma; which is unknown. The Frailty models were unable to account for competing risks; this means that the absolute risks given to patients from this risk prediction model are conditional on patients surviving for the length of the corresponding period and not dying from melanoma or other causes. However, few patients (9.6%) died over the follow-up period without developing a subsequent melanoma and we estimated minimal impact of not accounting for competing risks (shown in the Supplementary online material).
Data were collected for both in situ and invasive melanomas during follow-up. Mostly, other studies have shown that people diagnosed with melanoma in situ have a similar risk of subsequent melanoma as those with invasive disease.25,30,33 As this was an observational study, patient follow-up intervals were not standardised and so our findings reflect the spectrum of follow-up practices in the community. However, the likelihood of detecting a new primary melanoma would be higher for people who had more frequent skin examinations and this may be influenced by multiple factors.
It is a weakness of this study that, whilst histological characteristics of each subsequent primary melanoma were taken into account in the model, other risk factors were measured at one timepoint only. For variables that do not change, such as sex, skin colour or genomic factors, this is not a problem. However, some variables, specifically previous keratinocyte cancer, family history, naevus density, and average hours/day of recreational sun exposure since age 15, may change over time. These risk factors were assessed between 2000–2003 at entry to the GEM study, which may have been at the time of their first or a subsequent (multiple) primary melanoma. The impact on the model of changes to risk factors over time would require data collection at multiple timepoints, which we do not have. We recommend prospective evaluation of this risk prediction model in clinical practice to enable such refinements and to assess calibration in different populations.
Supplementary Material
What’s already known about this topic?
Guidelines for the frequency and length of follow-up for patients with melanoma to detect new primary melanoma are based on limited evidence.
People with a previous primary melanoma have, on average, a higher risk of developing another primary invasive melanoma compared with the general population, but an accurate way of estimating individual risk is needed.
What does this study add?
A comprehensive risk prediction model for subsequent melanoma based on population-based data from 1,266 participants with melanoma (2,613 melanomas) over a median 14 years of follow-up. The risk prediction model includes 12 risk factors comprising demographic, phenotypic, histopathologic, and genomic factors, and recreational sun exposure in beach and water activities from age 15.
The risk prediction model and nomograms enable estimation of absolute risk of subsequent melanoma, and can be used to tailor surveillance intensity, communicate individual risk and provide patient education.
ACKNOWLEDGMENTS
We are grateful Himali Patel, Pampa Roy, Emily La Pilla and Vikram Mavinkurve for assistance with genotyping. We acknowledge the NSW Ministry of Health, NSW Cancer Registry and Australian Institute of Health and Welfare for access to linked follow-up data.
FUNDING
This work was supported by Melanoma Institute Australia, The University of Sydney; National Cancer Institute (P01CA206980 to NET and MB, R01CA112243 to NET, U01CA83180 and R01CA112524 to MB, R01CA098438 to CBB, R03CA125829 and R03CA173806 to IO, P30CA016086, P30CA014089, P30CA008748); National Institute of Environmental Health Sciences (P30ES010126). AEC received a National Health and Medical Research Council (NHMRC) Career Development Fellowship (1147843) and Cancer Institute NSW Career Development Fellowship (15/CDF/1–14). JFT and BKA were supported by the Melanoma Foundation of the University of Sydney. DWO was supported by the James and Jesse Millis Distinguished Professorship at the University of North Carolina, Chapel Hill. The funders had no role in the study design, data collection, data analysis, data interpretation, writing of the report, or decision to submit the paper for publication.
Footnotes
Conflicts of interest: There are no conflicts of interest to declare.
REFERENCES
- 1.van der Leest RJ, Flohil SC, Arends LR et al. Risk of subsequent cutaneous malignancy in patients with prior melanoma: a systematic review and meta-analysis. J Eur Acad Dermatol Venereol 2015; 29:1053–62. [DOI] [PubMed] [Google Scholar]
- 2.Gershenwald JE, Scolyer RA, Hess KR et al. Melanoma staging: Evidence-based changes in the American Joint Committee on Cancer eighth edition cancer staging manual. CA Cancer J Clin 2017; 67:472–92. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Watts CG, Cust AE, Menzies SW et al. Cost-Effectiveness of Skin Surveillance Through a Specialized Clinic for Patients at High Risk of Melanoma. J Clin Oncol 2017; 35:63–71. [DOI] [PubMed] [Google Scholar]
- 4.Watts CG, Dieng M, Morton RL et al. Clinical practice guidelines for identification, screening and follow-up of individuals at high risk of primary cutaneous melanoma: a systematic review. Br J Dermatol 2015; 172:33–47. [DOI] [PubMed] [Google Scholar]
- 5.Begg CB, Hummer AJ, Mujumdar U et al. A design for cancer case-control studies using only incident cases: experience with the GEM study of melanoma. Int J Epidemiol 2006; 35:756–64. [DOI] [PubMed] [Google Scholar]
- 6.Amorim LD, Cai J. Modelling recurrent events: a tutorial for analysis in epidemiology. Int J Epidemiol 2015; 44:324–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.du Bois RM, Weycker D, Albera C et al. Ascertainment of individual risk of mortality for patients with idiopathic pulmonary fibrosis. Am J Respir Crit Care Med 2011; 184:459–66. [DOI] [PubMed] [Google Scholar]
- 8.Harrell FE. Regression Modeling Strategies With Applications to Linear Models, Logistic Regression, and Survival Analysis. New York: Springer; 2001. [Google Scholar]
- 9.Steyerberg EW. Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating. New York, USA: Springer; 2009. [Google Scholar]
- 10.Collins GS, Reitsma JB, Altman DG et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ 2015; 350:g7594. [DOI] [PubMed] [Google Scholar]
- 11.Meads C, Ahmed I, Riley RD. A systematic review of breast cancer incidence risk prediction models with meta-analysis of their performance. Breast Cancer Res Treat 2012; 132:365–77. [DOI] [PubMed] [Google Scholar]
- 12.Usher-Smith JA, Walter FM, Emery JD et al. Risk Prediction Models for Colorectal Cancer: A Systematic Review. Cancer Prev Res (Phila) 2016; 9:13–26. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Poyet C, Nieboer D, Bhindi B et al. Prostate cancer risk prediction using the novel versions of the European Randomised Study for Screening of Prostate Cancer (ERSPC) and Prostate Cancer Prevention Trial (PCPT) risk calculators: independent validation and comparison in a contemporary European cohort. BJU Int 2016; 117:401–8. [DOI] [PubMed] [Google Scholar]
- 14.Olsen CM, Pandeya N, Thompson BS et al. Risk Stratification for Melanoma: Models Derived and Validated in a Purpose-Designed Prospective Cohort. J Natl Cancer Inst 2018. [DOI] [PubMed] [Google Scholar]
- 15.Vuong K, McGeechan K, Armstrong BK et al. Risk Prediction Models for Incident Primary Cutaneous Melanoma: A Systematic Review. JAMA Dermatol 2014; 150:434–44. [DOI] [PubMed] [Google Scholar]
- 16.Cust AE, Drummond M, Kanetsky PA et al. Assessing the incremental contribution of common genomic variants to melanoma risk prediction in two population-based studies. J Invest Dermatol 2018; 138:2617–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Katki HA, Kovalchik SA, Berg CD et al. Development and Validation of Risk Models to Select Ever-Smokers for CT Lung Cancer Screening. JAMA 2016; 315:2300–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Tammemagi MC. Application of risk prediction models to lung cancer screening: a review. J Thorac Imaging 2015; 30:88–100. [DOI] [PubMed] [Google Scholar]
- 19.Barbour A, Guminski A, Liu W et al. What is the ideal setting, duration and frequency of follow-up for melanoma patients? In: Clinical practice guidelines for the diagnosis and management of melanoma. Sydney: Cancer Council Australia; 2018; Available from: https://wiki.cancer.org.au/australia/Guidelines:Melanoma. [Google Scholar]
- 20.Coit DG, Andtbacka R, Anker CJ et al. Melanoma, version 2.2013: featured updates to the NCCN guidelines. J Natl Compr Canc Netw 2013; 11:395–407. [DOI] [PubMed] [Google Scholar]
- 21.Chen T, Fallah M, Forsti A et al. Risk of Next Melanoma in Patients With Familial and Sporadic Melanoma by Number of Previous Melanomas. JAMA Dermatol 2015; 151:607–15. [DOI] [PubMed] [Google Scholar]
- 22.Muller C, Wendt J, Rauscher S et al. Risk Factors of Subsequent Primary Melanomas in Austria. JAMA Dermatol 2018; doi: 10.1001/jamadermatol.2018.4645 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Nosrati A, Yu WY, McGuire J et al. Outcomes and risk factors in patients with multiple primary melanomas. J Invest Dermatol 2019. 139:195–201. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Moloney FJ, Guitera P, Coates E et al. Detection of primary melanoma in individuals at extreme high risk: a prospective 5-year follow-up study. JAMA Dermatol 2014; 150:819–27. [DOI] [PubMed] [Google Scholar]
- 25.Youlden DR, Youl PH, Soyer HP et al. Distribution of subsequent primary invasive melanomas following a first primary invasive or in situ melanoma Queensland, Australia, 1982–2010. JAMA Dermatol 2014; 150:526–34. [DOI] [PubMed] [Google Scholar]
- 26.Bradford PT, Freedman DM, Goldstein AM et al. Increased risk of second primary cancers after a diagnosis of melanoma. Arch Dermatol 2010; 146:265–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Watts CG, Cust AE, Menzies SW et al. Specialized surveillance for individuals at high risk for melanoma: a cost analysis of a high-risk clinic. JAMA Dermatol 2015; 151:178–86. [DOI] [PubMed] [Google Scholar]
- 28.Damude S, Hoekstra-Weebers JE, Francken AB et al. The MELFO-Study: Prospective, Randomized, Clinical Trial for the Evaluation of a Stage-adjusted Reduced Follow-up Schedule in Cutaneous Melanoma Patients-Results after 1 Year. Ann Surg Oncol 2016; 23:2762–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Lim WY, Turner RM, Morton RL et al. Use of shared care and routine tests in follow-up after treatment for localised cutaneous melanoma. BMC Health Serv Res 2018; 18:477. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Siskind V, Hughes MC, Palmer JM et al. Nevi, family history, and fair skin increase the risk of second primary melanoma. Journal of Investigative Dermatology 2011; 131:461–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Titus-Ernstoff L, Perry AE, Spencer SK et al. Multiple primary melanoma: two-year results from a population-based study. Arch Dermatol 2006; 142:433–8. [DOI] [PubMed] [Google Scholar]
- 32.Slingluff CL Jr., Vollmer RT, Seigler HF. Multiple primary melanoma: incidence and risk factors in 283 patients. Surgery 1993; 113:330–9. [PubMed] [Google Scholar]
- 33.Schuurman MS, de Waal AC, Thijs EJM et al. Risk factors for second primary melanoma among Dutch patients with melanoma. Br J Dermatol 2017; 176:971–8. [DOI] [PubMed] [Google Scholar]
- 34.Giles G, Staples M, McCredie M et al. Multiple primary melanomas: an analysis of cancer registry data from Victoria and New South Wales. Melanoma Res 1995; 5:433–8. [PubMed] [Google Scholar]
- 35.Helgadottir H, Tuominen R, Olsson H et al. Cancer risks and survival in patients with multiple primary melanomas: Association with family history of melanoma and germline CDKN2A mutation status. J Am Acad Dermatol 2017; 77:893–901. [DOI] [PubMed] [Google Scholar]
- 36.Wray NR, Yang J, Hayes BJ et al. Pitfalls of predicting complex traits from SNPs. Nat Rev Genet 2013; 14:507–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Ferrone CR, Ben Porat L, Panageas KS et al. Clinicopathological features of and risk factors for multiple primary melanomas. JAMA 2005; 294:1647–54. [DOI] [PubMed] [Google Scholar]
- 38.Gordon LG, Elliott TM, Olsen CM et al. Multiplicity of skin cancers in Queensland and their cost burden to government and patients. Aust N Z J Public Health 2018; 42:86–91. [DOI] [PubMed] [Google Scholar]
- 39.Doubrovsky A, Menzies SW. Enhanced survival in patients with multiple primary melanoma. Arch Dermatol 2003; 139:1013–8. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.