Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/ou_press/risk-stratification-for-melanoma-models-derived-and-validated-in-a-agNVJT4sub
References (45)
-
T. Fears, D. Guerry, R. Pfeiffer, R. Sagebiel, D. Elder, A. Halpern, E. Holly, P. Hartge, M. Tucker (2006)
Identifying individuals at high risk of melanoma: a practical predictor of absolute risk.Journal of clinical oncology : official journal of the American Society of Clinical Oncology, 24 22
-
C. Fortes, S. Mastroeni, L. Bakos, G. Antonelli, L. Alessandroni, M. Pilla, M. Alotto, A. Zappalà, Thomas Manoorannparampill, R. Bonamigo, P. Pasquini, F. Melchi (2010)
Identifying individuals at high risk of melanoma: a simple toolEuropean Journal of Cancer Prevention, 19
-
E. Richtig, E. Santigli, R. Fink‐Puches, W. Weger, R. Hofmann-Wellenhof (2008)
Assessing melanoma risk factors: how closely do patients and doctors agree?Public health, 122 12
-
G. Guy, Cheryll Thomas, T. Thompson, M. Watson, Greta Massetti, L. Richardson (2015)
Vital Signs: Melanoma Incidence and Mortality Trends and Projections — United States, 1982–2030Morbidity and Mortality Weekly Report, 64
-
Guy (2015)
Vital signs: Melanoma incidence and mortality trends and projections - United States, 1982-2030MMWR Morb Mortal Wkly Rep, 64
-
M. Goldberg, J. Doucette, H. Lim, J. Spencer, J. Carucci, D. Rigel (2007)
Risk factors for presumptive melanoma in skin cancer screening: American Academy of Dermatology National Melanoma/Skin Cancer Screening Program experience 2001-2005.Journal of the American Academy of Dermatology, 57 1
-
J. Sterne, I. White, J. Carlin, Michael Spratt, P. Royston, M. Kenward, A. Wood, J. Carpenter (2009)
Multiple imputation for missing data in epidemiological and clinical research: potential and pitfallsThe BMJ, 338
-
A. Vickers, A. Cronin, E. Elkin, M. Gonen (2008)
Extensions to decision curve analysis, a novel method for evaluating diagnostic tests, prediction models and molecular markersBMC Medical Informatics and Decision Making, 8
-
D. English, B. Armstrong (1988)
Identifying people at high risk of cutaneous malignant melanoma: results from a case-control study in Western AustraliaBritish Medical Journal (Clinical research ed.), 296
-
H. Uno, T. Cai, M. Pencina, R. D’Agostino, Lee-Jen Wei (2011)
On the C‐statistics for evaluating overall adequacy of risk prediction procedures with censored survival dataStatistics in Medicine, 30
-
D. Whiteman, A. Green, C. Olsen (2016)
The Growing Burden of Invasive Melanoma: Projections of Incidence Rates and Numbers of New Cases in Six Susceptible Populations through 2031.The Journal of investigative dermatology, 136 6
-
Jiali Han, G. Colditz, D. Hunter (2006)
Risk factors for skin cancers: a nested case-control study within the Nurses' Health Study.International journal of epidemiology, 35 6
-
C. Watts, A. Cust, S. Menzies, G. Mann, R. Morton (2017)
Cost-Effectiveness of Skin Surveillance Through a Specialized Clinic for Patients at High Risk of Melanoma.Journal of clinical oncology : official journal of the American Society of Clinical Oncology, 35 1
-
S. Guther, Katrin Ramrath, D. Dyall‐Smith, M. Landthaler, Wilhelm Stolz (2012)
Development of a targeted risk‐group model for skin cancer screening based on more than 100 000 total skin examinationsJournal of the European Academy of Dermatology and Venereology, 26
-
C. Olsen, A. Green, R. Neale, P. Webb, Rebekah Cicero, L. Jackman, Suzanne O'Brien, Susan Perry, B. Ranieri, D. Whiteman (2012)
Cohort profile: the QSkin Sun and Health Study.International journal of epidemiology, 41 4
-
PH.D. YOUDEN (1950)
Index for rating diagnostic testsCancer, 3
-
C. Olsen, R. Neale, A. Green, P. Webb, The Study, The Study, D. Whiteman (2015)
Independent validation of six melanoma risk prediction models.The Journal of investigative dermatology, 135 5
-
G. Quéreux, D. Moyse, Y. Lequeux, O. Jumbou, A. Brocard, D. Antonioli, B. Dréno, J. Nguyen (2011)
Development of an individual score for melanoma riskEuropean Journal of Cancer Prevention, 20
-
K. Wernli, N. Henrikson, Caitlin Morrison, Matt Nguyen, G. Pocobelli, Paula Blasi (2016)
Screening for Skin Cancer in Adults: Updated Evidence Report and Systematic Review for the US Preventive Services Task Force.JAMA, 316 4
-
F. Moloney, P. Guitera, E. Coates, N. Haass, K. Ho, R. Khoury, R. O’Connell, Leo Raudonikis, H. Schmid, G. Mann, S. Menzies (2014)
Detection of primary melanoma in individuals at extreme high risk: a prospective 5-year follow-up study.JAMA dermatology, 150 8
-
Collins (2015)
Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): The TRIPOD statementBr J Cancer., 112
-
K. Vuong, K. Vuong, B. Armstrong, E. Weiderpass, E. Lund, H. Adami, H. Adami, M. Veierød, J. Barrett, J. Davies, D. Bishop, D. Whiteman, C. Olsen, J. Hopper, G. Mann, A. Cust, K. McGeechan (2016)
Development and External Validation of a Melanoma Risk Prediction Model Based on Self-assessed Risk Factors.JAMA dermatology, 152 8
-
R. MacKie, T. Freudenberger, T. Aitchison (1989)
PERSONAL RISK-FACTOR CHART FOR CUTANEOUS MELANOMAThe Lancet, 334
-
Centers for Disease Control and Prevention (2011)
Cancer survivors—United States, 2007MMWR Morb Mortal Wkly Rep, 60
-
C. Garbe, S. Krüger, C. Orfanos, P. Büttner, J. Weiss, H. Soyer, U. Stocker, M. Roser, J. Weckbecker, R. Panizzon, F. Bahmer, W. Tilgen, I. Guggenmoos‐Holzmann (1994)
Risk factors for developing cutaneous melanoma and criteria for identifying persons at risk: multicenter case-control study of the Central Malignant Melanoma Registry of the German Dermatological Society.The Journal of investigative dermatology, 102 5
-
J. Fine, R. Gray (1999)
A Proportional Hazards Model for the Subdistribution of a Competing RiskJournal of the American Statistical Association, 94
-
V. Mar, R. Wolfe, J. Kelly (2011)
Predicting melanoma risk for the Australian populationAustralasian Journal of Dermatology, 52
-
L. Marrett, W. King, S. Walter, L. From (1992)
Use of host factors to identify people at high risk for cutaneous malignant melanoma .CMAJ : Canadian Medical Association journal = journal de l'Association medicale canadienne, 147 4
-
Friederike Erdmann, J. Lortet-Tieulent, J. Schüz, H. Zeeb, R. Greinert, E. Breitbart, F. Bray (2013)
International trends in the incidence of malignant melanoma 1953–2008—are recent generations at higher or lower risk?International Journal of Cancer, 132
-
G. Collins, J. Reitsma, D. Altman, K. Moons (2015)
Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statementBMJ : British Medical Journal, 350
-
H. Welch, W. Black (2010)
Overdiagnosis in cancer.Journal of the National Cancer Institute, 102 9
-
P. Royston, D. Altman (2013)
External validation of a Cox prognostic model: principles and methodsBMC Medical Research Methodology, 13
-
K. Nielsen, A. Måsbäck, H. Olsson, C. Ingvar (2012)
A prospective, population‐based study of 40,000 women regarding host factors, UV exposure and sunbed use in relation to risk and anatomic site of cutaneous melanomaInternational Journal of Cancer, 131
-
P. Royston (2015)
Tools for Checking Calibration of a Cox Model in External Validation: Prediction of Population-Averaged Survival Curves Based on Risk GroupsThe Stata Journal, 15
-
Kirsten Bibbins-Domingo, David Grossman, Susan Curry, Karina Davidson, M. Ebell, J. Epling, Francisco García, Matthew Gillman, Alex Kemper, A. Krist, Ann Kurth, C. Landefeld, Carol Mangione, William Phillips, Maureen Phipps, Michael Pignone, Albert Siu (2016)
Screening for Skin Cancer: US Preventive Services Task Force Recommendation Statement.JAMA, 316 4
-
J. Usher-Smith, J. Emery, A. Kassianos, F. Walter (2014)
Risk Prediction Models for Melanoma: A Systematic ReviewCancer Epidemiology, Biomarkers & Prevention, 23
-
A. Vickers, E. Elkin (2006)
Decision Curve Analysis: A Novel Method for Evaluating Prediction ModelsMedical Decision Making, 26
-
E. Breitbart, A. Waldmann, Sandra Nolte, Marcus Capellaro, R. Greinert, B. Volkmer, A. Katalinic (2012)
Systematic skin cancer screening in Northern Germany.Journal of the American Academy of Dermatology, 66 2
-
K. Moons, R. Donders, T. Stijnen, F. Harrell (2006)
Using the outcome for imputation of missing predictor values was preferred.Journal of clinical epidemiology, 59 10
-
P. Marcus, N. Pashayan, T. Church, V. Doria-Rose, M. Gould, R. Hubbard, M. Marrone, D. Miglioretti, P. Pharoah, P. Pinsky, K. Rendle, H. Robbins, Megan Roberts, Betsy Rolland, M. Schiffman, Jasmin Tiro, A. Zauber, D. Winn, Muin Khoury, Muin Khoury (2016)
Population-Based Precision Cancer Screening: A Symposium on Evidence, Epidemiology, and Next StepsCancer Epidemiology, Biomarkers & Prevention, 25
-
C. Morze, C. Olsen, Susan Perry, L. Jackman, B. Ranieri, Suzanne O'Brien, Rebekah Cicero, D. Whiteman (2012)
Good test-retest reproducibility for an instrument to capture self-reported melanoma risk factors.Journal of clinical epidemiology, 65 12
-
E. Cho, B. Rosner, D. Feskanich, G. Colditz (2005)
Risk factors and individual probabilities of melanoma for whites.Journal of clinical oncology : official journal of the American Society of Clinical Oncology, 23 12
-
A. Shain, B. Bastian (2016)
From melanocytes to melanomasNature Reviews Cancer, 16
-
Gayle Masri, W. Clark, D. Guerry, A. Halpern, C. Thompson, D. Elder (1990)
Screening and surveillance of patients at high risk for malignant melanoma result in detection of earlier disease.Journal of the American Academy of Dermatology, 22 6 Pt 1
-
K. Vuong, K. McGeechan, B. Armstrong, A. Cust (2014)
Risk prediction models for incident primary cutaneous melanoma: a systematic review.JAMA dermatology, 150 4
- Publisher
- Oxford University Press
- Copyright
- © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please email: [email protected].
- ISSN
- 0027-8874
- eISSN
- 1460-2105
- DOI
- 10.1093/jnci/djy023
- Publisher site
- See Article on Publisher Site
Abstract
Abstract Background Risk stratification can improve the efficacy and cost-efficiency of screening programs for early detection of cancer. We sought to derive a risk stratification tool for melanoma that was suitable for the general population using only self-reported information. Methods We used melanoma risk factor information collected at baseline from QSKIN, a prospective cohort study of Queensland adults age 40 to 69 years at recruitment (n = 41 954). We examined two separate outcomes: 1) invasive melanomas and 2) all melanomas (invasive + in situ) obtained through data linkage to the cancer registry. We used stepwise Cox proportional hazards modeling to derive the risk models in a randomly selected two-thirds sample of the data set and assessed model performance in the remaining one-third validation sample. Results After a median follow-up of 3.4 years, 655 (1.6%) participants developed melanoma (257 invasive, 398 in situ). The prediction model for invasive melanoma included seven terms. At baseline, the strongest predictors of invasive melanoma were age, sex, tanning ability, number of moles at age 21 years, and number of skin lesions treated destructively. The model for “all melanomas” (ie, invasive and in situ) included five additional terms. Discrimination in the validation data set was high for both models (C-index = 0.69, 95% confidence interval [CI] = 0.62 to 0.76, and C-index = 0.72, 95% CI = 0.69 to 0.75, respectively), and calibration was acceptable. Conclusions Such a tool could be used to target surveillance activities to those at highest predicted risk of developing melanoma over a median duration of 3.4 years. Melanoma incidence has increased steadily over the past three decades in most fair-skinned populations (1) and is projected to continue rising in the United States and other nations for several decades (2,3). The two main strategies employed to reduce the burden of melanoma are primary prevention and early detection. Whereas primary prevention seeks to reduce population exposure to sunlight and thereby reduce melanoma incidence, efforts to detect melanomas at an earlier stage (either through organized screening or other case-finding activities) aim to minimize the risk of invasion and metastasis, and thus reduce mortality from melanoma. The efficacy of population-based screening for melanoma remains unproven (4,5). Instead, the advice in the United States and other countries is to offer screening only to individuals at high risk for melanoma, although no definition of “high risk” has been agreed on (6–10). Targeted screening of those at very high risk for melanoma (ie, family history and/or personal history and/or dysplastic nevus syndrome) has been shown to be both effective (11,12) and cost-efficient (13). A number of models have been developed to estimate a person’s future risk of melanoma, mostly using case–control data, which are prone to recall and other biases (14,15). While several models have been developed in prospective cohorts (16–19), only one of those cohorts collected comprehensive information on melanoma risk factors at baseline (17), and it was restricted to women. To date, none of the models developed from prospective data have been made publicly available. Here, we report the development of a tool intended for community use, with which to stratify people based on their predicted risk of melanoma into appropriate early detection activities. Methods Study Population The QSkin Sun and Health Study is a prospective cohort study of men and women age 40 through 69 years at recruitment in 2011 who were sampled randomly from the Queensland population (n = 43 794). A description of the study design and methods has been published previously (20). Participants with a prior history of invasive or in situ melanoma (n = 1811) or who withdrew from the study after baseline (n = 29) were excluded; the eligible cohort included 41 954 participants. The Human Research Ethics Committee at the QIMR Berghofer Medical Research Institute approved the study, and all participants gave written consent to take part. We adhered to the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) guidelines for reporting (Supplementary Materials, Tripod checklist, available online) (21). Data Collection At baseline, in addition to demographic items and general medical history, we asked participants to estimate their perceived risk of developing melanoma. Comprehensive information relating to known or suspected melanoma risk factors (including ethnicity, pigmentary and phenotypic characteristics, sun exposure, sun protection, use of tanning beds, skin examination practices, family history, and history of skin cancer) was collected via a self-completed survey (available online: https://qskin.qimrberghofer.edu.au/page/About/Baseline_survey). Melanoma Case Confirmation Data on all melanoma diagnoses from baseline up to December 31, 2014, were obtained via record linkage to the Queensland Cancer Registry (almost complete ascertainment), supplemented by pathology records when available. We examined two separate outcomes: invasive cutaneous melanomas and all cutaneous melanomas (invasive + in situ). Candidate Predictor Variables We selected 28 candidate predictor variables a priori. The demographic characteristics considered were age, sex, education, ethnicity, and private health insurance coverage. Phenotypic factors considered were self-reported eye, hair, and unexposed skin color; skin burning tendency and tanning ability; freckling density and nevus burden at age 21 years. Measures of sun exposure considered were sunburns as a child (<10 years), teenager/youth (10–20 years), and adult (>20 years), cumulative lifetime sun exposure, place of birth (Australia/other), latitude of place of birth (<±45º/≥±45º), and use of tanning beds. Medical history variables considered were number of skin cancers (not melanomas) excised surgically, number of actinic skin lesions treated destructively, and family history of melanoma. Other factors considered were sunscreen and hat use, smoking status, and alcohol consumption. These items were self-reported at baseline; measures of repeatability and validity have been published (22). Personal history of other cancer (not skin; registry confirmed) was also considered as a candidate predictor. Imputation of Missing Data Missing values for most candidate predictor items occurred at a prevalence of less than 1%, but they were higher for the sunburn variables (Table 1; Supplementary Table 1, available online). To avoid potential bias, we imputed missing values for candidate predictors using the fully conditional specification method in PROC MI in SAS v9.4 (SAS Institute, Cary, NC), assuming that data were missing at random. We included all predictor variables and the outcome variable (23) in the imputation step, specifying logistic regression to impute ordinal variables. Imputation was run over five cycles to generate five data sets. Table 1. Distribution of selected characteristics of study participants in the QSkin cohort (n = 41 954) Characteristic Invasive melanoma cases (n = 257)No. (%) In situ melanoma cases (n = 398)No. (%) Noncases(n = 41 299)No. (%) Total(n = 41 954)No. (%) Age group, y 40–49 37 (14.4) 71 (18.4) 11248 (27.2) 11356 (27.1) 50–50 93 (36.2) 137 (34.4) 15732 (38.1) 11962 (38.1) 60+ 127 (49.4) 190 (47.7) 14319 (34.7) 14636 (34.9) Sex Female 111 (43.2) 161 (40.5) 22693 (55.0) 22965 (54.7) Male 146 (56.8) 237 (59.6) 18606 (45.1) 18989 (45.3) Ethnicity Nonwhite 3 (1.2) 7 (1.8) 2666 (6.5) 2676 (6.4) White 253 (98.4) 389 (97.7) 38213 (92.5) 38855 (92.6) Missing 1 (0.4) 2 (0.5) 420 (1.0) 423 (1.0) Private health insurance No 69 (26.9) 96 (24.1) 13850 (33.5) 14015 (33.4) Yes 187 (72.8) 301 (75.6) 27245 (65.9) 27733 (66.1) Missing 1 (0.4) 1 (0.3) 204 (0.5) 206 (0.5) Tanning ability Tan deeply 26 (10.1) 74 (18.6) 9714 (23.5) 9814 (23.4) Tan moderately 122 (47.5) 191 (48.0) 20230 (49.0) 20543 (49.0) Tan a little 70 (27.2) 97 (24.4) 8463 (20.5) 8630 (20.6) Not tan 36 (14.0) 34 (8.5) 2557 (6.2) 2627 (6.3) Missing 3 (1.2) 2 (0.5) 335 (0.8) 340 (0.8) Moles at age 21 y None 48 (18.7) 94 (23.6) 11638 (28.2) 11780 (28.1) A few 128 (49.8) 186 (46.7) 21452 (51.9) 21766 (51.9) Some 59 (23.0) 83 (20.9) 5896 (14.3) 6038 (14.4) Many 17 (6.6) 28 (7.0) 1189 (2.9) 1234 (2.9) Missing 5 (2.0) 7 (1.8) 1124 (2.7) 1136 (2.7) Hair color Black 10 (3.9) 38 (9.6) 4195 (10.2) 4243 (10.1) Dark brown 74 (28.8) 113 (28.4) 13805 (33.4) 13992 (33.4) Light brown 109 (42.4) 149 (37.4) 15119 (36.6) 15377 (36.7) Blond 29 (11.3) 54 (13.6) 5705 (13.8) 5788 (13.8) Red/auburn 34 (13.2) 39 (9.8) 2228 (5.4) 2301 (5.5) Missing 1 (0.4) 5 (1.3) 247 (0.6) 253 (0.6) No. of previous skin cancers removed surgically 0 97 (37.7) 144 (36.2) 25839 (62.6) 26080 (62.2) 1 37 (14.4) 67 (16.8) 5497 (13.3) 5601 (13.4) 2–10 91 (35.4) 141 (35.4) 8115 (19.7) 8347 (19.9) 11–20 19 (7.4) 23 (5.8) 931 (2.3) 973 (2.3) 20+ 11 (4.3) 19 (4.8) 608 (1.5) 638 (1.5) Missing 2 (0.8) 4 (1.0) 309 (0.8) 315 (0.8) No. of previous skin lesions destroyed 0 55 (21.4) 102 (25.6) 19381 (46.9) 19538 (46.6) 1–5 65 (25.3) 93 (23.4) 10796 (26.1) 10954 (26.1) 6–10 35 (13.6) 62 (15.6) 3741 (9.1) 3838 (9.2) 11–20 33 (12.8) 52 (13.1) 2996 (7.3) 3081 (7.3) 21–49 35 (13.6) 35 (8.8) 2285 (5.5) 2355 (5.6) 50+ 33 (12.8) 51 (12.8) 1867 (4.5) 1951 (4.7) Missing 1 (0.4) 3 (0.8) 233 (0.6) 237 (0.6) Sunscreen use (past year) Never 30 (11.7) 63 (15.8) 8728 (21.1) 8821 (21.0) Less than 50% of the time 121 (47.1) 172 (43.2) 17393 (42.1) 17686 (42.2) More than 50% of the time 79 (30.7) 129 (32.4) 11207 (27.1) 11415 (27.2) All the time 27 (10.5) 33 (8.3) 3722 (9.0) 3782 (9.0) Missing 0 1 (0.3) 249 (0.6) 250 (0.6) Skin checks by a doctor (past 3 y) None 38 (14.8) 45 (11.3) 11604 (28.1) 11687 (27.9) Once 62 (24.1) 104 (26.1) 12183 (29.5) 12349 (29.4) 2–5 times 119 (46.3) 187 (47.0) 14053 (34.0) 14359 (34.2) More than 5 times 36 (14.0) 59 (14.8) 2477 (6.0) 2572 (6.1) Missing 2 (0.8) 3 (0.8) 982 (2.4) 987 (2.4) Family history of melanoma No 151 (58.8) 231 (58.0) 25859 (62.6) 26241 (62.6) Yes 77 (30.0) 120 (30.2) 9131 (22.1) 9328 (22.2) Missing 29 (11.3) 47 (11.8) 6309 (15.3) 6385 (15.2) Characteristic Invasive melanoma cases (n = 257)No. (%) In situ melanoma cases (n = 398)No. (%) Noncases(n = 41 299)No. (%) Total(n = 41 954)No. (%) Age group, y 40–49 37 (14.4) 71 (18.4) 11248 (27.2) 11356 (27.1) 50–50 93 (36.2) 137 (34.4) 15732 (38.1) 11962 (38.1) 60+ 127 (49.4) 190 (47.7) 14319 (34.7) 14636 (34.9) Sex Female 111 (43.2) 161 (40.5) 22693 (55.0) 22965 (54.7) Male 146 (56.8) 237 (59.6) 18606 (45.1) 18989 (45.3) Ethnicity Nonwhite 3 (1.2) 7 (1.8) 2666 (6.5) 2676 (6.4) White 253 (98.4) 389 (97.7) 38213 (92.5) 38855 (92.6) Missing 1 (0.4) 2 (0.5) 420 (1.0) 423 (1.0) Private health insurance No 69 (26.9) 96 (24.1) 13850 (33.5) 14015 (33.4) Yes 187 (72.8) 301 (75.6) 27245 (65.9) 27733 (66.1) Missing 1 (0.4) 1 (0.3) 204 (0.5) 206 (0.5) Tanning ability Tan deeply 26 (10.1) 74 (18.6) 9714 (23.5) 9814 (23.4) Tan moderately 122 (47.5) 191 (48.0) 20230 (49.0) 20543 (49.0) Tan a little 70 (27.2) 97 (24.4) 8463 (20.5) 8630 (20.6) Not tan 36 (14.0) 34 (8.5) 2557 (6.2) 2627 (6.3) Missing 3 (1.2) 2 (0.5) 335 (0.8) 340 (0.8) Moles at age 21 y None 48 (18.7) 94 (23.6) 11638 (28.2) 11780 (28.1) A few 128 (49.8) 186 (46.7) 21452 (51.9) 21766 (51.9) Some 59 (23.0) 83 (20.9) 5896 (14.3) 6038 (14.4) Many 17 (6.6) 28 (7.0) 1189 (2.9) 1234 (2.9) Missing 5 (2.0) 7 (1.8) 1124 (2.7) 1136 (2.7) Hair color Black 10 (3.9) 38 (9.6) 4195 (10.2) 4243 (10.1) Dark brown 74 (28.8) 113 (28.4) 13805 (33.4) 13992 (33.4) Light brown 109 (42.4) 149 (37.4) 15119 (36.6) 15377 (36.7) Blond 29 (11.3) 54 (13.6) 5705 (13.8) 5788 (13.8) Red/auburn 34 (13.2) 39 (9.8) 2228 (5.4) 2301 (5.5) Missing 1 (0.4) 5 (1.3) 247 (0.6) 253 (0.6) No. of previous skin cancers removed surgically 0 97 (37.7) 144 (36.2) 25839 (62.6) 26080 (62.2) 1 37 (14.4) 67 (16.8) 5497 (13.3) 5601 (13.4) 2–10 91 (35.4) 141 (35.4) 8115 (19.7) 8347 (19.9) 11–20 19 (7.4) 23 (5.8) 931 (2.3) 973 (2.3) 20+ 11 (4.3) 19 (4.8) 608 (1.5) 638 (1.5) Missing 2 (0.8) 4 (1.0) 309 (0.8) 315 (0.8) No. of previous skin lesions destroyed 0 55 (21.4) 102 (25.6) 19381 (46.9) 19538 (46.6) 1–5 65 (25.3) 93 (23.4) 10796 (26.1) 10954 (26.1) 6–10 35 (13.6) 62 (15.6) 3741 (9.1) 3838 (9.2) 11–20 33 (12.8) 52 (13.1) 2996 (7.3) 3081 (7.3) 21–49 35 (13.6) 35 (8.8) 2285 (5.5) 2355 (5.6) 50+ 33 (12.8) 51 (12.8) 1867 (4.5) 1951 (4.7) Missing 1 (0.4) 3 (0.8) 233 (0.6) 237 (0.6) Sunscreen use (past year) Never 30 (11.7) 63 (15.8) 8728 (21.1) 8821 (21.0) Less than 50% of the time 121 (47.1) 172 (43.2) 17393 (42.1) 17686 (42.2) More than 50% of the time 79 (30.7) 129 (32.4) 11207 (27.1) 11415 (27.2) All the time 27 (10.5) 33 (8.3) 3722 (9.0) 3782 (9.0) Missing 0 1 (0.3) 249 (0.6) 250 (0.6) Skin checks by a doctor (past 3 y) None 38 (14.8) 45 (11.3) 11604 (28.1) 11687 (27.9) Once 62 (24.1) 104 (26.1) 12183 (29.5) 12349 (29.4) 2–5 times 119 (46.3) 187 (47.0) 14053 (34.0) 14359 (34.2) More than 5 times 36 (14.0) 59 (14.8) 2477 (6.0) 2572 (6.1) Missing 2 (0.8) 3 (0.8) 982 (2.4) 987 (2.4) Family history of melanoma No 151 (58.8) 231 (58.0) 25859 (62.6) 26241 (62.6) Yes 77 (30.0) 120 (30.2) 9131 (22.1) 9328 (22.2) Missing 29 (11.3) 47 (11.8) 6309 (15.3) 6385 (15.2) Table 1. Distribution of selected characteristics of study participants in the QSkin cohort (n = 41 954) Characteristic Invasive melanoma cases (n = 257)No. (%) In situ melanoma cases (n = 398)No. (%) Noncases(n = 41 299)No. (%) Total(n = 41 954)No. (%) Age group, y 40–49 37 (14.4) 71 (18.4) 11248 (27.2) 11356 (27.1) 50–50 93 (36.2) 137 (34.4) 15732 (38.1) 11962 (38.1) 60+ 127 (49.4) 190 (47.7) 14319 (34.7) 14636 (34.9) Sex Female 111 (43.2) 161 (40.5) 22693 (55.0) 22965 (54.7) Male 146 (56.8) 237 (59.6) 18606 (45.1) 18989 (45.3) Ethnicity Nonwhite 3 (1.2) 7 (1.8) 2666 (6.5) 2676 (6.4) White 253 (98.4) 389 (97.7) 38213 (92.5) 38855 (92.6) Missing 1 (0.4) 2 (0.5) 420 (1.0) 423 (1.0) Private health insurance No 69 (26.9) 96 (24.1) 13850 (33.5) 14015 (33.4) Yes 187 (72.8) 301 (75.6) 27245 (65.9) 27733 (66.1) Missing 1 (0.4) 1 (0.3) 204 (0.5) 206 (0.5) Tanning ability Tan deeply 26 (10.1) 74 (18.6) 9714 (23.5) 9814 (23.4) Tan moderately 122 (47.5) 191 (48.0) 20230 (49.0) 20543 (49.0) Tan a little 70 (27.2) 97 (24.4) 8463 (20.5) 8630 (20.6) Not tan 36 (14.0) 34 (8.5) 2557 (6.2) 2627 (6.3) Missing 3 (1.2) 2 (0.5) 335 (0.8) 340 (0.8) Moles at age 21 y None 48 (18.7) 94 (23.6) 11638 (28.2) 11780 (28.1) A few 128 (49.8) 186 (46.7) 21452 (51.9) 21766 (51.9) Some 59 (23.0) 83 (20.9) 5896 (14.3) 6038 (14.4) Many 17 (6.6) 28 (7.0) 1189 (2.9) 1234 (2.9) Missing 5 (2.0) 7 (1.8) 1124 (2.7) 1136 (2.7) Hair color Black 10 (3.9) 38 (9.6) 4195 (10.2) 4243 (10.1) Dark brown 74 (28.8) 113 (28.4) 13805 (33.4) 13992 (33.4) Light brown 109 (42.4) 149 (37.4) 15119 (36.6) 15377 (36.7) Blond 29 (11.3) 54 (13.6) 5705 (13.8) 5788 (13.8) Red/auburn 34 (13.2) 39 (9.8) 2228 (5.4) 2301 (5.5) Missing 1 (0.4) 5 (1.3) 247 (0.6) 253 (0.6) No. of previous skin cancers removed surgically 0 97 (37.7) 144 (36.2) 25839 (62.6) 26080 (62.2) 1 37 (14.4) 67 (16.8) 5497 (13.3) 5601 (13.4) 2–10 91 (35.4) 141 (35.4) 8115 (19.7) 8347 (19.9) 11–20 19 (7.4) 23 (5.8) 931 (2.3) 973 (2.3) 20+ 11 (4.3) 19 (4.8) 608 (1.5) 638 (1.5) Missing 2 (0.8) 4 (1.0) 309 (0.8) 315 (0.8) No. of previous skin lesions destroyed 0 55 (21.4) 102 (25.6) 19381 (46.9) 19538 (46.6) 1–5 65 (25.3) 93 (23.4) 10796 (26.1) 10954 (26.1) 6–10 35 (13.6) 62 (15.6) 3741 (9.1) 3838 (9.2) 11–20 33 (12.8) 52 (13.1) 2996 (7.3) 3081 (7.3) 21–49 35 (13.6) 35 (8.8) 2285 (5.5) 2355 (5.6) 50+ 33 (12.8) 51 (12.8) 1867 (4.5) 1951 (4.7) Missing 1 (0.4) 3 (0.8) 233 (0.6) 237 (0.6) Sunscreen use (past year) Never 30 (11.7) 63 (15.8) 8728 (21.1) 8821 (21.0) Less than 50% of the time 121 (47.1) 172 (43.2) 17393 (42.1) 17686 (42.2) More than 50% of the time 79 (30.7) 129 (32.4) 11207 (27.1) 11415 (27.2) All the time 27 (10.5) 33 (8.3) 3722 (9.0) 3782 (9.0) Missing 0 1 (0.3) 249 (0.6) 250 (0.6) Skin checks by a doctor (past 3 y) None 38 (14.8) 45 (11.3) 11604 (28.1) 11687 (27.9) Once 62 (24.1) 104 (26.1) 12183 (29.5) 12349 (29.4) 2–5 times 119 (46.3) 187 (47.0) 14053 (34.0) 14359 (34.2) More than 5 times 36 (14.0) 59 (14.8) 2477 (6.0) 2572 (6.1) Missing 2 (0.8) 3 (0.8) 982 (2.4) 987 (2.4) Family history of melanoma No 151 (58.8) 231 (58.0) 25859 (62.6) 26241 (62.6) Yes 77 (30.0) 120 (30.2) 9131 (22.1) 9328 (22.2) Missing 29 (11.3) 47 (11.8) 6309 (15.3) 6385 (15.2) Characteristic Invasive melanoma cases (n = 257)No. (%) In situ melanoma cases (n = 398)No. (%) Noncases(n = 41 299)No. (%) Total(n = 41 954)No. (%) Age group, y 40–49 37 (14.4) 71 (18.4) 11248 (27.2) 11356 (27.1) 50–50 93 (36.2) 137 (34.4) 15732 (38.1) 11962 (38.1) 60+ 127 (49.4) 190 (47.7) 14319 (34.7) 14636 (34.9) Sex Female 111 (43.2) 161 (40.5) 22693 (55.0) 22965 (54.7) Male 146 (56.8) 237 (59.6) 18606 (45.1) 18989 (45.3) Ethnicity Nonwhite 3 (1.2) 7 (1.8) 2666 (6.5) 2676 (6.4) White 253 (98.4) 389 (97.7) 38213 (92.5) 38855 (92.6) Missing 1 (0.4) 2 (0.5) 420 (1.0) 423 (1.0) Private health insurance No 69 (26.9) 96 (24.1) 13850 (33.5) 14015 (33.4) Yes 187 (72.8) 301 (75.6) 27245 (65.9) 27733 (66.1) Missing 1 (0.4) 1 (0.3) 204 (0.5) 206 (0.5) Tanning ability Tan deeply 26 (10.1) 74 (18.6) 9714 (23.5) 9814 (23.4) Tan moderately 122 (47.5) 191 (48.0) 20230 (49.0) 20543 (49.0) Tan a little 70 (27.2) 97 (24.4) 8463 (20.5) 8630 (20.6) Not tan 36 (14.0) 34 (8.5) 2557 (6.2) 2627 (6.3) Missing 3 (1.2) 2 (0.5) 335 (0.8) 340 (0.8) Moles at age 21 y None 48 (18.7) 94 (23.6) 11638 (28.2) 11780 (28.1) A few 128 (49.8) 186 (46.7) 21452 (51.9) 21766 (51.9) Some 59 (23.0) 83 (20.9) 5896 (14.3) 6038 (14.4) Many 17 (6.6) 28 (7.0) 1189 (2.9) 1234 (2.9) Missing 5 (2.0) 7 (1.8) 1124 (2.7) 1136 (2.7) Hair color Black 10 (3.9) 38 (9.6) 4195 (10.2) 4243 (10.1) Dark brown 74 (28.8) 113 (28.4) 13805 (33.4) 13992 (33.4) Light brown 109 (42.4) 149 (37.4) 15119 (36.6) 15377 (36.7) Blond 29 (11.3) 54 (13.6) 5705 (13.8) 5788 (13.8) Red/auburn 34 (13.2) 39 (9.8) 2228 (5.4) 2301 (5.5) Missing 1 (0.4) 5 (1.3) 247 (0.6) 253 (0.6) No. of previous skin cancers removed surgically 0 97 (37.7) 144 (36.2) 25839 (62.6) 26080 (62.2) 1 37 (14.4) 67 (16.8) 5497 (13.3) 5601 (13.4) 2–10 91 (35.4) 141 (35.4) 8115 (19.7) 8347 (19.9) 11–20 19 (7.4) 23 (5.8) 931 (2.3) 973 (2.3) 20+ 11 (4.3) 19 (4.8) 608 (1.5) 638 (1.5) Missing 2 (0.8) 4 (1.0) 309 (0.8) 315 (0.8) No. of previous skin lesions destroyed 0 55 (21.4) 102 (25.6) 19381 (46.9) 19538 (46.6) 1–5 65 (25.3) 93 (23.4) 10796 (26.1) 10954 (26.1) 6–10 35 (13.6) 62 (15.6) 3741 (9.1) 3838 (9.2) 11–20 33 (12.8) 52 (13.1) 2996 (7.3) 3081 (7.3) 21–49 35 (13.6) 35 (8.8) 2285 (5.5) 2355 (5.6) 50+ 33 (12.8) 51 (12.8) 1867 (4.5) 1951 (4.7) Missing 1 (0.4) 3 (0.8) 233 (0.6) 237 (0.6) Sunscreen use (past year) Never 30 (11.7) 63 (15.8) 8728 (21.1) 8821 (21.0) Less than 50% of the time 121 (47.1) 172 (43.2) 17393 (42.1) 17686 (42.2) More than 50% of the time 79 (30.7) 129 (32.4) 11207 (27.1) 11415 (27.2) All the time 27 (10.5) 33 (8.3) 3722 (9.0) 3782 (9.0) Missing 0 1 (0.3) 249 (0.6) 250 (0.6) Skin checks by a doctor (past 3 y) None 38 (14.8) 45 (11.3) 11604 (28.1) 11687 (27.9) Once 62 (24.1) 104 (26.1) 12183 (29.5) 12349 (29.4) 2–5 times 119 (46.3) 187 (47.0) 14053 (34.0) 14359 (34.2) More than 5 times 36 (14.0) 59 (14.8) 2477 (6.0) 2572 (6.1) Missing 2 (0.8) 3 (0.8) 982 (2.4) 987 (2.4) Family history of melanoma No 151 (58.8) 231 (58.0) 25859 (62.6) 26241 (62.6) Yes 77 (30.0) 120 (30.2) 9131 (22.1) 9328 (22.2) Missing 29 (11.3) 47 (11.8) 6309 (15.3) 6385 (15.2) Statistical Analysis We randomly selected a two-thirds sample of the data set to derive the prediction models (hereafter “derivation sample”; n = 27 975); and assessed model performance in the remaining one-third sample (“validation sample”; n = 13 979). Risk Model Development We used Cox proportional hazards modeling to derive the risk model. We tested the proportional hazards assumption for the final model in each imputed data set using the scaled Schoenfeld residuals test. We first assessed the association between self-perceived risk and first primary melanoma development, and then proceeded to derive predictive models from the pool of 28 candidate variables. To select variables from five imputed development data sets, we used a stepwise selection process consisting of alternating forward selection and backward elimination steps. For each imputation, best subset variables were selected based on a 10% level of statistical significance (two-sided) using the type III test. The final model was run in all five imputed data sets, and we then combined the regression coefficients using a modified version of Rubin’s rules suitable for categorical variables (24). We examined the estimates for the final models with and without taking into account competing risk of death (25). We tested two-way interactions between covariates. In our primary analyses for invasive melanoma, we ignored events during follow-up because our tool estimates future risk; subsequent events cannot be known at the time of risk assessment. To examine the influence of this approach, we conducted a sensitivity analysis in which we censored in situ melanoma cases diagnosed during follow-up. Finally, to examine whether predictive factors differed according to age (<60 years, ≥60 years) and sex, we repeated the process within strata of these variables. Model Validation We assessed the model performance using tests for discrimination and calibration. We evaluated discrimination using the C-index and its 95% confidence interval (CI) (26). Rubin’s rules were applied to pool the C-index from the five imputed data sets. A C-index of 0.5 indicates discrimination no better than chance, and a C-index of 1.0 indicates perfect discrimination. To assess calibration, we calculated a linear predictor score for each individual in the development and validation samples based on our model coefficients, and then defined five risk groups based on the quintiles of score among cases, as described by Royston (27). We estimated cumulative hazard at observed time points for each risk group in both the development and validation samples (27) and then plotted the expected vs observed cumulative hazard curves for each risk group (28). To assess the tool’s utility to guide targeted screening (ie, risk stratification), we calculated the “number needed to screen” for different scenarios in which melanoma screening was restricted to groups of individuals defined by their predicted risk score. We calculated the Youden index (29) for each decile of predicted risk to identify the “optimal” cutoff point at which both sensitivity and specificity were maximized. (29). Finally, we conducted decision curve analyses to evaluate the clinical utility of the model across a range of risk thresholds for screening (30,31). Results Patient Characteristics Of 41 954 eligible participants, 54.7% were women, and the mean age was 56 years. Most participants reported white European ancestry (92.6%), and most were born in Australia (80.2%). The mean follow-up time was 3.5 years (median = 3.4 years, interquartile range = 3.3–3.8 years). Distributions of factors that were statistically significant in the final models (P < .05) are presented in Table 1 by outcome status; the distribution of statistically nonsignificant candidate predictor items by outcome status is presented in Supplementary Table 1 (available online). During follow up, 655 (1.6%) participants developed melanoma (257 invasive melanomas, 398 in situ melanomas). Of the invasive cases, 75.9% were less than 1 mm and 60.7% were of the superficial spreading subtype (Supplementary Table 2, available online). Key variables were distributed similarly in development and validation samples (Supplementary Table 3, available online). Self-perceived Risk We first examined whether self-perceived risk of developing melanoma alone was predictive of invasive melanoma during follow-up. The group perceiving themselves to be “highly likely” to develop melanoma had statistically significantly higher risks than the other four risk groups (hazard ratio = 2.53, 95% confidence interval [CI] = 1.23 to 5.19, for “highly likely” vs “highly unlikely,” P = .01) (Figure 1). Self-perceived risk discriminated cases from noncases only slightly better than chance, however (C-index = 0.54, 95% CI = 0.48 to 0.61). Figure 1. View largeDownload slide Cumulative hazard of developing invasive melanoma according to self-perceived risk. Figure 1. View largeDownload slide Cumulative hazard of developing invasive melanoma according to self-perceived risk. Prediction of Invasive Melanomas Seven statistically significant predictors were retained in the final model (viz. age, sex, hair color, tanning ability, number of moles at age 21 years, number of skin lesions treated destructively, and sunscreen use) (Table 2). The strongest predictors for invasive melanoma were age, sex, tanning ability, number of moles at age 21 years, and number of skin lesions treated destructively (Table 2). There were no statistically significant interactions between any covariates in the final model. Table 2. Specification and performance of the model to predict risk of invasive melanoma: Derivation sample (n = 27 975) Risk factor HR (95% CI) P* Age group, y ≤45 1.00 (ref) 46–50 0.60 (0.28 to 1.32) 51–55 0.76 (0.38 to 1.53) 56–60 1.84 (1.01 to 3.36) 61–65 1.55 (0.83 to 2.91) >65 2.34 (1.27 to 4.31) <.001 Sex Female 1.00 (ref) Male 2.13 (1.56 to 2.92) <.001 Tanning ability Deeply tan 1.00 (ref) Moderately tan 2.39 (1.32 to 4.30) Lightly tan 3.50 (1.89 to 6.48) Not tan 4.79 (2.37 to 9.69) <.001 Moles at age 21 y None 1.00 (ref) A few 1.87 (1.24 to 2.81) Some 3.29 (2.07 to 5.23) Many 4.42 (2.14 to 9.10) <.001 Hair color Black 1.00 (ref) Brown/blond 2.54 (1.12 to 5.78) Red/auburn 4.46 (1.79 to 11.14) .003 No. of previous skin lesions destroyed 0 1.00 (ref) 1–5 1.51 (0.96 to 2.39) 6–20 2.26 (1.44 to 3.53) 21+ 2.51 (1.56 to 4.05) <.001 Sunscreen use when outside in the sun (past year) Never 1.00 (ref) Ever 2.01 (1.24 to 3.27) .005 C-index† pooled (95% CI) Derivation 0.79 (0.72 to 0.85) Validation 0.69 (0.62 to 0.76) Risk factor HR (95% CI) P* Age group, y ≤45 1.00 (ref) 46–50 0.60 (0.28 to 1.32) 51–55 0.76 (0.38 to 1.53) 56–60 1.84 (1.01 to 3.36) 61–65 1.55 (0.83 to 2.91) >65 2.34 (1.27 to 4.31) <.001 Sex Female 1.00 (ref) Male 2.13 (1.56 to 2.92) <.001 Tanning ability Deeply tan 1.00 (ref) Moderately tan 2.39 (1.32 to 4.30) Lightly tan 3.50 (1.89 to 6.48) Not tan 4.79 (2.37 to 9.69) <.001 Moles at age 21 y None 1.00 (ref) A few 1.87 (1.24 to 2.81) Some 3.29 (2.07 to 5.23) Many 4.42 (2.14 to 9.10) <.001 Hair color Black 1.00 (ref) Brown/blond 2.54 (1.12 to 5.78) Red/auburn 4.46 (1.79 to 11.14) .003 No. of previous skin lesions destroyed 0 1.00 (ref) 1–5 1.51 (0.96 to 2.39) 6–20 2.26 (1.44 to 3.53) 21+ 2.51 (1.56 to 4.05) <.001 Sunscreen use when outside in the sun (past year) Never 1.00 (ref) Ever 2.01 (1.24 to 3.27) .005 C-index† pooled (95% CI) Derivation 0.79 (0.72 to 0.85) Validation 0.69 (0.62 to 0.76) * Type 3 P value (two-sided) from development data set (first imputation). CI = confidence interval; HR = hazard ratio. † Pooled C-index for five imputed data sets. Table 2. Specification and performance of the model to predict risk of invasive melanoma: Derivation sample (n = 27 975) Risk factor HR (95% CI) P* Age group, y ≤45 1.00 (ref) 46–50 0.60 (0.28 to 1.32) 51–55 0.76 (0.38 to 1.53) 56–60 1.84 (1.01 to 3.36) 61–65 1.55 (0.83 to 2.91) >65 2.34 (1.27 to 4.31) <.001 Sex Female 1.00 (ref) Male 2.13 (1.56 to 2.92) <.001 Tanning ability Deeply tan 1.00 (ref) Moderately tan 2.39 (1.32 to 4.30) Lightly tan 3.50 (1.89 to 6.48) Not tan 4.79 (2.37 to 9.69) <.001 Moles at age 21 y None 1.00 (ref) A few 1.87 (1.24 to 2.81) Some 3.29 (2.07 to 5.23) Many 4.42 (2.14 to 9.10) <.001 Hair color Black 1.00 (ref) Brown/blond 2.54 (1.12 to 5.78) Red/auburn 4.46 (1.79 to 11.14) .003 No. of previous skin lesions destroyed 0 1.00 (ref) 1–5 1.51 (0.96 to 2.39) 6–20 2.26 (1.44 to 3.53) 21+ 2.51 (1.56 to 4.05) <.001 Sunscreen use when outside in the sun (past year) Never 1.00 (ref) Ever 2.01 (1.24 to 3.27) .005 C-index† pooled (95% CI) Derivation 0.79 (0.72 to 0.85) Validation 0.69 (0.62 to 0.76) Risk factor HR (95% CI) P* Age group, y ≤45 1.00 (ref) 46–50 0.60 (0.28 to 1.32) 51–55 0.76 (0.38 to 1.53) 56–60 1.84 (1.01 to 3.36) 61–65 1.55 (0.83 to 2.91) >65 2.34 (1.27 to 4.31) <.001 Sex Female 1.00 (ref) Male 2.13 (1.56 to 2.92) <.001 Tanning ability Deeply tan 1.00 (ref) Moderately tan 2.39 (1.32 to 4.30) Lightly tan 3.50 (1.89 to 6.48) Not tan 4.79 (2.37 to 9.69) <.001 Moles at age 21 y None 1.00 (ref) A few 1.87 (1.24 to 2.81) Some 3.29 (2.07 to 5.23) Many 4.42 (2.14 to 9.10) <.001 Hair color Black 1.00 (ref) Brown/blond 2.54 (1.12 to 5.78) Red/auburn 4.46 (1.79 to 11.14) .003 No. of previous skin lesions destroyed 0 1.00 (ref) 1–5 1.51 (0.96 to 2.39) 6–20 2.26 (1.44 to 3.53) 21+ 2.51 (1.56 to 4.05) <.001 Sunscreen use when outside in the sun (past year) Never 1.00 (ref) Ever 2.01 (1.24 to 3.27) .005 C-index† pooled (95% CI) Derivation 0.79 (0.72 to 0.85) Validation 0.69 (0.62 to 0.76) * Type 3 P value (two-sided) from development data set (first imputation). CI = confidence interval; HR = hazard ratio. † Pooled C-index for five imputed data sets. In the sensitivity analysis, we censored participants diagnosed with in situ melanoma during follow-up: the final model included the same seven terms as the risk model derived from our primary analysis. Prediction of All Melanomas (Invasive + In Situ) The final model for predicting the risk for all melanomas (ie, invasive + in situ) retained the seven terms identified for invasive melanomas, with five new terms: history of skin checks by a doctor (past three years), number of excisions for skin cancers, private health insurance, ethnicity, and family history of melanoma (Table 3). The strongest predictors were age, sex, moles at age 21 years, and hair color (Table 3). Table 3. Specification and performance of the model to predict risk of any melanoma (invasive or in situ): Derivation sample (n = 27 975) Risk factor HR (95% CI) P* Age group, y ≤45 1.00 (ref) 46–50 0.80 (0.50 to 1.26) 51–55 0.86 (0.56 to 1.31) 56–60 1.48 (1.00 to 2.18) 61–65 1.56 (1.04 to 2.32) >65 1.93 (1.30 to 2.87) <.001 Sex Female 1.00 (ref) Male 1.86 (1.53 to 2.28) <.001 Ethnicity Nonwhite 1.00 (ref) White 2.49 (1.10 to 5.62) .03 Private health insurance No 1.00 (ref) Yes 1.24 (0.99 to 1.54) .05 Tanning ability Deeply tan 1.00 (ref) Moderately tan 1.46 (1.09 to 1.96) Lightly tan 1.69 (1.22 to 2.34) Not tan 1.56 (1.02 to 2.40) .02 Moles at age 21 y None 1.00 (ref) A few 1.27 (1.00 to 1.62) Some 2.09 (1.57 to 2.78) Many 3.58 (2.34 to 5.48) <.001 Hair color Black 1.00 (ref) Brown/blond 1.13 (0.78 to 1.62) Red/auburn 2.02 (1.27 to 3.19) .001 No. of previous skin lesions destroyed 0 1.00 (ref) 1–5 1.24 (0.92 to 1.66) 6–20 1.68 (1.23 to 2.30) 21+ 1.60 (1.12 to 2.30) .006 Family history of melanoma No 1.00 (ref) Yes 1.24 (1.00 to 1.53) .03 Sunscreen use when outside in the sun (past year) Never 1.00 (ref) Ever 1.51 (1.13 to 2.02) .006 Past history of excisions for skin cancer 0 1.00 (ref) 1 1.30 (0.97 to 1.75) 2+ 1.42 (1.10 to 1.84) .04 Skin checks by a doctor (past 3 y) None 1.00 (ref) Once 1.46 (1.05 to 2.04) 2 or more times 1.71 (1.24 to 2.36) .006 C-index†, pooled (95% CI) Derivation 0.74 (0.71 to 0.78) Validation 0.72 (0.69 to 0.75) Risk factor HR (95% CI) P* Age group, y ≤45 1.00 (ref) 46–50 0.80 (0.50 to 1.26) 51–55 0.86 (0.56 to 1.31) 56–60 1.48 (1.00 to 2.18) 61–65 1.56 (1.04 to 2.32) >65 1.93 (1.30 to 2.87) <.001 Sex Female 1.00 (ref) Male 1.86 (1.53 to 2.28) <.001 Ethnicity Nonwhite 1.00 (ref) White 2.49 (1.10 to 5.62) .03 Private health insurance No 1.00 (ref) Yes 1.24 (0.99 to 1.54) .05 Tanning ability Deeply tan 1.00 (ref) Moderately tan 1.46 (1.09 to 1.96) Lightly tan 1.69 (1.22 to 2.34) Not tan 1.56 (1.02 to 2.40) .02 Moles at age 21 y None 1.00 (ref) A few 1.27 (1.00 to 1.62) Some 2.09 (1.57 to 2.78) Many 3.58 (2.34 to 5.48) <.001 Hair color Black 1.00 (ref) Brown/blond 1.13 (0.78 to 1.62) Red/auburn 2.02 (1.27 to 3.19) .001 No. of previous skin lesions destroyed 0 1.00 (ref) 1–5 1.24 (0.92 to 1.66) 6–20 1.68 (1.23 to 2.30) 21+ 1.60 (1.12 to 2.30) .006 Family history of melanoma No 1.00 (ref) Yes 1.24 (1.00 to 1.53) .03 Sunscreen use when outside in the sun (past year) Never 1.00 (ref) Ever 1.51 (1.13 to 2.02) .006 Past history of excisions for skin cancer 0 1.00 (ref) 1 1.30 (0.97 to 1.75) 2+ 1.42 (1.10 to 1.84) .04 Skin checks by a doctor (past 3 y) None 1.00 (ref) Once 1.46 (1.05 to 2.04) 2 or more times 1.71 (1.24 to 2.36) .006 C-index†, pooled (95% CI) Derivation 0.74 (0.71 to 0.78) Validation 0.72 (0.69 to 0.75) * Type 3 P value (two-sided) from development data set (first imputation). CI = confidence interval; HR = hazard ratio. † Pooled C-index for five imputed data sets. Table 3. Specification and performance of the model to predict risk of any melanoma (invasive or in situ): Derivation sample (n = 27 975) Risk factor HR (95% CI) P* Age group, y ≤45 1.00 (ref) 46–50 0.80 (0.50 to 1.26) 51–55 0.86 (0.56 to 1.31) 56–60 1.48 (1.00 to 2.18) 61–65 1.56 (1.04 to 2.32) >65 1.93 (1.30 to 2.87) <.001 Sex Female 1.00 (ref) Male 1.86 (1.53 to 2.28) <.001 Ethnicity Nonwhite 1.00 (ref) White 2.49 (1.10 to 5.62) .03 Private health insurance No 1.00 (ref) Yes 1.24 (0.99 to 1.54) .05 Tanning ability Deeply tan 1.00 (ref) Moderately tan 1.46 (1.09 to 1.96) Lightly tan 1.69 (1.22 to 2.34) Not tan 1.56 (1.02 to 2.40) .02 Moles at age 21 y None 1.00 (ref) A few 1.27 (1.00 to 1.62) Some 2.09 (1.57 to 2.78) Many 3.58 (2.34 to 5.48) <.001 Hair color Black 1.00 (ref) Brown/blond 1.13 (0.78 to 1.62) Red/auburn 2.02 (1.27 to 3.19) .001 No. of previous skin lesions destroyed 0 1.00 (ref) 1–5 1.24 (0.92 to 1.66) 6–20 1.68 (1.23 to 2.30) 21+ 1.60 (1.12 to 2.30) .006 Family history of melanoma No 1.00 (ref) Yes 1.24 (1.00 to 1.53) .03 Sunscreen use when outside in the sun (past year) Never 1.00 (ref) Ever 1.51 (1.13 to 2.02) .006 Past history of excisions for skin cancer 0 1.00 (ref) 1 1.30 (0.97 to 1.75) 2+ 1.42 (1.10 to 1.84) .04 Skin checks by a doctor (past 3 y) None 1.00 (ref) Once 1.46 (1.05 to 2.04) 2 or more times 1.71 (1.24 to 2.36) .006 C-index†, pooled (95% CI) Derivation 0.74 (0.71 to 0.78) Validation 0.72 (0.69 to 0.75) Risk factor HR (95% CI) P* Age group, y ≤45 1.00 (ref) 46–50 0.80 (0.50 to 1.26) 51–55 0.86 (0.56 to 1.31) 56–60 1.48 (1.00 to 2.18) 61–65 1.56 (1.04 to 2.32) >65 1.93 (1.30 to 2.87) <.001 Sex Female 1.00 (ref) Male 1.86 (1.53 to 2.28) <.001 Ethnicity Nonwhite 1.00 (ref) White 2.49 (1.10 to 5.62) .03 Private health insurance No 1.00 (ref) Yes 1.24 (0.99 to 1.54) .05 Tanning ability Deeply tan 1.00 (ref) Moderately tan 1.46 (1.09 to 1.96) Lightly tan 1.69 (1.22 to 2.34) Not tan 1.56 (1.02 to 2.40) .02 Moles at age 21 y None 1.00 (ref) A few 1.27 (1.00 to 1.62) Some 2.09 (1.57 to 2.78) Many 3.58 (2.34 to 5.48) <.001 Hair color Black 1.00 (ref) Brown/blond 1.13 (0.78 to 1.62) Red/auburn 2.02 (1.27 to 3.19) .001 No. of previous skin lesions destroyed 0 1.00 (ref) 1–5 1.24 (0.92 to 1.66) 6–20 1.68 (1.23 to 2.30) 21+ 1.60 (1.12 to 2.30) .006 Family history of melanoma No 1.00 (ref) Yes 1.24 (1.00 to 1.53) .03 Sunscreen use when outside in the sun (past year) Never 1.00 (ref) Ever 1.51 (1.13 to 2.02) .006 Past history of excisions for skin cancer 0 1.00 (ref) 1 1.30 (0.97 to 1.75) 2+ 1.42 (1.10 to 1.84) .04 Skin checks by a doctor (past 3 y) None 1.00 (ref) Once 1.46 (1.05 to 2.04) 2 or more times 1.71 (1.24 to 2.36) .006 C-index†, pooled (95% CI) Derivation 0.74 (0.71 to 0.78) Validation 0.72 (0.69 to 0.75) * Type 3 P value (two-sided) from development data set (first imputation). CI = confidence interval; HR = hazard ratio. † Pooled C-index for five imputed data sets. The prediction models for both outcomes met the proportional hazards assumption. Final models that incorporated death as a competing risk resulted in negligible change to the estimates (data not shown). Several variables were close to the level of statistical significance required for retention in the models, including history of cancer, and for the invasive melanoma model only, history of physician skin checks and self-perceived melanoma risk. Including these factors in the final models did not change the pooled C-statistic or model fit. Thus we have presented the most parsimonious models that retained only statistically significant variables. Discrimination and Calibration Model discrimination in both the derivation and validation samples was high. The mean C-index across the five imputed data sets for the invasive melanoma model was 0.79 (95% CI = 0.72 to 0.85) in the development sample and 0.69 (95% CI = 0.62 to 0.76) in the validation sample (Table 2). The C-index was slightly lower for the “all melanoma” model in the development data set (pooled C-index = 0.74, 95% CI = 0.71 to 0.78) but higher in the validation sample (pooled C-index = 0.72, 95% CI = 0.69 to 0.75) than for the invasive risk model (Table 3). When we plotted observed and predicted cumulative hazards of invasive melanoma for each of five risk strata of cohort participants in the validation sample, we observed reasonable calibration for the three lowest-risk groups, but the model overestimated risk of melanoma for the two highest-risk groups (Figure 2A). We observed better calibration for the model predicting risks of all melanomas (invasive and in situ) (Figure 2B). Figure 2. View largeDownload slide Calibration of the stratification tool across the five risk strata in the validation sample: predicted and observed cumulative hazards of: (A) invasive melanoma (n = 80 melanoma cases) and (B) all melanoma (n = 210 melanoma cases). Black lines represent the predicted cumulative hazard curves, and gray lines represent the observed cumulative hazard curves. Figure 2. View largeDownload slide Calibration of the stratification tool across the five risk strata in the validation sample: predicted and observed cumulative hazards of: (A) invasive melanoma (n = 80 melanoma cases) and (B) all melanoma (n = 210 melanoma cases). Black lines represent the predicted cumulative hazard curves, and gray lines represent the observed cumulative hazard curves. Stratified Analyses We conducted stratified analyses to determine whether predictive factors differed according to age and sex and found some variation in the terms retained in the models. The data became very sparse in some subgroups, however, with low numbers of melanoma cases in the validation sample (Supplementary Tables 4–7, available online). Optimizing Melanoma Screening We assessed the sensitivity and specificity of the tool within deciles of predicted risk (Table 4). The Youden index was optimized (J = 0.35) in the seventh decile, with a sensitivity of 74.2% and specificity of 60.7%. If this level of predicted risk was the threshold for eligibility to enter a melanoma screening program, then of 34 people meeting or exceeding this threshold, one would develop melanoma (invasive or in situ) over a median duration of 3.4 years. However, a strategy based on that threshold would fail to detect 25.8% of future cases. If the entry criteria for melanoma screening were relaxed so that it was open to all those with predicted risks in the sixth decile or higher, we estimate that 82.1% of melanoma cases would be detected, and the number needed to screen would be 39. Table 4. Performance of melanoma screening within deciles of predicted risk score (all melanoma) Predicted risk score decile Observed absolute risk at 1 y†, % Observed absolute risk at 2 y†, % Observed absolute risk at 3 y†, % Sensitivity, % Specificity, % Youden index, J Cumulative No. needed to screen* 10th 1.34 2.83 4.16 31.5 90.4 0.22 20 9th 0.65 1.61 2.43 49.5 80.5 0.30 26 8th 0.65 1.35 1.93 63.5 70.7 0.34 30 7th 0.36 0.82 1.35 74.2 60.7 0.35 34 6th 0.22 0.84 1.11 82.1 50.7 0.33 39 5th 0.19 0.51 0.73 87.6 40.8 0.28 44 4th 0.21 0.38 0.62 91.9 30.6 0.23 49 3th 0.24 0.43 0.72 96.9 20.6 0.18 53 2th 0.02 0.12 0.21 98.6 10.4 0.09 58 1st 0.07 0.19 0.21 100.0 0.0 0.00 64 Predicted risk score decile Observed absolute risk at 1 y†, % Observed absolute risk at 2 y†, % Observed absolute risk at 3 y†, % Sensitivity, % Specificity, % Youden index, J Cumulative No. needed to screen* 10th 1.34 2.83 4.16 31.5 90.4 0.22 20 9th 0.65 1.61 2.43 49.5 80.5 0.30 26 8th 0.65 1.35 1.93 63.5 70.7 0.34 30 7th 0.36 0.82 1.35 74.2 60.7 0.35 34 6th 0.22 0.84 1.11 82.1 50.7 0.33 39 5th 0.19 0.51 0.73 87.6 40.8 0.28 44 4th 0.21 0.38 0.62 91.9 30.6 0.23 49 3th 0.24 0.43 0.72 96.9 20.6 0.18 53 2th 0.02 0.12 0.21 98.6 10.4 0.09 58 1st 0.07 0.19 0.21 100.0 0.0 0.00 64 * Numbers needed to screen will depend on the prevailing incidence of melanoma in the population to which the tool is applied. Numbers cited here were derived from the QSkin cohort (age 40–69 years; crude melanoma incidence 172 × 10-5 person-years). In populations with lower melanoma incidence, the numbers needed to screen will be higher. † Observed absolute risk of melanoma calculated using the Kaplan-Meier method for groups classified according to the predicted risk score deciles. Table 4. Performance of melanoma screening within deciles of predicted risk score (all melanoma) Predicted risk score decile Observed absolute risk at 1 y†, % Observed absolute risk at 2 y†, % Observed absolute risk at 3 y†, % Sensitivity, % Specificity, % Youden index, J Cumulative No. needed to screen* 10th 1.34 2.83 4.16 31.5 90.4 0.22 20 9th 0.65 1.61 2.43 49.5 80.5 0.30 26 8th 0.65 1.35 1.93 63.5 70.7 0.34 30 7th 0.36 0.82 1.35 74.2 60.7 0.35 34 6th 0.22 0.84 1.11 82.1 50.7 0.33 39 5th 0.19 0.51 0.73 87.6 40.8 0.28 44 4th 0.21 0.38 0.62 91.9 30.6 0.23 49 3th 0.24 0.43 0.72 96.9 20.6 0.18 53 2th 0.02 0.12 0.21 98.6 10.4 0.09 58 1st 0.07 0.19 0.21 100.0 0.0 0.00 64 Predicted risk score decile Observed absolute risk at 1 y†, % Observed absolute risk at 2 y†, % Observed absolute risk at 3 y†, % Sensitivity, % Specificity, % Youden index, J Cumulative No. needed to screen* 10th 1.34 2.83 4.16 31.5 90.4 0.22 20 9th 0.65 1.61 2.43 49.5 80.5 0.30 26 8th 0.65 1.35 1.93 63.5 70.7 0.34 30 7th 0.36 0.82 1.35 74.2 60.7 0.35 34 6th 0.22 0.84 1.11 82.1 50.7 0.33 39 5th 0.19 0.51 0.73 87.6 40.8 0.28 44 4th 0.21 0.38 0.62 91.9 30.6 0.23 49 3th 0.24 0.43 0.72 96.9 20.6 0.18 53 2th 0.02 0.12 0.21 98.6 10.4 0.09 58 1st 0.07 0.19 0.21 100.0 0.0 0.00 64 * Numbers needed to screen will depend on the prevailing incidence of melanoma in the population to which the tool is applied. Numbers cited here were derived from the QSkin cohort (age 40–69 years; crude melanoma incidence 172 × 10-5 person-years). In populations with lower melanoma incidence, the numbers needed to screen will be higher. † Observed absolute risk of melanoma calculated using the Kaplan-Meier method for groups classified according to the predicted risk score deciles. Using decision curve analysis, we found that the net benefit of the risk prediction model was superior to “screen all” or “screen none” approaches when the threshold probability fell between 0.3% and 8.9% (Figure 3). This range of values includes the probability at which the Youden index was maximized (1.37%). Figure 3. View largeDownload slide Decision curve analysis of the effect of the prediction model for all melanoma (invasive or in situ). Net benefit is plotted against the continuous range of threshold probabilities, which refer to the absolute risk of melanoma in the next 3.4 years, at which screening would be offered. Fixed lines demonstrate the net benefit of “screening all” and “screening none.” Figure 3. View largeDownload slide Decision curve analysis of the effect of the prediction model for all melanoma (invasive or in situ). Net benefit is plotted against the continuous range of threshold probabilities, which refer to the absolute risk of melanoma in the next 3.4 years, at which screening would be offered. Fixed lines demonstrate the net benefit of “screening all” and “screening none.” Discussion Risk stratification tools can improve the efficacy and cost-effectiveness of early detection by identifying those individuals who are likely to benefit from screening (32). We used data from a large cohort to develop a tool to predict risk of developing melanoma over a median duration of 3.4 years. We confirm that self-perceived risk of melanoma correlates poorly with actual risk and is not useful for guiding early detection practices (33). We found that information on seven items yielded a risk prediction index for invasive melanoma with high discrimination (C-index = 0.79, 95% CI = 0.72 to 0.85) and reasonable calibration. The strongest predictors were age, sex, tanning ability, and number of moles at age 21 years, and number of skin lesions treated destructively. Hair color and sunscreen use also independently predicted risk. There is debate as to whether in situ melanomas should be considered intended end points for early detection. Some argue that in situ lesions are legitimate targets for a screening program as they represent premalignant melanomas. Others argue that most are indolent and that their numbers are increasing due to overdiagnosis (34), a phenomenon that could possibly extend to thin invasive melanomas also. Our study had too few melanomas with thickness greater than 1 mm (47 in total, 30 in the development sample) to derive separate models for thick vs thin melanomas, although such an analysis would clearly be of interest. We contend that these arguments represent the extremes of opinion, and neither wholly represents the reality that some in situ melanomas have malignant potential, whereas others do not (35). We therefore developed models accommodating both points of view by separately examining the predictors of invasive and “all melanomas.” Information on 12 items predicted risk of “all melanomas” with high discrimination (C-index = 0.74). Seven of the 12 items were also identified as statistically significant predictors of invasive melanomas, and the additional five items included three host characteristics (ethnicity, family history of melanoma, number of excisions for skin cancers) and two items related to higher levels of medical surveillance (private health insurance status and history of physician skin checks). Only two prediction models for melanoma have been developed in prospective cohort studies. The first model was developed in the Nurses’ Health Study (NHS I and II)/Health Professionals Follow-up Study (HPFS) (16); the second was developed in the Melanoma Inquiry of Southern Sweden (MISS) study (17), a population-based cohort of women. The NHS/HPFS prediction model for invasive melanoma considered 13 factors for men and 19 for women; in situ melanomas were not considered. The final model included six items (age, sex, hair color, number of moles, history of severe/painful sunburn, and family history of melanoma). For the NHS, information on some factors was collected in a retrospective survey conducted 16 years after recruitment (36). The discrimination of our risk model for invasive melanoma was higher than the NHS/HPFS risk model (C-index = 0.62). The MISS model predicted risk of all melanomas (invasive and in situ) and included three items (family history of melanoma, number of nevi, and hair color). No C-index or calibration was reported, nor was it internally validated. Other models have been developed using data from case–control studies (37–43) (reviewed by Usher-Smith et al. [14] and Vuong et al. [15]), some of which have been externally validated (44,45). These models, however, are not able to incorporate baseline incidence of melanoma in the populations in which they were derived, and they are also prone to biases inherent in the case–control design. We have previously validated six melanoma risk prediction tools in a population-based sample from Queensland (45), including the model by Cho et al. (16), four derived from case–control data (38–40,44), and one that used registry data and published risk estimates from meta-analysis (46). Although the ability of the models to discriminate between cases and controls was generally high, the calibration was poor. Our risk stratification model for invasive melanoma thus has advantages over previous models. Strengths include the population-based sampling frame, prospective design, comprehensive capture of salient risk factors and health behaviors at baseline (including factors not assessed in previous studies), prior validation of the baseline questionnaire, complete ascertainment of melanoma events during follow-up (including in situ melanomas), and the analytical approach, which included splitting the cohort into derivation and validation samples. Population screening for melanoma has been implemented in Germany (48), but it is not advocated by authorities in other jurisdictions. Rather, the advice is for doctors to identify high-risk patients and offer them targeted surveillance (6–10). The tool developed here could be used to identify such patients, although practical implementation would require decisions to be made about an acceptable threshold for action. For example, we have shown that a risk score sufficient to correctly identify 82.1% of melanoma patients would require screening 39 people to detect one case. Setting the threshold for screening to a higher risk score would result in lower sensitivity. Optimizing the threshold for action will depend on the resources available and the tolerance for unscreened cases in the population. As opposed to causal modeling, which aims to determine whether a factor has a direct effect on an outcome, prediction modeling aims to derive a model to best explain the occurrence of an outcome. Thus, the factors retained in prediction models should not be interpreted as being causal. Limitations of our analyses include reliance upon self-reported factors and relatively short duration of follow-up resulting in a modest number of incident invasive melanoma cases. We sought to validate our risk stratification model using external data sets; however, no other studies have captured information on all of the factors that were statistically significant predictors in our models, including some factors that were highly predictive (eg, prior treatment for actinic skin lesions). In summary, we have developed a risk stratification tool for melanoma using information that can be captured easily by self-report, and it is thus suitable for use in the general population. Such a tool could be used to guide targeted screening in line with recommendations (6–10). In future research, we plan to implement the stratification tool in clinical settings and assess its effectiveness. Funding This work was supported in part by the National Health and Medical Research Council (NHMRC) of Australia (grant numbers 552429, 1073898). DW, PW, and RN are supported by Research Fellowships from the NHMRC. Notes Affiliations of authors: Department of Population Health, QIMR Berghofer Medical Research Institute, Queensland, Australia (CMO, NP, BST, JCD, PMW, ACG, REN, DCW); School of Public Health, University of Queensland, Queensland, Australia (CMO, NP, PMW, REN, DCW); Cancer Research UK Manchester Institute and University of Manchester, Manchester Academic Health Science Centre, Manchester, UK (ACG). The funders had no role in the design of the study; the collection, analysis, or interpretation of the data; the writing of the manuscript; or the decision to submit the manuscript for publication. We thank Ms. Carly Scott and Ms. Peta Gordon at the Queensland Cancer Registry for their assistance with preparing the cancer data. References 1 Erdmann F , Lortet-Tieulent J , Schuz J , et al. . International trends in the incidence of malignant melanoma 1953-2008—are recent generations at higher or lower risk? Int J Cancer. 2013 ; 132 2 : 385 – 400 . Google Scholar Crossref Search ADS PubMed 2 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 6 : 1161 – 1171 . Google Scholar Crossref Search ADS PubMed 3 Guy GP Jr , Thomas CC , Thompson T , et al. . Vital signs: Melanoma incidence and mortality trends and projections - United States, 1982-2030 . MMWR Morb Mortal Wkly Rep . 2015 ; 64 21 : 591 – 596 . Google Scholar PubMed 4 Wernli KJ , Henrikson NB , Morrison CC , et al. . Screening for skin cancer in adults: Updated evidence report and systematic review for the US Preventive Services Task Force . JAMA. 2016 ; 316 4 : 436 – 447 . Google Scholar Crossref Search ADS PubMed 5 US Preventive Services Task Force Bibbins-Domingo K , Grossman DC , et al. . Screening for skin cancer: US Preventive Services Task Force recommendation statement . JAMA. 2016 ; 316 4 : 429 – 435 . Google Scholar Crossref Search ADS PubMed 6 Lebwohl MG. American Academy of Dermatology (AAD) statement on skin cancer screening (Statement from the President, AAD). https://www.aad.org/media/news-releases/aad-statement-on-skin-cancer-screening. Accessed September 19, 2017. 7 American Cancer Society . Skin cancer prevention and early detection. https://www.cancer.org/cancer/skin-cancer/prevention-and-early-detection/uv-protection.html. Accessed September 19, 2017. 8 Cancer Research UK. Melanoma - screening. http://www.cancerresearchuk.org/about-cancer/melanoma/getting-diagnosed/screening. Accessed September 19, 2017. 9 Cancer Council Australia . Screening for melanoma. http://www.cancer.org.au/about-cancer/types-of-cancer/skin-cancer/melanoma.html. Accessed September 19, 2017. 10 Australian Cancer Network Melanoma Guidelines Revision Working Party . Clinical Practice Guidelines for the Managment of Melanoma in Australia and New Zealand. Cancer Council Australia and Australian Cancer Network, Sydney and New Zealand Guidelines Group, Wellington, New Zealand; 2008 . 11 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 8 : 819 – 827 . Google Scholar Crossref Search ADS PubMed 12 Masri GD , Clark WH Jr , Guerry Dt , et al. . Screening and surveillance of patients at high risk for malignant melanoma result in detection of earlier disease . J Am Acad Dermatol. 1990 ; 22 ( 6 Pt 1 ): 1042 – 1048 . Google Scholar Crossref Search ADS PubMed 13 Watts CG , Cust AE , Menzies SW , Mann GJ , Morton RL. Cost-effectiveness of skin surveillance through a specialized clinic for patients at high risk of melanoma . J Clin Oncol. 2017 ; 35 1 : 63 – 71 . Google Scholar Crossref Search ADS PubMed 14 Usher-Smith JA , Emery J , Kassianos AP , Walter FM. Risk prediction models for melanoma: A systematic review . Cancer Epidemiol Biomarkers Prev. 2014 ; 23 8 : 1450 – 1463 . Google Scholar Crossref Search ADS PubMed 15 Vuong K , McGeechan K , Armstrong BK , Cust AE. Risk prediction models for incident primary cutaneous melanoma: A systematic review . JAMA Dermatol. 2014 ; 150 4 : 434 – 444 . Google Scholar Crossref Search ADS PubMed 16 Cho E , Rosner BA , Feskanich D , Colditz GA. Risk factors and individual probabilities of melanoma for whites . J Clin Oncol. 2005 ; 23 12 : 2669 – 2675 . Google Scholar Crossref Search ADS PubMed 17 Nielsen K , Masback A , Olsson H , Ingvar C. A prospective, population-based study of 40,000 women regarding host factors, UV exposure and sunbed use in relation to risk and anatomic site of cutaneous melanoma . Int J Cancer. 2012 ; 131 3 : 706 – 715 . Google Scholar Crossref Search ADS PubMed 18 Goldberg MS , Doucette JT , Lim HW , et al. . Risk factors for presumptive melanoma in skin cancer screening: American Academy of Dermatology National Melanoma/Skin Cancer Screening Program experience 2001-2005 . J Am Acad Dermatol. 2007 ; 57 1 : 60 – 66 . Google Scholar Crossref Search ADS PubMed 19 Guther S , Ramrath K , Dyall-Smith D , Landthaler M , Stolz W. Development of a targeted risk-group model for skin cancer screening based on more than 100,000 total skin examinations . J Eur Acad Dermatol Venereol. 2012 ; 26 1 : 86 – 94 . Google Scholar Crossref Search ADS PubMed 20 Olsen CM , Green AC , Neale RE , et al. . Cohort profile: The QSkin Sun and Health Study . Int J Epidemiol. 2012 ; 41 4 : 929 – 929i . Google Scholar Crossref Search ADS PubMed 21 Collins GS , Reitsma JB , Altman DG , Moons KG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): The TRIPOD statement . Br J Cancer. 2015 ; 112 2 : 251 – 259 . Google Scholar Crossref Search ADS PubMed 22 Morze CJ , Olsen CM , Perry SL , et al. . Good test-retest reproducibility for an instrument to capture self-reported melanoma risk factors . J Clin Epidemiol. 2012 ; 65 12 : 1329 – 1336 . Google Scholar Crossref Search ADS PubMed 23 Moons KG , Donders RA , Stijnen T , Harrell FE Jr. Using the outcome for imputation of missing predictor values was preferred . J Clin Epidemiol. 2006 ; 59 10 : 1092 – 1101 . Google Scholar Crossref Search ADS PubMed 24 Sterne JA , White IR , Carlin JB , et al. . Multiple imputation for missing data in epidemiological and clinical research: Potential and pitfalls . BMJ. 2009 ; 338 : b2393 . Google Scholar Crossref Search ADS PubMed 25 Fine JP , Gray RJA. A proportional hazards model for the subdistribution of a competing risk . J Am Stat Assoc. 1999 ; 94 446 : 496 – 509 . Google Scholar Crossref Search ADS 26 Uno H , Cai T , Pencina MJ , D'Agostino RB , Wei LJ. On the C-statistics for evaluating overall adequacy of risk prediction procedures with censored survival data . Stat Med . 2011 ; 30 10 : 1105 – 1117 . Google Scholar PubMed 27 Royston P. Tools for checking calibration of a Cox model in external validation: Prediction of population-averaged survival cures based on risk groups . Stata J . 2015 ; 15 1 : 275 – 291 . 28 Royston P , Altman DG. External validation of a Cox prognostic model: Principles and methods . BMC Med Res Methodol. 2013 ; 13 : 33 . Google Scholar Crossref Search ADS PubMed 29 Youden WJ. Index for rating diagnostic tests . Cancer. 1950 ; 3 1 : 32 – 35 . Google Scholar Crossref Search ADS PubMed 30 Vickers AJ , Cronin AM , Elkin EB , Gonen M. Extensions to decision curve analysis, a novel method for evaluating diagnostic tests, prediction models and molecular markers . BMC Med Inform Decis Mak. 2008 ; 8 : 53 . Google Scholar Crossref Search ADS PubMed 31 Vickers AJ , Elkin EB. Decision curve analysis: A novel method for evaluating prediction models . Med Decis Making 2006 ; 26 6 : 565 – 574 . Google Scholar Crossref Search ADS PubMed 32 Marcus PM , Pashayan N , Church TR , et al. . Population-based precision cancer screening: A symposium on evidence, epidemiology, and next steps. Cancer Epidemiol Biomarkers Prev. 2016 ; 25 11 : 1449 – 1455 . Google Scholar Crossref Search ADS 33 Richtig E , Santigli E , Fink-Puches R , Weger W , Hofmann-Wellenhof R. Assessing melanoma risk factors: How closely do patients and doctors agree? Public Health. 2008 ; 122 12 : 1433 – 1439 . Google Scholar Crossref Search ADS PubMed 34 Welch HG , Black WC. Overdiagnosis in cancer . J Natl Cancer Inst. 2010 ; 102 9 : 605 – 613 . Google Scholar Crossref Search ADS PubMed 35 Shain AH , Bastian BC. From melanocytes to melanomas . Nat Rev Cancer. 2016 ; 16 6 : 345 – 358 . Google Scholar Crossref Search ADS PubMed 36 Han J , Colditz GA , Hunter DJ. Risk factors for skin cancers: A nested case-control study within the Nurses' Health Study . Int J Epidemiol. 2006 ; 35 6 : 1514 – 1521 . Google Scholar Crossref Search ADS PubMed 37 Vuong K , Armstrong BK , Weiderpass E , et al. . Development and external validation of a melanoma risk prediction model based on self-assessed risk factors . JAMA Dermatol. 2016 ; 152 8 : 889 – 896 . Google Scholar Crossref Search ADS PubMed 38 MacKie RM , Freudenberger T , Aitchison TC. Personal risk-factor chart for cutaneous melanoma . Lancet. 1989 ; 2(8661 ): 487 – 490 . 39 Fears TR , Guerry Dt , Pfeiffer RM , et al. . Identifying individuals at high risk of melanoma: A practical predictor of absolute risk . J Clin Oncol. 2006 ; 24 22 : 3590 – 3596 . Google Scholar Crossref Search ADS PubMed 40 Quereux G , Moyse D , Lequeux Y , et al. . Development of an individual score for melanoma risk . Eur J Cancer Prev. 2011 ; 20 3 : 217 – 224 . Google Scholar Crossref Search ADS PubMed 41 Marrett LD , King WD , Walter SD , From L. Use of host factors to identify people at high risk for cutaneous malignant melanoma . CMAJ . 1992 ; 147 4 : 445 – 453 . Google Scholar PubMed 42 English DR , Armstrong BK. Identifying people at high risk of cutaneous malignant melanoma: Results from a case-control study in Western Australia . Br Med J (Clin Res Ed). 1988 ; 296 6632 : 1285 – 1288 . Google Scholar Crossref Search ADS PubMed 43 Garbe C , Buttner P , Weiss J , et al. . Risk factors for developing cutaneous melanoma and criteria for identifying persons at risk: Multicenter case-control study of the Central Malignant Melanoma Registry of the German Dermatological Society . J Invest Dermatol. 1994 ; 102 5 : 695 – 699 . Google Scholar Crossref Search ADS PubMed 44 Fortes C , Mastroeni S , Bakos L , et al. . Identifying individuals at high risk of melanoma: A simple tool . Eur J Cancer Prev. 2010 ; 19 5 : 393 – 400 . Google Scholar Crossref Search ADS PubMed 45 Olsen CM , Neale RE , Green AC , et al. . Independent validation of six melanoma risk prediction models . J Invest Dermatol. 2015 ; 135 5 : 1377 – 1384 . Google Scholar Crossref Search ADS PubMed 46 Mar V , Wolfe R , Kelly JW. Predicting melanoma risk for the Australian population . Australas J Dermatol. 2011 ; 52 2 : 109 – 116 . Google Scholar Crossref Search ADS PubMed 47 Centers for Disease Control and Prevention . Cancer survivors—United States, 2007 . MMWR Morb Mortal Wkly Rep . 2011 ; 60 9 : 269 – 272 . PubMed 48 Breitbart EW , Waldmann A , Nolte S , et al. . Systematic skin cancer screening in Northern Germany . J Am Acad Dermatol. 2012 ; 66 2 : 201 – 211 . Google Scholar Crossref Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please email: [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)
Journal
JNCI: Journal of the National Cancer Institute – Oxford University Press
Published: Oct 1, 2018