Table A1.

Basic characteristics of studies reporting risk prediction models for melanoma. Studies are ordered according to year of publication. (N = 40 studies).

Author (Year)	Study Design	Size of Study Sample	Year(s) of Data Collection	Country	Analytic Model	Risk Measure	Variables in Final Model (1)
English and Armstrong (1988) [20]	Case-control study	511 cases, 511 controls	1980–1981	Australia	Logistic regression	Risk score	Number of raised nevi on arms, age on arrival in Australia, mean time spent outdoors in summer aged 10–24, family history, personal history of non-melanoma skin cancer
Garbe et al. (1989) [25]	Case-control study	200 cases, 200 cases	1987	Germany	Logistic regression	Relative risks	Number of melanocytic common nevi, number of atypical nevi, actinic lentigines, occupational sun exposure, skin type
MacKie et al. (1989) [30]	Case-control study	280 cases, 280 controls	1987	Scotland	Logistic regression	Relative risk (risk groups)	Benign nevi >2 mm, freckling, atypical nevi >5 mm, episodes of severe sunburn
Augustsson et al. (1991) [40]	Case-control study	121 cases, 379 controls	1986–1988	Sweden	Logistic regression	Relative risks	Skin type, hair color, eye color, total body nevus ≥2 mm count, number of dysplastic nevi
Marret et al. (1992) [32]	Case-control study	583 cases, 608 controls	1984–1986	Canada	Logistic regression	Relative risk	Hair color, skin reaction to repeated sun exposure, freckle density, nevi density
Garbe et al. (1994) [24]	Case-control study	513 cases, 498 controls	1990–1991	Germany	Logistic regression	Relative risk estimates (risk groups)	Number of melanocytic common nevi, actinic lentigines, atypical nevi, skin type
Barbini et al. (1998) [17]	Case-control study	150 cases, 546 controls	1992–1995	Italy	Linear discriminant analysis	Risk score (negative score → low risk)	Colorimetric variables, Fitzpatrick
Landi et al. (2001) [29]	Case-control study	183 cases, 179 controls	1994–1999	Italy	Logistic regression	Odds ratios	Dysplastic nevi, skin color, tanning ability, eye color
Harbauer e al. (2003) [28]	Case-control study	202 cases, 202 controls	2001	Austria	Logistic regression	Odds ratios	Skin type, UV damage, number of nevi
Dwyer et al. (2004) [19]	Case-control study	244 cases, 483 controls	1998–1999	Australia	Logistic regression	Odds ratios	MC1R genotype, melanin density
Fargnoli et al. (2004) [21]	Case-control study	100 cases, 200 controls	2000–2001	Italy	Logistic regression	Relative risk estimates (high risk: ≥ 2 (3) risk factors in model 3 (1))	Model 1: Hair color, eye color, skin type Model 2: Hair color, eye color, skin type, occupational sun exposure, atypical nevi Model 3: Skin type, sun exposure, nevi, atypical nevi Model 4: Skin type, occupational sun exposure, nevi, atypical nevi
Cho et al. (2005) [18]	Cohort study	535 cases, total 178,155	1976, 1986, 1989	United States	Gail method	Risk score and 10-years-absolute risk	Sex, age, family history, sunburns, number of nevi on arms, hair color
Whiteman and Green (2005) [35]	Published case-control studies	NA	NA	Several countries	Not reported	10-years absolute risk	Age, place of residence, number of melanocytic nevi, skin color, MC1R genotype
Fears et al. (2006) [22]	Case-control study	718 cases, 945 controls	1991–1992	United States	Gail method	5-year-absolute risk (high risk: p ≥ 0.15%)	Sex, skin color, sunburns, number of moles >5 mm (only men), number of moles ≥2 mm, freckling, severe sun damage (only men), tanning ability (only women)
Goldberg et al. (2007) [26]	Cohort study	3329 cases, total 362,804	2001–2005	United States	Logistic regression	Risk score (high risk: score 4–5)	Sex, regular dermatologist, history of previous melanoma, mole changing, age
Fortes et al. (2010) [23]	Case-control study	304 cases, 305 controls	2001–2003	Italy	Risk score was calculated using effect estimates from meta-analysis	Individual risk score (high risk: risk score ≥ 3)	Freckles in childhood, skin color, number of common nevi, hair color, sunburns in childhood
Mar et al. (2011) [31]	Published meta-analysis and registry data	NA	NA	Australia	Gail method	5-year-absolute risk	Common nevi, atypical nevi, freckles, hair color, family history, non-melanoma skin cancer, personal melanoma history
Nielsen et al. (2011) [33]	Cohort study	215 cases, total 29,520	1990–1992 (Followup: –2007)	Sweden	Cox regression	Hazard ratios for each risk factor	Family history, number of nevi, hair color, sunbathing vacations, sunbed use
Quéreux et al. (2011) [37]	Case-control study	171 cases, 1390 controls	2007	France	Gail method, logistic regression and combinatorial analysis	Risk Score	Gail method: Sunburn in childhood, family history, number of common nevi on arms, density of freckles, skin type, recalled total sun exposure Logistic regression and combinatorial analysis: Sex, age, skin type, presence of freckles, number of nevi on arms, severe blistering sunburn in childhood, life in a country at low latitude, family history
Williams et al. (2011) [36]	Case-control study	386 cases, 727 controls	1997	United States	Logistic regression	Risk score (high risk: top 15%)	Sex, age, number of severe sunburns, hair color, freckles, number of raised moles, non-melanoma skin cancer history
Guther et al. (2012) [27]	Cohort study	250 cases, total 108,281	2005–2006	Germany	Logistic regression	Risk score (high risk: >0.0034)	Age, hair color, personal history of melanoma, suspicious melanocytic lesions
Smith et al. (2012) [39]	Case-control study	923 cases, 813 controls	Not reported	United States	Not reported	Not reported	Model A: Sex, age, hair color, eye color, mole count, freckling, family melanoma history Model B: Model A + outdoor UV, indoor UV, MC1R
Bakos et al. (2013) [16]	Case-control study	53 cases, 66 controls	2005–2008	Brazil	Risk score calculated using effect estimates from meta-analysis	Risk score (high risk: >3)	Presence of freckles in childhood, skin color, eye color, hair color, sunburn episodes throughout life
Cust et al. (2013) [45]	Case-control study	413 cases, 263 controls	2000–2002	Australia	Logistic regression	Odds ratios	Base model: Age, sex, city Self-reported model: MC1R genotype, nevi, pigmentation score (2), sun and sunbed exposure (3), family history, non-melanoma skin cancer Physician-measured model: Nevi, MC1R genotype, non-melanoma skin cancer, solar lentigines, family history, pigmentation score (4)
Fang et al. (2013) [59]	Multiple case-control studies	2298 cases, 6652 controls	NA	United States	Logistic regression	Odds ratios	Model 1: Single SNP Model 2: PRS (5) Model 3: Sex + age Model 4: Sex + age + pigmentation Model 5: Sex + age + pigmentation + PRS
Stefanaki et al. (2013) [34]	Case-control study	284 cases, 284 controls	NA	Greece	Logistic regression	Odds ratios	Model A: Eye color, hair color, skin color, skin type, tanning, sunburns Model B: All predictors in model A + 3 strongest SNPs Model C: All predictors in model A + all SNPs
Nikolic et al. (2014) [57]	Case-control study	341 cases, 356 controls	2001–2012	Serbia	Logistic regression + decision tree	Absolute risk	Level of education, intermitted exposure, use of sunbeds, HCT, solar damage of skin, Fitzpatrick, hair color, eye color, number of common nevi, number of dysplastic nevi, congenital nevi
Penn et al. (2014) [56]	Case-control study	875 cases, 765 controls	2004–2007	United States	Logistic regression	Odds ratios	Base model: Age, sex, hair color, eye color, skin color, freckles, mole phenotype Full model: Base model + sun burns, indoors tanning, MC1R genotype
Sneyd et al. (2014) [61]	Case-control study	368 cases, 270 controls	1992–1994	New Zealand	Logistic regression + Gail method	5-year-absolute risk	Women: Skin color, 1st degree relative with large or unusual moles, number of moles, personal history of nonmelanoma skin cancer Men: Number of moles, personal history, age at diagnosis, occupation, birthplace
Davies et al. (2015) [51]	Multiple case-control studies	NA	NA	Several countries	Logistic regression	Risk score (with risk categories)	Hair color, skin type, freckles, family history, nevi distribution, number of large nevi, sunburn
Kypreou et al. (2016) [54]	Case-control study	800 cases, 800 controls	2000–2014	Greece	Logistic regression	Odds ratios	Genetic risk score (6), age, sex, eye color, hair color, skin color, phototype, tanning ability
Vuong et al. (2016) [49]	Case-control study	629 cases, 535 controls	2000–2002	Australia	Gail method	20-year-absolute risk	Hair color, nevi density, family history, personal history of non-melanoma skin cancer, sunbed use
Cho et al. (2018) [58]	Cohort study	422/289 cases (lifetime/incident melanoma); total 19,102	NA–2015	United States	Logistic regression + Cox regression	Odds ratios and hazard ratios	Genetic risk score (7)
Cust et al. (2018) [62]	Case-control study	629 cases, 535 controls	2000–2002	Australia	Logistic regression	Odds ratios	Base model: Family history, hair color, nevi, personal history of non-melanoma skin cancer, sunburns in childhood, sunbed sessions, freckles, eye color, sun exposure Full model: Base model + PRS (8)
Gu et al. (2018) [60]	Case-control study	15,976 cases, 25,504 controls	NA	Several countries	Logistic regression	10- and 20-year-absolute risk	Model 1: Age, sex, country Model 2: (1) + eye color, hair color, skin type, common nevi Model 3: (1) + PRS (9) Model 4: (2) + PRS
Hübner et al. (2018) [53]	Cohort study based on data from SCREEN project	585 cases, total 354,635	2003–2004	Germany	Logistic regression	Odds ratios	Sex, age, personal melanoma history, family history, multiple common nevi, atypical nevi, congenital nevi
Olsen et al. (2018) [52]	Cohort study	655 cases, total 41,954	2011–2014	Australia	Cox regression	Hazard ratios	Model 1 (invasive melanoma): Age, sex, tanning ability, moles at age 21, hair color, number of previous skin lesions treated destructively, sunscreen use Model 2 (all melanoma): (1) + ethnicity, private health insurance, family history, past history of excisions for skin cancer, skin checks in past 3 years
Richter and Khoshgoftaar (2018) [55]	Cohort study based on EHR data	17,246 cases, total 9,531,408	2011–2017	United States	Logistic regression, decision tree + random forest	Risk score	Not reported
Tagliabue et al. (2018) [46]	Case-control study	3830 cases, 2619 controls	NA	Several countries	Logistic regression	Odds ratios	Base model: Age, sex, sunburns, number of common nevi, RH-phenotype Base model + MC1R genotype
Vuong et al. (2019) [50]	Case-control study	461 cases, 329 controls	2000–2002	Australia	Logistic regression	Relative risks	Number of nevi, solar lentigines, hair color, personal history of keratinocytic cancer

Abbreviations: UV = ultraviolet, MC1R = melanocortin 1 receptor, SNP = Single Nucleotide Polymorphism, PRS = Polygenic Risk Score, HCT = Hormonal Contraceptive Therapy, SCREEN = Skin Cancer Research to provide Evidence for Effectiveness of Screening in Northern Germany, EHR = Electronic Health Records, RH-Phenotype = Red Hair-Phenotype. ⁽¹⁾ For studies with multiple models: Models included in analysis are highlighted in bold. ⁽²⁾ Score calculated from the variables: tanning ability, propensity to sunburn, skin color, eye color, hair color and freckles. ⁽³⁾ Term for the individual variables total childhood sun exposure, blistering sunburns and lifetime sunbed sessions. ⁽⁴⁾ Score was calculated from the following variables: hair color, eye color, skin reflectance, tanning ability, propensity to sunburn and freckles. ⁽⁵⁾ Comprised of 11 SNPs that demonstrated association with melanoma risk in previous studies. ⁽⁶⁾ Based on SNPs that showed genome-wide significant association with melanoma in previous studies. ⁽⁷⁾ Calculated using 21 genome-wide association study—significant SNPs. ⁽⁸⁾ Derived from 21 gene regions associated with melanoma. ⁽⁹⁾ Combines 204 common SNPs.