Outcomes for smokers who develop melanoma: a systematic review and meta-analysis

Summary

Background

There is compelling evidence that the incidence of melanoma in cigarette smokers is substantially lower than in non-smokers. However, the risks of both recurrence and death appear to be higher in smokers if melanoma does develop. The magnitude of these increased risks is poorly documented. This systematic review aimed to analyse melanoma survival outcomes among smokers compared to never-smokers using published studies, and report the magnitude of any survival differences.

Methods

Searches of Medline, Embase and Cochrane CENTRAL (to 11/03/2024) using terms for melanoma and smoking were conducted. Included studies were those reporting outcomes including disease severity at presentation, risk of death or adverse effects from treatment in smokers and never-smokers with melanoma. No study design or language restrictions were imposed. Risk of bias was assessed using the Newcastle–Ottawa tool. The review protocol was registered with PROSPERO (ID CRD42024518505).

Findings

Forty-six studies involving 164,166 melanoma patients, of whom 70,766 had smoked cigarettes, were identified. The pooled individual Hazard Ratios (HR) from multivariable analyses, showed that death from melanoma was 33% higher in current compared to never-smokers (HR 1.33, 95% CI 1.14–1.55, p = 0.0002, 13,971+ patients) with limited heterogeneity (I² 11%). However, former-smokers and never-smokers had very similar rates of death from melanoma (HR 1.04, 95% CI 0.94–1.14, p = 0.52, 16,307+ patients) with moderate heterogeneity (I² 63%). From univariable analyses, current-smokers had a higher risk of sentinel node-positivity compared to never-smokers (HR 1.35 95% CI 1.13, 1.62, p = 0.001, 5163 patients). Ever-smokers had a greater risk of complications from sentinel node biopsy (Odds Ratio (OR) 2.0 95% CI 1.41–2.85, p = 0.0001, 3745 patients) and lymph node dissection (OR 1.7, 95% CI 1.23–2.20, p = 0.0007, 4596 patients) than never-smokers based on risks from multivariable analyses.

Interpretation

Current smokers are more likely to die from their melanoma than never-smokers, while former-smokers appear to have similar risks to never-smokers. Smokers have higher risks of sentinel node-positivity and of complications from node surgery. Study limitations included reliance on self-reporting of smoking status. In only seven studies did patients receive modern systemic therapies, limiting the ability to assess their relative efficacy in smokers and non-smokers.

Funding

None.

Research in context.

Evidence before this study

There is consistent evidence that current-smokers have a significantly-reduced risk of developing melanoma compared to never-smokers but the impact of smoking on those who develop melanoma has been poorly documented. A search of Medline, Embase and the Cochrane database of systematic reviews on 29 Nov 2022 using the medical subject heading melanoma with terms for smoking, without language restrictions, identified 154 relevant primary studies. No systematic review of smoking and melanoma outcomes was found. Many single centre studies, often small, reported a worse prognosis, more advanced disease and poorer treatment outcomes in smokers compared to never-smokers. However, there were no studies providing precise estimates of the effect of smoking on melanoma outcomes from analyses that adjusted for other prognostic features; this prompted the current meta-analysis.

Added value of this study

Given the paradoxical pattern of a reduced risk of developing melanoma in smokers, it is important to recognize the negative effects of smoking on melanoma outcomes. By more precisely quantifying the increased risk of death from melanoma, presentation with worse prognostic features and complications from node surgery in smokers after controlling for other risk factors, clinicians can have a more focused discussion with melanoma patients who are smokers, explaining how smoking cessation can impact their clinical course.

Implications of all the available evidence

The study highlights the need to better understand the interplay between smoking tobacco and melanoma progression. The study also highlights the considerable need for further reporting of smoking and use of contemporary systemic therapies for melanoma. Our findings underline the importance of collecting data on smoking status (including e-cigarette and marijuana use) at multiple time points, as well as quantity smoked and dates of commencement and cessation, in both routine clinical care and in clinical trials.

Introduction

Smoking is the single largest cause of cancer in the world, with the International Agency for Research on Cancer stating that smoking is a cause of 18 cancer types.¹ Remarkably, however, there is compelling evidence that the incidence of primary melanoma amongst smokers is substantially lower than in non-smokers.2, 3, 4, 5 Nevertheless, as with other cancers where higher disease recurrence rates and poorer survival outcomes are associated with smoking, the prognosis of smokers in whom melanoma does develop has been reported to be worse than the prognosis in non-smokers, and smokers may present with more advanced disease than non-smokers.6, 7, 8

In 1979 a group of Australian researchers hypothesised that because smoking leads to aberrant immunological function, it may affect the course of disease for melanoma patients.⁹ In 1908 patients with melanoma, significantly more smokers than non-smokers were >50 years old (p < 0.01 men, p < 0.001 women) and presented with regional or distant metastases (men: 22.4 vs. 17.9%; women: 12.2 vs. 9.0%). Five-year disease-free survival (DFS) was significantly worse for male smokers compared to non-smokers (53.5 vs. 63.0%, p < 0.02).⁹ More contemporary data support a strong association between current smoking and regional melanoma metastases.¹⁰ Smoking may also influence the outcome of surgery, with higher rates of wound infection and breakdown reported, and a greater number of post-operative systemic complications.

The lack of a rigorous analysis of all available evidence in the literature prompted us to undertake the current review. We sought to assess all available data on the risk of death (from melanoma or any cause), disease features at diagnosis and treatment outcomes in smokers who developed melanoma.

Methods

The review protocol was registered with PROSPERO (ID CRD42024518505), and PRISMA reporting guidelines were followed (Checklist in Supplementary Data File).¹¹

Search strategy

Medline and Embase were searched up to 11th March 2024 by combining the medical subject heading (MeSH) melanoma with MeSH and text words for smoking, cigarettes, tobacco, nicotine, vaping and cannabis smoke (Supplementary Table S1).

Selection criteria

No study design or language restrictions were used. Titles and articles were assessed independently by two authors (GJW and EBF) with discrepancies resolved by trial data extraction. Articles were included if they reported raw frequencies or risk estimates, with p-values or 95% confidence intervals from either univariable analyses (UVA) or multivariable analyses, for the outcomes of interest in smokers and never-smokers. Studies with mixed cancer types were included if data for melanoma patients were reported separately. Single case reports were excluded. One author (GJW) performed data extraction and included fields for study identifiers, data source, study design, methods and outcomes (deaths, adverse events, primary melanoma features at presentation) and aggregate risks. For studies with ambiguous data, an email was sent to the corresponding author for clarification.

Data analysis

Cochrane Review Manager¹² was used to produce forest plots for analyses where three or more studies reported the same outcome. When studies reported patient numbers for an outcome within smoking and non-smoking groups, a risk was generated using the dichotomous data-type option, and pooled with a random effects model. Aggregate risk data (hazard ratios or odds ratios) were log transformed and standard errors estimated from 95% CIs or p-values¹³ and analysed using the generic inverse variance option, with fixed effects for risks generated from multivariable analyses and random effects models for univariable analyses. Mean differences were calculated for Breslow thickness. Heterogeneity assessments used the I² measure,¹⁴ and summary risks were reported only when the I² value was ≤60%, indicating that the level of heterogeneity was low to moderate. All study data were included with exclusions only when duplication of patient data was identified. Where possible, subgroup analyses were performed for sex (males and females), study design (cohort, case–control or cross sectional), by country of patient recruitment (USA, UK, Other) and for risk of bias (high; ≥3 fields high risk of bias and low; ≤2 fields at high risk of bias). Funnel plots of the main analyses were generated to assess publication bias. Risk of bias was assessed by one author using the Newcastle–Ottawa scale for non-randomised studies, modified for study design.¹⁵

Role of funding source

There was no funding source for the study, author salaries were provided by their institutions. Institutions had no role in the study processes.

Results

The search identified 46 studies from 14 countries (Fig. 1) that reported outcomes for melanoma patients who were smokers or never-smokers (Table 1). Two studies used the MSLT I and II international randomised trial cohorts.^8,58 Studies were published between 1981 and 2024 and all were written in English. Seven were published as conference abstracts only.

Fig. 1.

Flow chart of search results with the number of included studies, excluded studies and reasons for study exclusion.

Table 1.

Details of the included studies.

Study ID	Publication type	Population	Country	Recruitment period	Outcomes	Factors in multivariable analyses	Melanoma (N pts)	Smokers with melanoma (N pts)
Shaw 1979/819,16	Full	Melanoma	Australia	1950–1979	Death (Melanoma) + Severity (stage, local vs. visceral recurrence site)	Age, sex, stage, site, Breslow, surgical treatment (WLE alone vs. WLE + SNB)	2583	1301
Rigel 19816	Full	Melanoma	USA	1972–1980	Severity (Breslow, metastases)	Age, sex, smoking	178	116
Keeney 198217	Full	Melanoma of choroid	USA	1958–1978	Death (Melanoma) + Severity	Sex	94	44
Koh 198418	Full	Melanoma, stage I	USA	1972–1977	Death (Melanoma) + Disease features (Breslow)	Age, sex, site, Breslow, Clark, Subtype, invasion level, ulceration, lymphocyte response, regression, mitotic rate, dermal nevus, microsatellites, lymph node status	196	44
Van Durme 200019	Full	Melanoma	USA	1994	Disease features (stage at diagnosis)	Age, sex, marital status, smoking, education level, income, urban/rural residence	1884	252
Batty 200820	Full	Mixed	UK	1967–1970	Death (Melanoma), Incidence	Not done	35	24
DeLancey 201121	Full	Melanoma	USA	1982	Death (Melanoma), Incidence	Not done	1238	661
Hinz 201222	Full	Melanoma (SLNB –ve)	Germany	2010–2011	Disease features (Breslow, node measurements)	Age, BMI, Breslow, Quantity of LNs, Volume LNs, Largest diameter of LN, Max hypoechoic margin	79	44
Kanth 201223	Abstr	Melanoma (+aCTLA4)	USA	2005–2012	Adverse/complications	Sex, IL-2 treatment, laboratory results	30	14
Rudnicka 201224	Abstr	Melanoma	Poland	NR	Incidence + Disease features (Breslow)	Alcohol, Smoking	54	19
Smyth 201225	Full	Melanoma, with lung lesion biopsied	USA	1996–2009	Disease features (metastases)	Age, sex, stage, smoking, prior non-melanoma cancer, cutaneous vs. not, site	229	117
Ul Mulk 201226	Full	Melanoma (+LND)	Denmark	2008–2011	Adverse/complications	BMI, co-morbidity, seroma puncture, smoking	96	23
Warren 201327	Full	Mixed	USA	1982–1998	Death (All cause, Melanoma)	Age, sex, race, stage, Diagnosis date, BMI, total pack years	524	273
Gould 201428	Full	Melanoma	USA	1987–1989	Death (Melanoma)	Age, sex, Breslow thickness, ulceration, microsatellites, lesion site, fruit consumption, red meat consumption, fish consumption, alcohol, hair colour, Mitotic index, Clark level, Regression, education, marital status, eye colour	249	155
Stuiver 201429	Full	Melanoma (+LND)	Netherlands	2003–2013	Adverse/complications	Age, BMI, smoking, diabetes, comorbidities- non-diabetes, palpable disease, days of bedrest	138	58
Moore 201530	Abstr	Melanoma	USA	1981–2009	Death (All cause)	Sex, race, ethnicity, SES, urban/rural setting, smokers	54,296	33,121
Mercante 201631	Abstr	Melanoma (+LND)	USA	2005–2015	Adverse/complications	Sex, BMI, inguinal/axillary, smoking	554	NR
Ascha 201732	Full	Melanoma (+SNB, LND)	USA	2005–2014	Adverse/complications	Age, sex, BMI, Hypertension, diabetes, other co-morbidities (heart, lungs), laboratory values (not further details)	3006	NR
Cauci 201733	Full	Melanoma	Italy	NR	Severity (metastases) Incidence	Age, gender, phototype 1 + 2, total body nevi> 50, lifelong sunburn >10	120	59
Jones 20178	Full	Melanoma ≥1.2 mm	Multi-national (MSLT I trial, part of MSLT-II data)	1994–2014	Disease features (Breslow, Ulceration, SLNB status)	Age, sex, site (Breslow for SNB + outcome)	4231	1819
Postlewait 201734	Abstr	Melanoma (+LND)	USA	2003–2015	Adverse/complications	Age, BMI, diabetes, smoking, lesion site, therapeutic LND	239	81
Espinosa-Pereiro 201935	Full	Melanoma (+SNB)	Spain	2011–2017	Adverse/complications	Many, not all listed	124	15
Persa 201936	Full	Melanoma (+SNB)	Germany	2011–2016	Adverse/complications	Age, sex, diabetes, hypertension, smoking, BMI, Breslow, Lymphovascular invasion, Histologic type, Ulceration, Axillary/inguinal, N nodes excised, size of metastasis in node, extracapsular extension	615	114
Pozniak 201937	Full	Melanoma	UK	2000–2012	Death (Melanoma) + Disease features	Age, sex, site, smoking, vitamin D, deprivation score	703	346
Tejera-Vaquerizo 201938	Full	Melanoma (+SNB)	Spain	2000–2016	Death (All cause, Melanoma) + Disease features (metastases, disease free survival)	Age, sex, site, Breslow, year, histology type, ulceration, mitotic index, regression, lymphocyte infiltrate, microsatellites, vascular invasion, Clark level, SNB status	752	288
Wang 201939	Full	Mixed	USA	1997–2013	Death (All cause, Melanoma)	Age, sex, race, education, marital status, BMI, physical activity, alcohol intake, income, self-rated health, hypertension, diabetes, CHD, stroke, duration of cancer	2208	1172
Gibson 202040	Full	Melanoma	UK	2000–2015	Incidence + Death (All cause, Melanoma)	Age, sex, SES	7124	4525
Hardie 202041	Full	Melanoma	UK	2000–2012	Death (All cause, Melanoma) + Disease features (Breslow thickness)	Age, sex, lesion site, Vitamin D, deprivation score, housing problems, negative life events, depression/anxiety,	2183	1058
Maas 202042	Full	Melanoma	USA	2001–2009	Death (All cause)	Age, sex, marital status, health insurance, race, ethnicity, geographical area, tobacco, histology, stage at Diagnosis, site	36,578	13,973
Mahamat Saleh 202043	Full	Melanoma, stages I or II	Australia	2010–2014	Disease features (Breslow)	Age, sex, phototype, education, BMI, education, smoking, sun protection measures, frequency of skin checks, family history of melanoma, dietary patterns	634	47
Sars 202044	Full	Melanoma (+LND)	Sweden	2005–2014	Adverse events/complications	Age, sex, smoking, diabetes, BMI, Inguinal/axillary site	144	46
Swami 202045	Full	Melanoma (+aPD-1)	USA	2012–2017	Death (All cause) + Adverse events	Age, sex, BMI, metastases, labs, RT, other drugs	169	99
Harrell Shreckengost 202146	Full	Melanoma, stages 0-III	USA	2010–2017	Death (All cause, Melanoma) + Disease features (Breslow)	Age, sex, race, alcohol, tobacco, stage, Breslow, ulceration, lymph node status, site, income estimates, BRAF status, BMI	1756	734
Karlsson 202147	Full	Mixed	Finland	2009–2018	Death (Melanoma), Incidence	For death; Age, sex, ECOG, BMI, comorbidities	955	355
Scanlon 202148	Full	Melanoma (+SNB)	Ireland	1994–2014	Death (All cause)	Age, sex, smoking, marital status, anatomical location, stage, SNB results, region of residence	13,302	NR
Stang 202149	Full	Mixed	USA	2008–2013	Death (All cause)	Age, sex, stage, ethnicity	9576	2533
Hata 202250	Full	Mixed (+IMT)	Japan	2014–2020	Adverse events	Age, gender, smoking, BMI, N met sites, met location, line of therapy, lab results, cancer types, ECOG	38	19
Qin 202251	Full	Mixed (+IMT)	USA	2011–2017	PFS, Overall response (e.g. CR), Disease control (CR, PR or stable)	Age, sex, ECOG, BMI, Family history, Smoking, GFR, locally adv disease, ≥3 met sites, bone metastases, cancer type	73	38
Swigert 202252	Full	BCC, SCC, Melanoma	USA	2009–2019	Disease features (Breslow)	Age, race, smoking, alcohol, drug use, HIV, hepatitis C, mental illness, immunosuppression	15	15
Tu 202253	Full	Mixed	USA	1998–2014	Death (All cause)	Not done	7369	3224
Wang 202254	Full	Melanoma ( ± IMT)	Australia	2015–2020	Adverse events	Age, sex, history of Major Atherosclerotic and Cardiovascular Events (MACE), smoking, disease at baseline, BRAF/MEK use	646	297
Chua 202355	Full	Mixed (+ Inguinal LND)	USA	2005–2018	Adverse events/complications	Not done	892	158
El Saadany 202356	Abstr	Mixed (+IMT)	Switzerland	NR	Death (All cause)	NR	88	34
Mattila 20237	Full	Melanoma	Finland	2005–2019	Death (All cause, melanoma), Recurrence free survival	Age (< or >70), sex, comorbidity score, stage	1359	518
Pan 202357	Abstr	Melanoma (+IMT)	USA	2010–2021	Adverse events	NR	672	303
Jackson 202458	Full	Melanoma	Multinational (MSLT I and II)	1994–2014	Death (Melanoma) + Disease features (SLNB)	Age, sex, Breslow, Ulceration, Primary site, SLNB status	6279	2771

Full; full text article, Abstr; Conference abstract, WLE; wide local excision, SNB; sentinel lymph node biopsy, BMI; body mass index, LN; lymph nodes, aCTLA4; anti-cytotoxic T-lymphocyte-associated antigen 4, IL-2; Interleukin 2, IMT; immunotherapy treatment, SES; socio economic status, MSLT I; multicenter selective lymphadenectomy trial, phase I and II, LND; lymph node dissection, RT; radiation therapy, aPD-1; anti programmed cell death protein, BCC; basal cell carcinoma, SCC; squamous cell carcinoma, HIV; human immunodeficiency virus, UV; ultravoiolet, NR; not reported, ECOG; Eastern Cooperative Oncology group, GFR; glomerular filtration rate, CHD; chronic heart disease.

Smoking and death from melanoma

Ten studies that included 3958 current-smokers and 10,013 never-smokers with melanoma reported a risk of death from melanoma using multivariable analyses. Pooled analyses showed a significantly-increased risk of death from melanoma in current-smokers (HR 1.33, 95% CI 1.14–1.55, p = 0.002) with limited heterogeneity (I² 11%) (Fig. 2A). Only two studies^21,27 reported this risk in men and women separately and results were variable, with small sample sizes (Table 2). Eight studies, involving 6734 former-smokers and 9573 never-smokers with melanoma, reported a risk of death from melanoma using multivariable analyses. Analyses showed no difference in risk between the two groups (HR; 1.04, 95% CI 0.94–1.14, p = 0.44) but with moderate heterogeneity (I² 63%) (Fig. 2B). A single study²¹ reported risk separately in men and women, with confidence intervals crossing 1.0 for both sexes, suggesting no difference in risk. Only two studies^7,38 compared risk of melanoma death in current-smokers compared to former-smokers, and one study²⁷ separated the sexes, with all three studies showing no difference in risk but with small sample sizes and considerable imprecision. Subgroup analyses by study design, country of recruitment and risk of bias showed no significant differences in risk estimates between any of the subgroups (Supplementary Fig. S1A–F).

Fig. 2.

Risk of melanoma death in smokers vs. non-smokers, using multivariable analyses.

Table 2.

Findings from studies comparing risk of death, complications and adverse events from treatment in smokers and non-smokers with melanoma using multivariable analyses (except where specified).

Comparison group	Studies, n	Number of patients per group, n	Pooled estimate (95% CI)c Individual study data	I2c (P value)
Death from melanoma			Hazard ratio
Current vs. Never	97,20,28,37, 38, 39, 40, 41,58	3958 vs. 10,013a	1.33 (1.14, 1.55)	11% (0.0002)
Males	221,27	71 vs. 385	0.77, 1.21	–
Females	221,27	58 vs. 443	1.04, 1.36	–
Former vs. Never	87,20,38, 39, 40, 41,47,58	6937 vs. 10,173a	1.03 (0.94, 1.13)	57% (0.52)
Males	121	371 vs. 263	0.86	–
Females	121	242 vs. 314	0.83	–
Current vs. Former	27,38	331 vs. 475	0.99, 1.03	–
Males	127	39 vs. 115	0.99	–
Females	127	42 vs. 77	1.30	–
All cause deaths			Hazard ratio
Current vs. Never	738, 39, 40, 41, 42,49,59	7,835b vs. 52,355	1.32 (1.26, 1.38)	60% (<0.00001)
Males	227,49	435 vs. 1853	1.44, 1.20	–
Females	227,49	343 vs. 1750	1.29, 1.64	–
Current vs. Never (UVA)	37,42,53	5934 vs. 4986	1.38 (1.28, 1.49)	2% (<0.00001)
Former vs. Never	620,38, 39, 40, 41,47	14,967 vs. 30,494	1.11 (1.04, 1.18)	0% (0.001)
Males	149	1161 vs. 1731	1.19	–
Females	149	675 vs. 1621	1.15	–
Former vs. Never (UVA)	37,42,53	11,781 vs. 27,591	1.08, 1.31, 1.36	82%
Current vs. Former	138	114 vs. 174	1.12	–
Males	227,49	435 vs. 1276	1.22, 1.15	–
Females	227,49	343 vs. 752	1.37, 1.47	–
Ever vs. Never	151	38 vs. 36	1.1	–
Complications after nodal surgery			Odds ratiof
SNB, Ever vs. Never	332,35,36	129 vs. 610e	2.01 (1.41, 2.85)	10% (0.0001)
LND, Ever vs. Never (All) (Unique data only)	431,32,44,55 331,44,55	204 vs. 824d 204 vs. 824d	1.59 (1.22, 2.06) 1.65 (1.23, 2.20)	0% (0.0005) 0% (0.0007)
Rate of SNB positivity			Hazard ratio
Current vs. Never (unique data onlyg)	38,38,58 (3 cohorts)	1191 vs. 3972	1.35 (1.13, 1.62)	0% (0.001)
Former vs. Never (unique data onlyg)	38,38,58 (3 cohorts)	1868 vs. 3972	0.99 (0.82, 1.19)	41% (0.90)

UVA: univariable analysis, SNB: sentinel node biopsy, LND: lymph node dissection.

^aOne study (Batty 2008²⁰) did not report the number of melanoma cases.

^bOne study (Moore 2015⁵⁹) did not report numbers of current smokers.

^cP value reported only when I² ≤ 60%.

^dTwo studies (Ascha 2017³² and Mercante 2016³¹) did not report the number of patients in each group.

^eOne Study (Ascha 2017³²) did not report the number of patients in each smoker category.

^fOdds ratios could not be converted to risk ratios because P0 could not be estimated for all studies.

^gTwo studies used the MSLT-I and II data, so only one study, Jackson 2024⁵⁸ was analysed and that study reported MSLT-I results separate to MSLT-II results, thus 3 cohorts of patients were included.

Smoking and death from all causes

Seven studies reported the risk of all-cause death in 7835 current-smokers and 52,355 never-smokers with melanoma using multivariable analyses. Pooled analyses showed a significantly-increased risk of all-cause death in melanoma patients who were current-smokers relative to never-smokers (HR 1.32, 95% CI 1.26–1.38, p < 0.00001) but with some heterogeneity (I² 60%) (Table 2, Supplementary Figure S1A). Risks pooled from UVA resulted in a similar estimate (HR 1.38, 95% CI 1.28–1.49) (Supplementary Figure S2E). Two studies^27,49 reported all-cause risk of death from multivariable analyses in males and females separately, with both showing a non-significant increase in risk in current-smokers.

Six studies with 14,967 former-smokers and 30,494 never-smokers with melanoma reported a risk of all-cause death using multivariable analyses. Analysis showed a significantly-increased risk of all-cause death in former-smokers relative to never-smokers (HR 1.11, 95% CI 1.04–1.18, p = 0.001) with no heterogeneity (I² 0%) (Supplementary Figure S2F). A single study⁴⁹ reported all-cause risk of death from multivariable analyses in males and females separately, with risk estimates from both sexes suggesting a non-significant, slightly increased risk in former-smokers.

No differences in risk estimates were seen with subgroup analyses by study design, country of recruitment or risk of bias for most analyses (Supplementary Figure S2B, C, G, H, I). For risk of bias, in current vs. never smokers there was a significantly lower risk of death from all causes in studies with low risk of bias; HR 1.31 (95% CI 1.24, 1.37) compared to studies with high risk of bias; HR 1.99 (95% CI 1.43, 2.75) (Supplementary Figure S2D).

Comparisons of current-smokers versus former-smokers, or ever-smokers versus never-smokers, were reported in three studies, one that combined sexes³⁸ and two^27,49 with sexes separated (Table 2). One small study with 74 melanoma patients compared ever-smokers with never-smokers.⁶⁰ None of these studies found a statistically-significant difference in all-cause risk of death between groups.

Quantity of cigarettes smoked and risk of death

Few studies reported analyses of quantity smoked and risk of death; three studies reported death from melanoma^18,21,39 and a single study²⁰ reported death from any cause, in varying groupings stratified by quantity of cigarettes smoked (Supplementary Table S2). Pooled analyses were not possible due to different quantity groupings and gender-specific data. Studies generally did not report the number of patients in each quantity group and estimates were highly variable. A study including 2208 patients with melanoma reported HRs of 2.1 (95% CI 1.02–4.44) and 2.3 (95% CI 1.30–4.22) from multivariable analyses for patients smoking 1–10 and ≥10 cigarettes/day, respectively, compared to never-smokers.³⁹ Another study of 121,106 patients with melanoma, in which 613 never-smokers and 227 current-smokers died, reported a HR from multivariable analyses that suggested no increased risk of death from melanoma with increasing number of cigarettes per day.²¹

Smoking and node surgery

Three studies with 129 ever-smokers and 610 never-smokers with melanoma reported the risk of complications after sentinel node biopsy (SNB) using multivariable analyses.^32,35,36 Complications reported included; hospital admission, seroma, scarring, wound infections, lymphoedema, haematoma, wound dehiscence, lymphorrhagia, altered sensitivity and urinary tract infection. Pooled analysis showed a doubling of the complication risk in ever-smokers (OR 2.01, 95% CI 1.41–2.85, p = 0.003) with limited heterogeneity (I² 10%) (Fig. 3A and Table 2). Four studies compared melanoma patients who were ever-smokers and never-smokers to generate a risk of complications after lymph node dissection (LND) using multivariable analyses^31,32,44,55 but two of these^32,55 used the same data source (NSQIP) with overlapping recruitment periods. Complications reported included; hospital re-admission, re-operation, sepsis, myocardial infarction, unplanned intubation, acute renal failure, pneumonia, urinary tract infection, blood transfusion, lymphoedema, seroma and wound dehiscence. Analysis after exclusion of the smaller study³² showed a significantly increased risk of complications in smokers undergoing LND (OR 1.65, 95% CI 1.23–2.20, p = 0.0007) with no heterogeneity I² 0% (Fig. 3B). Exclusion of the alternative duplicate dataset resulted in no major change in the estimate (Supplementary Figure S3). Other surgical complications, such as seroma, wound dehiscence and rate of frozen shoulder, were reported in single studies only (Supplementary Table S3).

Fig. 3.

Complications from lymph node surgery, using risk estimates from multivariable analyses. ∗ Ascha 2017 and Chua 2023 both use NSQIP data with overlapping recruitment periods so duplication in patients analysed, Ascha 2017 excluded from analyses, for shorter recruitment period and less detail in reporting numbers.

No studies reported complications from wide local excision or radiation therapy in smokers compared to never-smokers.

Smoking and systemic therapy

Seven studies including 1358 melanoma patients (646 smokers, 712 never-smokers) treated with contemporary immunotherapy (anti CTLA4 or anti PD-1 reported outcomes for these groups. Three studies reported risk of all-cause death in IO-treated smokers compared to IO-treated never-smokers, but each differed in smoker-groupings (ever vs. never or current vs. never) or analysis method (UVA, multivariable analyses) so a pooled analysis was not appropriate. Each study reported an increased risk of all-cause death in smokers relative to never-smokers, but none was statistically-significant (Table 3). Single studies reported outcomes of progression-free survival and response rates,⁶⁰ median survival⁵⁶ as well as adverse events^23,50,57,61 (Table 3).

Table 3.

Individual study data for outcomes reported in melanoma patients treated with immunotherapy agents, stratified by smoking status.

Outcome	Smoker comparison	Analysis type	Current smoker	Ever smoker	Former smoker	Never smoker	Risk estimate (95% CI)	P value	Ref
All cause death	Ever vs. Never	UVA	–	34		54	1.4 (0.72, 2.53)	NS	56
	Ever vs. Never	MVA	–	38		36	1.1 (0.73, 1.70)	NS	60
	Current vs. Never	UVA	32	–		67	1.3 (0.66, 2.36)	NS	45
	Former vs. Never	UVA			67	70	1.1 (0.63, 1.76)	NS	45
Median survival	Ever and Never			34		54	Ever; 1.7 years Never; 3.4 years	–	56
Progression free survival	Ever vs. Never	MVA		38		35	0.77 (0.57, 1.1)	NS	60
Major atherosclerotic cardiovascular events with IO or no IO treatment	Ever vs. Never	UVA		139 (IO) 158 (no IO)		150 (IO) 199 (no IO)	3.5 (1.0, 12.5)	0.05	54
Risk of progression with aPD-1 treatment	Current vs. Never	UVA	32			70	1.0 (0.60, 1.77)	NS	45
	Former vs. Never	UVA			67	70	1.1 (0.77, 1.70)	NS	45
Overall Response Rate, with IO treatment	Ever vs. Never	MVA		38		35	1.5 (0.8, 2.88)	NS	60
Disease Control Rate (CR + PR + Stable) with IO treatment	Ever vs. Never	MVA		38		35	1.8 (0.97, 3.15)	NS	60
High grade endocrine toxicity	Ever vs. Never	MVA		303		369	6.3 (1.2, 51.0)	0.05	57
Gastrointestinal toxicity	Ever vs. Never	MVA		303		369	2.7 (1.1, 6.9)	0.03	57
Discontinuation of treatment for immune related adverse events	Ever vs. Never	MVA		303		369	1.2 (NR)	0.012	57
Immune related adverse events	Ever vs. Never	MVAa		19		18	2.8 (1.41, 5.36)	0.003	50
Risk of colitis (aCTLA4)	Ever vs. Never	UVA		14		16	0.3 (0.05, 1.79)	NS	23

NS; not significant, IO: immunotherapy agents aCTLA4 and aPD-1, UVA; univariable analyses, MVA; multivariable analyses.

^aAnalysis is within mixed tumour types but adjusts for tumour type in multivariable analyses.

Smoking and prognostic features at the time of diagnosis of primary melanomas

Five studies reported the differences in mean Breslow thickness of melanomas at diagnosis in current-smokers compared with never-smokers,6, 7, 8^,41,58 but two studies^8,58 used overlapping data from MSLT-I and MSTL-II trial participants. Pooled analysis showed high heterogeneity (I² 85–94%, Supplementary Fig. S4A–C) irrespective of which duplicate dataset was excluded. The melanomas were 11% thicker in current-smokers compared to never-smokers in two studies,^41,58 15% thicker in current-smokers in one study,⁸ 28% thicker in current-smokers in a fourth study⁶ and not thicker in the fifth study.⁷ Four studies^7,8,41,58 using three datasets reported the mean difference in Breslow thickness in former-smokers compared to never-smokers, giving a pooled estimate of 3.8% (95% CI 1.17, 6.51%, p = 0.005, I² 0%) thicker melanomas in former-smokers compared to never-smokers (Supplementary Figure S4D). Five studies^{18,22,24,37,41} reported the mean difference in Breslow thickness in ever-smokers versus never-smokers, with highly variable results (I² 98%, Supplementary Figure S4E), ranging from 20% thinner²⁴ to 63% thicker melanomas in ever-smokers compared to never-smokers.

Three studies^8,38,58 reported the risk of SNB-positivity in current-smokers compared to never-smokers, with two using the same data source (MSLT-I and MSLT-II) but one⁵⁸ reported the two trial datasets separately. Pooled analysis showed an increased risk of SNB-positivity in current-smokers compared to never-smokers; HR 1.35 (95% CI 1.13–1.62, p = 0.001) with no heterogeneity (I² 0%) (Table 2, Fig. 4A.). Analysis of former-smokers compared to never-smokers showed no increase in SNB-positivity risk; HR 0.99 (95% CI 0.82–1.19, p = 0.90) with some heterogeneity (I² 41%) (Table 2, Fig. 4B.).

Fig. 4.

Risk of positive sentinel node biopsy in smokers compared to never-smokers, from univariable analyses.

Risk of bias

Summarised findings from risk of bias assessments are provided in Fig. 5. The critical issue of demonstrating how similar or different smokers (current, ever or former) and never-smokers were for the well-recognised prognostic features for melanoma survival was very poorly documented, with 39 of the 46 studies not adequately reporting prognostic features for the smokers and never-smokers. The remainder of the studies were moderately or highly biased on this matter based on their descriptive data for the two groups showing moderate or major differences in important prognostic features, often in age and Breslow thickness, between the current-smoker and never-smoker groups. Ascertainment of smoking status was similarly unclear. Sixteen studies stated that smoking status was self-reported, but it was unclear in the remainder of studies. Most studies suggested that smoking status was collected once but this was rarely detailed. Bias in relation to the representativeness of the sample was low in only 16 studies, while in nine there was a high risk of bias and 18 studies were moderately biased in the representativeness of their sample. Follow-up duration, assessment of the primary outcome, the influence of the outcome on patient inclusion and the influence of a patient's smoking status on inclusion were much better reported and were at low-risk of bias for most studies. Publication bias was assessed using funnel plots of the main analyses and showed reasonable symmetry given the limited number of studies (9, 7, 4, and 3 studies), suggesting that publication bias was limited (Supplementary Fig. S5A–F).

Fig. 5.

Summarised risk of bias assessments in included studies.

Grade of evidence

All included studies were non-randomised, observational studies, meaning that they provided low-level evidence. Additionally, many studies had moderate risk of bias. However, the consistency of the findings from multivariable analyses was high, most studies provided direct evidence, precision was good for the main outcomes and publication bias was not obvious, leading to an overall moderate grade of evidence.

Discussion

Despite the intriguing evidence that the incidence of melanoma in smokers is lower than in non-smokers,2, 3, 4, 5 this review confirms the suggestion from previous studies that once a smoker develops melanoma, survival from it is significantly reduced and complications from node surgery are more likely. Quantitation of these adverse effects is novel to this meta-analysis. The analyses confirmed that after adjusting for other prognostic factors, the risk of death from melanoma for current-smokers is 33% higher than the risk for never-smokers. The risk of death from any cause in smokers was similarly increased (32%) and therefore is probably explained by an excess of melanoma deaths as opposed to death from non-melanoma causes.

An important and encouraging finding of the present study was reversion to the baseline risk of death from melanoma in people who managed to stop smoking. However, there was high variability in quantification of the risk of death from melanoma in former-smokers, and this is likely to be explained by variability in the time since smoking cessation and possibly also quantity smoked prior to cessation. It would be valuable to document date of smoking cessation as part of each patient's clinical record to allow for better quantification of risk reduction in the future.

Although melanoma management has evolved significantly over the last decade, surgery remains the cornerstone of treatment for early disease. Smoking has the potential to increase surgical complication rates as well as increasing rates of systemic complications that can greatly hinder post-operative recovery. Though we identified no studies reporting data on complication rates from wide local excision in smokers, the by-products of tobacco metabolism and their indirect effect on local oxygen delivery to tissues have been implicated in delayed wound healing and infection in smokers in many other scenarios.⁶² The available data in the present study support this theory, with ever-smokers being twice as likely to experience complications after a SNB and 1.6 times more likely to do so after a LND.

The interplay between smoking and drug efficacy may be particularly relevant in the modern era of melanoma treatment that is centred on immunotherapy. Although very little information is currently available documenting the effects of smoking on melanoma treatment using immunotherapy agents, there can be little doubt that smoking has important effects on the immune system. A number of studies have explored various aspects of the immune response in relation to smoking and their results suggest that smoking reduces the number and activity of natural killer cells and lowers IgG and IgA immunoglobulin levels^63,64 while also reducing T cell anergy and inducing immunosuppression.⁶⁵ A state of immunosuppression induced by smoking may reduce the efficacy of immune checkpoint inhibitors, a class of drugs that act to modify the dysregulation of T cells by tumours to enhance anti-tumour immunity.⁶⁶ Given this possibility, it will be important for clinicians to collect data on smoking status at the start of treatment and in an ongoing fashion to better elucidate the effects of smoking on the response to modern systemic therapies.

Limited data exist on the differences in prognostic features of melanomas in smokers versus non-smokers at the time of diagnosis, but they suggest that current-smokers present later than non-smokers, with thicker primary melanomas, and subsequently are more likely to be SNB-positive and have a worse prognosis. It is therefore important that clinicians encourage smokers to present for regular skin checks.

The exhaustive literature search and compilation of 46 studies from 14 countries brought together the largest dataset to date documenting melanoma outcomes in smokers and non-smokers. This allowed more precise risk estimates to be generated. The analyses of risk of death from melanoma using multivariable analyses estimates were highly consistent and many used multivariable analyses adjusting for known prognostic factors, providing reassurance that the disparate studies were generating very similar results. While only limited data were available to assess the effects of smoking on complications from node surgery, the studies that did report data were mostly similar, indicating increased complication rates, with one exception possibly due to chance effects in a very small sample (15 smokers in 124 patients).³⁵

Though our search yielded 46 different studies, 4 institutions contributed 9 of the available studies, with some duplication of patients. However, our analyses included only unique datasets. While 41 studies with unique datasets were available for analysis, few studies reported the same outcomes using the same analysis methods (UVA, multivariable analyses, etc), limiting the maximum number of studies included in any one analysis to ten. Our analyses focused on pooled risks generated from multivariable analyses and each of the studies adjusted for various factors that appeared significant in their univariate analyses. Issues adjusted for in the multivariable analyses differed across studies and it was therefore surprising that such consistent risks were reported, given the variability in factors for which adjustments were made across the studies. This probably reflects the different influence that features such as body mass index, alcohol intake and socioeconomic status have on survival in various countries and differing health facilities. Only seven studies reported the use of contemporary systemic treatments and few reported similar outcomes, preventing the ability to pool results. Lastly, in all studies that reported the method by which smoking status was determined it was by self-reporting, an inherently flawed means of data collection. Limitations to the review process included a single author (GJW) performing data extraction and risk of bias assessment. Data extraction by a single, non-clinical author may illicit errors uncorrected by a second reviewer, however these errors are likely to be random and non-differential, limiting the impact these would have on findings.

In conclusion, this systematic review and meta-analysis showed that melanoma patients who are smokers at the time of diagnosis are more likely to die from their melanoma, are more likely to develop complications from node surgery and are more likely to have a positive SNB. However, those who stop smoking have a similar risk of dying from their melanoma as never-smokers. Clinicians treating melanoma patients should therefore encourage them to cease smoking. To better understand the effects of smoking on melanoma development, progression and patient responses to treatment, it will be important to collect information on smoking status for all patients, and particularly for research study participants, to enable more reliable assessment of the impact of smoking on treatment efficacy and patient outcomes.

Contributors

EBF; Conception of study, article title review, drafted Introduction and Discussion sections, manuscript review, GJW; Electronic searches, article title review, data extraction, analyses, tables and figures, drafted Methods and Results sections, manuscript review, SNL; Review and comment on Methods section, manuscript review, JFT; Conception of study, team coordination, manuscript review. All authors had access to the data and final responsibility for the decision to submit the publication. All authors read and approved the final version of the manuscript.

Data sharing statement

All data extracted and analysed in this review are available upon request. Data for the main analyses are presented in the paper and supplementary data file.

Declaration of interests

JFT has received honoraria for advisory board participation from BMS Australia, MSD Australia, GSK and Provectus Biopharmaceuticals, and travel support from GSK and Provectus Biopharmaceuticals. SNL has received fees for professional services from SkylineDx BV and an honorarium for editorial duties from The British Association of Dermatologists. The other authors have no competing interests.