Weights are from a random-effects model. I2 indicates heterogeneity estimate; dashed vertical line, estimated pooled effect size estimate; open diamond, a visual summary of the overall 95% CI of the effect estimate of atopic dermatitis on the risk of cancers. BCC indicates basal cell carcinoma; CNS, central nervous system; KS, Kaposi sarcoma; NA, not applicable; NMSC, nonmelanoma skin cancer; SCC, squamous cell carcinoma; SIR, standardized incidence ratio; lines with solid squares, SIRs for the individual study and 95% CIs.
aI2 values are not applicable for groups of cancers in which only 1 study is presented, and therefore no subgroup I2 value could be calculated.
Weights are from a random-effects model. I2 indicates heterogeneity estimate; dashed vertical line, estimated pooled effect size estimate; and open diamond, a visual summary of the overall 95% CI of the effect estimate of atopic dermatitis on the risk of cancers. CNS indicates central nervous system; OR, odds ratio; and lines with solid squares, ORs for the individual study and 95% CIs.
Weights are from a random-effects model. I2 indicates heterogeneity estimate; dashed vertical line, estimated pooled effect size estimate; and open diamond, a visual summary of the overall 95% CI of the effect estimate of atopic dermatitis on the risk of cancers. GU indicates genitourinary; MM, multiple myeloma; OR, odds ratio. Lines with solid squares represent ORs for the individual study and 95% CIs.
bOverall I2, OR (95% CI), and weight for the data of Figure 2 and 3 combined.
eFigure 1. PRISMA Diagram of Searched and Included Studies.
eFigure 2. Atopic Dermatitis and Cancer Risk in Cohort Studies With Substantial Heterogeneity (I2 Values >50%)
eFigure 3. Atopic Dermatitis and Cancer Risk in Case-Control Studies With Substantial Heterogeneity (I2 Values >50%)
eTable 1. Risk of Bias in Nonrandomized Studies of Interventions (ROBINS-I) Assessment of Included Cohort Studies
eTable 2. Risk of Bias in Nonrandomized Studies of Interventions (ROBINS-I) Assessment of Included Case-Control Studies
eFigure 4. Funnel Plot and Contour-Enhanced Funnel Plot of Standardized Incidence Ratios for the Effect of Atopic Dermatitis on Central Nervous System Cancer Risk in Case-Control Studies
Customize your JAMA Network experience by selecting one or more topics from the list below.
Wang L, Bierbrier R, Drucker AM, Chan A. Noncutaneous and Cutaneous Cancer Risk in Patients With Atopic Dermatitis: A Systematic Review and Meta-analysis. JAMA Dermatol. 2020;156(2):158–171. doi:10.1001/jamadermatol.2019.3786
What is the risk of noncutaneous and cutaneous cancers in patients with atopic dermatitis (AD) compared with the general population?
This systematic review and meta-analysis included 8 population-based cohort studies (n = 5 726 692) and 48 case-control studies (n = 14 136). In population-based cohort studies, AD was statistically significantly associated with keratinocyte carcinoma (5 studies) and kidney cancer (2 studies), whereas case-control studies demonstrated lower odds of lung and respiratory system cancers (4 studies).
Despite evidence suggesting potential associations between AD and the risk of some cancers, more research is needed to address heterogeneity and biases across existing studies to make definitive conclusions.
Impaired skin barrier and aberrant immune function in atopic dermatitis (AD) may alter immune response to malignant cancer. Conflicting data exist on the risk of cancer in patients with AD.
To assess the risk of noncutaneous and cutaneous cancers in patients with AD compared with the general population without AD.
Studies identified from searches of MEDLINE and Embase that were published from 1946 and 1980, respectively, to January 3, 2019. The following search terms were used: [(exp NEOPLASMS/ OR neoplas*.tw. OR tumo*.tw. OR cancer*.tw. OR malignanc*.tw.) AND (exp Dermatitis, Atopic/ OR (atopic adj1 (dermatit* or neurodermatit*)).tw. OR eczema.tw. OR disseminated OR neurodermatit*.tw.)].
Included were observational studies (cohort and case-control designs) reporting a risk estimate for cancer in patients with AD compared with a control group (general population or patients without AD).
Data Extraction and Synthesis
Two independent reviewers extracted data and assessed the risk of bias using the Risk of Bias in Nonrandomized Studies of Interventions (ROBINS-I) assessment tool, modified for observational exposure studies. Data were pooled using a random-effects model and expressed as standardized incidence ratios (SIRs) or odds ratios (ORs) with 95% CIs. Heterogeneity was assessed using the Cochrane Q statistic and the I2 statistic.
Main Outcomes and Measures
The main outcome of the study was risk of cancer measured by SIRs or ORs.
This systematic review and meta-analysis included 8 population-based cohort studies (n = 5 726 692 participants) and 48 case-control studies (n = 114 136 participants). Among cohort studies, a statistically significant association was found between AD and keratinocyte carcinoma (5 studies; pooled SIR, 1.46; 95% CI, 1.20-1.77) as well as cancers of the kidney (2 studies; pooled SIR, 1.86; 95% CI, 1.14-3.04), central nervous system (2 studies; pooled SIR, 1.81; 95% CI, 1.22-2.70), and pancreas (1 study; SIR, 1.90; 95% CI, 1.03-3.50). Among 48 case-control studies, pooled effects showed patients with AD had statistically significantly lower odds of central nervous system cancers (15 studies; pooled OR, 0.76; 95% CI, 0.70-0.82) and pancreatic cancer (5 studies; pooled OR, 0.81; 95% CI, 0.66-0.98), contrary to the higher incidence found in cohort studies. Case-control studies also demonstrated lower odds of lung and respiratory system cancers (4 studies; pooled OR, 0.61; 95% CI, 0.45-0.82). No evidence of association was found between AD and other cancer types, including melanoma. There was substantial heterogeneity between studies for many other cancers, which precluded pooling of data, and there was moderate to serious risk of bias among included studies.
Conclusions and Relevance
Observational evidence suggests potential associations between AD and increased risk of keratinocyte carcinoma and kidney cancer as well as lower odds of lung and respiratory system cancers. Further research is needed to address the heterogeneity and limitations of current evidence and to better understand the mechanisms underlying a possible association between AD and cancer risk.
Atopic dermatitis (AD) is a common, chronic inflammatory skin condition. Impaired skin barrier and aberrant immune function are central to the pathogenesis of AD. Defects of the epidermal barrier1 may facilitate transcutaneous penetration of carcinogenic agents and viruses, and systemic abnormalities in cell-mediated immunity may alter the immune response to cancer.2 Treatment of severe, acute flares and refractory cases with immunosuppressive medication may also increase susceptibility to cancer.
There are conflicting data on the risk of cancer in patients with AD. For example, American studies of the association between AD and pancreatic cancer found protective associations3 and no association,4 whereas a Swedish study5 found an increased risk. A study6 conducted in Denmark found a decreased risk of melanoma associated with AD, whereas a study7 conducted in Taiwan found an increased risk. Given the heterogeneous findings in the literature and the large global burden of AD,8 a systematic review of the literature is needed to provide a better understanding of the association between AD and the risk of various cancers.
The planning and conduct of this systematic review and meta-analysis followed the Cochrane Handbook for Systematic Reviews of Interventions.9 We prospectively registered our systematic review and meta-analysis on PROSPERO (CRD42018092929).
MEDLINE and Embase were searched for articles published from 1946 and 1980, respectively, to January 3, 2019. The following search terms were used: [(exp NEOPLASMS/ OR neoplas*.tw. OR tumo*.tw. OR cancer*.tw. OR malignanc*.tw.) AND (exp Dermatitis, Atopic/ OR (atopic adj1 (dermatit* or neurodermatit*)).tw. OR eczema.tw. OR disseminated OR neurodermatit*.tw.)]. The search excluded animal trials. We focused on peer-reviewed, published studies and did not search the grey literature. Reference lists of included studies and related systematic reviews were scanned for additional studies. We included observational studies (cohort and case-control studies) reporting a risk estimate for cancer in patients with AD or eczema compared with a control group (general population or patients without AD). There were no limitations on age or type of cancer. In instances in which multiple studies used overlapping patient samples, the study with the most recent sample was included.
Two of us (L.W. and R.B.) independently screened titles and abstracts for potentially eligible studies. The full text of studies retained after the initial screening was reviewed in duplicate for eligibility, with reasons for exclusion documented. Relevant data were extracted in duplicate using a standardized extraction form. Any discrepancies in screening and extraction were resolved by discussion between the 2 reviewers. The following items were extracted from the qualifying reports: authors, publication year, study design, cohort name, study period, sample size, patient characteristics (age, sex, and race), covariates used for adjustment, cancer type, follow-up time, country of study, methods of identifying AD and cancer, and adjusted effect estimates for the association between AD and cancer.
Two of us (L.W. and R.B.) assessed the risk of bias of included studies using the Risk of Bias in Nonrandomized Studies of Interventions (ROBINS-I)10 assessment tool, modified for studies evaluating exposures rather than intervention. The ROBINS-I tool comprises 7 domains through which bias might be introduced. Studies were categorized as low risk, moderate risk, serious risk, and critical risk of bias under each domain. For meta-analyses with at least 10 included studies, publication bias was assessed using the Egger regression test11 and visual inspection of funnel plots.
Effect estimates for cancer incidence observed in cohort and case-control studies were pooled separately using the inverse variance method of the DerSimonian and Laird random-effects model to produce standardized incidence ratios (SIRs) from cohort studies and odds ratios (ORs) from case-control studies, with 95% CIs. When reported by case-control studies, we used the adjusted OR to account for confounders. Heterogeneity was assessed by the Cochrane Q statistic, where 2-sided P < .10 was considered statistically significant, and quantified by the I2 statistic, where I2 values of 50% or greater were considered to indicate substantial heterogeneity.9 Meta-analysis was only performed if substantial heterogeneity was not observed across study estimates for a given cancer. Data analysis was conducted using Stata, version 13.0 statistical software (StataCorp LP).
The Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) diagram of searched and included studies is shown in eFigure 1 in the Supplement. The comprehensive literature search identified a total of 7280 articles from MEDLINE (n = 1582) and Embase (n = 5698). After full-text review, 147 studies met our inclusion criteria. A further 27 studies were identified and included based on a manual search of references listed in the included articles. A total of 56 studies (8 cohort studies5-7,12-16 and 48 case-control studies4,17-63) were ultimately included.
Characteristics of the cohort studies are summarized in Table 1. Eight population-based cohort studies5-7,12-16 were included in this systematic review, with a total of 5 726 692 participants from the United Kingdom (2 studies12,16), Sweden (2 studies5,15), Taiwan (2 studies7,13), and Denmark (2 studies6,14). Study periods in the 8 studies ranged from 1886 to 2011, whereas the mean follow-up was between 5 and 42 years. The mean age of participants ranged from 15 to 58 years across study cohorts. Six studies identified cases of AD using International Classification of Diseases, Ninth Revision codes in hospital medical records (2 studies5,12) or national patient registers (4 studies6,7,14,16). For the remaining studies, one used a questionnaire15 and another used periodic medical examination.16 Four studies5,6,14,15 used a national cancer registry to ascertain cancer diagnoses, whereas the remainder used self-report (1 study16), a national insurance database (1 study7), Registry for Catastrophic Illness Patient Database (1 study13), and electronic medical records (1 study12).
Characteristics of the case-control studies are listed in Table 2. Overall, 48 case-control studies4,17-63 were included, with a total of 114 136 participants. Most participants were from the United States (17 studies4,17,20,21,25,26,28,30,34,37,39,43,48,53-56) and United Kingdom (10 studies17,19,23,24,48-51,59,63). Study periods ranged from 1956 to 2014. The mean age of participants ranged from 3 to 69 years. Cancer cases were identified through regional or national cancer registries or surveillance systems (17 studies19,20,29,37,38,42,45,46,49-54,56,59,60), in-person hospital or clinic recruitment (20 studies4,17,21,23-25,27,28,31-36,40,41,43,58,61,63), other mechanisms (7 studies18,30,39,47,55,57,62), or multiple methods (4 studies22,26,44,48). The most frequently studied cancers included central nervous system cancers (15 studies17,18,21,27,34,38,45,47-52,59,63) and hematologic cancers (15 studies19,20,23,28,30,32,41,44,46,53,55-57,60,62). Other cancer types included keratinocyte carcinomas (5 studies25,26,36,42,43), melanomas (3 studies33,37,40), pancreatic cancer (5 studies24,35,41,54,63), and others (5 studies24,31,39,45,61). Seven studies38,39,46,55,56,62,63 were focused exclusively on childhood cancers. Controls without cancer were identified through hospital or clinic recruitment (10 studies19,21,23,24,26,33,36,49-51), electoral roll or population database extraction (19 studies18,22,25,27,30-32,35,40,41,43-47,52,57,60,61), random digit dialing (8 studies28,29,37-39,53,54,56), recruitment through personal relations to the case (3 studies33,34,42), neighborhood recruitment (1 study20), retrieval from cohort study (1 study58), multiple methods (4 studies4,17,48,59), or not reported (1 study62). The exposure of AD was ascertained using a self-administered questionnaire or professionally administered questionnaire or interview (45 studies4,17-25,27-29,31-54,56-63) or medical record abstraction (3 studies26,30,55).
In pooled results from cohort studies without substantial heterogeneity, there were statistically significant associations between AD and a higher incidence of keratinocyte carcinoma (5 studies5-7,12,14; pooled SIR, 1.46; 95% CI, 1.20-1.77) as well as cancers of the kidney (2 studies5,7; pooled SIR, 1.86; 95% CI, 1.14-3.04), central nervous system (2 studies5,7; pooled SIR, 1.81; 95% CI, 1.22-2.70), and pancreas (1 study5; SIR, 1.90; 95% CI, 1.03-3.50). These results are shown in Figure 1.
In pooled results from case-control studies without substantial heterogeneity, AD was associated with statistically significantly lower odds of central nervous system cancers (15 studies17,18,21,27,34,38,45,47-52,59,63; pooled OR, 0.76; 95% CI, 0.70-0.82) and pancreatic cancer (5 studies22,29,35,54; pooled OR, 0.81; 95% CI, 0.66-0.98) (Figure 2 and Figure 3), contrary to the higher incidence found in cohort studies. Case-control studies also demonstrated a statistically significant association between AD and lower odds of lung and respiratory system cancers (4 studies24,58,61,62; pooled OR, 0.61; 95% CI, 0.45-0.82).
In both cohort and case-control studies, there was no evidence of an association between AD and other cancer types with pooled estimates, including breast, colorectal, head and neck, male genitourinary, myeloma, and melanoma. These results are shown in Figure 1, Figure 2, and Figure 3.
There was substantial heterogeneity (I2 values ≥50%) across studies for many types of cancer, precluding meta-analysis. Among cohort studies5-7,12-16(eFigure 2 in the Supplement), leukemia and lymphomas, melanomas, female and male genitourinary, gastrointestinal, lung, and respiratory system cancers had high heterogeneity. Among case-control studies4,17-63 (eFigure 3 in the Supplement), there was evidence of substantial heterogeneity for keratinocyte carcinomas and leukemia and lymphomas.
Risk of bias assessment found that 3 of the 8 cohort studies7,13,14 had a serious risk of bias, and 5 studies5,6,12,26,28 had a moderate risk of bias (eTable 1 in the Supplement). Among case-control studies, 10 studies7,22,23,33-35,40,43,46,54 were found to have a serious risk of bias, and 38 studies4,17-21,24-26,28-32,36-39,41,42,44,45,47-53,55-63 had a moderate risk of bias (eTable 2 in the Supplement). Publication bias assessment was undertaken for case-control studies reporting effect estimates for cancers of the central nervous system because this was the only subgroup with at least 10 included studies (15 studies). The Egger regression test (bias, −0.12; 95% CI, −1.19 to 0.94; P = .81) suggests no small-study associations, although there was evident asymmetry on visual inspection of the contour-enhanced funnel plot (eFigure 4 in the Supplement).
Our systematic review and meta-analysis of population-based cohort studies revealed a statistically significant association between AD and increased risk of keratinocyte carcinoma and kidney cancer. Respiratory system cancer was statistically significantly less common among patients with AD in case-control studies. Conflicting, qualitatively different evidence was found for central nervous system cancers and pancreatic cancer, with cohort studies showing a higher risk of cancer associated with AD, whereas case-control studies showed lower odds. For many cancers, there was substantial heterogeneity of effect estimates in the literature, making firm conclusions difficult.
Halling-Overgaard et al64 recently published a letter reporting a systematic review of 32 studies that evaluated the risk of solid organ cancer in patients with AD. They reported an inverse association between AD and brain cancer (n = 12 studies), which we only found among case-control studies (n = 15 studies); an opposite positive association was shown among cohort studies in the present systematic review and meta-analysis (n = 2 studies). For pancreatic cancer, Halling-Overgaard et al64 found no association with AD (n = 4 studies), in contrast to our finding of an inverse association based on case-control studies (n = 5 studies). The differing results between the 2 systematic reviews and meta-analyses can be attributed to several factors, including the larger number of included studies in the present systematic review and meta-analysis (n = 56 studies vs 32 studies) because we used a more comprehensive search strategy (eFigure 1 in the Supplement). Although Halling-Overgaard et al64 pooled unadjusted ORs, we used the adjusted OR reported by each primary study as recommended by the Cochrane Handbook for Systematic Reviews of Interventions.9
Multiple theories are proposed to explain the association between AD and keratinocyte carcinoma development. Surveillance may increase detection of skin cancer among patients with AD because of frequent follow-up for the chronic skin disease.65 The associated loss of function of the filaggrin gene (OMIM 135940) in some patients with AD may also contribute to pathogenesis because filaggrin abnormalities increase susceptibility of keratinocytes to UV radiation–induced damage.66-68 This conclusion is supported by an increase in self-reported squamous cell carcinoma precursors, actinic keratoses, in patients with filaggrin mutations.69 Kaae et al69 support this theory by concluding that there may be an association between squamous cell carcinoma development and filaggrin mutation.
Development of cancer in patients with AD may also be associated with reduced immune surveillance through immunologic dysfunction associated with the disease itself or related to treatments for the condition.70 The chronic inflammatory state and resultant epidermal abnormalities may also lead to the development of keratinocyte cancer through associated increased cell turnover.71 Chronic inflammation may also explain the positive association between AD and solid organ tumors, including brain, pancreatic, and kidney cancers.5,7
Conversely, overactive immune states in patients with AD may also be a protective factor for some types of cancer, including lung, pancreatic, and central nervous system cancers, as found in the case-control studies in our meta-analysis. This finding is potentially associated with increased immune surveillance and clearance of cells with malignant potential.3 Pompei et al71 demonstrated that patients with known allergy had increased response to cancer therapy and cure rates compared with patients without allergy. This finding suggests that the helper T-cell subtype 2 (TH2) hyperreactivity in allergic patients may have a role in cancer cell identification and clearance, challenging the belief that cancer clearance is a mainly TH1-driven response.72,73 IgE levels may be associated with antitumor activity in patients with atopy (ie, patients with diagnoses of any of the triad of asthma, AD, or allergic rhinitis). IgE is effective in triggering immune responses to carcinogenic cells, such as in colorectal cancer.12
The present study has strengths. There was no statistically significant publication bias among case-control studies reporting risk of AD on central nervous system cancers (Egger test P = .81). There was also no notable asymmetry on visible inspection of the standard funnel plot. These outcomes suggest there were no small-study associations. However, studies appear to be missing in areas of statistical significance on the right-hand side of the contour-enhanced funnel plot, suggesting factors in addition to publication bias may be contributing to observed outcomes, such as variable study quality (eFigure 4 in the Supplement).
Our systematic review and meta-analysis and the underlying evidence base have several limitations, and the results should be interpreted with caution. First, our literature search was limited to MEDLINE and Embase, as well as reference lists of included studies, meaning that relevant studies indexed in other databases may have been missed. Our inclusion of only peer-reviewed published studies confers an assumption of higher-quality studies being included but may have resulted in inflated effect estimates because of publication bias. Second, observational study designs are particularly susceptible to selection, attrition, ascertainment, and reporting biases. Most studies (76.8% [43 of 56 total studies]) were rated as having a moderate risk of bias, particularly because of confounding from factors like increased medical surveillance and race. This bias may explain the conflicting results from cohort vs case-control studies. Third, despite stratifying the meta-analyses by study design to account for major methodological differences between cohort and case-control studies, the systematic review found substantial heterogeneity of associations across studies for multiple types of cancer, which precluded meta-analysis. This heterogeneity may have occurred because of clinical and methodological differences across the observational studies in the meta-analyses, including the racial and genetic composition of the populations, as well as variation in how AD and cancer diagnoses were defined and assessed. No cohort studies included information on the criteria used to diagnose AD, and many of the studies used self-report questionnaires to ascertain an AD diagnosis. Seven5-7,12,14-16 of 8 included cohort studies did not collect information about disease severity, which could allow for a more comprehensive evaluation of the association between AD and malignant cancer. Furthermore, none of the studies analyzed the association between AD treatment and cancer development. The frequency and intensity of topical or systemic medication use may alter the level of immunosuppression and cancer risk.
In this systematic review and meta-analysis, we found that AD may be associated with an increased risk of keratinocyte carcinoma and kidney cancer in population-based cohort studies. Our results also suggest a protective role of AD in lung and respiratory system cancers. Given the statistically significant heterogeneity of outcomes and the substantial risk of bias among included studies, further research with clearly defined criteria for AD diagnosis, adjustment for key confounders, and details on AD severity and treatment is required to better understand the mechanisms underlying the possible association between AD and cancer risk.
Accepted for Publication: October 13, 2019.
Corresponding Author: Lily Wang, MD Program, Faculty of Medicine, University of Toronto, Medical Sciences Building, One King’s College Circle, Toronto, ON M5S 1A8, Canada (email@example.com).
Published Online: December 11, 2019. doi:10.1001/jamadermatol.2019.3786
Author Contributions: Ms Wang and Dr Chan had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Wang, Drucker, Chan.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Wang, Bierbrier.
Critical revision of the manuscript for important intellectual content: Wang, Drucker, Chan.
Statistical analysis: Wang.
Supervision: Drucker, Chan.
Conflict of Interest Disclosures: Dr Drucker reported serving as an investigator for and receiving research funding from Sanofi and Regeneron; being a consultant for Sanofi, RTI Health Solutions, Eczema Society of Canada, and Canadian Agency for Drugs and Technology in Health; receiving honoraria from Prime Inc, Spire Learning, CME Outfitters, Eczema Society of Canada, and the Canadian Dermatology Association; and reported that his institution has received educational grants from Sanofi and AbbVie. No other disclosures were reported.