Prevalence of dermatological toxicities in patients with melanoma undergoing immunotherapy: Systematic review and meta-analysis

Background Checkpoint inhibitors have revolutionized advanced melanoma care; however, their cutaneous side effects have not been definitively elucidated. Objective To identify the prevalence of cutaneous toxicity in patients with melanoma treated with immune checkpoint inhibitors as monotherapy and/or in combination with chemotherapy and/or radiotherapy. Materials and methods We performed a systematic review and meta-analysis, which encompassed both clinical trials and observational studies describing the dermatological toxicities in patients treated with immune checkpoint inhibitors. The protocol was registered in the International Prospective Register of Systematic Review under the number CRD42018091915. The searches were performed using the CINAHL, Cochrane CENTRAL, LILACS, LIVIVO, PubMed, Scopus, and Web of Science databases. The methodological quality of the studies was evaluated with the JBI Critical Appraisal Checklist for Studies Reporting Prevalence Data Results A total of 9,802 articles were identified in the databases. The final sample comprised 39 studies. The evaluated drugs were ipilimumab, tremelimumab, pembrolizumab, and nivolumab. The results suggest that the most prevalent side effect was grade 1 and 2 pruritus (24%), followed by grade 1 and 2 rash (21%) and grade 1 and 2 vitiligo (10%). Conclusion The most prevalent side effects in patients treated with checkpoint inhibitors are pruritus, rash, and vitiligo, and they are rated mostly as grades 1 and 2 adverse events. Remarkably, vitiligo is most commonly found in patients treated with PD-1 inhibitors.

International Prospective Register of Systematic Review (PROSPERO) [21] under the number CRD42018091915.

Eligibility criteria
This review aimed to answer the following guiding question, based on the PECO strategy-Population, Exposure, Control and Outcomes: "What is the prevalence of dermatological toxicity (O) in patients with melanoma (P) undergoing treatment with ICIs (E)? " We included clinical trials (randomized and non-randomized) and observational studies that evaluated melanoma cancer patients undergoing treatment with a single ICI, a combination of ICIs, or a combination of an ICI with chemotherapy and/or radiotherapy and that described the prevalence of dermatological toxicity. There were no restrictions regarding the language or publication period.
The studies were analyzed for inclusion and exclusion criteria in two phases. In phase 1 (screening of the titles and abstracts), we excluded studies that evaluated children and adolescents with cancer; adult oncology patients treated with ICIs with an associated autoimmune disease; and adult oncology patients treated with ICIs who also received vaccines, target therapy, or other therapies. Further, we excluded studies with cutaneous toxicity not associated with the use of ICIs, studies showing other (non-dermatological) toxicities associated with the use of ICIs, literature reviews, letters, case reports, personal opinions, conferences, abstracts, and book chapters.
In phase 2 (full text reading), we excluded studies that presented complementary data from previously published investigations, qualitative studies, studies that did not present complete data and/or did not allow the extraction of the data of interest, and studies that evaluated other types of cancer (not melanoma).

Information sources
The searches were carried out in the following electronic databases: CINAHL, Cochrane CEN-TRAL, LILACS, LIVIVO, PubMed, Scopus, and Web of Science. A search strategy was developed for each of the databases (S1 File). The gray literature was also checked using Google Scholar and Open Gray. Further, a manual search was performed on the list of references of the included studies. All searches were performed on January 23, 2019.

Study selection
The study selection was carried out in two phases. In phase 1, two reviewers (N.F.M.S.G. and A.C.C.C.) independently assessed the titles and abstracts of all citations identified in the electronic databases. Articles that did not meet the inclusion criteria were excluded. This step was performed on a Rayyan web application [22]. In phase 2, the same reviewers applied the inclusion criteria to the full text of the articles. Any disagreements in the first or second phase were resolved by discussion until an agreement was reached between the two authors. When consensus could not be reached, the third reviewer (C.I.V.) was asked to evaluate the article and make the final decision.
References, including the removal of duplicates, were managed in EndNoteBasic [23].

Data collection process
Two reviewers (N.F.M.S.G. and A.C.C.C.) independently collected data from the selected studies. A third reviewer (C.I.V.) evaluated the accuracy of the collected information. For all included studies, the following information was recorded: study characteristics (author, year, country of publication, study design, and purpose); sample characteristics (sample size and mean age of the participants); duration of the drug treatment and follow-up; and characteristics of the results (type of dermatological toxicity and main conclusions).

Risk of bias in individual studies
To evaluate the risk of bias, the JBI Critical Appraisal Checklist for Studies Reporting Prevalence Data [24], composed of nine items, was used. For each of the items, it was possible to check "yes", "no", "unclear", or "not/applicable". The methodological quality of the studies was categorized as low risk of bias (70% or more "yes" responses), moderate risk of bias (50%-69% "yes" responses), and high risk of bias (up to 49% "yes" responses).

Summary measures
The frequency of dermatological toxicity described by the studies was considered the primary outcome. The frequency was expressed as a percentage, which corresponds to the number of cases present among all patients treated with ICIs. We analyzed the toxicities with highest prevalence in the studies, allowing associative measures such as rash, pruritus, and vitiligo.

Synthesis of the results
The meta-analysis was performed using the Jamovi software, version 1.6, which offers a variety of statistical techniques [25]. The prevalence was estimated by the number of events out of the total sample. We considered the homogeneity of the studies in relation to the type of exposure and outcome. Heterogeneity was calculated using I 2 . An I 2 value greater than 50% was considered a substantial indicator of heterogeneity among the studies, and a random effect model could be used. When I 2 is less than 50%, a fixed effect model is recommended. Jamovi provides fixed and random effect models for each analysis, and we therefore chose between the two based on the I 2 values. Beside the I 2 values, we also considered the Confidence Interval and the p value. The level of significance was set at 0.05. Funnel plot was not considered, since it is a prevalence systematic review, and the graphic results could be non interpretability [26].

Study selection
In phase 1, 9,802 studies were identified in the seven electronic databases. Among the 7,484 articles that remained after the removal of duplicates, 218 were selected. Another five studies that were identified in the gray literature were added and continued to phase 2. No additional study was identified in the list of references of the included articles.
Among the 223 studies that followed into Phase 2, 184 articles were excluded (S2 File). Thus, 39 studies [27-65] that met the eligibility criteria were included in the qualitative synthesis. Among these, 35 studies were considered adequate for the meta-analysis. The process of identification, inclusion, and exclusion of studies is described in Fig 1.
Regarding the anti-PD-1 class, the drug nivolumab was evaluated as monotherapy at doses of 2 mg/kg and 3 mg/kg [50,54,59,60] and compared with chemotherapy [40,62]. Higher frequencies of the AEs were observed at higher doses. When compared with chemotherapy, it was observed that the group treated with nivolumab also had dermatological AEs more frequently than the chemotherapy group in both grades 1 and 2, and in grades 3 and 4. Pembrolizumab was evaluated as monotherapy at doses of 2, 10, and 200 mg [33, 49,58,61] and compared with chemotherapy [42].
Vitiligo. Vitiligo was observed only with the administration of the anti-PD-1 class. Studies evaluating monotherapy with nivolumab and pembrolizumab showed the occurence of vitiligo grade 1 and 2 [33, 49,58,61]. When in comparison with chemotherapy, the group treated with

PLOS ONE
nivolumab or pembrolizumab experienced a higher frequency of vitiligo than the group undergoing chemotherapy [40,42]. Other toxicities. Dry skin grade 1 and 2 was observed in patients using pembrolizumab alone [58], and in patients who were treated with pembrolizumab or nivolumab in comparison with chemotherapy [42,62]. Higher frequency of dry skin was observed on the group treated with immunotherapy.
Among the observational studies, the risk of bias was considered low, moderate, and high in one [54], two [32,53], and eight studies [34,36,39,49,50,52,57,65], respectively. The studies were considered to have moderate risk of bias, when it was not possible to evaluate whether the sample represented the population or harm [32,53], or whether the statistical analysis of the data was performed adequately [53]. Among the studies with high risk of bias, most inconsistencies referred to the uncertain description of the response rate or lack of its management [34,36,39,49,50,52,57,65], unrepresentative and unstratified samples [34, 39,49,50,52,57], lack of description of the statistical analyses [36], use of an unusual statistical analysis [39], or unclear adequacy of the statistical analysis [34].
The detailed evaluation of each study is presented in S3 File.

Synthesis of the results
Among the included 39 studies, 34 were grouped to perform a meta-analysis. Four studies were excluded, because they did not grade the AEs (n = 3) [36,37, 64,65] or because the reported ipilimumab AEs were caused by combined administration with chemotherapy (n = 1) [30]. We analyzed the dermatological toxicities which have been reported by two or more studies: rash, pruritus, vitiligo, dry skin, and erythema multiforme. Despite two or more studies had reported skin hypopigmentation and DRESS syndrome, it was not possible to perform the meta-analysis for these toxicities because they were not graduated [36,59] or were reported in different grades [36,55]. The heterogeneity among the studies that evaluated grades 1 to 4 rash, grades 1 and 2 pruritus, and grades 1 and 2 vitiligo was high (rash grades 1 and 2-I 2 = 96.93%, p = 0.001, CI = 0.20-0.31; rash grades 3 and 4-I 2 = 79.68%, p = 0.001, CI = 0.01-0.02; pruritus grades 1 and 2-I 2 = 87.21%, p = 0.001, CI = 0.21-0.27; and vitiligo grades 1 and 2-I 2 = 88.37%, p = 0.001, CI = 0.07-0.13). Therefore, we opted to use random effect models for the statistical analysis. The heterogeneity could be explained by the high number of included studies (n = 39), observed by the between-study differences regarding proportion of events.
The results of the meta-analyses for grades 1 and 2 rash, pruritus, vitiligo, erythema multiforme, and dry skin are shown in Figs 2-6. The results of the meta-analysis for grades 3 and 4 rash outcomes and pruritus are shown in S1 and S2 Figs, respectively.
The prevalence of grades 1 and 2 rash and grades 3 and 4 rash was 26% (Fig 2) and 2% (S1 Fig), respectively. The prevalence of grades 1 and 2 pruritus (Fig 3) and grades 3 and 4 pruritus were 25% and 1%, respectively (S2 Fig). However, the prevalence of grades 1 and 2 vitiligo was 10% (Fig 4). We could not evaluate the prevalence of grades 3 and 4 vitiligo because only one study [27] described this event. The prevalence of erythema multiforme grade 1 and 2 was 4% (Fig 5), and dry skin grade 1 and 2 was 4% (Fig 6). The studies did not describe higher grades of erythema multiforme or dry skin. For this reason, we could not evaluate the prevalence of these events in grades 3 and 4.

Discussion
Cutaneous AEs are often the first toxicities to occur with the use of ICIs. Despite being selflimited, these toxicities may lead to ICI dose interruption or treatment discontinuation [66]. Dermatological toxicities might be mediated by a shared antigen which is coexpressed by the tumor cells and dermoepidermal junction [10,12].

PLOS ONE
Furthermore, cutaneous AEs are related to T-cell activation mediated by the blockade of PD-1 (or the PD-L1 ligand) and CTLA-4 receptors as well as the cross-reactivity between T cells directed against a tumor and T cells directed against normal tissue antigens [67]. Several hypotheses investigate factors that influence the risk of irAE development, including genetic factors, cytokines, and the composition of the patient's gastrointestinal microbiological flora [11].
The most frequent cutaneous AEs identified in this meta-analysis were grade 1 and 2 pruritus (24%), grade 1 and 2 rash (21%), grade 1 and 2 vitiligo (10%), grade 1 and 2 erythema multiforme (4%), and grade 1 and 2 dry skin. There was a higher prevalence of grades 1 and 2 compared to grades 3 and 4 of these AEs. According to Postow [68], approximately 50% of the patients treated with ICIs, mainly ipilimumab, may develop pruritus and rash. Pruritus is considered the most frequent AE reported by patients treated with ICIs. The prevalence of pruritus is high with ipilimumab administered alone or in combination with other ICIs and, usually appears concomitantly with rash; however, pruritus may also precede the rash or even appear with intact skin [67,69]. Like pruritus, rash is also one of the most frequently observed AEs in patients treated with ICIs [69] and, often start after few treatment cyles occuring mainly on the trunk and extremities [70].
Erythema multiforme (EM) is not a very commom AE observed in patients undergoing immunotherapy. However, there are some case reports showing the development of EMrelated to the administration of nivolumab and ipilimumab [75,76] that required immunotherapy discontinuation, and corticosteroid treatment. Nevertheless, the studies in this review reported grade 1 and 2 EM which did not lead to treatment discontinuation. Meantime, dry skin is a commom AE related to both chemotherapy and immunotherapy. It is important to educate patients on the use of moisturizers to prevent itching.

PLOS ONE
Combined anti-CTLA-4 and anti-PD-1 therapy is associated with the development of more frequent cutaneous irAEs of greater severity with an earlier onset than monotherapy with ICIs [60]. Combination treatment with nivolumab and ipilimumab reportedly exacerbates their AEs and triggers high rates of grades 3 and 4 AEs. As a result, patients may repeatedly seek emergency care and may need hospitalizations and systemic immunosuppression. Further, the combination treatment has a toxicological profile that combine the side effects of both agents [63][64][65]. Although this combination is effective for cancer treatment, the high irEA rates are concerning [67].
Therefore, the management of adverse events requires early recognition, in addition to monitoring and classifying the grade of toxicity. The early identification of these events is critical to define the most appropriate intervention, such as treatment with the use of corticosteroids, temporary or permanent interruption of the use of ICI [80].
High-degree AEs may be potentially fatal; thus, patients should be carefully evaluated for symptoms consistent with Stevens-Johnson syndrome, pemphigus or toxic epidermal necrolysis [81]. In our review, one study identified Stevens-Johnson syndrome [52], classified as grade 4, affecting only one patient. Pemphigoid lesion grade 4 was also reported by one study [55], affecting only one patient. Though was not possible to perceive wheter these toxicities led to treatment discontinuation. However, toxic epidermal necrolysis was not listed in any of the studies. Both events are considered dermatological emergencies with high morbidity power which require an immediate intervention, hospitalization, and treatment discontinuation [82].
The frequency of irAEs increases with treatment exposure and requires long-term monitoring. Patients receiving PD-1 inhibitors may present with cutaneous AEs up to one year after the initiation of treatment [83]. Thus, follow-up periods longer than 12 months would be ideal to identify the incidence of irAEs. Overall 14 studies included in our review presented followup periods ranging from 1 to 5 years, which is considered an adequate time for the identification of irAEs [30,35,38,39,44,47,48,51,55,56,[60][61][62][63].
Further, authors did not always report the irAEs in the most appropriate manner, since important information such as the classification [36,65] and type of scales used to classify the AEs [27,36,65] was often not reported.
A positive aspect of the irAE report was the use of the Commom Toxicity Criteria for Adverse Events (CTC-AE) scale [28-35,37-64] to classify and grade irAEs by most of the studies included in this review. The CTC-AE scale was developed to report AEs manifested by patients participating in oncological clinical trials [84]. The use of the same scale by most of the studies allowed the standardization of the irAEs reported in this review. However, Spain & Larkin [79] point out that even though the CTC-AE scale is a good scale that can also be used in clinical practice, it may lead researchers and professionals to underestimate some of the irAEs, such as pituitary gland dysfunction, because it is a scale created for a pre-immunotherapy time.
Other dermatological manifestations were identified in the studies included in this review such as maculopapular exanthema, erythema multiforme, dermatitis, acneiform rash, lichenoid exanthema, folliculitis, rosacea, eczema, leukoderma, seborrheic dermatitis, and alopecia. The majority of them were classified as grade 1 and 2.

Limitations
The heterogeneity of the samples, lack of grading of the AEs in some studies, and unknown time of onset of the AEs may have impaired the analysis of the outcomes. It was not possible to identify subsets of patients with high possibility of displaying cutaneous side effects. Thus, future studies should classify and grade the immunomediated AEs accurately, present the time of onset of the manifested AEs, and report the management and reversibility of the described AEs.

Conclusion
The results of this review and meta-analysis show that the most prevalent irAEs are pruritus and rash. Even though mild and moderate irAEs were reported more frequently than severe irAEs in the included studies, there was also a significant representation of more severe AEs. Grade 3 or 4 irAEs have been associated with the use of ipilimumab. Although it is possible to manage these AEs in most cases, early identification plays a key role in the prevention of severe cases, avoiding treatment interruption.