Identification of TNF-α as Major Susceptible Risk Locus for Vitiligo: A Systematic Review and Meta-Analysis Study in the Asian Population Free

Vitiligo is a common complex skin depigmentation disorder with a prevalence rate of ∼0.5–2% worldwide, often leading to social discrimination and a subsequent psychological threat. The highest incidence is reported from India (8.8%), followed by Mexico (2.6–4%) and Japan (1.68%) [1]. A large cohort-based study from the USA diagnosed 27.49% vitiligo patients with mental health disturbances, with a significantly higher prevalence rate in female patients [2]. Similar studies conducted from Chandigarh and Gujarat in India have, however, reported male preponderance [3]. Familial studies among Caucasian and Koreans have reported an elevated risk of developing vitiligo in members of affected families with an overall frequency of 7% and 6.2%, respectively, suggesting genetic component as a major contributor to disease pathogenesis [4]. In a population-based study conducted with the German adult population that included a working cohort of 121,783 samples (57% male, mean age 43 years), vitiligo has been found to be significantly more common in people with fair skin type [5]. In India, a multicentric case-control study has reported 3,962 vitiligo cases, where family history, consanguinity, hypothyroid disorders, and depressed mood were significantly (p < 0.001) higher among the cases. Around 46% of the cases reported the disease onset age to be <20 years [3].

There are four subtypes of vitiligo: non-segmental vitiligo with a bilateral symmetry onset; segmental vitiligo (SV) with a unilateral dermatomal distribution; mixed vitiligo where both non-segmental vitiligo and SV co-exist; and unclassifiable vitiligo which includes focal vitiligo, vitiligo punctate, follicular vitiligo, and vitiligo minor. The distinction between non-segmental and SV is crucial to prognosis and treatment [6].

Thus far, the aetiology of vitiligo has been found to be multifactorial, where autoimmune disease and melanocytorrhagy have been considered important factors contributing to the disease, with evidence of hyperactive inflammatory dendritic cells in the patients [7]. Factors like sunburn, oxidative stress, chemical trauma, agricultural and industrial pollution, intrinsic defects in melanocytes eventually activate the autoinflammation system, leading to melanocytic death [8]. Reports have shown that a patient with an autoimmune disease is more prone to develop vitiligo. Previous studies characterizing the associated comorbidities reported 61.8% patients with vitiligo had history of at least 1 atopic disease [9], 7.7% had psoriasis [10], 6.4% had thyroid disease [11], 54% had autoimmune disease, and 46% had other rheumatic diseases including degenerative, metabolic, and other diseases [12]. Interestingly, a study identified another internal risk factor of cancer development in patients with autoimmune and chronic inflammatory diseases like Vitiligo [13]. Notably, a 73-year-old patient with late onset of vitiligo developed oesophageal cancer following the diagnosis of vitiligo [14]. Additionally, two cases of vitiligo have been reported to be associated with breast cancer [15].

It has been presumed that a complex genetic architecture of DNA variants may contribute to vitiligo risk. Linkage analysis performed in patients with generalized vitiligo has proven the involvement of different target genes and varied susceptibility loci co-segregating in different populations, characterizing it as a typical polygenic disease [16]. GWAS and candidate gene-based association studies have also identified several underlying genes and risk variants. Since 2010, nine GWAS have been performed in patients with vitiligo, including various ethnic populations, confirming genetic associations of around 55 susceptible loci [17‒22] including both common and non-overlapping variants across populations. Interestingly, several susceptible risk loci associated with vitiligo have also been found to be associated with other autoimmune diseases, suggesting co-existing molecular pathways. Several candidate gene association studies on relatively smaller sample size compared to GWAS have often reported contradictory results of association, thus increasing the risk of false positive associations. For instance, while a study performed on the Saudi and Turkish populations shows strong association of rs1800896 (A/G) of IL-10 and rs1800925 of IL-13 with vitiligo risk, the same polymorphisms showed no signs of association in the South Indian population [18]. In a population-based cohort study from North India, conducted on 100 patients and 160 healthy controls, rs1799930, rs1799931, and rs1799929 of NAT gene showed significant association with vitiligo, whereas another study from North India [19] and one from Turkey [20] showed otherwise. A study conducted on 1,000 Iraqi individuals showed no statistically significant association between rs1800629/TNF-α G>A and vitiligo, in opposition to two other studies that found patients with vitiligo to have higher levels of cytokine, corresponding to rs1800629-A allele [21] than the cases. Studies in the Jordanian Arab population showed that PTPN22 rs2476601C>T was not associated with an increased risk of vitiligo, whereas the reverse is true in Caucasians and Indians [22]. The discrimination in genetic association study on the same variant could be related to small sample size, population stratification artifact [23], racial and/or ethnic differences, clinical and environmental factors, and genetic heterogeneity between populations [18]. Interestingly, great many numbers of studies have been conducted in Asia compared to the other continents. To overcome the contradictory outcomes of candidate gene association studies, meta-analysis may serve as an important method to assess the combined effect of the ambiguous candidate associations on vitiligo pathogenesis in the Asian population.

In addition, in silico approaches using various web tools could provide insight into the biological significance of the statistically associated variants with vitiligo pathogenesis. The purpose of this study was thus to perform a meta-analysis to determine the overall effect of the reported candidate associations with vitiligo in the Asian population and to characterize the significant variant(s) based on in silico analysis. It would thus provide a plausible biological mechanism behind the statistical associations in vitiligo risk assessment, providing insight into the actual effect of variants reported on vitiligo risks.

The study was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines. A systematic and comprehensive search was done to identify the most relevant studies in PubMed using the following keywords: Vitiligo polymorphism/polymorphisms association Asia/India/China/Jordan/Kazakhstan/Iraq/Turkey/Japan/Iran/Taiwan/North Korea/SouthKorea/Indonesia/Pakistan/Bangladesh/Philippines/Vietnam/Thailand/Myanmar/Afghanistan/Saudi Arabia/Uzbekistan/Malaysia/Yemen/Nepal/Sri-Lanka/Syria/Cambodia/Azerbaijan/United ArabEmirates/Tajikistan/Israel/Laos/Lebanon/Kyrgyzstan/Singapore/Oman/Palestine/Kuwait/Georgia/Mongolia/Armenia/Qatar/Bahrain/Timor-Leste/Cyprus/Bhutan/Maldives/Brunei. Only the case-control studies comprising genotype specific data were included for further analysis.

The studies were included in the meta-analysis based on the following inclusion criteria: (1) studies reported results on vitiligo risk for affecting the Asian population; (2) for each variant, case-control genotypes were reported; (3) studies were full research articles or review articles presenting new results; (4) studies were published in English. We included studies till January 2023. Exclusion criteria were as follows: (1) review articles, letters, editorials, comments, and case reports that did not include new data or complete information; (2) repetition of the similar studies conducted on the same population; (3) studies without genotype counts among cases or controls; (4) studies in languages other than English; (5) unpublished data; and (6) systematic reviews based on previously published data and meta-analysis.

However, we have not found any studies in languages other than English or those bearing unpublished data. For our study, however, variants reported in three or more independent studies were considered. The complete search protocol is provided in Figure 1.

For the meta-analysis, the following data were extracted from the studies: (1) first author; (2) year of publication; (3) mean age with standard deviation; (4) age of onset (early/late); (5) sex: male/female; (6) progressive/active or stationary/stable vitiligo; (7) family history of vitiligo; (8) presence of autoimmune disease, if any; (9) autoimmune diseases in family history; (10) comorbidities; (11) vitamin B12 level; (12) folate level; (13) homocysteine level; (14) haptoglobin level; (15) previous treatment regime; (16) extent of lesions, if any; (17) mean duration of the disease with standard deviation; (18) vitamin D intoxication (VDI); (19) vitamin D level; (20) segmental/non-segmental/generalized/localized/focal/acrofacial/vulgaris/universal/mixed vitiligo; (21) anti-melanocyte antibody level; (22) presence of leukotrichia; (23) auto-antibodies status; (24) mean age greater than 20 or not; (25) short or long course of vitiligo development; (26) drinking status; (27) smoking status; (28) cytokine level: TNF-α level, IL-6 level, IL-10 level; (29) if they are newly diagnosed cases; (30) enlargement of macules, if any; (31) repigmentation; (32) presence of associated disease: obesity, stress, sunburn; (33) vitiligo disease activity score (VIDA); (34) vitiligo area scoring index (VASI); (35) vitiligo extent intensity score (VETI); (36) if patients are from rural or urban background; (37) drugs used anti-hypersensitive drugs, cholesterol-modifying drugs, anti-diabetic (oral/insulin), anti-rheumatic drugs, allergic drugs, anti-thyroid drugs, oral-contraceptive drugs; (38) Fitzpatrick skin type 1/2/3/4; (39) genotype-specific case-control data.

For each continuous variable, like average age, age of onset, etc., we extracted mean value and SD value from each study for continuous variables. For each categorical variable, we converted it into multiple binary variables and recorded the percentage values of each category. For example, “associated disease” was coded as obesity (yes/no), stress (yes/no), and sunburn (yes/no), and the corresponding percentages were recorded. The details of the covariate data extracted can be found in online supplementary file SFI (for all online suppl. material, see https://doi.org/10.1159/000536480).

For each variant, the study-level unadjusted odds ratio (OR), standard errors (SE), and 95% confidence intervals (95% CIs) were calculated by using logistic regression with additive (see online suppl. file SFII.A), dominant (see online suppl. file SFIII.A), and recessive effect (see online suppl. file SFIV.A) models (using R function “glm”). Covariates, such as autoimmune diseases, vitamin D level, extent of lesion, anti-melanocyte antibody, autoantibody, family history of vitiligo, could not be adjusted since some selected studies did not provide sufficient data.

A meta-analysis was conducted using R software package (R, 3.4.2) [https://www.R-project.org/.] “metafor” on vitiligo genetic association reports from populations of Asia. For combining study-level estimates (using the R package “metafor”) [24], both fixed-effect (FE) using inverse-variance weighting and random-effects (RE) meta-analysis were performed using the DerSimonian-Laird estimator of heterogeneity [25]. Logistic regression of vitiligo status on variant genotypes was used to derive study-level ORs and SEs. Both FE and RE models were used to determine the overall evidence of statistical association of the reported variants with vitiligo risk. Benjamini-Hochberg method was used to correct multiple testing and significance was assessed at a false discovery rate (FDR) less than 5% level (p_FDR < 0.05) [26]. All variants were selected according to their p values in the FE meta-analysis. However, for variants with significant heterogeneity, the summary estimates from the RE model are more reliable. Inter-study heterogeneity was evaluated using Cochran’s Q test (pHet < 0.1) and heterogeneity index (I²) which measures the degree of inconsistencies across studies using the formula: I² = (Q − (n − 1) n)/Q × 100%, where n is the number of studies selected [27, 28]. According to the I² value, a grade cut-off of 25%, 50%, or 75% indicates the presence of low-, mid-, or high-grade heterogeneity (see online suppl. files SFII−SFIV).

To study the association of the eleven polymorphic variants in more homogeneous strata, subgroup analysis was performed based on various covariates, like generalized vitiligo versus localized vitiligo, male versus female, segmental versus non-segmental stratified data, etc. In order to generate study-level summary data, logistic regression (generalized linear model; “glm”) of disease status on subgroup-specific genotype counts was done. A meta-analysis was conducted within each subgroup, using the methods outlined earlier. Finally, a fixed-effect meta-regression to test for effect modification (interaction) by subtypes of vitiligo was performed. For this, vitiligo subtypes were used as a moderator variable using the “rma.uni” function in R package “metafor.”

A visual inspection of funnel plots as well as Egger’s regression test [29] were performed to evaluate the asymmetry of funnel plots (p < 0.05) for the estimation of publication bias, if any, among the selected studies. A funnel plot should have the X-axis corresponding to the study effect, and the Y-axis corresponding to the SE. Depending on the sample size, a more scattering distribution is seen with smaller samples, while a more concentrated distribution is seen with larger samples. Egger's regression test was used to quantify the funnel plot asymmetry (p < 0.05). If the funnel plot is symmetrical, it indicates no publication bias; if it is asymmetrical, it indicates publication bias. For three or more studies, Egger’s test can detect publication bias effectively. In this study, Egger’s regression test was used to determine funnel plot asymmetry using a weighted regression model with multiplicative dispersion. All eleven variants were evaluated based on three different genetic models: additive, dominant, and recessive. Four studies were not used as sufficient genotype data were not available for the recessive model.

In silico functional analysis of vitiligo-associated variant(s) obtained through the meta-analysis were conducted by functional data-based and independent algorithm-based freely available web tools in order to understand their likely biological function in vitiligo pathogenesis. As will be clarified in the Results section, we found only one intergenic single nucleotide polymorphism (SNP) to be associated after the meta-analysis. Thus, the regulatory potential of the statistically significant variant rs1800629 was examined using rSNPBase 3.1 and RegulomeDB, followed by assessment of linkage disequilibrium (LD) analysis using HaploReg v4.1. The regulatory potential of the LD variants was also analysed, and they were found to be regulatory using rSNPBase 3.1. GTEx Portal evidence was used to infer the skin tissue-specific regulatory function(s) of the noncoding SNVs. We have also tried to annotate their functional contribution to vitiligo pathogenesis based on functional data available in the public domain database.

Systematic mining of the databases with the search strings mentioned above revealed 108 hits, from which 31 studies were included for meta-analysis following the specific inclusion/exclusion criteria set for the proposed study. These 31 studies included 11 polymorphisms associated with 6 genes and a total of 8,911 cases and 10,320 controls (online suppl. Table a). Covariate-specific case-control data, particularly vitamin B12, vitamin D, mean age, homocysteine level, folate level, and geographical region of the subjects, were recorded (online suppl. file SFI).

In the fixed-effect meta-analysis of the 11 variants, rs1800629/TNF-α was found to be associated with vitiligo risk significantly (p < 0.01), as shown in Table 1 in additive, dominant model and recessive model viz GG versus. AG versus. AA; OR 2.166; 95% CI: 1.5584–3.0129; p = 4.26E−06; GG versus. GA + AA; OR 2.590; 95% CI: 1.8138–3.6997; p = 1.65E−07; AA versus. AG + GG; OR 3.232; 95% CI: 1.6779–6.2272; p = 0.0004). The forest plot of the significantly associated variants is shown in Figure 2. After Benjamini-Hochberg FDR correction, rs1800629/TNF-α was found to be significant at 5% FDR in dominant and recessive models (p < 0.05). In order to estimate heterogeneity among studies, Cochran’s Q test and Heterogeneity Index (I²) were used (Tables 1, 2).

From the forest plot of rs1800629, it was found that most of the signal was driven by the point estimate of a single study [30], as denoted by the CI intervals (Fig. 2), and it is largely different from the point estimates of the remaining six studies. Omitting this study, the analysis also revealed rs1800629 to hold a significant association (p = 0.01) with a reduced occurrence of heterogeneity (I² < 20%).

The p_Het value for rs1800629 in additive model (Fig. 2a), dominant model (Fig. 2b), and recessive model (Fig. 2c) were 0.084368, 0.007334, and 0.000343, respectively. Based on I² metric, the polymorphic variant rs1800629/TNF-α is found to exhibit moderate grade heterogeneity (I² < 50%) among studies in the additive effect model only (online suppl. file SFII.C). Whereas, based on I² metric, the polymorphic variant rs1800629/TNF-α is found to exhibit medium-grade heterogeneity (I² > 50%) among studies in the dominant effect model and recessive effect model (online suppl. files SFIII.C and SFIV.C).

For each variant, funnel plots were constructed to assess publication bias. According to visual inspection of the plots (Fig. 2), none of the studies exhibit publication bias. Furthermore, Egger’s regression test can be used to assess funnel plot asymmetry (critical p < 0.05) which revealed that none of the variants show any evidence of publication bias (Table 3).

Stratified subgroup analysis could not be performed as no sufficient genotype data were available for other covariates among the case-versus-control studies. But subgroup analysis was done among the vitiligo cases based on the covariates using male versus female, and different clinical subtypes like focal vitiligo and so on. No significant effect modification was obtained due to the smaller number of studies available (online suppl. file SFV).

Out of fifty four covariates, overall mean age of vitiligo patients has shown a significant association with vitiligo patients (OR: 1.78, CI: 1.66–1.90, p value = 9.27E−61) (online suppl. Table b). But stratification of genotype based on the covariates was not done due to lack of sufficient data on Asian populations.

SNP rs1800629 that was found to be associated with vitiligo through our meta-analysis, was further assessed in silico to determine the probable biological mechanism behind such statistical associations. The 5′ upstream variant, rs1800629/TNF-α, was submitted in rSNPBase 1.0 and was found to be a potentially regulatory single nucleotide variant (online suppl. file SFVI). RegulomeDB analysis revealed rs1800629, located 321bp upstream of TNF-α, is likely to affect binding of the transcription factor SP1, altering polymorphic variant with a score of 1d (see online suppl. file SFVI). HaploReg and rSNPBase 3.1 revealed seven variants to be in LD with our query SNP rs1800629 (see online suppl. file SFVI). By using eQTL based evidence from GTEx Portal, we tried to ascribe skin tissue-specific regulatory role(s) to both the significant variants, if any. We identified fifteen regulatory target gene[s]/loci targeted by rs1800629 and expressed in skin-specific tissue with m value >0.9 (keratinocyte, melanoma, fibroblast, sun-exposed skin, non-sun-exposed skin, as obtained from GTEx RNA-seq data (see online suppl. Table c). The risk allele rs1800629-A showed association with higher level of circulatory cytokine TNF level and this is already known that TNF-α gene expression is likely to increase in vitiligo patients [30‒32]. The clinical significance from dbSNP has also shown rs1800629-A allele results in increased susceptibility to psoriatic arthritis, leading to the formation of erythematous hypopigmented lesions, which often form the basis of vitiligo. A summary of the in silico functional assessments of the polymorphic variant rs1800629 is presented in online supplementary file SFVI.

Vitiligo is the most common hypopigmentary complex disorder that causes progressive melanocyte loss. In almost 50% cases, vitiligo starts before the age of 20 years [32], and it affects men and women equally [33]. Genetic predisposition has been known to contribute to vitiligo pathogenesis, and it has also been associated with autoimmune diseases like autoimmune thyroid disease and type 1 diabetes [34]. However, a complete understanding of the genetic factors involved in vitiligo is yet to be elucidated.

Our study presents the first meta-analysis of candidate gene-based vitiligo association studies in the Asian population conducted on 11 variants of 6 genes across 31 studies from 48 different countries of Asia. The study aimed to evaluate the combined effect of each variant on overall and covariate stratified vitiligo risk in Asia. Interestingly, out of the six genes selected for meta-analysis, CTLA4 [35], TNF [36], VDR [37], APEX1 [38], ACE [39] have been implicated in regulation of immune response leading to the development and/or progression of autoimmune disease. MTHFR [40] and VDR [41] are required for physiological responses to oxidative stress mediated by reactive oxygen species. Vitiligo pathogenesis may be influenced by immunological pathways and oxidative stress mechanisms. Even under high-stress conditions, genes susceptible to various inflammatory responses and autoimmune diseases are involved in maintaining T cells, thereby preventing disease progression [42]. In addition to causing oxidative stress and persistent inflammation, some genes of the immunomodulatory pathway act as autoantigens and could contribute to vitiligo [43]. The reports suggest that oxidative stress-induced unfolded protein responses in the endoplasmic reticulum connect oxidative stress with autoimmunity in vitiligo [44]. Following the inclusion criteria, detailed text mining revealed that oxidative stress regulator genes and immune-responsive genes are associated with vitiligo risk.

The meta-analysis revealed a TNF-α SNP rs1800629G>A to have a strong risk association with vitiligo pathogenesis in Asian population. TNF-α is a proinflammatory melanocyte-inhibiting cytokine which plays a central role to many autoimmune diseases. A case-control study reported that vitiligo patients were diagnosed with hypertension and stress due to elevated catecholamines more often than the normal population, and catecholamines released in excess can damage melanocytes [45]. Increased TNF-α transcripts in the lesional and non-lesional skin are also well documented in vitiligo patients [46]. It can up-regulate intercellular adhesion molecule-1 (ICAM-1) levels on the melanocyte cell surface, thereby enhancing T-cell/melanocyte adherence to the skin that results in melanocyte destruction [47]. In fact, that TNF-α secretion promotes the cytotoxic T-cell and T_H1-cell-mediated melanocyte destruction leading to the depigmentation process in vitiligo has been reported [48]. Studies have also suggested the anti-inflammatory role of TNF-α in vitiligo [36]. In vivo, TNF-α promotes proliferation of T-regulatory cells, which suppress the activity of autoreactive T cells against melanocytes in vitiligo [49]. Thus, in patients with vitiligo, TNF-α levels have been found to be dramatically increased. TNF-α is also one of the several paracrine factors secreted by the keratinocytes that act on melanocytes and inhibit melanocyte proliferation and melanogenesis [50]. Additionally, TNF-α inhibits the development and differentiation of melanocyte stem cells and is known to suppress tyrosinase and its associated proteins, thus pulling a leash on melanogenesis [48].

There is conflicting evidence regarding the association between TNFA-308(G > A) and vitiligo susceptibility [21, 30, 51‒57]. Variations in ethnicity and smaller sample sizes may have contributed to the difficulty in conclusively establishing a genetic association between TNFA-308(G > A) SNP and vitiligo susceptibility risk [58, 59]. In light of recent associations between TNFA-308(G > A) SNP and vitiligo susceptibility, meta-analysis should be performed to enhance statistical power and elucidate the exact role of this SNP in vitiligo susceptibility [21, 56, 60]. The variant rs1800629-A allele that has been identified as a genetic risk factor for vitiligo susceptibility in our meta-analysis, has been associated with a significantly higher serum TNF-α level compared to the G allele in Gujarat [30], Egyptian [56], South Indian Tamil [52], and Iraqi population too [21]. The A allele of rs1800629 has been found to confer higher risk of developing vitiligo compared to the controls in the Iranian population [53], Egyptian population [56], South Indian Tamil population [52, 60]. Interestingly, patients with active vitiligo had higher plasma TNF-α levels than patients with stable vitiligo [61]. In macrophages, rs1800629-A allele has been found to significantly increase the level of TNF transcription, contributing to several human disease susceptibility such as autoimmune hepatitis, type 1 and type 2 diabetes mellitus, acne vulgaris, psoriasis, and Sjogren’s syndrome [56].

It was intriguing that the seven LD variants with our query SNP rs1800629 were also found to have regulatory potential using rSNPBase. When each LD variant was investigated separately for its role in vitiligo, none were found to be associated directly with vitiligo pathogenesis. However, rs2734583 showed association with epidermal necrolysis, and a higher level of serum TNF-α has been reported in patients with toxic epidermal necrolysis, whereas TNF-α gene expression is likely to increase in vitiligo patients. It would be important, thus, to find out if rs2734583, being in LD with our prioritized significant variant, might influence vitiligo risk in various populations and if rs1800629 has any impact on epidermal necrolysis.

Interestingly, out of eighteen regulatory target gene[s]/loci that were found to be in cis-eQTL from GTEx for rs1800629, eight loci (PRRC2A, ZBTB41, MCCD1, RPL15P4, SNORD117, NCR3, NFKBIL1, and ENSG00000201207) are known to play significant role in the regulation of the immune response, complement system, and autoimmune disorder [62‒65]. CRB1 mutation has been found to be associated with severe form of retinitis pigmentosa [66], and LTB regulates our target gene TNF cytokine signalling pathway (see online suppl. Table c).

Covariate-based subgroup analyses for the majority of the covariates were not feasible due to lack of sufficient data in the selected studies. Our results indicate significant association of overall age with the vitiligo patients in all three models as discussed above, but stratification of genotype based on age was not done due to lack of sufficient data on the Asian population.

Apart from the statistical heterogeneity, there is considerable admixture among different ethnicities in Asia, which might lead to heterogeneity between the studies, modifying the risk signatures. In fact, a meta-analysis of rs1800629 across the world population also indicates a strong association with vitiligo risk [67]. Despite limitations, our study reveals insights into fundamental biological pathways involved in vitiligo risk by identifying genes and loci associated with vitiligo in the Asian population and corroborating results from previous studies.

In summary, we have performed a meta-analysis to clarify the complex nature of vitiligo by determining the pooled effect of various vitiligo-associated risk loci on disease phenotype. Along with the statistically significant genetic variant rs1800629, we identified other variants that were found to be in LD with rs1800629 that might have the potential to modulate vitiligo development directly or indirectly. In our current study, we functionally annotated the relevant significant SNVs associated with vitiligo using public datasets which would enrich the current molecular diagnosis knowledge for vitiligo. A meta-analysis could become a very important tool to estimate the true effect of these variants on vitiligo, considering covariate stratifications of the population, when more studies are conducted on the genetic association with vitiligo on different populations and subpopulations with contradictory outcomes. For a study of this kind to be more statistically accurate and powerful, more analyses of the global population are needed, along with covariate-adjusted summary data from individual studies.

An ethical statement of approval is not applicable because this study is based exclusively on published literature.

The authors declare no competing interests.

We would like to thank the Department of Science and Technology-Promotion of University Research and Scientific Excellence (DST-PURSE), Government of India, for providing funds to the University of Calcutta for academic infrastructural facilities. Tithi Dutta and Arpan Saha are supported by Senior Research Fellowship from University Grant Commission [UGC], Government of India. We have not received any extramural funding for the preparation of the data or the manuscript.

T.D., S.B., and M.S. conceptualized and designed the study. S.S., S.A., S.N., and M.S. contributed to the literature searches and data curation. S.A. and A.S. contributed to the data arrangement and statistical calculations. R.M. and D.S. also helped in the study design. T.D. drafted the manuscript with important intellectual contributions from S.B. and M.S. S.B. and M.S. critically revised the manuscript, figures, tables, and supplementary data and supervised the entire work. All authors provided critical feedback and helped in shaping the research, analysis, and manuscript. None of the authors has received extramural funding for the preparation of the data or the manuscript.

All data generated or analysed during this study are included in this article and its supplementary information files. The additional raw input data files will be available from the corresponding authors on request. The R scripts/codes used for the analysis will be available from the corresponding author on request.