Moving from nature to nurture: a systematic review and meta-analysis of environmental factors associated with juvenile idiopathic arthritis

Abstract Objectives JIA is the most common paediatric rheumatic disease, thought to be influenced by both genetics and the environment. Identifying environmental factors associated with disease risk will improve knowledge of disease mechanism and ultimately benefit patients. This review aimed to collate and synthesize the current evidence of environmental factors associated with JIA. Methods Four databases (MEDLINE, Embase, Web of Science and Cumulative Index to Nursing and Allied Health Literature) were searched from inception to January 2020. Study quality was rated using the Newcastle-Ottawa Scale. Pooled estimates for each environmental factor were generated using a random-effects, inverse-variance method, where possible. The remaining environmental factors were synthesized in narrative form. Results This review includes 66 environmental factors from 39 studies (11 cohort and 28 case-control studies) over 45 years. Study sample sizes ranged from 41 to 1.9 million participants. Eight environmental factors from ten studies were meta-analysed. Caesarean section delivery was associated with increased JIA risk [pooled odds ratio (OR) 1.11, 95% CI: 1.01, 1.22]. Conversely, presence (vs absence) of siblings (pooled OR 0.60, 95% CI: 0.44, 0.81) and maternal prenatal smoking (pooled OR 0.70, 95% CI: 0.58, 0.84) were associated with decreased JIA risk. Conclusion This review identifies several environmental factors associated with JIA and demonstrates the huge breadth of environmental research undertaken over five decades. We also highlight the challenges of combining data collected over this period due to limited between study comparability, evolution in healthcare and social practices, and changing environment, which warrant consideration when planning future studies.


Introduction
JIA is the most common rheumatic disease of childhood with a pooled prevalence of 32.6/100 000 in Caucasians [1]. International consensus criteria currently define seven subtypes of JIA based on their clinical features [2]; however, diagnostic criteria have evolved over the years [3]. JIA is considered a complex disease influenced by both non-Mendelian genetics and the environment. The genetic contribution to JIA, examined both in familial studies and in genome-wide association studies, has been estimated to be between 13% and 32% [4][5][6][7]. A substantial contribution to disease risk is likely attributed to gene-environment interactions and environmental factors.
Studies of other autoimmune diseases demonstrate the potential for environmental exposures to induce epigenetic changes [8], modulate the immune system [9] or alter the microbiome [10]. While the role of environmental factors in JIA is less clear, it is likely that similar mechanisms influence JIA risk. Two narrative reviews of environmental factors associated with JIA have been published in recent years [11,12] that highlight several key limitations of research to date. Firstly, the availability of high-quality data in JIA is limited, often of modest sample size and lags behind that of other autoimmune conditions. Secondly, many environmental factors studied so far, such as breastfeeding, infection and smoking, require further scrutiny. Finally, several environmental factors, such as diet, remain relatively unexplored.
Identifying environmental factors associated with JIA, and quantifying their effects, has the potential to benefit patients by guiding research priorities and informing future studies probing causality and disease mechanism. Such developments could inform new therapeutic modalities and assist in patient counselling and risk stratification. Finally, environmental associations may be readily and/or feasibly modifiable, which could translate into lower disease burden within populations.
The aim of this systematic review and meta-analysis was to collate, quantify and evaluate the current evidence for environmental factors that influence JIA risk and highlight areas of unmet research need.

Methods
This review follows guidance from the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement [13] and the protocol was preregistered on PROSPERO (ID: CRD42017078306) [14].

Literature searching
Expert advice from a literature searching specialist was sought prior to designing the search strategy. A systematic search of MEDLINE (Ovid), Embase (Ovid), Cumulative Index to Nursing and Allied Health Literature (CINAHL, EBSCOhost) and Web of Science (WOS, Clarivate Analytics) was performed, with English language restriction, from database inception to 18 January 2020. The MEDLINE search strategy (Table 1)  was  amended  for  use  in  other  databases  (Supplementary Table S1, available at Rheumatology online). Bibliographies of excluded review articles were hand-searched to identify further relevant studies.

Study selection
All aspects of study selection and risk of bias assessment were undertaken independently by two reviewers (S.L.N.C. and either K.S.M. or I.M.). Any discrepancies were resolved by discussion then involvement of an additional reviewer (A.V.R.). References were downloaded into Endnote (version X9.3.2, Clarivate Analytics) and de-duplicated as recommended [15]. Blind title and abstract screening was performed using Rayyan (QCRI) [16] against the inclusion/exclusion criteria ( Table 2). The full texts of potentially relevant studies were retrieved and further reviewed against the inclusion/exclusion criteria.

Data extraction
Data items (Supplementary Table S2, available at Rheumatology online) were extracted from each included study by S.L.N.C. into a predesigned template. This was independently checked by either K.S.M. or I.M., with involvement of an additional reviewer (A.V.R.) where necessary. Where a study included multiple case or control groups, data was extracted from (i) the most matched case and control groups and (ii) community

Risk of bias (quality) assessment
We used the Newcastle-Ottawa Scale (NOS) for casecontrol and cohort studies to assess the risk of bias and methodological quality of included studies [17]. The scale scores three study domains (selection criteria, comparability and outcome) with a maximum score of nine indicating the highest quality. A risk of biases table was created, summarizing study quality.

Data synthesis and statistical analysis
Data was analysed using the statistical software R (version 3.6.1) in RStudio. The data was synthesized into ten environmental factor 'domains' (maternal, paternal, perinatal, early life, smoking, environmental, socioeconomic, living environment, dietary and infections). For consistency, unadjusted odds ratios (OR) and 95% CI were derived for all variables where sufficient raw data was available. To aid synthesis, the data was condensed in two ways. Firstly, although all relevant data reported by a study was extracted, only the most inclusive evidence for each environmental factor from each study was included in our synthesis, as jointly determined by S.L.N.C. and A.V.R. For example, sexstratified or JIA subtype specific analyses were not included in the narrative synthesis where a mixed-sex or pan-JIA subtype analysis, respectively, was reported. Secondly, a single point estimate with 95% CIs was reported for each variable from each study. This consisted of, in order of priority and subject to data availability, the most adjusted point estimate, the derived unadjusted estimate or the study reported unadjusted estimate. Our complete dataset is included in Supplementary Table S3, available at Rheumatology online, illustrating the data inclusion decisions. Forest plots were constructed summarizing a single point estimate with 95% CI for each environmental factor. Pooled ORs were generated for environmental factors examined in multiple studies in a comparable manner (clinically and statistically). Where possible, ordinal categorical data was converted to a single estimate using generalized least squares of trend [18] implemented in the glst module [19] in Stata (version 14.2) prior to meta-analysis (Supplementary Data S1, available at Rheumatology online). Only studies reporting adjusted estimates were included in meta-analysis. All meta-analyses used the inverse weighted variance method in the meta R package using a random effects model. Statistical heterogeneity was assessed using Cochran's Q test (v 2 test) and Higgin's I 2 value [20].

Study selection
Literature searches identified 5933 unique database records (Fig. 1A). The majority (5731/5933) were excluded during title and abstract screening. Of the 202 full-text records examined, 39 studies met the prespecified inclusion criteria.

Paternal factors
Compared with maternal factors, there was little data regarding paternal factors associated with offspring JIA (Supplementary Fig. S1B, available at Rheumatology online). Two studies [27,28] examining paternal age found no association with JIA. One study [27] found increased JIA risk associated with paternal civil status (OR 2.48, 95% CI: 1.06, 5.8), consistent with the corresponding maternal estimate.      Evidence from single studies suggests that ideal or less than ideal maternal weight gain during pregnancy [32] and Apgar score <6 [26] at delivery are negatively associated with JIA. One study [38] reported a distinct pattern in month of birth of JIA patients (with peak in November to March) vs the general population but this was unsupported by other studies examining season of birth or pregnancy [26,32,37]. There was little evidence of association between JIA and multiple birth [26,30], neonatal vitamin D levels [29] and congenital malformation [26].

Early life factors
Five studies [26,28,37,39,40] examined the association between JIA and sibling-related factors ( Supplementary Fig. S1D, available at Rheumatology online). Meta-analysis of two studies [37,39] [26,39] and there was limited evidence of a dosedependent association between JIA and sibling number [26,39,40]. Evidence of an association between JIA and birth order was inconsistent [28,39,40]. One study associated child obesity (BMI >30) with increased risk JIA (adjusted OR 3.98, 95% CI: 1.82, 8.34) but evidence was limited for other BMI categories [28]. Two studies examined the number and type of stressful life events [28,41] with largely inconsistent estimates; for example, household unemployment [28] and maternal employment outside of the home [41] showed opposing effects

Living environment
Six studies [28,32,34,37,41,46] examined factors associated with residential and family environment ( Supplementary Fig. S1F, available at Rheumatology online). One study found day-care attendance was strongly associated with decreased JIA risk (adjusted OR 0.12, 95% CI: 0.04, 0.44) [32]; however, this was unsupported by others [28,34,46]. There was little consistent evidence of association between residential area and JIA; the strongly positive association between living in a flat (vs a farm) and JIA (adjusted OR 2.69, 95% CI: 1.19) [37] was unsupported by residential area data from other studies [28,34,41] or data regarding early life contact with animals/pets [28,34].

Environmental factors
Two studies [32,59] examined factors related to the external environment ( Supplementary Fig. S1J, available at Rheumatology online). Limited evidence suggests that higher maternal sunlight exposure during pregnancy is negatively associated with JIA [59]. Similarly, estimated childhood ultraviolet radiation exposure by quartile was also associated with decreased JIA risk in a dosedependent manner (OR 0.19, 95% CI: 0.04, 0.85 for the highest quartile) [59]. With regard to pollutants, one study [32] reported that prenatal pollution exposure (adjusted OR 27.40, 95% CI: 6.85, 109.70) and exposure to second, but not third tertile, ozone in the first two years of life (adjusted OR 6.50, 95% CI: 2.15, 20.53 and adjusted OR 1.00, 95% CI: 0.54, 2.90, respectively) were risk factors for JIA.

Discussion
Summary of the evidence The literature search strategy was designed to capture environmental factors in their broadest sense and identify all relevant studies regarding the association between these factors and JIA. A large number of articles were screened identifying 39 studies reporting 66 environmental factors for inclusion. Only eight environmental factors from ten studies were suitable for meta-analysis, highlighting the unmet need for replicated environmental research in JIA. Meta-analysis found strong evidence that CSD was associated with increased JIA risk whereas sibling status and maternal prenatal smoking appeared protective. There was limited evidence of increased JIA risk associated with postterm birth and SGA, and decreased JIA risk associated with maternal smoking, !3 siblings, birth weight (high or low) and preterm birth. No evidence of association between JIA and maternal age was found.
Of the 58 environmental factors not included in metaanalysis, some themes emerged. Consistent with sibling status, there was some evidence that increasing parity/ number of prior births is protective of JIA whereas maternal fertility problems were associated with increased risk. However, data regarding number of siblings and birth order was conflicting. There was some evidence (largely from unadjusted data) of a positive association between JIA and both antibiotic exposure and early life infections but little evidence for specific pathogens. We found little evidence of association between JIA and maternal/child dietary factors, with the potential exception of fish consumption. Data regarding the association between JIA and breastfeeding status or duration of exclusive/total breastfeeding was highly conflicting. Low vitamin D levels have been causally implicated in multiple sclerosis risk [60]. However, only one study examined the association between vitamin D levels and JIA risk, finding no protective effect [29]. Other factors that influence vitamin D levels such as season of pregnancy/ birth and maternal/offspring sunlight exposure showed conflicting findings.
Interpretation of findings JIA is considered a complex disease, influenced by genetic and environmental factors. The environmental factors identified in this review support several hypotheses explaining the pathological process leading to autoimmunity in JIA. The hygiene hypothesis proposed that the presence of siblings and subsequent early life exposure to infections was protective of atopy and autoimmunity due to effects on the immune repertoire [61,62]. Accordingly, meta-analysis found decreased JIA risk in association with sibling status. In contrast, infectious agents have been proposed to contribute to autoimmunity through several mechanisms such as molecular mimicry, epitope spreading and bystander activation [63]. However, we found no consistent evidence of specific pathogen risk factors for JIA. More recently, there is a growing interest in the contribution of the microbiome to human health. An estimated 22-36% of inter-person microbiome variation is attributed to environmental influences [64], with early life environmental factors such as mode of delivery and breastfeeding influencing the microbiome later in life [65]. We identified CSD as a risk factor for JIA. CSD is postulated to affect the microbiome in multiple ways, including lack of exposure to vaginal flora, which begins at rupture of membranes (indeed, Kristensen and Henriksen reported increased JIA risk associated with elective but not acute CSD [36]), or routine administration of prophylactic intrapartum antibiotics which can pass to the foetus [66]. Changes in the microbiome may provide a unifying hypothesis to explain the association between CSD and JIA plus the weaker evidence of an association between antibiotic use and early life infections and JIA. However, CSD may be associated with JIA independent of any effects on the microbiome or may be mediating another exposure rather than itself being a primary risk factor for JIA. We were unable to determine a robust association between JIA and other perinatal/maternal factors known influence mode of delivery (e.g. pre-/post-term delivery, high/low birth weight, increasing maternal age and conception via assisted reproductive technology) in this review; however, we cannot exclude other indications for CSD as the primary exposure influencing JIA risk. Furthermore, dissecting direction of association from observational data is challenging and it remains possible that any association between JIA and infectious agents/ the microbiome is in reverse; that JIA represents an end point of immune perturbation that also renders children more susceptible to infection and/or resultant exposure to antibiotics, or that active JIA disease/JIA treatments impact the microbiome.
Finally, environmental exposures have the potential to alter the epigenome and affect gene regulation [67]. Prenatal smoking has largely been associated with adverse offspring health and can alter immune responses [68], with altered DNA methylation a potential mediator [69]. Thus, our finding of a negative association with JIA was unanticipated. Explanations include specific effects of in utero smoke exposure on JIA risk or residual confounding, such as socioeconomic position or other measured/unmeasured confounders. However, the majority of studies of prenatal smoking did adjust for socioeconomic position. Furthermore, the effect of prenatal smoking on JIA risk is larger than for any other socioeconomic factor examined. Objective measures of prenatal smoking would be helpful to disentangle this association and limit potential reporting bias. Ultimately, the development of JIA is likely the result of complex and multifactorial immune interactions in genetically susceptible individuals, with many environmental factors making small but cumulatively important contributions to disease risk.

Limitations of individual studies
The studies included in this review have some inherent limitations. Many studies had small sample size, thus imprecise findings cannot be assumed to represent no effect. Several studies examined multiple environmental factors and/or undertook subgroup analyses but did not account for findings potentially resulting from chance due to multiple testing. Fourteen studies diagnosed JIA based on International Classification of Diseases coding rather than physician-defined criteria, which may lead to misclassification bias. Confounder adjustment was variable, with 15/39 studies performing no adjustment. Although JIA can occur throughout childhood, only 1/11 cohort studies followed patients up to age 16 years. Substantial (>5%), or absent reporting of, loss to followup affected 8/11 cohort studies, potentially biasing estimates. With regard to case-control studies, 10/28 did not use community control subjects, limiting the generalizability of the results. Additionally, half of the casecontrol studies did not report the case and control response rate; therefore, we cannot exclude non-response bias from these studies.

Strengths and limitations
Narrative reviews of environmental associations of JIA have recently been published [11,12]. However, to our knowledge, this is the first study to systematically review environmental factors associated with JIA and attempt to quantify their effects. To its strength, this study conforms to PRISMA guidelines and was conducted in accordance with a pre-registered protocol [14]. We used a comprehensive and inclusive search strategy, composed with input from a literature searching specialist and implemented across multiple databases. To ensure the reliability of study selection, scoring and data extraction, all aspects involved a second, and where necessary a third, reviewer. The methodological quality of included studies was independently assessed using a validated scoring system. The included studies varied in quality; however, because JIA is rare we opted for inclusivity and no study was excluded based on quality assessment. Finally, to aid future research, in addition to the data reported in the main text, we have made our full dataset available.
The main limitation of this review is bias due to heterogeneity, which has arisen for several reasons. Firstly, the diagnostic criteria for JIA have evolved over the review period [3]. Secondly, individual studies varied in their inclusion/exclusion of specific JIA subtypes. Thirdly, because the review period spans 45 years, it is likely that social practices, and thus environment influences, have changed. Collectively, historic JIA cohorts may differ from those diagnosed using more recent criteria. However, two-thirds of studies were published in the past decade, and half within the past five years. Despite searching multiple databases and handsearching bibliographies of excluded review articles, the breadth of this review means we may not have captured all relevant studies. Additionally, we excluded studies where we could not extract JIA-specific data from composite outcomes; such data may be informative. Due to resource constraints, only studies in English were included; however, we felt it unlikely that this led to omission of major relevant studies. We were unable to formally assess potential publication bias using funnel plots due to the small numbers of studies investigating each environmental factor. Bias may also result from studies rarely adjusting for the same factors (if at all) and the different lengths of follow-up between studies. Environmental factors across studies were seldom identified, defined and/or measured in the same way, or could be transformed into comparable measures. This resulted in sufficient data to assess only eight factors using meta-analysis. Of these, three included only two studies. Accordingly, for most meta-analyses the pooled estimates were imprecise (encompassing weak evidence of both positive or negative effects) and should be interpreted with caution rather than taken as evidence of no effect. Several studies were excluded where the outcome was JIA course/severity rather than incidence. Whether environmental factors for disease course/severity overlap with those for incidence is an important research question but is outside the scope of this review. Finally, because this review includes only observational studies, the inevitable question remains as to whether our findings represent correlation or causation.

Implications of findings
Identifying robust risk or protective factors associated with JIA has enormous capability to alter patient counselling, aid risk stratification and inform future (causal) research. Furthermore, environmental factors may be modifiable with implications for population JIA risk reduction. The identification of risk (CSD) and protective (sibling status and prenatal smoking) factors associated with JIA necessitates further study into causality and mechanism, and the identification of other putative risk factors (e.g. antibiotics) highlights areas for research priority. The inability to pool the majority of environmental data underscores the need for research reproducibility and standardization of study design. Validating the findings of this review and identifying novel risk/protective factors will require further studies in large populations and likely require international collaboration and co-operation to align this work. Close consideration should be given to the measurement of and adjustment for confounders. Observational studies need to be integrated with mechanistic studies and other data modalities (such as genomic or microbiomic data) to attempt to delineate correlation from causation and improve our understanding of JIA aetiopathogenesis.

Conclusions
This review highlights the plethora of environmental research undertaken in JIA over the last five decades and the challenges posed by using data from historical cohorts. The environmental factors identified here will assist in planning future studies to probe the extent of these associations and understand JIA aetiopathogenesis more broadly, which will ultimately translate into patient benefit.
Funding: This research was funded in whole, or in part, by the Wellcome Trust [grant number 211030/Z/18/Z to SLNC as part of grant number 203918/Z/16/Z]. For the purpose of Open Access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. This work is also supported by the Medical Research Council and the University of Bristol [MC_UU_00011/5 to CLR]. The funders had no input in the protocol design or the undertaking of this systematic review.
Disclosure statement: G.S. has received speaker fees/ Honoraria from Abbvie, and Novartis. A.V.R. has received speaker fees/honoraria/consultancy from Abbvie, Eli Lilly, Pfizer, Roche, SOBI, Novartis and UCB. The remaining authors have declared no conflicts of interest.

Data availability statement
All data underlying the results are from publicly available sources and are included within the article or uploaded as supplementary information. No additional source data are required.

Supplementary data
Supplementary data are available at Rheumatology online.