Environmental risk factors associated with ANCA associated vasculitis: A systematic mapping review

A B S T


Introduction
Anti-neutrophil cytoplasm antibody (ANCA)-associated vasculitis (AAV) is a rare multi-system autoimmune disease comprising four clinico-pathological syndromes: Microscopic polyangiitis (MPA), Granulomatosis with polyangiitis (GPA), Eosinophilic granulomatosis with polyangiitis (EGPA) and renal-limited vasculitis (RLV). It is characterised by a pauci-immune necrotising small-vessel vasculitis with a relapsing and remitting course. AAV is associated with the development of characteristic autoantibodies directed against myeloperoxidase (MPO) or proteinase-3 (PR3) molecules (ANCA), which are abundant proteins in neutrophils and monocytes. While outcomes have improved since the advent of immunosuppression in the 1960s, significant treatment related morbidity and mortality remain a problem [1]. Personalised risk assessment and treatment strategies are a priority.
Classification of systemic vasculitis has traditionally been confusing, with little consensus. Initial diagnostic criteria were developed by the American College of Rheumatology (ACR) in 1990 [2]. The Chapel Hill Consensus Conference (CHCC) criteria, focusing on histology, were then published in 1994, and revised in 2012 [3] to include the serological hallmark, ANCA. The European Medicines Agency (EMEA) algorithm incorporates both of these in a pragmatic consensus classification [4], in an effort to standardise their application clinically. It is anticipated that the 'Diagnostic and Classification of the Systemic Vasculitides (DCVAS)' [5] study will aid in uniforming the classification going forward to facilitate large-scale studies and comparisons.
Like many autoimmune diseases (AIDs), the exact aetiology of AAV, and the factors that influence relapse are unknown. Evidence suggests a complex interaction of polygenic genetic susceptibility [6], epigenetic influences [7] and environmental triggers. The aetiological role of the environment has received least attention. Several microbial, occupational and climatic exposures have been implicated, but most studies thus far have been small, underpowered single centre epidemiological studies, primarily due to the rarity of disease. Physicians receive limited training in environmental medicine, while environmental scientists receive scanty medical training, resulting in a collaborative knowledge gap. This, with the lack of validated environmental exposure biomarkers and standardised tools to quantify exposure, has further hampered efforts. The expansion of disease registries and increasing collaboration over the last decade has assisted in more comprehensive case ascertainment, but more concerted and standardised efforts are critical going forward. This review focuses on the environmental risk factors associated with AAV -defined as any non-genetic aetiological factor. Given the diverse nature of both the triggers reported and the study designs utilised, a pragmatic decision to complete a 'systematic mapping review' was made [8], to identify gaps in the literature, thus informing further research.

Eligibility criteria
Articles that examined any environmental risk factor (defined by any non-genetic aetiological factor) in AAV disease activity (new onset disease or relapse) were included. Studies had to make explicit reference to AAV, which includes the 3 clinico-pathological phenotypes (GPA, MPA and EGPA), rather than isolated ANCA-positivity. All articles identified were English-language, full manuscripts involving adult humans (> 16 years). There was no restriction on publication date and all study designs, except single case reports, were included. Articles which duplicated the findings of other studies were excluded.

Search strategy
The systematic search was performed on 9 th December 2019, using the following databases: EMBASE, Medline (Ovid), Cochrane Library, CINAHL and Web of Science. The search strategy for EMBASE was devised by JS, in consultation with an expert librarian, DM. A comprehensive search strategy can be found in the supplementary material (Supplementary Appendix 1). The search strategy was subsequently adapted for other databases by DM. Duplicates were excluded. Using Covidence [9], JS and JH independently screened titles and abstracts for relevance. Any discrepancies regarding title and abstract eligibility were reviewed by a third independent reviewer, ML. Full texts of articles satisfying eligibility criteria were obtained for full review. Based on expert knowledge and a review of the reference lists of articles, additional relevant articles were identified.

Data collection, data items and risk of bias
The articles included in this review were evaluated for environmental risk factors associated with AAV disease activity. These risk factors were subsequently grouped into common themes: geoepidemiology, pollution, infection and "other". Each article was assessed for: author, year of publication and country of origin, type of study design and method, environmental risk factor(s) explored and method of determining exposure and outcome of the study ( Tables 1, 2 and 3). The quality and risk of bias of each study was reviewed using an appropriate tool (Supplementary Tables 1 and 2): the Modified Newcastle-Ottawa Scale for Case-control and cohort studies (> 6/9 is considered good quality) [10], the International Narrative Systematic Assessment (INSA) scoring system for narrative reviews (> 5/7 is considered good) [11], and the respective checklists from the Centre for Evidence-Based Medicine for randomised controlled trials and meta-analyses. In order to produce a complete map of the literature no study was excluded, but the quality of each was considered in the interpretation of results. Germany [32] Case series. 445 consecutive GPA patients  in the University Hospital Schleswig-Holstein Lubeck Season. Questionnaire of symptom onset, with cross-check of medical records.
No seasonal variance in symptom onset (p 0.3). Carruthers et al. 1996 Norfolk, UK [27] Prospective cohort. All patients attending rheumatology, ENT, respiratory or nephrology departments (Norwich Health Authority) between 1988-1994 with new diagnosis of GPA, OR, those with histopathological diagnosis of GPA were entered into a vasculitis registry (n=21).
Season. Medical records review. Symptom onset was most common in winter (Dec-Feb) (43%), and least common in autumn (Sep-Nov) (0%). Cotch et al. 1996 New York, USA [19] Population-based cohort study. 571 patients with first hospitalisation with GPA identified from SPARCS (discharge database for all nonfederal, nonpsychiatric facilities in New York State) between 1986-1990.
Season. Review of discharge database for season of hospitalisation. Season. Medical records review regarding symptom onset.
There was significantly higher incidence of symptom onset in winter (Jan-Mar) (chisquare = 13.36, p 0.003). Periodic fluctuations of frequencies with peaks every 10-12 months were observed, when Eigen decomposition was applied. Falk et al. 1990 GDCN, SE USA [24] Inception cohort study. 70 patients with PIGN and ANCA positivity were included between Sep 1984 -Jan 1989.
Season of symptom onset via medical records.
There was a seasonal variation in symptom onset. The frequency of symptom onset was statistically higher than expected in winter (Dec-Feb, 38% vs. 25% expected, p < 0.05) and was lower than expected in summer (June-Aug, 11% vs. 25% expected, p < 0.05) in AAV patients with renal involvement. Koldingsnes et al. 2000 Norway [33] Retrospective cohort study. 55 GPA patients (ACR criteria) identified from hospital discharge records from all 11 hospitals in the North of Norway and 2 pathology databases (three 5-year periods between 1984-1998).
Season. Medical records review regarding symptom onset.
No seasonal variance in symptom onset of GPA. Lane et al. 2007 Norfolk, UK [34] Retrospective cohort study. 51 GPA patients (ACR criteria) diagnosed between 1989-1998 identified from a prospective vasculitis register at the Norfolk and Norwich University Hospital.
Season. Medical records review regarding symptom onset, verified by interview of 47 surviving patients.
There was a non-significant trend towards a higher onset of GPA symptoms onset in autumn (Sep-Nov, 35.3%), and a trough in summer (June-Aug, 15.7%), p < 0.5. Li et al. 2018 China [29] Retrospective cohort. 299,906 inpatients (split between 2010-12 and 2013-15) from 878 tertiary hospitals across 31 provinces of China, identified by ICD-10 discharge diagnoses in an electronic database (Hospital Quality Monitoring System) with relevant details obtained from corresponding medical records.
Season determined by frequency of AAV per season. Medical record review.  Latitude, UV radiation. Using the time period and latitude-longitude of the largest city in each study, the mean daily-and winter-ambient UV radiation levels were obtained from satellite data. Ambient UV radiation level weighted to erythemally effective wavelengths and to vitamin D effective wavelengths were then estimated -these various measures were highly correlated.
There was a strong positive correlation between incidence of both GPA (r 0. 62 Germany [23] Registry based prospective cohort study in north and south Germany (1998)(1999). 473 incident PSV cases (CHCC) identified from a population of 4,880,543, of which 90 were AAV (56% urban residence).
Latitude. There was no difference in the incidence of any AAV subtype between north and south Germany.
There was no clusters of PSV in urban/rural areas.
(continued on next page)

Seasonality and temporal clustering
Reports on seasonality are conflicting, although the majority of hospital- [24][25][26][27][28] and population-based studies [29,30] report a peak onset in winter. These studies report a statistically higher symptom onset in winter (December-February, 30.2% -43%, expected: 25%), and lower than expected symptom onset in summer (June-August, 11%). Paradoxically, a French study reported a summer peak, particularly in August [31], while an increased incidence during autumn (March-May) was noted in New Zealand [21]. In contrast, five studies were unable to find any consistent seasonal pattern [19,[32][33][34][35]. Notably, Lane et al [34] re-examined the Norfolk data and found a trend towards higher symptom onset of GPA in autumn, contrasting to the winter peak found by Carruthers et al when analysing a very similar cohort [27]. Interestingly, an Italian group found different seasonal association between GPA and MPA [28]. The symptom onset was mainly in autumn and winter for GPA patients (74%), but in spring and summer for MPA patients (69%). This is consistent with the increased prevalence of MPA and GPA in hot and cold countries, respectively [22].
This argument is further enhanced by studies that display temporal clustering. GPA occurs in a cyclical nature, with non-random peaks, every 4-5 years and 7.6 years in Britain [36] and Norway [33], respectively -this effect was not observed for MPA. A similar 3-to 5-year and 3-year periodicity was noted in Scandinavian SVV patients with renal involvement between 1975-95 [25] and Spanish AAV patients [37], respectively. A shorter periodicity (10-12 months) was noted in Europe [22] Extension of study comparing incidence of SVV in Norwich (UK) and Lugo (Spain) [40].
Utilising the same methodology, incident SVV cases from the University of Tromsø, Norway were compared.
Latitude, as described above [40], by comparing age-and sex-specific incidence rates of SVV between Tromsø, Norway (latitude 70°N), Norwich, UK (52°N) and Lugo, Spain (43°N).  Silicon-containing chemicals (silica dioxide and asbestos), obtained via a questionnaire designed by an occupational health physician. Extent of exposure was estimated by the intensity x frequency x duration. ANCA serology and smoking status were included as confounders.
Occupational exposure to silica dioxide (12.9%) and asbestos (9.7%) were reported by the AAV group but not (0%) in the control group (p < 0.05).

Farquhar et al. 2017
Christchurch, New Zealand [61] Retrospective cohort. N= 52 AAV cases. Cases were identified by positive ANCA serology (all tests referred to 1 of 2 laboratories) and AAV (EMA) confirmed by review of medical records. No statistically significant difference between the incidence rates of AAV pre-(1.87/100,000/ annum) and post-(1.73/100,000/annum) the 2011 Christchurch earthquake. There was also no difference when stratifying by ANCA serotype or restricting to those residing in the urban area. Li et al. 2018 China [29] Retrospective cohort. 299,906 inpatients (split between 2010-12 and 2013-15) from 878 tertiary hospitals across 31 provinces of China, identified by ICD-10 discharge diagnoses in an electronic database (Hospital Quality Monitoring System) with relevant details obtained from corresponding medical records.
Air pollutants (average concentrations of SO 2 , NO 2 , CO 2 and particulate matter such as PM 2.5) (adjusted for annual average temperature, humidity and precipitation) and water pollutants (e.g. ammonia nitrogen) obtained from National Bureau of Statistics of China.
There was a positive correlation between exposure to carbon monoxide and the frequency of AAV (R 2 0.172, p 0.025). (continued on next page) Earthquake related environmental exposures (e.g. asbestos, silica), following the 1995 Kobe earthquake (magnitude 7.2). The environmental bureau of Kobe city reported a 20-fold increase in asbestos levels in the air following the earthquake, which lasted for 30 months. There is no data on silica.
There was a cluster of MPO-AAV cases observed in the Kobe area within the 3-year period following the 1995 earthquake, equating to a more than doubling in annual frequency (14 cases in Kobe vs. 6 cases in Kyoto between 1995-97).
The patients post the Kobe earthquake had higher rates of URT (p < 0.01), and displayed a more severe phenotype, with more patients requiring renal replacement therapy at diagnosis (p < 0.02) and a high rate of severe pulmonary involvement (p < 0.01).  There was a significant association between any reported exposure to dust (specifically silica and grain dust) and GPA (OR 3.6, 95 % CI 1.5, 8.3, p 0.003). There was a positive relationship between the intensity of lifetime occupational exposures of silica (p 0.001), OSs (p 0.001) and metals (p 0.003) and GPA.

SMOKING
Occupation as a farm worker was associated with an increased GPA risk (OR 3.43, 95% CI 1.5, 7.5, p 0.002). Visiting a farm in the 12 months prior to the index date was also significantly associated with developing GPA ( For all findings above, on subgroup analysis, these associations were maximal for GPA, while there were no associations with MPA and EGPA. There was no association between pets and AAV (OR 1.15, 95% CI 0.73, 1.91, p 0.54).
There were no significant differences in the occupations reported between those with AAV and controls, including no increased number of silica associated occupations among patients. Albert et al. 2005 Philadelphia, USA [46] Case series (n=7) of GPA cluster in Dublin, Pennsylvania, in a 3-year period, within a 10-mile radius of an Environmental Protection Agency Superfund toxic waste site.

OTHER POLLUTANTS
Exposure questionnaire of 32 environmental variables (e.g. solvents).
This cluster of GPA patients were potentially exposed to high levels of industrially generated contaminants including: - the Catalan region of Spain, which included mainly MPO-AAV patients with renal involvement [30]. Trends in infection rates has been hypothesized as an explanation of this observed periodicity. However, there was no evidence of cyclic fluctuations in either AAV or infection rates in other studies [23,27,34]. This discrepancy in seasonal variability and periodicity likely reflects the highly variable prodrome, resulting in uncertainty in assigning the date of symptom onset. Furthermore, there is considerable heterogeneity in study methodology, with varying use of disease classifications and definition of disease onset (first symptom or clinical diagnosis). The impact of these variations is evident when one compares two conflicting analyses of essentially the same Norfolk cohort: an autumn peak of GPA diagnoses was described by Lane et al. [34] which differs to a winter predominance in symptom onset reported by Carruthers et al. [27]. It is also plausible that inconsistencies in study findings relate to real differences in triggering factors in diverse geographic regions, such as location-specific infectious trends or UV radiation.

Latitude
There are marked variations in PR3-ANCA:MPO-ANCA and GPA:MPA incidence ratios with latitude, which further supports the potential effect of an environmental influence on disease phenotype. UV radiation has been postulated as a potential latitude-sensitive factor, due to its strong inverse correlation [38], especially in winter. Lower levels of UV radiation at higher latitudes have been implicated in other AIDs, such as multiple sclerosis and type 1 diabetes [39].
Epidemiological studies suggest a North-South declining gradient in GPA risk in the Northern hemisphere, with the inverse observed for MPA [22,40]. The reciprocal finding is true for GPA, but not for MPA, in the Southern hemisphere [21]. Gatenby et al noted that for each degree increase in latitude the incidence of GPA and EGPA increased by 3.5% (IRR 1.04, 95% CI 1.00, 1.07, p 0.03) and 3.4% (IRR 1.04, 95% CI 0.99, 1.08, p 0.11) respectively [41]. Furthermore, by overlaying models of ambient UV radiation over incidence of vasculitis, this group found a strong and modest inverse correlation between both erythemally and vitamin D effective wavelengths and GPA and EGPA, respectively [41]. Given that PR3-ANCA are highly specific for GPA, it is not unsurprising that a recent multicentre study reported increasing odds of PR3 positivity with increasing latitude and decreasing UV radiation [38]. Nonetheless, this finding is confounded by the known genetic associations of MPO-and PR3-AAV and the genetic variability between Europeans in different regions. In contrast, there was no difference in the incidence of any AAV subtype between north and south Germany [23], but the latitudinal difference here may be too slight. UVB radiation is a fundamental step in vitamin D synthesis via skin exposure, and hence is one proposed mechanism in which it may trigger immune dysregulation, resulting in AAV. 1, 25-dihydrohydroxyvitamin D3 (1, 25(OH) 2 D 3 ), its active form, has immunomodulatory effects, including suppression of Th1 and Th17 cells and upregulation of Treg cells, resulting in decreased production of inflammatory cytokines and a more tolerogenic environment [41]. Th1, Th17 and Treg cells are crucial to granuloma formation. This may explain why the latitudinal and inverse UV radiation gradient is strongest for GPA and EGPA (less so), but not for MPA, where granulomas are not found. It is important to recognise, however, that multiple other factors play a role in vitamin D synthesis and subsequent serum levels: season, skin type, clothing, Collaborative Network registry vs. 65 age-, sex-, and race-matched disease controls (nephrology outpatient clinic).
history) obtained from a patient-completed self-reported questionnaire. Of note, influence of silica in this study is included in the meta-analysis elsewhere and so is not considered here [50].
There was no association found between AAV and smoking (OR 0. 66 Case-control study. All surviving (n=28) GPA/ MPA patients diagnosed at the Liverpool Regional renal Unit between 1980-1994 vs. 28 healthy community-based age-, gender-, social class-, and residence-matched controls (randomly selected from blood donors).
Hydrocarbon exposure via structured questionnaire administered by single blinded interviewer. Type, duration and intensity of exposure obtained. Hydrocarbon exposure score calculated from intensity x duration (hours).
The mean hydrocarbon exposure was significantly greater in cases (15,231) vs. controls (4,989), p < 0.01. SA nasal colonisation via culture of bilateral swab of anterior nares. Endonasal activity was assessed using direct endoscopy.
There was significantly higher rates of nasal colonisation with SA in GPA patients vs. CRS patients and HCs outside the hospital. The rate was also substantially (but NS) higher vs. RA patients and hospital staff.
Nasal carriage of SA was associated with significantly higher endonasal activity (p 0.01) and an over 6-fold increased risk of relapse (OR 6.68, p 0.052) compared to GPA patients who were not colonised with SA. There was no correlation between SA nasal carriage and TMP-SMX. Pavone et al. 2006 Italy [28] Retrospective cohort. 75 AAV (ACR/CHCC criteria) patients (with 2 ≤ organ involvement) identified by the Secondary and Primary Vasculitides study group (1985 -2003  Composition of nasal microbiome via swab of middle meatus, using culture-independent techniques: V1-V2 region 16s rRNA (bacterial) and internal transcribed spacer gene sequencing (fungal).
There was a significant difference in the nasal microbiome between GPA patients and controls (p 0.04). These differences were maximal in those off non-CCS IS and with active disease.
GPA patients displayed nasal dysbiosis with significantly less Propionibacterium acnes and Staphylococcus epidermidis than controls (p 0.02 for both). They also displayed a significantly reduced abundance of Malasseziales vs. controls (p 0.04), which was maximal in those with active disease (vs. those in remission, p 0.04).
There was no significant difference in the abundance of nasal SA between GPA and controls (p 0.49). Richter et al. 2009 Birmingham, UK [72] Case-control. 33   There was no association between the number nor type of infection and GPA relapse.
EBV and CMV exposure via IgG and IgM antibody panels using Bio-Rad's BioPlex 2200.
There was a significant increase EBV viral capsid antigen IgG seropositivity (signifying prior infection) in GPA patients compared to healthy controls.
Increased titres of CMV IgM antibodies, but not IgG antibodies, were found in many AIDs, including GPA and EGPA. Cook et al. 1997 Liverpool, UK and Erlangen, Germany [96] Case-control. Sera from 6 GPA and 3 EGPA patients from a AID cohort (n=47) vs. sera from HIV +/-Kaposi sarcoma patients were used as positive controls.  Atopic history, obtained using a structured interviewer-administered questionnaire.
GPA and EGPA patients reported a significantly higher rate of atopy when compared to healthy (p 0.0003 and p 0.002, respectively) and disease controls (p 0.0008 and p 0.004, respectively).
(continued on next page)  [43]. Unsurprisingly, given the seasonal trend in 25(OH)D levels, serum titres fell significantly in those who relapsed, which occurred more often in autumn (difference -6.3 ± CI 14.4 nmol/l, p 0.017), but trended upwards in those who remained in remission (difference 2.7 ± 16.3 nmol/l, p 0.43) [43]. It remains to be seen whether these changes in 25(OH)D levels are biologically significant. An in vitro study revealed that 1,25(OH) 2 D 3 (p 0.02), all-trans retinoic acid (p 0.001) and 9-cis retinoic acid (p < 0.0001) (1 μmol) all inhibited lipopolysaccharide stimulated TNF∝ production in AAV [42]. TNF∝ is important in AAV pathogenesis via its neutrophil priming effect. Vitamin D supplementation may therefore be useful in down-regulating TNF∝ production, in AAV. Population data on vitamin D levels from inception cohorts is required to further explore this hypothesis.

Residence
Many studies have explored the impact of urban versus rural living on AAV with divergent findings. There were no significant differences in the frequency of AAV between rural and urban areas of New York state, Germany and New Zealand [19,21,23]. A separate North German study, found that there was a higher prevalence of GPA (114/million vs. 85/million) and MPA (43/million vs. 16/million) in urban compared to rural areas, while the inverse was true for EGPA (35/million rural vs. 10/million urban) [44]. This is the opposite to what was reported by Kanecki's group in Poland, where the prevalence of EGPA was higher in urban areas (69% vs. 31% respectively, p < 0.001). It also contrasts to the findings from Australia where the incidence of MPA was higher in rural compared to urban areas (14.4/year per million vs. 1.6/year per million), with a similar trend noted for GPA (prevalence in rural areas 129/million vs. 82/million in urban areas). Given the rarity of AAV and the variation in defining urban/rural status between studies, there is currently no clear consensus. As with other geoepidemiological factors, the influence of urban/rural living may vary in different regions.

Pollution
There is evidence for the role of silica, pesticides and organic solvents in AIDs such as rheumatoid arthritis, SLE and systemic sclerosis [45]. Many occupational and industrially generated exposures [46] have been implicated in the pathogenesis of AAV, often thought to cause effect through inhalation, as high levels of upper respiratory tract involvement are observed. Conflictingly, a large Swedish case-control study found no association between 32 selected occupations and GPA, with odds ratios ranging from 0.6 to 1.9 [47]. Borderline occupations included bakers (OR 1.6, 95% CI 1, 2.6) and miners (OR 1.9, 95% CI 1, 3.5), although these were not statistically significant [47]. Similarly, there were no differences in the occupations reported between AAV cases and controls in a large German study [48]. However, of particular note are exposure to silica, other environmental dusts and farming, which have the most convincing evidence to date ( Table 2).

Environmental dusts, including silica
Of all potential environmental risk factors, the evidence for silica is most consistent. Since 1951, researchers have reported several cases of pulmonary silicosis associated with proliferative glomerulonephritis. When immunofluorescence became used to analyse renal biopsies in the 1980s, it was noted that these cases were largely pauci-immune (few immune deposits). Later, they were also found to be ANCA positive [49]. These findings are consistent with case-control studies that report 22-46% of AAV patients have previously been exposed to silica. These studies culminated in a meta-analysis that found that 'ever being exposed' to silica was associated with a 2.5-fold increased risk of AAVthis risk was even higher for patients with renal involvement, GPA or MPA [50]. It is important to note that there was moderate heterogeneity among the included studies (I 2 48.4%), although subgroup analysis did not identify a single study influencing the overall pooled odds ratio. There was also a possible publication bias identified by asymmetry of the funnel plot. Some studies also reported a dose-response relationship, particularly with regards to intensity, rather than duration, of exposure [51][52][53][54].
Silica is abundant on earth, occurring in both crystalline and noncrystalline forms. Sand, soil and rock are composed of the crystalline form and consequently workers involved in 'dusty trades' may be exposed when handling raw materials [53]. The most common occupation associated with silica exposure in an ANCA-positive French group was bricklaying [52]. Others include farming, mill and textile work, and dusty construction labour [55]. The mechanism by which silica triggers AAV has yet to be fully elucidated. It stimulates an inflammatory reaction through accelerated apoptosis of neutrophils and alveolar macrophages in a dose-dependent manner [56]. MPO released by neutrophils can be taken up by activated alveolar macrophages and may be presented to T-and B-lymphocytes resulting in the development of anti-MPO antibodies and a dysregulated immune response in a genetically predisposed individual, culminating in AAV [57].
The role of silica and other environmental dusts is further supported by the outbreaks of AAV, mainly MPO-ANCA positive, observed following 3 large earthquakes in Asia [29,58,59]. A cluster of MPO-AAV cases was observed in the Kobe area in the 3-year period following the 1995 earthquake, equating to a more than doubling in annual frequency [58]. The patients diagnosed post the Kobe earthquake had higher rates of upper respiratory tract involvement and displayed a more severe phenotype, with more requiring renal replacement therapy and exhibiting severe pulmonary involvement. The environmental bureau of Kobe city reported a 20-fold increase in asbestos levels in the air following the earthquake, which lasted for 30 months. There was no measured data regarding silica but it was postulated that large amounts were released into the atmosphere as a result of large-scale building destruction [58]. Similarly, the annual incidence of MPO-AAV doubled after the Great East Japan Earthquake of 2011, with a more severe phenotype noted by higher BVAS scores and higher silica concentrations in the bronchoalveolar lavage compared to pre-earthquake [59]. In addition to silica, the deposited marine sludge contained sand and soil, microorganisms, hydrocarbons, heavy metals and other industrial pollutants. Analogously, a 1.37-fold increase in AAV frequency (from 0.19% in 2013 to 0.26% in 2014) was noted following the 2014 Yunnan earthquake in China, despite no significant increase in PM10 from 2010 to 2015 [29]. Most recently, Webber et al. performed a nested casecontrol study among world-trade centre (WTC) rescue workers following the 9/11 terrorist attacks -prolonged work at the WTC site was an independent predictor of subsequent AIDs [60]. GPA and EGPA were included in the analysis but very small numbers precluded subgroup analysis. In contrast, no increase in AAV was detected following the 2011 Christchurch earthquake, despite increased particulate matter and overall reduced air quality noted [61]. The earthquake in NZ was less severe than those in Asia, with a magnitude of 6.1 on the Richter scale, compared to 6.5 to 9.0. Furthermore, the population studied differed from the Asian disasters, in that the population density in Christchurch is less than a quarter of that of Kobe and the Asian MPO-ANCA predominance is not seen. To our knowledge, the psychosocial impact of these natural disasters has not been studied in detail -it has been proposed that psychological stress may also trigger neutrophil activation and the subsequent inflammatory cascade responsible for AAV.

Other environmental pollutants
Several other small studies report conflicting relationships with other environmental pollutants.
Solvents are chemical compounds that are routinely used in industry. They include aromatic hydrocarbons, alcohols, paint thinners, glues and many other toxic products. Lane et al observed that any history of high organic solvent exposure was positively associated with AAV, specifically GPA. This was not substantiated in three other studies [52,53,62] which also reported no association between AAV and fuels, lead, welding fumes or other heavy metals. This is also in keeping with the findings of Nuyts et al, who found that exposure to hydrocarbons or welding fumes were not risk factors for GPA. However, they did identify non-significant associations between GPA and lead and cadmium [63]. In contrast, Pai et al. found a significantly higher mean hydrocarbon exposure in GPA and MPA cases compared to controls [64] and mercury exposure was associated with a 10-fold increased risk of GPA in Philadelphia [62].
Li et al identified a positive correlation between carbon monoxide (CO) exposure and the frequency of AAV in China [29]. Interestingly, they found no significant relationship with any other air (e.g. PM 2.5, PM 10, other particulate matter, sulphur dioxide or nitric dioxide) or water (e.g. ammonia nitrogen) pollutant measured, adjusted for annual average temperature, humidity and precipitation. Likewise, Albert et al uncovered a positive association between CO and GPA, which approached statistical significance [62]. Given the availability of granular location specific CO data, its influence on AAV disease activity should be investigated in a larger cohort.

Farming and animals
Farming and various associated aspects have been postulated as a trigger for AAV disease activity, via the exposure to innumerable inhaled antigens from soil, dusts, animals and pesticides. Lane et al. first reported the association between farming and AAV -specifically in those with GPA and less so MPA (but not EGPA), who visited a farm in the 12 months prior to the index date, or had significant livestock exposure [52]. This association was not observed for those living on a farm, possibly due to their immune tolerance from prolonged and early exposure. A similar observation was reported in a case series [65]. The findings were also replicated by two other groups [48,54], although specific animal types varied: cows and chickens in the UK [52], cattle and pigs in Germany [48], while sheep conferred the highest risk in New Zealand [54]. These variations likely reflect differences in local farming practices. There was no association identified between AAV and pet ownership [52,54]. The increased risk of GPA associated with gardening, specifically digging, mowing and planting [54,65], also supports the hypothesis that inhaled antigens from the soil are a potential culprit. It is possible that other specific triggers within the farming environment are responsible, but have not been sufficiently studied. One such example is the non-significant trend towards an increased risk of AAV with renal involvement following pesticide exposure (OR 2.32, 95% CI 0.47, 11.5, p 0.303) [53].
The relationship between farming and AAV is not a universal finding -several studies found no association [35,47,62]. A large highquality Swedish registry study involving 2288 GPA cases and a 10-fold larger control group found no association with livestock farming [47]. They used linked occupation registry data capturing a less detailed exposure history compared to a structured questionnaire and excluded exposures outside of the workplace (e.g. leisure farming or gardening). In the remainder of studies, numbers were often small, and participants were exposed to multiple factors, making it difficult to draw conclusions. While the notably large Swedish study essentially eliminates livestock farming as a significant player, it is possible that a subtle relationship exists between exposure to a farm in the year before diagnosis and GPA occurrence.

Infection
There is broad agreement that infection plays a role in triggering AAV disease activity, but the exact microorganism(s) has yet to be determined. Pinching et al. first described the potential role of infection, primarily respiratory, in triggering disease relapse in 1980 [66] in a case series of 18 patients with GPA. There are many case reports proposing various organisms (e.g. Rickettsiae, Enterococcus, etc. [12]) that are temporally related to the occurrence of ANCA positivity and less so to the associated clinical syndrome, AAV. However, these findings often conflict with the negative results of larger observational studies [31]. Here, we will focus only on studies that report overt clinical ANCA-associated vasculitis, and exclude single case reports ( Table 3).
Supporting the idea of an underlying infectious agent, several studies have demonstrated the spatial and temporal clustering in AAV onset, in addition to seasonality, with incidence peaking in the winter when upper respiratory tract infections are most common (see 3.1. seasonality and temporal clustering). Anecdotally, the disease prodrome often includes flu-like symptoms and disease relapses not infrequently coincide with infection. A large Danish registry study demonstrated a dose-response effect between increasing number of hospitalisations for infection and increased GPA risk, which was maximal for those hospitalised in close proximity to their diagnosis [67]. Similarly, Pearse et al found that respiratory tract and sinus infections were associated with increased odds of developing GPA in the 1-2 and 1-4 years prior to the index date, respectively [68], but they were unable to find evidence of infection as a long-term triggering factor [68]. It is possible that these infectious episodes act as stochastic triggering events, however, it is more likely that the overlapping symptoms are an undiagnosed grumbling prodrome.

Staphylococcus aureus
The role of Staphylococcus aureus (SA) in the pathogenesis of AAV has been extensively explored by many research groups in both clinical and mechanistic studies. Multiple studies have illustrated higher rates of chronic nasal SA colonisation in GPA patients compared to healthy and disease controls [69][70][71], as well as higher rates of the small-colony variant phenotype of SA [71] and SA in bronchoalveolar fluid [72]. This finding is possibly due to underlying damage of the nasal mucosa observed in GPA, or as a consequence of the significantly deranged nasal microbiome [73]. In contrast, a study by Rhee et al did not detect a significant difference in the abundance of nasal SA between GPA patients and controls [73]. This may be due to more sensitive cultureindependent identification techniques, which do not differentiate between viable and non-viable SA. Furthermore, samples in the latter study by Rhee et al were obtained from the middle meatus, in contrast to the anterior nares in prior studies.
It is widely accepted that chronic nasal SA carriage (CNSAC) is associated with an increased risk of relapse in GPA (aRR 7.16, 95% CI 1.63, 31.5, p 0.009), since the seminal trial by Stegeman et al [74]. This has been replicated in multiple subsequent studies [70,[75][76][77]. In addition to confirming a 6-fold increased risk of relapse for GPA patients with CNSAC, Laudien et al also demonstrated its association with significantly higher levels of endonasal activity compared to non-colonised GPA patients [70]. Zycinska et al found a slightly higher risk of relapse for those colonised with Methicillin-resistant SA (MRSA) compared to Methicillin-sensitive SA (MSSA) [77]. While the risk appears comparable for those with superantigen (SAg) positive and negative SA [76], of those with SAg-positive SA, SAg toxic-shock-syndrome-toxin-1 was associated with the highest risk for relapse [76]. On the contrary, Tan et al found no association between CNSAC and AAV relapse [71], although this study included all AAV patients, not just those with GPA. The low relapse rate observed in this study (11.8% relapsed over 4 years) may also have prevented the detection of any association [71].
Similarly, a small Norwegian study of GPA patients receiving rituximab maintenance therapy did not find any association between CNSAC and relapse risk [79], likely due to methodological differences: concomitant antibiotics were allowed in this study and the overall frequency of SA carriers was significantly lower(20-28% vs. 63%). Interestingly, an Italian group reported a reduced relapse risk associated with nasal SA colonisation in GPA patients -these results are largely confounded by all patients with nasal SA receiving trimethoprim-sulfamethoxazole (TMP-SMX) and topical mupirocin until culture negative [28], further supporting the protective effect of TMP-SMX.
The elevated relapse risk does not extend to MPA patients [80], but there is some evidence in EGPA patients not receiving antimicrobials [28]. Interestingly, Salmela et al found an unexpectantly low rate of CNSAC in MPA patients (2%), who typically avoid destructive nasal disease [80]. This suggests that those with CNSAC may reflect a subgroup of GPA, which in itself is associated with an increased relapse risk, rather than SA itself being responsible.
With the advent of more sophisticated culture-independent techniques, there is increasing evidence for disruption of the nasal microbiome in GPA patients, known as dysbiosis [69,73], particularly in those with active disease and those off non-glucocorticoid immunosuppression. These novel studies suggest that SA is not the only player in the altered microbial composition of the nasal cavity in GPA patients -there are significant differences in the relative abundance of multiple bacteria at a class and family level, including a reduction in negative competitors of SA [69,73]. Further research is required to assess the clinical implication of this.
There has been much mechanistic work performed into how SA may cause its effect. The concept of autoantigen complementarity has been well documented, whereby a pathogenic microbe (e.g. SA) introduces a peptide, which is a mimic of the reverse-transcribed PR3 peptide (cPR3, the peptide translated from the anti-sense DNA strand of the PR3 gene). Antibodies to both the sense and anti-sense peptide are produced, resulting in PR3 autoantibodies, which trigger an inflammatory cascade in a primed individual, although this finding has not yet been replicated [81].
There has been more debate surrounding the role for infection and molecular mimicry. Kain et al originally demonstrated that 90% of patients with active focal PIGN have intermittent autoantibodies to lysosomal membrane protein-2 (anti-LAMP-2) [82]. This was not reproduced by Roth et using cohorts from Boston and North Carolina [83], although they did not stratify patients by disease activity, which appears to be important for this autoantibody. Kain et al later confirmed the high frequency of Anti-LAMP-2 in acute AAV patients, in three independent European cohorts [84] -they become rapidly undetectable on initiation of induction treatment and often reappear during relapse. Anti-LAMP-2 antibodies cross-react with the bacterial adhesin, FimH, which is derived from many gram negative fimbriated organisms, due to the strong sequence homology between it and a LAMP-2 epitope. Rats immunised with FimH were found to develop AAV-GN due to anti-LAMP-2 antibodies [85]. This provides evidence for the role of infection with fimbriated bacteria in the aetiopathogenesis of AAV, via the production of Anti-LAMP-2 autoantibodies, through molecular mimicry. Recently, Kitching's group demonstrated molecular mimicry mediated by a peptide, 6PGD 391-410 , which is part of an immunogenic plasmid found in certain SA strains. Bacterial plasmids may act as a vehicle to transfer the propensity for autoimmunity. Ooi et al. demonstrated that by injecting mice with either 6PGD 391-410 or genetically engineered SA containing 6PGD 391-410 , this cross-reactive peptide can precipitate loss of tolerance to MPO and subsequent development of anti-MPO glomerulonephritis in an experimental model [86]. Other potentials mechanisms have been postulated and summarised in a review by Konstantinov et al [87]: enhanced autoantigen expression via epigenetic modulations, NETosis and the involvement of Toll-like receptors. Further mechanistic work is required to refine these mechanisms.

Trimethoprim-sulfamethoxazole
Two randomised controlled trials demonstrate the protective effect of trimethoprim-sulfamethoxazole (TMP-SMX) in reducing the rate of relapse in GPA. Stegeman et al initially reported a 68% relative risk reduction in relapse in those taking TMP-SMX (960mg twice daily) compared to those who received placebo (HR 0.32, 95% CI 0.13, 0.79, p 0.02), when adjusted for ANCA-positivity at the outset [88]. Subsequently, a Polish group reported a 60% relative risk reduction in relapse when treated with TMP-SMX (960mg three times weekly) compared to placebo (HR 0.4, 95% CI 0.12, 0.69, p 0.003) [77]. Another Polish cohort study confirmed this protective effect (OR 0.52, p < 0.0092). A re-analysis of two EUVAS trials (NORAM and CYCAZAREM trials) showed a non-significant trend towards a reduced relapse risk with TMP-SMX (960mg 3 times weekly). In contrast to the other trials, this study included both GPA and MPA patients. Similarly, a French study including all AAV subgroups found that while TMP-SMX reduced rates of nasal SA carriage, this did not translate to a reduce relapse risk [71]. The TMP-SMX dose in this study was not uniform, however, and the effect of TMP-SMX was a secondary endpoint, with a low event rate, resulting in the study being potentially underpowered to detect a difference. The benefit of TMP-SMX is multi-faceted: firstly its anti-microbial effect, but also its immunosuppressive effect through its antifolate metabolism.

Viral infections
The evidence to support the role of a viral pathogen in AAV is scanty, with the quality of studies being suboptimal. With regards to other autoimmune diseases, there is a strong association between Epstein-Barr virus (EBV) and both rheumatoid arthritis (RA) and SLE, and between parvovirus and RA [45].
Parvovirus B19 has received the most attention in the context of AAV, with no significant association found. It is recognised that acute parvovirus infection is often accompanied by transient ANCA seropositivity, which falls spontaneously with resolution of infection [89]. However, a group from France reported equal IgG parvovirus B19 seropositivity between cases and controls, with all cases being negative for IgM and DNA [90], suggesting that neither acute nor chronic parvovirus infection is a risk factor for AAV. Molecular studies support this finding in a small case series of MPA patients [91].
The only positive association identified between AAV and any virus is with EBV and cytomegalovirus (CMV), albeit the evidence is weak. Barzilai et al demonstrated a significant increase EBV viral capsid antigen Immunoglobulin G (IgG) seropositivity (signifying prior infection) in GPA patients compared to healthy controls [92]. They also found increased titres of CMV IgM, but not IgG, antibodies in GPA and EGPA patients compared to controls, suggesting recent CMV infection may play a role. However, the selection of cases and controls were not well defined in this study. No significant differences in CMV IgG positivity between cases and controls was later confirmed by a Swedish group [93].
The relationships between hepatitis B virus (HBV) and hepatitis C virus (HCV), and polyarteritis nodosa (PAN) and cryoglobulinaemia, respectively, are well recognised. Patients with PAN who have HBVassociated disease have been shown to display a more severe initial phenotype [96]. Similarly, EGPA patients with resolved HBV infection had more severe initial disease (BVAS p 0.03, FFS p < 0.001) and higher relapse rates (RR 16, p 0.016) compared to those without [95]. A trend towards increased severity of initial disease was also noted for AAV in general, but did not reach statistical significance [95]. There was no serological evidence for the involvement of prior human herpes virus (HHV) 8 [96] or hantavirus [95]. Overall, as yet, there is no convincing evidence that any viral agent plays a significant role in AAV aetiopathogenesis.

Other infections
There is sparse epidemiological data pertaining to other organisms. As for viral infections, the studies are small and the evidence is generally weak. Two studies demonstrate a trend towards higher levels of Chlamydia pneumoniae IgM seropositivity in AAV than controls [98`,99]. There was no difference in IgA or IgG seropositivity, suggesting that active Chlamydia pneumoniae infection may play a role. Furthermore, there is weak serological data to suggest a variety of other organisms are more commonly observed in GPA patients vs. controls: the prevalence of IgG antibodies to Helicobacter pylori and Toxoplasma gondii were significantly higher in GPA patients vs. age-, sex-, and ethnicity-matched healthy controls [100]. No difference in serology between GPA patients and controls was noted for Treponema pallidum [100]. An Italian group noted that prior TB infection, defined by an overt clinical or radiological diagnosis was more common in AAV patients with renal involvement compared to disease controls [101]. The robustness of clinical diagnosis is questionable however, as culture was not confirmed in all cases and this study preceded the use of Quantiferon testing. Finally, a Japanese case series by Ishiguro et al highlights the potential of various causative fungi including Aspergillus species, Candida species and Fusarium vasinfectum in the pathogenesis of EGPA due to the temporal relationship of onset noted with the disease [102]. At present, all of these microorganisms are speculative at best. Further large scale longitudinal studies are required to further investigate these organisms.

Vaccination
The onset of autoimmunity post vaccination is not infrequently recorded as an adverse event. Few studies concentrate on AAV specifically ( Table 3). Two case-control studies exploring multiple potential risk factors did not find any positive association between vaccination within 6 months of symptom onset and the occurrence of AAV -they may have been underpowered to detect this, however [52,54]. In a multi-centre Swiss study, no relapses of AAV occurred in patients who received the diphtheria/tetanus vaccine [103].
There are many case reports and series reporting the onset or relapse of AAV between 12-28 days post influenza vaccination [104,105]. This has not been substantiated by any large scale observational study [106,107]. In fact, Stassen et al performed a retrospective cohort study demonstrating its safety and protective effect: the relapse rate was lower in patients vaccinated against influenza compared to those not (3.4 vs. 6.3 per 100 patients at risk) [106].

Atopy
Atopy has been postulated as a potential risk factor due to the higher rate of symptom onset observed during summer months in France [31]. Furthermore, O'Donnell's group hypothesized that the 'hygiene hypothesis' of allergic and autoimmune disease was responsible for the significantly lower incidence of AAV among socially deprived groups compared to all others [21]. It is important to note, from the outset, that the quality of evidence was low for all studies exploring the role of atopy. Half of the studies found that patients with GPA [52,108] and EGPA [108] reported a significantly higher rate of atopy when compared to controls. Subgroup analysis identified drug allergies (particularly antibiotics) as most significant. These findings were not replicated by other studies [54]: on the contrary, one reported a reduced GPA risk associated with an allergic history (OR 0.25, 95% CI 0.1, 0.6, p 0.001) [62]. There is thus insufficient evidence to accurately conclude the role of atopy ( Table 3).

Smoking
Despite the toxic nature and recognised detrimental health consequences of smoking, the association between it and various AIDs is not strong. While there is a positive correlation between smoking and SLE, multiple sclerosis, Crohn's disease, Hasimotos thyroiditis and Graves' disease [109], smoking appears protective against ulcerative colitis [110]. A cross-sectional study involving two German vasculitis referral centres found that there was a lower prevalence of smokers among newly diagnosed AAV patients compared with the background population [110] ( Table 2). It is important to note that confounders, such as gender, were not considered in this study. Similarly, Pearse et al reported a slightly protective effect of current smoking on GPA risk, but an increased risk in former smokers, in the UK [68]. Nicotine's immunosuppressive effect has been proposed as a potential explanation for these findings [111]. A Japanese study, primarily including MPO-ANCA MPA patients with renal involvement, reported conflicting findings: smoking increased the risk of relapse (p 0.003) in a dose-dependent manner (p 0.004). Specifically, current smoking was associated with a 7.5-fold increased risk of relapse (aHR 7.48, 95% CI 2.73, 21.0). On the contrary, four case-control studies found no relationship between smoking and AAV [52][53][54]. Biological biomarkers to qualify smoking status were not used, thus the inherent bias in self-reported smoking status must be considered when interpreting these findings.

Drugs
Drug-induced AAV is often regarded as a separate entity to primary AAV. It usually exhibits a milder phenotype that reverses with cessation of the drug, with simultaneous fall of ANCA titres, negating the risk for prolonged immunosuppression and resulting in relatively good longterm outcomes. There are multiple recent reviews that explore implicated drugs: propylthiouracil (PTU), hydralazine, anti-TNF∝ agents, sulfasalazine, D-penicillamine, minocycline [112,113], other antithyroid drugs including carbimazole and methimazole [114], and levamisole-adulterated cocaine [115]. Most of these agents have potent epigenetic effects (e.g. hypomethylation), resulting in altered gene expression -this is likely the mechanism by which they induce autoimmunity. Furthermore, there are multiple case reports describing the occurrence of EGPA following the use of anti-leukotriene agents (e.g. montelukast). However, there is broad agreement that this phenomenon is attributed to the simultaneous withdrawal of corticosteroids in severe asthma cases, which unmasks undiagnosed EGPA [116][117][118][119]. Multiple other studies report increased ANCA positivity in the context of certain drugs, without overt clinical vasculitis. A recent review attempted to separate the risk of developing GPA from that conferred by its indication -allopurinol was the only drug that conferred a greater risk for developing GPA (OR 1.8, 95% CI 1.2, 2.8, p 0.01) than its indication, gout (OR 1.2, 95% CI 0.9, 1.8, p 0.3) [68]. Interestingly, the risk associated with sulfasalazine and anti-thyroid drugs (levothyroxine, carbimazole, PTU) were similar to those of rheumatoid arthritis and thyroid disease, respectively, suggesting no additional risk conferred by the drug itself [68].

Conclusion
The precise aetiology of AAV has yet to be elucidated. The current consensus is that there is a complex interplay between various Table 4 Summary of environmental risk factors in ANCA-associated vasculitis disease activity. observational studies reported a 2.5-fold increased risk of AAV with 'ever being exposed' to silica. A dose-dependent response is noted, particularly in relation to intensity of exposure, rather than duration of same.

Yes
High Natural disasters Cohort (3) Case series (1) Clusters of AAV, mainly MPO-ANCA positive, were noted after 3 large earthquakes in Asia, postulated to be due to increases in silica, asbestos and carbon monoxide. In contrast, no increased incidence of AAV was noted following the 2011 Christchurch earthquake.

High
Other industrial contaminants (e.g. organic solvents, OS) Case-control (5) Case series (1) Any history of high OS exposure was positively associated with AAV in 1 study, specifically GPA. Similarly, a significantly higher mean hydrocarbon exposure was found in GPA and MPA patients vs. controls. These findings were not reproduced in 3 other studies.

Unclear
High Mercury Case-control (1) Mercury was associated with a 10-fold increased risk of GPA in 1 study.

Low
Carbon Monoxide (CO) Cohort (1) Case-control (1) There is a trend towards an association between CO and AAV, specifically GPA.

Farming, Animals
Case-control (5) Case series (1) 4/6 studies report an increased risk of AAV with farming. The risk is maximal for those with GPA who were exposed in the year prior to diagnosis, particularly to livestock.
Yes Low (continued on next page) environmental and epigenetic factors, in a genetically susceptible individual. It is likely that different triggers and the degree to which they influence disease activity vary by subgroup (e.g. ANCA subtype, geographic region). The most notable findings support the role of various pollutants -primarily silica and other environmental antigens released during natural disasters and through farming. Assorted geoepidemiological triggers are also implicated including seasonality and latitudedependent factors such as UV radiation. Finally, infection is a likely culprit, yet the exact microorganism(s) are debated -Staphylococcus aureus is the most presently convincing. Table 4 synthesises the evidence and overall perceived risk for each identified factor. Given the lack of uniformity in disease identification and classification, as well as the challenge of obtaining adequate power in epidemiological studies, due to the rarity of disease, the overall quantity of evidence is small and the study quality is generally poor. The anticipated Diagnostic and Classification Criteria in Vasculitis (DCVAS) [5] may aid in categorising disease subgroups to allow uniformity in case ascertainment. There is a need for validated exposure biomarkers and more interoperable disease registries that can be integrated with other data streams (e.g. genetic, epigenetic and environmental) to facilitate international collaboration and large-scale epidemiological studies, with novel statistical and machine learning analytical techniques. Possible approaches include Bayesian Additive Regression Trees (BART) [120], Gaussian Processes [121], Artificial Neural Networks [122] and various case-crossover study designs [123], such as those recently utilised by Dixon et al. to explore the impact of weather on chronic pain [124].
AAV is one of over 100 AIDs, which when considered together affect approximately 10% of the population. Accordingly, the degree of individual-and population-level burden of AID makes the exploration and understanding of potential initiating factors critical. Discovering further insights into aetiopathogenesis is crucial to allow personalisation of risk stratification and treatment regimens, thereby optimising the riskbenefit balance associated with toxic immunosuppression. By identifying triggers of AAV relapse, we can learn to predict disease flares and hence develop a precision medicine approach in this cohort -a critically unmet need. Strides made in AAV have the potential to create a new paradigm for how AIDs are studied going forward.

Declaration of Competing Interest
JS, JH, DM, ML none declared.

Acknowledgements
Thank you to Professor John Browne who leads the 'Systematic Reviews for the Health Sciences' module (PG7016, University College Cork) for sharing the necessary skills to complete this review.  (2) Case-control (5) No association was found in > 50% of studies. 2 studies demonstrated a protective effect of current smoking, but an increased risk associated with former smoking. 1 study found a dose-dependent increased risk of MPA relapse associated with smoking, which was maximal for current smokers compared to 'ever having smoked'.

Unclear
High