Factors Associated with Sequelae of Campylobacter and Non-typhoidal Salmonella Infections: A Systematic Review

Despite the significant global burden of gastroenteritis and resulting sequelae, there is limited evidence on risk factors for sequelae development. We updated and extended previous systematic reviews by assessing the role of antibiotics, proton pump inhibitors (PPI) and symptom severity in the development of sequelae following campylobacteriosis and salmonellosis. We searched four databases, including PubMed, from 1 January 2011 to 29 April 2016. Observational studies reporting sequelae of reactive arthritis (ReA), Reiter's syndrome (RS), irritable bowel syndrome (IBS) and Guillain-Barré syndrome (GBS) following gastroenteritis were included. The primary outcome was incidence of sequelae of interest amongst cases of campylobacteriosis and salmonellosis. A narrative synthesis was conducted where heterogeneity was high. Of the 55 articles included, incidence of ReA (n = 37), RS (n = 5), IBS (n = 12) and GBS (n = 9) were reported following campylobacteriosis and salmonellosis. A pooled summary for each sequela was not estimated due to high level of heterogeneity across studies (I2 > 90%). PPI usage and symptoms were sparsely reported. Three out of seven studies found a statistically significant association between antibiotics usage and development of ReA. Additional primary studies investigating risk modifying factors in sequelae of GI infections are required to enable targeted interventions.


Introduction
Campylobacter and non-typhoidal Salmonella enterica (NTS) are important agents of human bacterial gastroenteritis, representing over 30% (174.3 million) of diarrhoeal illnesses globally in 2010. While Campylobacter was the most common cause of bacterial gastroenteritis, NTS accounted for most of the deaths caused by a bacterial foodborne agent (over 59,000) and with the highest rank for disability adjusted life years amongst foodborne disease hazards in 2010 (Havelaar et al., 2015). Globally, foodborne disease burden is not equally distributed amongst the World Health Organisation (WHO) sub regions, with the greatest burden falling on the sub regions in Africa. Nevertheless, both Campylobacter and NTS (henceforth gastrointestinal (GI) infections) still pose a significant disease and economic burden in developed countries (Scallan et al., 2011;Majowicz et al., 2010).
Gastroenteritis caused by Campylobacter jejuni/coli and most serotypes of NTS are characterised by a self-limiting illness without the need for medical intervention. Yet, a subset of patients develop sequelae such as reactive arthritis (ReA), Reiter's Syndrome (RS), irritable bowel syndrome (IBS), Guillain-Barré Syndrome (GBS), Inflammatory Bowel Disease (IBD), Crohn's disease (CD) and ulcerative colitis (UC) (Ajene et al., 2013;Keithlin et al., 2014Keithlin et al., , 2015. Evidence on the factors predisposing some patients to sequelae development is limited, with only one study assessing the factors for development of IBS following enteric infection (Thabane et al., 2007). The authors found that young age, prolonged fever, anxiety and depression were risk factors for post-infectious IBS, but they did not stratify those factors by the infecting pathogen. This is a drawback for burden of disease studies, as estimates of pathogen specific sequelae development are required for prioritization of public health interventions.
In a systematic review to assess the proportion of patients who develop chronic sequelae following GI infection, the authors found that study-level factors, such as diagnosis method for complications, EBioMedicine 15 (2017) [100][101][102][103][104][105][106][107][108][109][110][111] follow-up period from infection to sequelae development, and study size, contribute to the reported incidence of ReA and IBS following Campylobacter and NTS infection (Keithlin et al., 2014(Keithlin et al., , 2015. However, the association of clinical factors such as proton pump inhibitors (PPI) usage and antibiotics in the development of chronic sequelae were not investigated. These drugs, which commonly increase risk of gastroenteritis, may also have a role in sequelae development due to changes to the gut microbiome and gastric pH that can favour pathogenic organisms (Doorduyn et al., 2008).
In light of the existing gap in the evidence of factors contributing to sequelae development in patients with GI infections, this systematic review extends the previous reviews to assess the study-and patientlevel risk factors associated with the development of complications following Campylobacter and NTS infections. Specifically, we assess whether use of PPI, treatment with antibiotics and clinical symptoms such as duration of diarrhea and fever are risk factors for the development of ReA, RS, IBS, GBS, IBD, CD and UC in adults and children with a Campylobacter or NTS infection.

Methods
This systematic review and meta-analysis was conducted in line with the 'Meta-analysis of Observational Studies in Epidemiology' (MOOSE) guidelines (Stroup et al., 2000). The protocol was registered on PROSPERO (CRD 42015026042).

Search Strategy and Selection Criteria
We searched four electronic databases, PubMed, Agricola[http:// agricola.nal.usda.gov/], EMBASE [OvidSP] (1974-2016 and CabDirect [OvidSP] (2000[OvidSP] ( to 2016 Week 15) for studies reporting sequelae of ReA, RS, IBS, GBS, IBD, CD and UC following gastrointestinal infections (Campylobacter and NTS). The search strategies consisted of a combination of relevant subject headings and free-text words in title and abstract for exposure and outcome. We restricted our search to studies published between 01 January 2011 and 29 April 2016 as this was an update and extension of the previous reviews with searches up to July 2011 (Keithlin et al., 2014(Keithlin et al., , 2015. The detailed search strategies and results for each of the databases are presented in Supplementary Table S1. Additional eligible studies on GI infections and associated sequelae were sought by reviewing the reference lists of identified articles. No language restrictions were applied during the search. Two reviewers independently screened titles and abstracts for relevance. Studies were included if they were cohort (prospective or retrospective), case-control, surveillance report or cross-sectional studies, or outbreak investigations of people with Campylobacter or NTS infection, and reported the number or proportion of people who developed the sequelae of interest following Campylobacter or NTS infection. Studies were excluded if they: reported sequelae without evidence of a past exposure with Campylobacter or NTS infection; reported sequelae with only serological evidence of past exposure to pathogens; reported sequelae for multiple foodborne infections without a breakdown of the proportion/numbers by pathogen and sequelae; were case reports, case series or experimental studies such as randomised controlled trials and laboratory based studies. We added data for the period before July 2011 from the previous systematic reviews (Keithlin et al., 2014;Keithlin et al., 2015) and extracted additional variables as required.

Data Extraction and Bias Assessment
Data were extracted independently by two reviewers (OE and MP) using a standardized form. Inconsistencies were resolved through a consensus process, with any disagreement resolved by a third reviewer, TF. Data coding and categorisation was in accordance with the previous reviews or otherwise stated (Supplementary Table S2) (Keithlin et al., 2014(Keithlin et al., , 2015. The primary outcome was the number of cases who developed the specific sequelae of interest divided by the total number of Campylobacter or NTS cases. The Joanna Briggs Institute Prevalence Critical Appraisal Tool was adapted to evaluate the quality of each study (Munn et al., 2014). This tool was selected because of its flexibility to address the risk of bias across a variety of study designs, as commonly found in the study of incidence and prevalence. Initially a calibration exercise was performed by review members using a random sample of three studies. The items in the tool were applied to the selected studies to ensure consistency across reviewers and validity in assessing the risk of bias with this tool. Following this exercise two additional questions were included (Supplementary Table S3). Inconsistency was resolved through a consensus process.

Statistical Analysis
We performed meta-analysis in STATA version 13 (StataCorp LP) using "metaprop_one", a user written command for meta-analysing proportions (Nyaga et al., 2014;Freeman and Tukey, 1950). Heterogeneity was quantified using the I 2 measure (Higgins et al., 2003). Where heterogeneity was high (I 2 over 50%), no summary estimate was calculated.
It was possible to have multiple outcomes per study. Briefly, some studies reported multiple diagnostic methods for both pathogen and complication or multiple pathogen serotype/species. Each combination of pathogen diagnosis or sequelae diagnosis was considered as a separate outcome measure. Meta-analysis was performed using the most rigorous outcome measure based on the reference standard. For instance, where a study reported both laboratory-confirmed and probable diagnosis for a pathogen with multiple diagnostic methods for the sequelae of interest, such as self-reported diagnosis and further diagnosis by a specialist (rheumatologist for reactive arthritis), the combination of Campylobacter/NTS cases with a laboratory-confirmed diagnosis and the sequelae assessed by a specialist was used.
A priori subgroups to explore potential sources of heterogeneity were investigated based on relevant methodological characteristics (study design, study size, follow-up period and sequelae diagnosis) (Keithlin et al., 2014(Keithlin et al., , 2015 and clinical characteristics (healthcare facility visited, symptoms of GI infections, reported PPI and antibiotic usage), if data were available.

Results
Primary searches identified 4133 references. On removal of duplicates and after screening, five studies met the inclusion criteria reporting ReA, RS, IBS and GBS following Campylobacter infection (n = 4) and ReA, RS and IBS following NTS infection (n = 3) (Baker et al., 2012;Uotila et al., 2014;Porter et al., 2013aPorter et al., , 2013bTuompo et al., 2013). Data was extracted from all five studies and additional studies from the previous systematic reviews on ReA, RS, IBS and GBS following GI infections (n = 50). Hence 55 studies were included in the analysis ( Fig. 1 and Table 1). No eligible studies on IBD, CD and UC were identified by our database search, so these sequelae are not considered further in this review. Table 2 shows the characteristics of the included studies. The number of patients with GI infections varied widely (range 6 to 57,425 with Campylobacter and 24 to 34,664 with NTS infection) and represented all age groups. Studies reported diagnoses of the pathogen and complication according to standard practices for all cases of GI infections (87%, 48/55 and 75%, 41/55 respectively). Only 18% (10/55) reported adequate sample size calculation and, where response rate was low (49%, 27/55), only 15% (n = 4) adjusted for possible response bias in their analysis. An overall risk of bias score was not assigned as it was possible to have a high score (N70%, 8.5/12) without using a reliable method of pathogen and diagnosis for all patients.

INCLUDED STUDIES
Only studies reporting incidence of ReA considered the use of PPI and antibiotics as potential factors contributing to sequelae development. One study assessing the use of PPI in the development of ReA following Campylobacter and NTS infection found a significant association after adjustment for age, sex and degree of urbanization (adjusted OR 2.9 (95% CI 1.4-6.1)) (Doorduyn et al., 2008).
Clinical symptoms were scarcely reported by studies and, where available, studies used different definitions and thresholds for diarrhea and fever (data not shown). This prevented further analysis of the role of symptom severity in sequelae development.
For the methodological subgroup analysis, only consultation with a rheumatologist statistically reduced heterogeneity of studies within the "specialist" in Campylobacter and NTS triggered ReA (Tables 4A,  4B, 4C, 4D, Figs. 2 and 3). Follow-up may have contributed to heterogeneity in Campylobacter triggered IBS as studies within the 6-month follow-up stratum had a statistically significant reduction in heterogeneity (I 2 = 24.2%) (Tables 4A, 4B, 4C, 4D, Fig. 6). The only clinical characteristic considered in a subgroup analysis was the healthcare facility visited for the GI infection in those developing ReA and IBS. Studies reporting a "GP/Physician" visit for NTS infection were fairly homogenous (I 2 = 2.0%), nonetheless high heterogeneity remained in this stratum for patients with Campylobacter triggered ReA (Figs. 2 and 3).

Discussion
Previous systematic reviews considering incidence of ReA, RS, IBS and GBS conducted literature searches up until 2011 without assessing factors contributing to sequelae development. Five years after these searches were conducted, we only identified five new studies based on our inclusion criteria. We found that use of PPI and antibiotics may be possible factors associated with the development of ReA following GI infections. These factors were sparsely reported by studies and where information was available high heterogeneity (I 2 N 90%) prevented the pooling of data. Despite the over prescription of PPI in both primary and secondary settings (Forgacs and Loganayagam, 2008), only one study reported an association of PPI usage and ReA development in patients with gastroenteritis (Doorduyn et al., 2008). The authors demonstrated that PPI usage was independent of a single nucleotide polymorphism (SNP) in interferon gamma (IFN-γ) in cases with Campylobacter and NTS infections and the development of reactive arthritis (Doorduyn et al., 2008). IFN-γ is a cytokine crucial in the immune response against enteric infections. The combination of a SNP in IFN-γ with PPI usage could lead to increased susceptibility to enteric infections and subsequent prolonged or repeated episodes of GI infection. These sequential events may increase the susceptibility to reactive arthritis (Doorduyn et al., 2008).
Of all the sequelae considered, only risk of reactive arthritis was assessed following antibiotic usage in cases of Campylobacter and NTS infection. The associated risk is not clear as the studies report elevated, decreased or no risk of ReA following GI infection. The association may be dependent on the dose and type of antibiotics, as evidenced in a systematic review (Agger et al., 2015) evaluating the risk of hemolytic uremic syndrome following the use of antibiotics in patients with shiga toxin producing Escherichia coli (STEC) infections. Agger et al. (2015) showed that protein and cell wall synthesis class of antibiotics may be protective and improve recovery time and proposed a review of guidelines on contraindication of antibiotics for STEC infections.
Most gastroenteritis cases do not require treatment with antibiotics unless they are severe and occur in at risk groups i.e. elderly, children and those with underlying comorbidities. Due to insufficient information in the studies, we could not assess the potential reason for antibiotic treatment or the risk associated with sequelae development following antibiotic usage in a meta-analysis. Moreover, information on duration of treatment, dose of antibiotics, age and gender of all cases who received antimicrobial treatment were not available in all of the studies, thereby limiting any further comparisons or pooling of data.
Heterogeneity could not be explained by most of the subgroup analysis considered, except in the reported method of diagnosing reactive arthritis complication following Campylobacter and NTS infection, the type of healthcare facility visited for an NTS infection and follow-up period in Campylobacter triggered IBS. In studies reporting a visit to the GP/ Physicians in cases of NTS infection the heterogeneity was significantly reduced (I 2 = 2%). The three studies were all conducted in Finland, reported specialist diagnosis for reactive arthritis, had laboratory confirmation of the NTS infection and two assessed use of antibiotics in development of ReA in NTS patients. A GP/physician consultation may lead to a laboratory confirmed diagnosis of infection, prescription of antibiotics, and referral to a specialist; hence influencing the reported sequelae incidence. The small number of studies with inconsistent follow-up period from NTS infection to ReA development limits further interpretation of this finding.
Follow-up period may be crucial in the reported incidence of IBS following campylobacteriosis, due to a statistically significant reduction in A&E/Emergency Dept, Hospitalised Townes 2008Dworkin 2001 GP/Physician Ekman 2000Mattila 1994Hannu et. al 2002b GP/Physician, Hospitalised Mattila 1998 Hospitalised Rudwaleit 2001Helms 2006Urfer 2000Hakansson 1976Petersen 1996 Not reported Ternhag 2008EastmondB 1983McColl 2000Doorduyn 2008Tuompo 2013Buxton 2002Samuel 1995Thomson 1992 Fig. 3. Forest plot of studies reporting incidence of NTS triggered ReA stratified by healthcare facility. Studies reporting the incidence of reactive arthritis following non-typhoidal Salmonella infection stratified by the type of healthcare facility/practitioner visited/utilised. No summary estimate was calculated due to high heterogeneity across all studies (I 2 N 90%). heterogeneity in studies reporting a 6-month follow-up period, despite using different versions of Rome I, II and III criteria for diagnosis. The current gold-standard for IBS diagnosis is Rome III classification, which is symptom-based, requiring patients to be symptomatic both at 3 and 6 months after initial symptom onset (Longstreth et al., 2006) was used by only one study (Spence and Moss-Morris, 2007). Potential risk modifying factors in the development of IBS following gastroenteritis were not evaluated due to insufficient reporting and primary aim of included studies.

Proportion (%)
Data source may be a potential source of unexplained heterogeneity even though it was not significant in our subgroup analysis (data not shown). Using a combination of studies reporting outbreaks, population surveillance and hospital surveillance may have introduced additional heterogeneity. Outbreaks are usually caused by a single strain, while population surveillance identifies the circulating strains. In one of the studies included in this review, Tuompo et al. (2013) used phenotypic methods to determine potential differences in the O antigens of different NTS serotypes circulating in a population surveillance. However no significant differences in the arthritogenicity of the serotypes to trigger ReA were identified (Tuompo et al., 2013). The precision of genomic methods, such as whole genome sequencing, may provide further insights into differing potential for bacterial strains to trigger sequelae. Our study has a number of major strengths. Firstly, our study addresses an existing knowledge gap by assessing the available evidence on factors contributing to the development of sequelae following Campylobacter and NTS infection. This is important in understanding factors amenable to intervention. Secondly, our review highlights the need for investigators to consider remediable risk factors in sequelae development following GI infections.
The reliance of observational studies on "natural experiments" such as outbreaks makes them prone to low study quality and high level of bias and heterogeneity. Most outbreak investigations are set up as a rapid response to an emergency situation to determine the cause of the outbreak. Time constraints limits the opportunity to address secondary factors that may improve the overall evidence on GI triggered sequelae. Therefore, it was possible for a study to have a high quality score for the primary research but missing key items such as method of diagnosis of pathogen and sequelae, crucial for risk factor study. Thus, a quality criteria was not adopted for inclusion of studies for meta-analysis. Although heterogeneity was significantly reduced for studies reporting a specialist consultation for diagnosis of ReA following either Campylobacter or NTS infection, we did not report a pooled estimate. Specialist diagnosis was preceded by self-reported symptoms. Not all patients with symptoms were seen by a specialist. Further, the   studies reported different follow-up times indicating different definitions of ReA were adopted. Other limitations of our study include inconsistencies in the reporting of key features, such as age, definition of sequelae, and follow-up of gastroenteritis cases in included studies. The risk factors we considered were also sparsely reported. Further, unexplained sources of heterogeneity prevented the reporting of a summary statistic for factors contributing to sequelae development. These limit the application of estimates to the general population and made interpretation of results difficult.
In conclusion, the findings of this systematic review show that the factors contributing to sequelae development following GI infections remain unclear. This is a challenge for policy makers in targeting interventions to reduce the overall burden of GI infections. Researchers should consider reporting information that will improve the overall evidence of sequelae of GI infections when observational studies are conducted. Applying record linkage in following up patients' health journey to study sequelae of GI infections may improve reporting of observational studies.
Furthermore, primary research in risk modifiers of gastroenteritis triggered sequelae and the potential for different bacterial strains to cause sequelae may address the knowledge gap in the relationship between pathogen and host.

Conflict of Interest
Prof. Perera received grants from UK NIHR, during the conduct of the study.

Author Contributions
OE wrote the manuscript. OE conceived the initial idea for the study, NM, MV, RP and TF critically appraised the protocol, manuscript and also contributed to its development by revising different versions. NR assisted with the database searches and ran the updated searches. OE and OvH screened all abstracts; OE and DC completed the full text screening; OE and MP performed all data extraction and risk of bias was completed by OE, MP and TF. All authors approved the final version and take responsibility for its content.