Dr J. K. Marshall, Division of Gastroenterology (2F59), McMaster University Medical Centre, 1200 Main Street West, Hamilton, ON L8 N 3Z5, Canada. E-mail: email@example.com
Background Individual studies suggest that post-infectious irritable bowel syndrome is common, but symptoms gradually improve.
Aim To review evidence for an association between intestinal infection and development of irritable bowel syndrome, assess the prognosis of post-infectious irritable bowel syndrome and explore factors that increase the risk.
Methods MEDLINE (1966–2007) and EMBASE (1980–2007) databases were searched to identify the studies of post-infectious irritable bowel syndrome epidemiology. Data were extracted by two independent reviewers. Pooled odds ratios (POR) and corresponding 95% CI for incidence of irritable bowel syndrome were estimated among the exposed and unexposed groups.
Results Eighteen of 26 studies identified were eligible for inclusion. Intestinal infection was associated with increased odds of developing irritable bowel syndrome at study end (POR = 5.86; 95% CI: 3.60–9.54). In subgroup analysis, the odds of developing irritable bowel syndrome was increased at 3 months (POR = 7.58; 95% CI: 4.27–13.45), 6 months (POR = 5.18; 95% CI: 3.24–8.26), 12 months (POR = 6.37; 95% CI: 2.63–15.40) and 24–36 months (POR = 3.85; 95% CI: 2.95–5.02). Among all studies (controlled and uncontrolled), the pooled incidence of irritable bowel syndrome at study conclusion was 10% (95% CI: 9.4–85.6). Subjects with post-infectious irritable bowel syndrome were younger and more anxious and depressed than those without post-infectious irritable bowel syndrome.
Conclusion The odds of developing irritable bowel syndrome are increased sixfold after acute gastrointestinal infection. Young age, prolonged fever, anxiety and depression are risk factors for post-infectious irritable bowel syndrome.
Irritable bowel syndrome (IBS) is a gastrointestinal disorder characterized by abdominal pain and altered defecation, for which there is no apparent structural basis.1, 2 IBS is common, with a prevalence of up to 22% in community subjects.3 Although IBS is not associated with an increase in mortality, it causes significant morbidity and places a substantial burden on both healthcare systems and society at large. It accounts for 12% of all visits to primary care physicians3 and up to 50% of all out-patient referrals seen by gastroenterologists.1 The pathophysiology of IBS is poorly understood, but there is evidence to support both host and environmental factors.4–23
The role of enteric infection in the pathogenesis of IBS has been recognized for years.24 Several studies have characterized post-infectious (PI) IBS and explored risk factors for its onset.6–23 While the prognosis of PI IBS has not been assessed, systematically individual studies have suggested that the symptoms persist for years in most patients and gradually improve.25
A recent meta-analysis by Halvorson et al.26 demonstrated an increased risk of PI-IBS in subjects who experience acute gastroenteritis. However, this review excluded several recent studies and did not assess either the prognosis of PI-IBS or risk factors for its development like age, psychological profile or severity of the acute illness.10, 11, 13, 23, 27, 28 The objective of this systematic review and meta-analysis was to review the evidence of the association between intestinal infection and development of IBS, assess the prognosis of PI-IBS and explore the risk factors for its development.
Materials and methods
A comprehensive search for prospective cohort studies investigating the association between infectious gastroenteritis and development of IBS was conducted by using the MEDLINE database (1966 to February 2007). Search strategies used the following medical subject headings (MeSH): (‘postinfectious IBS’ or ‘postdysenteric IBS’‘IBS’ or ‘functional colonic diseases’ or ‘gastrointestinal disorder’ or ‘spastic colon’) and (‘bacterial infections’ or ‘dysentery’ or ‘gastroenteritis’ or ‘diarrhoea’ or ‘infectious diarrhoeal disease’). The search was replicated by using the EMBASE database (1980 to February 2007). The electronic search was supplemented by a manual review of reference lists. The search was limited to human studies published in English language.
Study selection criteria and inclusion
Full texts of potentially relevant citations were retrieved and assessed for eligibility by two independent reviewers (M.T. and D.K.) according to predetermined criteria. If relevant information was not reported or was unclear, study investigators were contacted to provide clarification. Inter-rater agreement was assessed by using Cohen’s kappa and disagreements were resolved by a third investigator (J.K.M.).
Standardized forms were used to assess the study eligibility according to the following predefined criteria: (i) study population described as adults and/or minimum age specified as 16 years; (ii) diagnoses of IBS made using standard criteria (Manning, Rome I/II/III) or by clinical impression; (iii) exposure to gastroenteritis defined by positive stool culture, acute gastrointestinal illness or both; (iv) outcome of IBS reported as prevalence, incidence, relative risk, or odds ratio; and (v) subjects included in the analysis determined to be free from premorbid IBS. Among controlled cohort studies, acceptable controls could be contemporaneous, historical or drawn from a population-based database.
A standardized form was used by two independent reviewers to extract information from eligible studies. Data fields included the following: total number of patients enrolled; number eligible for inclusion; number of responders/consenters; number of losses after consent; number of patients at baseline; number of subjects at risk of developing IBS in exposed and non-exposed groups; number of patients exposed/non-exposed with and without IBS at different time points; number of drop outs/losses to follow-up at each time point; and relative risks, odds ratios or incidence/prevalence rates.
It was anticipated that definitions of acute gastroenteritis and criteria for diagnosis of IBS would vary among studies. For the purposes of data pooling, the original authors’ definitions were respected. Similarly, no attempt was made to differentiate among various sources of controls when estimating odds ratios.
Methodological quality assessment
Methodological quality of each study was assessed by two independent reviewers (M.T. and D.T.K.) using a modified New Castle Ottawa Scale for non-randomized studies in meta-analysis (http://www.ohri.ca/programs/clinical_epidemiology/oxford.htm). Separate scales were used for uncontrolled vs. controlled cohort studies, and assessed factors critical to observational design.
For prospective controlled cohort studies, a scale from 0 to 28 evaluated 14 factors as to whether they were described fully (score 2), partially (score 1) or not at all (score 0). These factors included: (i) selection of unexposed controls, (ii) selection of exposed cohort, (iii) ascertainment of exposure, (iv) demographics, (v) exclusion criteria for premorbid IBS, (vi) appropriate analysis according to study design, (vii) assessment of confounding, (viii) assessment of selection/non-differential bias, (ix) appropriate statistical methods, (x) accounting for dropouts/losses to follow-up, (xi) assessment of outcome, (xii) data collection by standard methods, (xiii) justification of sample size, (xiv) comparative analysis of responders and non-responders. For uncontrolled studies, a similar scale with maximum score 26 points was used to assess all factors except the description of unexposed controls.
Pooled odds ratios (POR) and their 95% CI for developing IBS in exposed vs. non-exposed controls group were calculated based on a fixed effects model (Mantel-Haenstzel). In a subgroup analysis that included controlled and uncontrolled studies, a weighted pooled incident rate of IBS with corresponding 95% CI was calculated for exposed and non-exposed controls. Analysis for prognosis of PI-IBS was performed by estimating POR from studies reporting incidence of PI-IBS at different time points. Homogeneity within groups was tested by using chi-squared tests or plotting the PI-IBS event rate for each study on the L’Abbe plot. If heterogeneity existed, a random effects model was used. Publication bias was assessed by using Egger’s regression method. Statistical analyses were performed by using revman metaview (Version 4.2; The Nordic Cochrane Centre, Copenhagen, Denmark) and Comprehensive Meta-Analysis software (Version 2; Biostat, Englewood, NJ, USA).
Figure 1 depicts results of the literature search. Eighteen of the 26 studies identified were eligible for inclusion (kappa 0.90). Seven studies reported additional analysis of participants from another eligible study and one was an extended follow-up of subset of participants from another eligible study and were all excluded.9, 16, 29–34 Therefore, 18 distinct prospective cohort studies (10 controlled) were available for analysis. Gastroenteritis was self-reported in eight studies and confirmed by clinically or microbiologically in 10. Follow-up ranged from 3 months to 6 years. Table 1 provides characteristics of the included studies and their methodological quality assessment scores.
One controlled cohort study was excluded from estimates of relative risk because the control group was patients diagnosed with mononucleosis. Among the nine remaining studies with healthy controls, intestinal infection was associated with increased odds of developing IBS at study end (POR = 5.86; 95% CI: 3.60–9.54) (see Figure 2). Among studies reporting each specific time point, the odds of developing IBS was increased at 3 months (POR = 7.58; 95% CI: 4.27–13.45), 6 months (POR = 5.18; 95% CI: 3.24–8.26), 12 months (POR = 6.37; 95% CI: 2.63–15.40) and 24–36 months (OR = 3.85; 95% CI: 2.95–5.02) after infection (see Figures 3–6). As the test for heterogeneity was significant, a random effects model was used.
Among all studies (controlled and uncontrolled), the pooled incidence of IBS at study conclusion was 10% (95% CI: 9.4–85.6). Among controls, pooled incident rate of subjects with IBS was only 0.35% (95% CI: 0.33–0.36).
Only one study reported premorbid psychological profiles and therefore data for this study were not included in the analysis.35 Among (n = 4) studies that reported Hospital Anxiety and Depression Scale scores10, 11, 13, 21 at study assessment, subjects who developed PI-IBS were more anxious and depressed than those who did not. The mean differences in anxiety and depression scores between subjects with and without PI-IBS were 2.7 (95% CI: 1.75–3.68) and 1.77 (95% CI: 0.04–3.50), respectively. Subjects with PI-IBS were also younger than those without PI-IBS [mean age difference −2.9 years (95% CI: −5.00 to −0.79)] and had a longer duration of fever [mean difference 0.77 days (95% CI: 0.32–1.22)]. Among studies that reported other markers for severity of illness such as duration of abdominal pain, vomiting and mucous stools, diarrhoea, bleeding and weight loss, we found no statistical significant difference in the mean duration of illness among PI-IBS and non-PI-IBS groups. However, because only a limited number of studies evaluated these parameters, we may have been underpowered to detect the difference.
Assessment of heterogeneity
In post hoc analysis, the observed heterogeneity among eligible studies was attributed to a single outlier result. When the large study by Rodriguez and Ruigomez15 was excluded, I2; was reduced from 69% to 33% and the test for heterogeneity was no longer significant (χ2 = 10.43; d.f. = 7; P = 0.17). Upon excluding this study from analysis, the pooled OR for developing PI-IBS at study end dropped to 4.4 (95% CI: 3.48–5.55).
Assessment of publication bias
Figure 7 shows a funnel plot to assess the heterogeneity using Eggers’ regression. The results show no evidence of significant publication bias (intercept 0.66; 95% CI: –1.30 to 2.62; P = 0.453).
The findings of this systematic review are consistent with those of individual published studies of PI-IBS, and also with the systematic review by Halvorson et al.26. We have demonstrated a strong association between intestinal infection and development of symptoms consistent with IBS. A pooled estimate from nine published prospective studies suggests that the odds of developing IBS are increased about sixfold after acute gastrointestinal infection and remain significantly increased for up to 3 years. The estimated incidence of IBS after enteric infection is 10% (95% CI: 9.4–85.6).
This review supports the previously observed association between PI-IBS and psychological disturbance. Indeed, a recent landmark study by Spence and Moss-Morris35 suggests that combinations of psychological profiles such as emotion, cognitive and behavioral profiles are significant risk factors for development of IBS after GE, and support an interactive model of biology and psychology in the onset of IBS. While intestinal infection may trigger the initial symptoms, cognition, behaviour and emotions may help prolong and maintain them over time.35
We have confirmed an association between increasing risk of PI-IBS and younger age. Evidence from this review suggests that subjects who develop PI-IBS are younger than those who do not. In part, this difference could be explained by the age gender effect. Many functional disorders including bloating and constipation are more common among females, and could be partially attributed to gender-specific factors such as circulating female sex hormones and changes in rectal sensitivity during menstrual cycles.36–38 Furthermore, the increased prevalence of psychological distress among young women can also contribute to functional presentations.35, 39–41
The nature of the acute gastrointestinal illness may be an important determinant of PI-IBS risk. We found that prolonged fever during acute gastroenteritis increased the risk of PI-IBS, perhaps as a marker of illness severity. Associations between microbial virulence factors and subsequent development of IBS have been previously described.16, 27 Most recently, Soyturk et al.42 demonstrated an association between a parasite trichinella infection and development of IBS in humans. However, most such studies have considered differences within species rather than differences among a spectrum of pathogens. Halvorson et al.26 demonstrated in their meta-analysis that the risk of PI-IBS was increased with bacterial vs. non-bacterial pathogens, and that bacterial dysentery is more likely to be associated with fever, diarrhoea abdominal pain and vomiting.41 We did not find significant association between increasing risk of PI-IBS and duration of diarrhoeal illness, but we may have been underpowered to detect an effect as not all studies assessed this risk factor. Our pooled estimates were based on univariate analysis reported by individual studies and do not control for potential confounders.
When interpreting the results of this review, certain limitations need to be taken into account. Because studies of PI-IBS are initiated after exposure to an enteric pathogen, they rely heavily on participants’ recollection of their previous gastrointestinal symptoms and their episode of acute infection. In fact, the time interval from exposure to data collection among these studies ranges from 3 months13, 18, 21 to 2 years.23 Recall bias can misclassify both cases and controls, as people with more severe chronic symptoms are more motivated than controls to recall exposure and may exaggerate its intensity43 and overestimate risk. We observed no association between recall interval and risk of PI-IBS, but may have been underpowered to detect such an effect. Future studies should strive to initiate data collection immediately after exposure, and to access public health and primary care health records to confirm details of acute and antecedent illness. Prospective designs that enroll cohorts before infection, such as that by Ilnyckyj et al.18, may circumvent this problem but face other methodological challenges such as high drop rates out and non-response.
Some studies included in this review recruited patients with enteric infection from public health or general primary healthcare databases based on positive stool culture. Although this approach is convenient and offers the most reliable approach to confirming exposure to infection, it could suffer from ascertainment biases because of differences in reporting and/or healthcare access. Patients who take time to submit stool cultures or consult a physician for acute infection may not be representative of the broader population who suffer enteric infection. If patients with severe illness are disproportionately likely to seek health care and submit stool samples, the risk of PI-IBS could be inflated. Furthermore, at least one study has suggested that patients with premorbid IBS are more likely to seek health care for acute gastroenteritis.19 Without careful assessment of premorbid functional symptoms, this phenomenon will also serve to overestimate the risk of PI-IBS.
The estimated incidence of IBS can vary substantially depending on the criteria used to define the disorder. For example, Rome II criteria are more restrictive and yield lower estimates of prevalence than Rome I or Manning.44, 45 Rome II criteria require both abdominal pain and changes in bowel habit lasting 12 weeks over 12 months, while Rome I consider abdominal pain and changes in bowel habit independently. Among studies assessed in this review, the incidence of PI-IBS ranged from 3% to 14% among studies that used Rome II criteria, vs. 16% to 32% among those that used Rome I criteria (excluding Ilnyckyj et al., who reported 4%). The recent Rome III criteria may also be less restrictive than Rome II, as they require only 6 months of symptoms (active for at least 3 months).46 To date, no study of PI-IBS has used Rome III criteria.
Eight of the 18 studies included in this review did not assess a control group (see Table 1) and hence do not allow estimation of attributable risk. Seven of the 10 controlled studies selected both exposed and unexposed subjects from the same population. This approach has methodological merits because controls are better compared with cases on potential confounding factors and allows exposed cases to be compared with those of their source population, with controls having the same opportunity to develop the disease and are representative of the exposure distribution in that population. Two controlled studies matched controls from a research database. This approach is convenient and inexpensive and reduces observer bias as exposure and outcome data are collected independent of any specific hypothesis. It would be ideal for future studies to use multiple control groups to estimate attributable risk, but this approach is complex, costly and rarely feasible.
In summary, this review supports the strong association between acute gastroenteritis and new onset of IBS. However, we also identify strengths and limitations of study design that should provide direction to future research. There remains much to understand about the risk factors and mechanisms that lead to PI-IBS, so that we can better prevent and treat this common phenomenon.
Declaration of personal and funding interests: None.