Article Text

PDF

Original article
Incidence of irritable bowel syndrome and chronic fatigue following GI infection: a population-level study using routinely collected claims data
  1. Ewan Donnachie1,
  2. Antonius Schneider2,
  3. Michael Mehring2,
  4. Paul Enck3
  1. 1 Department of Internal Medicine VI: Psychosomatic Medicine and Psychotherapy, University Hospital Tübingen, Germany and Institute of General Practice, Klinikum rechts der Isar, Technische Universität München, Munich, Germany
  2. 2 Institute of General Practice, Klinikum rechts der Isar, Technische Universität München, Munich, Germany
  3. 3 Department of Internal Medicine VI: Psychosomatic Medicine and Psychotherapy, University Hospital Tübingen, Tübingen, Germany
  1. Correspondence to Ewan Donnachie, Institute of General Practice, Klinikum rechts der Isar, Technische Universität München, Orleansstraße 47, 81667 Munich, Germany; ewan{at}donnachie.net

Abstract

Objectives To investigate the occurrence of postinfectious IBS in routine outpatient care, comparing different types of GI infection and its interaction with psychosomatic comorbidity.

Design Retrospective cohort study using routinely collected claims data covering statutorily insured patients in Bavaria, Germany. Cases were defined as patients without prior record of functional intestinal disorder with a first-time diagnosis of GI infection between January 2005 and December 2013 and classed according to the type of infection. Each case was matched by age, sex and district of residence to a patient without history of GI infection. Prior psychological disorder (depression, anxiety or stress reaction disorder) was assessed in the 2 years prior to inclusion. Proportional hazards regression models were used to estimate the HRs for GI infection and psychological disorder. Chronic fatigue syndrome (CFS) was assessed as a comparator outcome.

Results A total of 508 278 patients with first diagnosis of GI infection were identified, resulting in a matched cohort of 1 016 556 patients. All infection types were associated with an increased risk of IBS (HR: 2.19–4.25) and CFS (HR 1.35–1.82). Prior psychological disorder was a distinct risk factor for IBS (HR: 1.73) and CFS (HR: 2.08). Female sex was a further risk factor for both conditions.

Conclusion Psychological disorder and GI infections are distinct risk factors for IBS. The high incidence of non-specific GI infection suggests that postinfectious IBS is a common clinical occurrence in primary care. Chronic fatigue is a further significant sequela of GI infection.

  • irritable bowel syndrome
  • postinfectious syndrome
  • biopsychosocial model
  • chronic fatigue syndrome

Statistics from Altmetric.com

Significance of this study

What is already known on this subject?

  • GI infection and psychological disorder are well-established risk factors for developing IBS, but understanding of their role and interaction remains poor.

  • Previous studies have concentrated on unusual events such as traveller’s diarrhoea and infectious outbreaks that are not representative of wider primary care.

What are the new findings?

  • The population-based data analysis suggests that GI infection and psychological disorder appear to be distinct risk factors, contributing additively to the risk of developing both IBS and chronic fatigue syndrome.

  • Non-specific GI infection, comprising the vast majority of cases in general practice, is associated with an equally high risk as compared with specific bacterial or viral infections.

How might it impact on clinical pratice in the forseeable future?

  • General practitioners should be aware of postinfectious sequela following all types of GI infection.

  • Independent of GI infection, psychosocial status should be considered to provide timely guidance to patients at high risk of IBS and chronic fatigue.

Introduction

IBS is a common condition with an estimated prevalence between 10% and 30% in western populations. It is characterised by the presence of chronic diarrhoea, constipation, bloating and/or abdominal pain after exclusion of other causes.1 2 Once viewed primarily as a psychogenic condition and defined based on the absence of organic disease, a wealth of research over the past 30 years has led to a more nuanced, multifactorial understanding of IBS. The newly published Rome IV guideline reflects the variety of physiological findings by defining functional GI disorders as disorders of gut–brain interaction related to motility disturbance, visceral hypersensitivity, altered mucosal and immune function, altered gut microbiota and altered central nervous system processing.3 While this definition establishes a neurogastroenterological basis for IBS, understanding of the interrelation between these factors is incomplete.

An important subset of IBS cases occurs following viral and bacterial GI infection. Recent reviews identified a variety of settings in which cohorts of individual patients (eg, with traveller’s diarrhoea) or collective groups (eg, after infectious outbreaks) are found to be at high risk of developing persistent IBS symptoms.4–6 Comparison of such studies is however difficult due to the variety of infections investigated, the diagnostic criteria applied and differing methods of data collection and analysis. Although an increased risk of IBS has been observed in many diverse studies, estimates of its incidence range widely from 4% to 39%.

Gwee and colleagues were the first to demonstrate that patients with high stress levels and psychological disorders have a higher risk of experiencing postinfectious IBS.7 8 This led to the formulation of a biopsychosocial model for IBS, where psychological traits predict the development of chronic symptoms following inflammatory and physiological disturbances.9 10 Postinfectious IBS may therefore be viewed as a specific case of a postinfectious syndrome. Similar postinfectious hypotheses have been proposed both for visceral syndromes such as dyspepsia and overactive bladder syndrome as well as for non-visceral syndromes such as fibromyalgia and chronic fatigue syndrome (CFS).11 12 Fatigue in particular is known to be a frequent comorbidity in patients with IBS,13 14 but it is unknown whether this commonality extends to their pathogenesis.

While the biopsychosocial approach to treatment is now widely accepted,3 it remains unclear what role psychological factors play in the pathogenesis of IBS.15 Patients with IBS exhibit more psychological distress and disorder than people with IBS who do not seek medical assistance, leading to the hypothesis that psychosocial factors cause illness behaviour but not IBS itself.16–18 The interdependence of psychosocial factors and physiological processes poses a central difficulty, yielding a system that is not amenable to causal analysis. Often, the information available to the researcher is dependent to some extent on the factors under investigation. For example, follow-up of patients after an infectious outbreak or case of traveller’s diarrhoea is usually subject to recall bias, where patients with psychological disorder are more likely to report having persistent IBS symptoms.19 Furthermore, in the context of the wider primary care setting, the inter-relation of infection models and biopsychosocial approaches have not to date been investigated. Our aim was therefore to use routinely collected ambulatory care data to evaluate the risk of IBS onset in relation to GI infection and psychosomatic comorbidity.

Objectives

The present study uses routinely collected claims data to observe a large population-based cohort of patients from their first record of gastroenterological infection. We first seek to reproduce the finding that both gastroenteritis and psychological comorbidities are risk factors for IBS. We then investigate three questions of interest that, to date, have not sufficiently been addressed. First, we use the German modification of the International Classification of Diseases, 10th revision(ICD-10) to differentiate between different types of GI infection, assessing whether the type and the severity of infection affect the risk of developing postinfectious IBS. Second, we investigate whether pre-existing psychological disorders are risk factors for IBS, and whether this risk is experienced to the same extent by patients with and without GI infection. Third, we assess the incidence of CFS as a secondary outcome. This provides important context regarding the role of psychological disorder and the specificity of IBS as a sequela of GI infection.

Methods

Study data

We analyse anonymous claims data held by the National Association of Statutory Health Insurance Physicians of Bavaria (German: Kassenärztliche Vereinigung Bayerns, KVB). The data cover approximately 85% of the population of Bavaria (2013: 10.5 million people with statutory health insurance) and are submitted by physicians primarily for the purpose of remuneration. Alongside the fees and procedures claimed, diagnoses are recorded using the German modification of the ICD-10 coding system. As detailed in the online supplementary information, the ICD codes were grouped to form relevant illness categories and account for inconsistent coding practices.

Supplementary Material

Supplementary data

Cohort

Figure 1 illustrates the selection of the cohort, comprising patients aged between 18 and 60 years with first recorded diagnosis of a gastroenterological infection (ICD-10 codes A01–A09) between January 2005 and December 2013 inclusive. Patients with a prior diagnosis of IBS or functional intestinal disorder were excluded. Infections were classified as Salmonella, Campylobacter, Escherichia coli, other bacterial infection, protozoan infection, viral infection or non-specific infection. Patients with a record of cholera were excluded because this condition is very rare in Bavaria and there are known data quality problems (A00 was used as a default value). Due to the high number of non-specific cases, a random sample of 5% of these patients was taken from each quarterly period and the remaining patients excluded.

Figure 1

Schematic diagram of the study flow.

Matching

Each case was matched by age, sex and district of residence to a patient consulting a physician in the same quarter but without prior record of IBS or GI infection. Matching by district helps minimise possible differences resulting from regional variation in utilisation patterns.20

Observation period

Time was measured in quarters, with t=0 corresponding to the quarter of first GI infection (for cases) or the quarter in which matched (for controls). Patients were followed up from this time for a maximum of 5 years (20 quarters). The 2 years prior to this follow-up period served as a baseline period, during which patient characteristics were assessed. In particular, the presence of prior depression, anxiety or stress reaction disorders (defined for the purpose of this paper as ‘psychological disorder’) was identified by the record of a corresponding diagnosis during the baseline period.

Outcomes

The outcomes of interest are IBS and CFS. These primary outcomes are defined as follows:

  1. IBS: defined as a secured diagnosis with K58 or F45.32 in one or more quarters during follow-up

  2. CFS: defined as a secured diagnosis with G93.3 or F48.0 in one or more quarters during follow-up.

In addition to the primary definition of IBS using formally correct ICD-10 coding, a second, ‘weakened’ IBS outcome was applied. This definition includes patients coded with one of (1) IBS-like symptoms (diarrhoea, constipation, bloating); (2) a secured or suspected functional intestinal disorder (K59); or (3) suspected IBS (K58 marked as a suspected diagnosis). These codes must be present in at least two quarters (not necessarily successively) during follow-up, with the first observation determining the event time. This is intended as a sensitivity analysis, including patients with possible IBS who did not receive a formally correct diagnosis. Finally, for cases only, the occurrence of recurring gastroenteritis was assessed.

Statistical analysis

Data processing and statistical analysis was conducted using an Oracle database (V.11g) and the R language and environment for statistical computing (V.3.1.1).

Baseline assessment

In a preliminary step, descriptive statistics were calculated to describe the cohort at baseline. Of interest is the age and sex distribution of the cohort, the comorbidities of the groups and any differences between the infection types. As a measure of prior healthcare utilisation, the number of treatment episodes (ie, utilisation of a given practice in a given quarter) was calculated. Patients lying in the upper empirical quartile were defined as ‘high utilisers.’

Kaplan-Meier estimation

The Kaplan-Meier estimator was used to describe the cumulative incidence of the outcomes while correcting for dropout.21 Stratification by type of infection and the presence of prior psychological comorbidity enabled a visual summary of the results. CIs are reported in the online supplementary information.

Proportional hazards regression

For each outcome of interest, the Cox proportional hazards model was used to measure the effect of GI infection and prior psychological disorder.22 The findings of the Kaplan-Meier estimators could thus be quantified with adjustment for further potential confounders. In order to account for the matching process, the models adjusted for age (four groups with reference 18–30) and gender, with stratification by district of residence to allow for differing baseline risks.23 Potential time effects (eg, due to the evolution of coding practices) are modelled by adding a variable for the time difference in quarters between the matching quarter and the first quarter of 2005. The effects of interest are the presence of psychological disorder at baseline (reference level: no psychological disorder), the type of infection (reference level: controls without infection) and the interaction of these two factors. If psychological disorder is primarily a predisposing or perpetuating factor, we would expect this to be expressed through a positive interaction effect rather than through the main effect. The stability of the model was assessed through comparison with alternative model specifications, which are detailed in the online supplementary information.

Results

Cohort

Of the approximately 17 million patients consulting ambulatory physicians between January 2005 and December 2013, 11.3 million were aged between 18 and 60. After applying the inclusion and exclusion criteria to these patients, a cohort of 508 278 patients with first-time record of GI infection was obtained (table 1). The bacterial cases comprised 14 234 patients with record of Salmonella infection, 6344 with Campylobacter, 1530 with E. coli and 38 273 other bacterial infections. Protozoan infections were documented in 9796 cases. Specific viral cases, including both rotavirus and norovirus, were recorded for 305 697 patients. After taking a 5% sample of the non-specific cases, 132 404 such patients were included in the cohort for investigation.

Table 1

Baseline assessment

After identifying a matched control group without prior GI infection or IBS, the study collective consisted of 1 016 556 patients. Each patient was labelled according to the infection type, with controls allocated to the infection of their matched pair. On average, patients were observed for three quarters at baseline and 15 quarters (3.75 years) at follow-up.

Baseline assessment

Table 1 compares the case and control groups at baseline, with the cases further subdivided by type of infection. Although cases and controls are matched strictly by age (mean 35 years) and sex (50.7% female), relevant differences are found between the infection groups. The mean age ranges from 34.5 years (non-specificinfections) to 39.5 years (E. coli), while the proportion of female patients ranges from 48.9% (Campylobacter) to 58.2% (E. coli). Despite the matching, the groups differ with respect to the unmatched baseline variables. Of particular interest is the higher administrative prevalence of psychological disorder in the cases (21.9% as opposed to 17.0% among controls). Within the case subgroups, this proportion ranges from 20.9% (non-specific) to 29.9% (E. coli). Similar patterns are observed for conditions such as back pain, headache and chronic fatigue. IBS-related symptoms were also more prevalent among cases, with 12.2% having a record of abdominal pain, as compared with 9.1% among controls.

Kaplan-Meier estimation

Figure 2 displays Kaplan-Meier incidence estimates for each outcome, stratified by type of infection, case/control status and the presence of psychological disorder at baseline. The corresponding tables are provided as online supplementary information. For the primary outcome IBS, it is immediately clear that cases have a higher incidence in the 5 years post infection. Likewise, patients with psychological disorder are under higher risk than cases or controls without psychological disorder. These findings are consistent across all types of infection. E. coli appears to be associated with a substantially higher risk of IBS. With the weakened definition of IBS, a similar pattern is seen with the possible exception of non-specific diagnoses. Here, the effect of infection appears to be large, but the effect of psychological disorder is small. Among the cases, the probability of consulting with a further episode of GI infection is high, reaching 30%–40% after 2 years and 50%–60% after 5 years. Interestingly, the highest rate of recurrence is seen for patients with non-specific infections. For the outcome CFS, the incidence is substantially greater both for cases and for patients with psychological comorbidity. As CFS was not an exclusion criterion, some patients, particularly those with pre-existing psychological comorbidity, had pre-existing CFS. Nevertheless, the postinfectious incidence of CFS is substantially higher than that for IBS, both among cases and among controls.

Figure 2

Kaplan-Meier estimation of incidence for each outcome (rows) and type of GI infection (columns), stratified by case group (continuous line for cases and dashed line for controls) and by the presence of psychological disorder at baseline (red denotes prior psychological disorder; blue denotes no priori psychological disorder). CFS, chronic fatigue syndrome.

Proportional hazards model

The results of the proportional hazards regression models are presented in table 2 and visualised in figure 3. For the outcome IBS, clear significant HRs are seen for all types of infection, with the HR varying between 2.19 (viral infections) and 4.25 (E. coli). The HR associated with psychological disorder is 1.73, slightly smaller than that for female sex (HR: 2.08). Interaction effects are negatively significant for Campylobacter, other bacterial infections, viral infections and non-specific infections. The interpretation is that GI infection increases risk to a lesser degree among patients who are already at higher risk due to psychological disorder. Patients over the age of 30 years had lower risk of IBS than the reference group of patients aged 18–30 years, with significant HR for the older age groups ranging from 0.89 to 0.92. The time effect was 1.00, indicating that the risk of a documented IBS diagnosis did not change within the period of study.

Figure 3

Results of the Cox proportional hazards models. ORs with 95% CIs show the increased risk of IBS and CFS attributable to each covariable. CFS, chronic fatigue syndrome.

Table 2

Hazard ratios with 95% confidence intervals for the proportional hazards models

The alternative, weakened definition of IBS yielded similar results with comparable HR for infection and for psychological disorder. The highest risk was identified for non-specific infection (HR: 4.42). Female sex was associated with a decreased risk of experiencing this weakened IBS outcome (HR: 0.92).

All types of GI infection were associated with an increased risk of being diagnosed with CFS. The HRs vary between 1.35 (Campylobacter) and 1.82 (non-specific infection) and are thus substantially lower than the corresponding ratios for IBS. The effects of psychological disorder (HR: 2.08) and female sex (HR: 1.72) were slightly higher than for IBS. The effect of age was reversed, with patients under 30 years experiencing lower risk, although this effect is relatively small. The HR for time is significant but small at 1.01.

Discussion

To our knowledge, this is the first study to compare the incidence of IBS following different types of GI infection in a population-based setting. Previous research has concentrated on unusual events such as disease outbreaks, or focused on relatively rare occurrences of specified bacterial or protozoan infections. However, our findings suggest that patients with unspecified GI infections are equally likely to receive a subsequent diagnosis of IBS. This is surprising because laboratory testing to identify a pathogen is usually reserved for more severe or persistent infections.24 While some studies have found symptom severity and persistence to be a risk factor for postinfectious IBS,25–27 the evidence is conflicting.28 29 As non-specific GI infection is a very common diagnosis in general practice, our results suggest that postinfectious IBS is a widespread problem of high clinical significance.

Our study also provides new insight into the role of psychological disorder in the pathogenesis of IBS. We find that patients with pre-existing psychological disorder are at particularly high risk of developing postinfectious IBS. As the baseline assessment was conducted before the first record of GI infection, the effect cannot be viewed as an epiphenomenon caused by somatic illness. Furthermore, we do not observe a positive interaction between infection and psychological disorder. This suggests that psychological disorder is not a latent predisposing factor that exacerbates a somatic insult, but a distinct precipitating factor acting in patients with and without GI infection. Our results are therefore consistent with studies that suggest distinct brain-gut and gut-brain pathways30 31 and could also explain why lower rates of psychological disorder are found in patients with IBS with prior GI infection.32 Our results are also largely in accord with a recent prospective study by Löwe and colleagues, who investigated patients attending a travellers clinic and were similarly able to assess psychological comorbidity prior to infection.19 Conducting multivariable logistic regression, they found that both travellers' diarrhoea (described as a triggering effect) and pre-existing illness anxiety (described as a predisposing factor) were significant predictors for the development of IBS. However, they did not test whether the presence of psychological disorder modulates the risk due to infection.

At baseline, patients with GI infection had a higher prevalence of psychological disorder than controls. Wouters and colleagues investigated a cohort of over 18 000 people exposed to contaminated drinking water, finding that pre-existing anxiety and somatisation were linked to a lower immune response and a greater risk of developing GI infection.33 In this respect, psychological disorder may be viewed as a predisposing factor for infection, which in turn is postulated as a cause of IBS. We account for this potential bias by using a control group without GI infection and by means of regression modelling. Our tentative finding of a negative interaction effect could conceivably arise if patients with psychological disorder are more likely to become ill or consult a physician. Further research is required to verify this result using independent data.

Approximately two-thirds of all patients with IBS are female and we find that female gender is a stronger predictor of postinfectious IBS than psychological disorder. With the ‘weakened’ IBS definition, however, a higher risk was seen for male patients. A possible explanation is the frequent occurrence of recurrent gastroenteritis, which would reduce the specificity of the weakened IBS definition. On the other hand, both age and gender are known to affect the perception of IBS by patients and physicians.34 35 It is therefore conceivable that some patients presenting with IBS symptoms (eg, in order to receive a sick note) receive a diagnosis of gastroenteritis without further investigation. Conversely, it is possible that female patients and patients with psychological disorder more often receive a diagnosis of IBS. The differences between these outcomes therefore suggest a tentative hypothesis but the data are insufficient to allow any strong conclusion. Further studies are required to investigate how patients with IBS experience primary care and how this depends on age, gender and other factors.

Fatigue is an acute symptom of infection and is known to persist following infection with both viral and non-viral organisms.36 However, whereas the link between GI infection and IBS is already well established, few studies have to date investigated the incidence of chronic fatigue in this patient group. We find that GI infection increases the risk of developing CFS, but to a lesser extent than for IBS. Furthermore, the risk attributable to psychological disorder appears to be greater for CFS than for IBS. These findings are compatible with the results of previous studies that have sought to compare the postinfectious occurrence of IBS and CFS.37 38 Additionally, whereas the risk for IBS was higher among younger patients, the risk for CFS appears to be higher among older patients. This is similar to the pattern found following a water-borne outbreak of Giardia infection in Bergen, Norway, where high rates of subsequent IBS and CFS were observed.39 40

Various hypotheses have been proposed to explain the role of GI infection and psychological disorder in the pathogenesis of the conditions. One hypothesis relates to the gut-brain axis, where neural, neuroimmune and neuroendocrine pathways facilitate a bidirectional interaction.41 Neurological processes can lead to long-term dysfunction in the gut and digestive system, providing a mechanism through which psychological disorder could lead to IBS. In the reverse direction, the gut microbiota is widely accepted to have a role in brain function and is therefore hypothesised to be a partial determinant of psychological comorbidity in patients with IBS.42 Reduced microbiome diversity and altered composition of the gut bacteria have been observed for patients with IBS and CFS, with connections made to gut inflammation, dysfunction of the mucosal barrier and increased immune activation.43–45 In a wider context, infection and the gut microbiota are the focus of current research into the pathogenesis of autoimmune diseases such as type 1 diabetes and coeliac disease46–48 and of allergic conditions such as asthma.49 50 As such, there are plausible hypotheses that may explain why GI infection leads to an increased risk of both IBS and CFS.

Two previous studies have used routinely collected data to investigate the association between gastroenteritis and IBS. Ruigómez et al used the UK General Practice Research Database to observe a cohort of 5894 patients from their first episode of bacterial gastroenteritis, matching by age, sex and year of consultation.51 Follow-up occurred between 1992 and 2001, with a mean follow-up time of 4.1 years. The adjusted relative risk of IBS due to bacterial gastroenteritis was 2.2, with an increased risk likewise shown for female sex, psychological disorders (depression, anxiety, stress, sleeping disorder) and other GI comorbidity. A second study by Kowalcyk et al analysed data from the Primary Care Network Utrecht in the Netherlands.52 A cohort of 2464 patients consulting with gastroenteritis was matched by age, sex and month of consultation with a group of randomly selected patients consulting for other conditions. The 5-year relative risk for IBS was 5.4, while the baseline comparison revealed that cases more often had a history of psychological problems, social problems, fear of disease and unexplained symptoms. The two studies are thus similar in design to the present study and congruent in their results. Thus, the associations between GI infection, psychological comorbidity and IBS are reproducible in different healthcare systems with different methods of disease classification (NHS read codes, International Classification of Primary Care and ICD-10 German modification, respectively).

Strengths and limitations

A major strength of the present study is the use of a large, routinely collected set of claims data. This enabled the investigation of over 500 000 patients from the general population with GI infection and provided a matched control group with which to assess the effects of interest. In particular, the patient collective is not restricted to, for example, patients in specialist care or travellers to exotic locations. The availability of ICD-10 diagnoses facilitated the classification and direct comparison of different bacterial, viral, parasitic and unspecified infections. Furthermore, psychological comorbidities were recorded before the time of infection, thus overcoming a central limitation of many primary studies.

The use of claims data may also be viewed as a limitation. Although the coding of ICD-10 diagnoses is mandatory, data quality is not audited and the accuracy of coding varies between physicians. This is most problematic in relation to the identification of IBS and CFS as outcome events, where, for example, some physicians may choose to code symptoms rather than the underlying syndrome. To a certain extent, this limitation is addressed by grouping ICD codes and by the consideration of a weak IBS definition as an alternative outcome. We cannot, however, rule out that patients with pre-existing IBS were included in the cohort. Similarly, some patients with non-infectious gastroenteritis may have been included in the non-specific infectious group due to miscoding.

A second limitation arising from the claims data is that potentially relevant covariables are not available. For example, the risk of infection has been associated with socioeconomic status53 54 and with the use of certain medication (in particular, proton pump inhibitors and antibiotics).55 56 It is unclear to what extent our primary outcome effects are modified by such factors. In general, however, claims data are not sufficient to confirm aetiological mechanisms.

Conclusion

In summary, we find that GI infection and psychological disorder are distinct risk factors for the occurrence of both IBS and CFS. These effects are observed with bacterial, viral, protozoan and non-specific infections. The high incidence of non-specific gastroenteritis suggests that postinfectious IBS is a common occurrence in primary care. General practitioners should be aware of such postinfectious sequela and assess psychosocial status to provide timely guidance to patients at high risk.

Literature

View Abstract

Footnotes

  • Contributors ED, AS and PE conceived the study. ED conducted data processing and analysis. All authors were involved in the interpretation of the data. ED drafted the article; all authors revised it critically for important intellectual content and approved the final version. ED is the guarantor.

  • Competing interests All authors have completed the ICMJE uniform disclosure form at and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement For reasons of data protection, no additional data are available.

  • Correction notice This paper has been amended since it was published Online First. Owing to a scripting error, some of the publisher names in the references were replaced with ’BMJ Publishing Group'. This only affected the full text version, not the PDF. We have since corrected these errors and the correct publishers have been inserted into the references.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.