Study design and patients

The ethics committees of all study sites approved the study protocol. Between July 2006 and February 2008, patients were consecutively recruited from virtually all places providing gastroenterological treatment in Switzerland (www.ibdcohort.ch). These included the study centres (ie, University Hospitals of Basel, Bern, Geneva, Lausanne, St Gallen and Zurich), regional hospitals and private practices. Recruitment was performed as follows: the treating gastroenterologists provided their patients with information about the study. When a patient had given written informed consent, gastroenterologists and study nurses completed the enrolment sheet asking about medical history, clinical status and laboratory parameters. More precisely, baseline data included gender, age, disease duration, previous hospitalisations, medication, height, weight, smoking status and the CD activity index.11 At the same time, patients received the PDS with the instruction to complete it and return it directly to the data centre (given that gastroenterologists assessed exacerbation, they had to be blind for post-traumatic stress). To facilitate generalisation of findings to the general population of CD patients treated in Switzerland, the only eligibility criterion was confirmation of a diagnosis of CD based on the criteria of Lennard–Jones12; the latter allows confirmation of a diagnosis of CD by radiological, endoscopic or histological findings, or by surgery.

Subsequent to enrolment and collection of baseline information, patients were followed over an observation period of 18 months. They were advised to inform their treating gastroenterologists whenever they experienced worsening or amelioration of symptoms, in which case an appointment was scheduled to clarify whether fluctuation of symptoms was caused by fluctuation of CD activity or by some other clinical state (eg, infections, gallstones). Regular meetings among gastroenterologists ensured the use of matched diagnostic and therapeutic criteria, as issued by the European Panel on the Appropriateness of CD Treatment.13 In addition, the authors informed the data centre in case of inconclusive or missing data in order to improve the enrolment sheet accordingly.

The analysis was performed on an a priori defined sample of 468 patients to assure accurate control for potential confounders without overfitting statistical models. A sample of 367 patients has a power of 90% to detect an OR of adverse events of 1.5 for an increase of 1 SD in post-traumatic stress at a significance level of α<0.05 (two tailed). We computed this power using Power and Precision software for an overall event rate of 25% over 18 months.14 However, we oversampled the calculated sample size by 40% and enrolled 597 patients (367 corresponds to 60% of 597) to account for projected loss in response rate, dropouts and missing values.

Measures

To assess post-traumatic stress we used the German and French version of the PDS,15 which is a self-rated questionnaire with 17 items. We replaced the term ‘event’ by ‘intestinal disease’ in the PDS questionnaire to ensure measurement of disease specific post-traumatic stress. Each of the three main characteristics—namely re-experiencing CD (“Did you have bad dreams or nightmares about the intestinal disease?”, four items), avoidance of cues related to CD (“Did you try not to think about, talk about or have feelings about the intestinal disease?”, seven items) and hyperarousal (five items)—is represented by a separate symptom cluster. All items are rated on a 4 point Likert Scale (0–3; total score 0–51) with a higher score indicating a higher level of post-traumatic stress. Validity measures in our study included a variance of item means of 0.39 and a variance of item variances of 0.01 (ie, good weighting) and a Cronbach's α of 0.84 (ie, good overall reliability). If no more than one item per symptom cluster was missing, replacing it by the mean of the non-missing items of the respective cluster did not affect the validity (personal communication with Stieglitz15 in 2009).

Patients scoring ≥15 points are likely to have fully developed PTSD, in which case a structured clinical interview is recommended.4 A structured clinical interview is mandatory to diagnose full PTSD. Our measure of interest was not PTSD. Instead, we examined post-traumatic stress along a continuum of severity (ie, already at levels far below those seen in fully developed PTSD). This approach has several advantages16 ,17: post-traumatic stress can easily be measured in larger samples by means of the PDS; post-traumatic stress can be detected in patients who may benefit from therapeutic interventions despite not having full PTSD; and ongoing revision of the diagnostic criteria of PTSD cannot affect the validity of the results (http://pn.psychiatryonline.org/content/45/10/8.full; last accessed 14 June 2010). We were primarily interested in the total PDS score but also in re-experience, avoidance and hyperarousal as separate scores.

Measures of exacerbation of CD were flares, extraintestinal manifestations, complications and non-response to therapy. These were selected because they have a high clinical impact, force patients to seek gastroenterological healthcare and, in combination, allow for a complete evaluation of the disease course. They were combined to one binary outcome which is referred to as adverse events (present/absent), as previously described.18 In brief, flares were defined as an increase of 100 points or more from baseline on the CD activity index. Any need for an increase in medication was considered a treatment failure. Extraintestinal manifestations were included if they occurred during the observation period. Complications included strictures, fistulae, malignancies or any other need for surgery.

Clinical experience has shown that disease duration, strictures and fistulae requiring hospitalisations, need for medication (ie, 5-aminosalicylates, sulfasalazin, steroids, immunosuppressors, antitumour necrosis factor α agents, antibiotics), low body mass index, baseline disease activity and smoking are predictive of the disease course.18 For this reason, we controlled not only for those factors but also for gender and age because these are the most important demographic parameters.

Data analysis

We analysed the data using SPSS 15 for Windows (Chicago, Illinois, USA) excluding the few cases with missing data.

Firstly, we described the total sample. We compared the groups with and without adverse events using standard methods. For mean and percentage differences, we computed asymptotic 95% CI and two sided p values. We used the t test for normally distributed metric variables (age, body mass index) and the Mann–Whitney U test for extremely skewed metric variables (diagnosis duration, hospitalisation days related to CD) or non-metric quantitative variables (scores). For categorical variables, we compared each category separately using Fisher's exact test.

Secondly, we divided the sample into three groups according to post-traumatic stress levels. Group 1 included all patients with a PDS score of 0, group 2 included all patients scoring <15 (hence this group also included patients in group 1) and group 3 comprised patients scoring ≥15 points. We compared group 3 with groups 1 and 2 in terms of relative risks and ORs of adverse events.

Thirdly, we performed four binary logistic regression models: a model for total post-traumatic stress (primary score), a model for re-experiencing symptoms, a model for avoidance symptoms and a model for hyperarousal symptoms (secondary scores). Each model contained the respective symptom score plus the above mentioned potential confounders as independent variables and adverse events (yes/no) as the outcome.

Fourthly, we performed 16 stratified binary logistic regressions by combing each symptom score with each measure of exacerbation (eg, re-experience and flares) in a different analysis without including potential confounders.

For the combination of total PDS score and adverse events, we set the level of significance at 0.05 and computed 95% CI for all ORs. For all other combinations, we set the level of significance at 0.01 to adjust for multiple comparisons and computed 99% CI for the ORs.