Long-term prevalence of post-traumatic stress disorder symptoms in patients after secondary peritonitis

A retrospective cohort of 278 consecutive patients, who were treated surgically for secondary peritonitis between January 1994 and January 2000, was the starting cohort in the study [1]. All patients were treated at the Department of Surgery in the Academic Medical Center at the University of Amsterdam, The Netherlands. All patients who were still alive at follow up were eligible for inclusion. These patients were informed about the study by telephone in order to improve the response rate. Because of the noninterventional nature of the study, the institutional review board waived the need for informed consent.

A separate questionnaire was included to collect relevant clinical data following discharge from the hospital for peritonitis (including readmissions since discharge after surgical treatment for secondary peritonitis and use of medication during the preceding few years, and newly developed diseases and their treatment).

Patients who returned incomplete questionnaires were contacted by phone within two weeks in an attempt to complete the questionnaire by phone. Patients who did not return the questionnaires were sent the questionnaires two more times within a six week period. After these attempts had been made, patients who had given initial telephone consent were contacted again to obtain information regarding their motivations for not responding.

Demographic and clinical data at the time of the index surgical procedure (the emergency laparotomy performed at initial presentation of peritonitis) were collected from hospital charts and computerized registration system. The following information was recorded: age, sex, comorbidity, use of medication, Acute Physiology and Chronic Health Evaluation [APACHE] II score before surgery and Mannheim Peritonitis Index (MPI). Disease and surgical characteristics recorded contained aetiology of peritonitis, origin of peritonitis, surgical treatment strategy and number of relaparotomies. Postoperative characteristics recorded included the number of days spent in hospital, the number of days spent in the ICU, days of mechanical ventilation, 'open abdomen' (laparostomy) during admission, number and type of complications, number of readmissions and the mean follow-up time. Patient recall was checked using the hospital information and medication system to check readmission and use of medication. Details regarding out-of-hospital medications, such as those prescribed by the family physician, were obtained only by questionnaire.

Ninety-five per cent confidence intervals around estimates of prevalence were calculated using the method of Wilson [20]. Clinical characteristics and the prevalence of PTSD symptoms between patients who were admitted to ICU during their initial stay and those who were treated solely on the surgical ward were compared. Depending on the nature of the clinical variables, we used Pearson χ², Student's t, or Mann-Whitney U tests.

We built multivariate logistic regression models to assess the association between ICU stay and the presence of PTSD symptomology (PTSS-10 score >35) after adjusting for other factors. We adjusted for factors related to patient characteristics (age [continuous] and sex), disease characteristics (APACHE-II score at baseline [continuous] and whether patients had undergone one or more relaparotomies [yes/no]) and postoperative characteristics (days spent in hospital [transformed to base 10 in order to improve the linear relationship with outcome]) [21]. These factors were chosen either because they were identified in earlier PTSD studies and literature [21] (for instance, age, sex and comorbidity) or because they exhibited univariate significance (P < 0.1) with the dependent factor (PTSD symptomology) in our study (APACHE II score, patients undergoing more than one relaparotomy and days in hospital). If factors were highly correlated, we selected only one of the correlated factors in the multivariate model to avoid the problem of co-linearity. Odds ratios with 95% confidence intervals (CIs) were used to quantify the strength of the association. To determine the fit of the final multivariate logistic model, we calculated the area under the receiver operating characteristic curve, also known as the concordance statistic, and performed the Hosmer-Lemeshow goodness-of-fit test.

To determine whether traumatic memories acquired during the stay in hospital or the ICU played a role in the development of PTSD symptomology, we examined the percentage of patients with PTSD symptomology within each level of response on the traumatic memories questions. Because of the ordered response on the traumatic memories questions, we used the χ² test for trend to examine this relation.

The median PTSS-10 score among all patients was 22, with 25% of the patients having a score below 13 and 25% of patients with a score above 33. Using the recommended cutoff value for PTSD symptomology of 35 points on the PTSS-10 questionnaire [17, 19, 22, 23], the overall prevalence of PTSD-related symptoms was 24% (95% CI 17% to 33%). The overall prevalence of PTSD symptomology including borderline patients who scored 27 points or more was 38% (95% CI 29% to 48%).

Patient, disease and operative characteristics for ICU patients (61%) and non-ICU patients (39%) are presented in Table 1. Patients who had had an ICU stay were on average 7.5 years older than patients who were not admitted to the ICU (P = 0.011). ICU patients also had higher APACHE II score (mean difference 2.2 points; P = 0.037) and MPI score (mean difference 3.2 points; P = 0.036). Of patients who had had an ICU stay 36% underwent laparostomy (open abdomen), whereas only 8% of the ward patients underwent laparostomy (in 92% of patients admitted to the surgical ward primary abdominal closure was done; P = 0.001). A relaparotomy was significantly more common in the ICU group than in the non-ICU group (72% versus 40%; P < 0.001).

With respect to postoperative characteristics, patients had a median stay in hospital of 37 days. ICU survivors had a longer hospital stay than did non-ICU survivors (median days: 49 versus 27; P = 0.001) and suffered more nonsurgical complications (57% versus 8%; P < 0.001). Fifty-four (89%) patients required mechanical ventilation during their ICU stay. These patients were ventilated for a median of 11 days. Four of the ICU-admitted patients suffered early ARDS (within 96 hours) following peritonitis.

In an univariate analysis, using a PTSS-10 score above 35 as the cutoff, we found a prevalence of PTSD symptomology of 18% (7/39) in the non-ICU group and 28% (17/61) in the ICU group (P = 0.21). We examined several factors to determine whether they confounded the strength of the relationship between ICU stay and the probability of having relevant PTSD symptomology (Table 2). After controlling for age, sex, APACHE II score, relaparotomy and length of hospital stay in a multivariate analysis, patients admitted to the ICU were more likely to report PTSD symptomology on the PTSS-10 questionnaire than patients admitted to the surgery ward only (odds ratio [OR] 4.3, 95% CI 1.11 to 16.5).

Other factors that were significantly associated with more PTSD symptoms in the multivariate model included gender, age, and severity of disease at initial surgery. Females were more likely to develop PTSD symptoms than were males (OR 3.5, 95% CI 1.2 to 10.6). With every one-year decrease in age, the likelihood of developing PTSD symptoms decreased (OR 0.93, 95% CI 0.89 to 0.98). Finally, with every point increase in APACHE-II score, the chances of developing PTSD symptoms increased (OR 1.1, 95% CI 1.002 to 1.25). Therefore, the main reason for finding a stronger relation between ICU stay and PTSD symptomology in the multivariate model is that older patients are less likely to develop PTSD symptoms. Because ICU patients on average were older than non-ICU patents, the unadjusted relationship underestimated the effect of ICU on PTSD symptoms. Males were also less likely to report PTSD symptomology (OR 0.95, 95% CI 0.91 to 0.98), but because of the comparable sex distribution in ICU and non-ICU patients it did not confound the relation between ICU stay and PTSD symptomology (Table 2). Length of hospital stay was associated with more PTSD symptomology, and it was therefore also a confounder for the relation between ICU stay and PTSD symptomology because hospital stay was markedly longer in ICU patients than in non-ICU patients. The area under the receiver operating characteristic curve for the final multivariate model was 0.77 (95% CI 0.66 to 0.88). This indicates that if we were to randomly choose one patient above the PTSS-10 cutoff value and one patient below, the probability that the patient above the cutoff would have a higher predicted risk for PTSD symptomology based on the model is 79%. Differences in observed versus predicted probabilities were small, with the Hosmer-Lemeshow test yielding a P value of 0.41.

In the total study population, traumatic memories were associated with more PTSD symptomology (Table 3). Patients reporting more traumatic memories during their ICU or hospital stay reported significantly more PTSD symptoms on the PTSS-10. Patients with nightmares, panic attacks, intense pain and difficulty breathing during their ICU or hospital ward stay had higher median scores than did patients reporting no traumatic memories from the ICU or hospital ward (Table 3). There were, however, no statistically significant differences between the ICU group and the non-ICU group of patients with respect to reporting of traumatic memories (nightmares: χ² = 5.84, P = 0.12; fear or panic attacks: χ² = 6.9, P = 0.075; pain: χ² = 1.01, P = 0.80; and difficulty breathing: χ² = 5.3, P = 0.15).

Ideally, PTSD is diagnosed using a SCID [23], in accordance with the DSM-IV [9]. SCID is a semi-structured diagnostic interview designed to allow clinicians and researchers to make reliable DSM-IV psychiatric diagnoses. In recent studies it has been established that a self-report PTSS-10 questionnaire can be as useful a tool in determining which patients are suffering from PTSD symptomology [5, 17]. These studies found significantly higher PTSS-10 scores in patients with a SCID-II PTSD diagnosis than in patients without. The sensitivity in these studies varied from 77% to 100%, and specificity from 92% to 98% when using a cutoff score of 35 [5, 17]. However, these estimates were imprecise because of the small sample sizes in these studies. It is unclear to what extent sensitivity and specificity of the PTSS-10 instrument for PTSD may vary according to disease and other characteristics [31].

Recognition of the distinction between PTSD symptoms captured by the PTSS-10 and a PTSD diagnosis is vital, because this questionnaire does not give a DSM-IV diagnosis but only an indication of the level of PTSD symptomology. Clinically, a score on the PTSS-10 above the cutoff should prompt the attending physician to refer the patient to a psychologist to conduct a SCID [9].

Our results suggest that the (persisting) presence of traumatic memories is likely to be relevant to development of PTSD-related symptoms following a traumatic event, and not the ICU stay alone, because we observed a strong linear relationship between traumatic memories and PTSS-10 score. We assessed these traumatic memories (or adverse experiences) in accordance with patients' recollections. This may limit the conclusions one can make, because it is possible that perception of a traumatic experience may contribute to long-term PTSD symptomology, hence making a causal conclusion impossible.

Information concerning other unrelated traumatic experiences or life events that may have occurred after hospital admission was not collected. Therefore, the influence of superimposed trauma cannot be ruled out [6]. Also, because this was a retrospective study, it was also not possible to collect PTSD data on patients before their peritonitis. However, considering the acute nature of peritonitis, it would be difficult to collect such data even in a prospective trial. Given the impact of a severe life-threatening illness such as peritonitis, a relationship with the development of PTSD symptoms is plausible, but causality cannot be established when no information is available on other life events.