ENTRACT was a prospective, multicenter, observational cohort study in patients admitted to nine ICUs in France and Belgium between November 1, 2014 to November 30, 2018. Approval was obtained from an independent ethics committee in both countries (Comité de Protection des Personnes CPP Ile de France XI, #14059 and Ethic Committee Erasme Hospital P2016/319). The study was registered at clinicaltrials.gov NCT02349074.
The trial complied with the Declaration of Helsinki and Good Clinical Practices, and French regulatory requirements. The patients were unable to provide informed consent at inclusion so written informed consent was obtained from their surrogates before inclusion; a written informed consent was obtained from the patients when they regained consciousness, in compliance with French and Belgian laws.
Patients
All adult patients (> 18 years) admitted after OHCA to one of the participating ICUs after OHCA were screened for eligibility by the ICU physicians. Patients who were hospitalized within the first 5 days after OHCA, had a temperature > 36 °C at the time of enrollment, and still required endotracheal intubation and mechanical ventilation were considered for inclusion. Exclusion criteria were in-hospital CA, extubation before gastroscopy, any suspicion of GI tract perforation, severe bleeding diathesis despite transfusion of coagulation products, patients with cardiac valver prostheses or previous endocarditis, suspicion of Creutzfeldt-Jakob disease, pregnancy or breast-feeding, and absence of medical insurance or being under guardianship (according to French legislation). If a patient had a severe coagulation disorder (platelet count < 30 G/L, International Normalized Ratio > 2) or was receiving heparin treatment or combined platelet inhibition treatment, but did not have intractable bleeding, inclusion was possible but GI biopsies were not performed.
Examination
Patients included in the study systematically underwent esophago-gastro-duodenoscopy (hereafter called gastroscopy) during their ICU stay after rewarming and before day-5. The clinicians were aware of the results of the gastroscopy and were free to modify the treatment accordingly.
Targeted temperature management to between 33 and 36 °C was strongly advised. Hemodynamic management was protocol driven. Briefly, if the mean blood pressure decreased under 65 mmHg, or if there were clinical signs of reduced tissue perfusion, fluid responsiveness was assessed (choice of method was left to the individual center); in case of hypovolemia, a crystalloid solution was infused until resolution. If the patient was fluid unresponsive, transthoracic echocardiography and/or invasive cardiac output monitoring were performed to assess need for inotropic agents. Vasopressor dose was titrated to maintain MAP between 65 and 75 mmHg. Other treatments were left to the treating physician’s discretion.
Data collection and definitions
Data regarding CA were collected according to Utstein style [9]. History of previous cardiovascular or GI disease and previous use of Proton Pump Inhibitors (PPI) or Non-steroidal anti-inflammatory drugs were noted.
Hypothermia was defined as a body temperature < 34 °C for at least 12 h within the first 24 h.
GI symptoms before gastroscopy were defined a priori as vomiting, GI bleeding, bloody or mucoid diarrhea and feeding intolerance. PPI administration and enteral feeding were collected.
Post-CA shock was defined as the need for continuous vasopressor infusion for more than 6 h despite adequate fluid resuscitation in the first 48 h after ICU admission [10]. The duration of shock was evaluated using the number of vasopressor-free days at day 10 [11] i.e. the number of days alive without vasopressors.
Outcomes
The primary outcome measure was the presence of macroscopic ischemic lesions in the upper GI tract, and was determined during gastroscopy by the gastroenterologist from each center.
Secondary outcomes were the location (esophagus, fundus, antrum, duodenum) and the severity of the lesions; the presence of post-CA shock, the SOFA score [12] during the first 3 days and at day-5 and 8 following CA; the ICU and hospital mortality; and the neurological outcome at hospital discharge assessed using the Cerebral Performance Category (CPC) scale.
The gastroenterologist was asked to complete a standardized form describing the results of the gastroscopy in the four locations. For mucosal lesions, the gastroenterologist was asked to state whether the lesions were ischemic or due to another cause. An ischemic mechanism was suspected if the mucosa was pale and inflamed, there was diffuse or giant ulceration not of a peptic origin, or there was necrosis. The severity of ischemic lesions was a priori classified as moderate if there was erythema or edema, and severe if there was ulceration, localized necrosis, or extensive necrosis [13]. Patients were classified according to the most severe lesion that was observed regardless of the location and the presence of other less severe lesions.
The neurological outcome was graded using the Cerebral Performance Category scale at hospital discharge (ranging from 1 [no or minor sequelae] to 5 [brain death]); a score of 1 or 2 was considered as a good neurological outcome [14].
Statistical analysis
Based on the only published study that has reported gastroscopy data following CA [15], we hypothesized that the incidence of upper GI lesions would be 20%. The estimated number of patients to include with a precision of 5% was therefore 246. Based on the hypothesis that gastroscopy would not be performed in 10% of the cases, we therefore planned to enroll 270 patients.
The incidence of ischemic injury at gastroscopy was calculated (number of new cases/ number of exposed patients) by using binomial proportions and the 95% confidence interval (95%CI) was estimated from the normal approximation of a binomial assumption. Patients characteristics and outcomes are given as mean (SD) or median (25th–75th centiles) values, or numbers (%).The three groups (moderate, severe or no ischemic lesions) were compared using an ANOVA or Kruskal–Wallis test, as appropriate, for quantitative variables, and the χ2 test for qualitative variables.
To assess risk factors for GI ischemic injury, associations were tested using multinomial logistic regression and estimated using crude and adjusted odds ratios (cORs and aORs, respectively) and their 95% confidence intervals. Candidate variables for the multivariate analysis were those associated with a p value < 0.20 in univariate analysis. Pairwise analyses were done to assess confounding factors and candidate variables were removed step-by-step to form the most parsimonious multivariate model with variables associated with ischemic injury with a p value < 0.05 (significant) or trend (p value between 0.05 and 0.10). We explored a potential center effect by adding the center in multilevel logistic regression models.
Survival with good neurologic outcome defined (CPC 1 or 2) at hospital discharge was estimated using the Kaplan–Meier method, and survival curves were compared according to the presence or absence of ischemic injury using the log-rank test. Factors associated with outcome were identified using univariate Cox proportional hazard models. The results are expressed as hazard ratios (HRs) and their 95% confidence intervals. We used multivariate modeling to determine whether ischemic injury was associated with outcome, taking into account the variables associated with a p value < 0.20 in univariate analysis. Pairwise analyses were done to assess confounding factors, and interactions were sought. The proportional hazards assumption was assessed statistically using the Schoenfeld residuals test. We verified the specification of the model with the Pregibon test (linktest command) to verify that any additional independent variables should be significant by chance. We bootstrap the multivariable model as a sensitivity analysis.
Analyses of the change in SOFA score over time were based on mixed linear regression models with random intercepts for repeated measurements. Factors associated with SOFA score changes were identified using univariate mixed linear regression models. The results are expressed as regression coefficients (β) with 95% confidence intervals. We explored a potential center effect by adding a third level to the mixed linear regression.
All tests were two-tailed and p values < 0.05 were considered statistically significant. Missing data were not imputed. Analyses were performed with STATA v15.0 (StataCorp, College Station, TX, USA).