BACKGROUND: Nonsteroidal anti-inflammatory drugs are a key component of contemporary perioperative analgesia. Recent experimental and observational clinical data suggest an associated increased incidence of anastomotic dehiscence in bowel surgery.
OBJECTIVE: The aim of this study was to conduct a systematic review and meta-analysis of anastomotic dehiscence in randomized, controlled trials of perioperative nonsteroidal anti-inflammatory drugs.
DATA SOURCES: Published and unpublished trials in any language reported 1990 or later were identified by searching electronic databases, bibliographies, and relevant conference proceedings.
STUDY SELECTION: Trials of adults undergoing bowel surgery randomly assigned to perioperative nonsteroidal anti-inflammatory drugs or control were included. The number of patients with a bowel anastomosis and the incidence of anastomotic dehiscence had to be reported or be available from authors for the study to be included.
INTERVENTION: At least 1 dose of a nonsteroidal anti-inflammatory drug was given perioperatively within 48 hours of surgery.
MAIN OUTCOME MEASURES: The primary outcome measured was 30-day incidence of anastomotic dehiscence as defined by authors.
RESULTS: Six trials comprising 480 patients having a bowel anastomosis met inclusion criteria. In 4 studies, anastomotic dehiscence rates were higher in the intervention groups. Overall rates were 14/272 participants (5.1%) in intervention arms vs 5/208 (2.4%) in control arms. Peto OR was 2.16 (95% CI 0.85, 5.53; p = 0.11), and there was no heterogeneity between studies (I2 statistic 0%).
LIMITATIONS: Sizes of available trials were small, preventing firm conclusions and subset analysis of drugs of different cyclooxygenase specificity. A precise and consistent definition of anastomotic dehiscence was not used across trials.
CONCLUSIONS: A statistically significant difference in incidence of anastomotic dehiscence was not demonstrated. However, the Peto OR of 2.16 (0.85, 5.53) and lack of heterogeneity between trials suggest that this finding may be due to a lack of power of the available data rather than a lack of effect.