The Association of Coloproctology of Great Britain and Ireland study of large bowel obstruction caused by colorectal cancer

Abstract

Background: This study was designed to investigate the early outcomes after surgical treatment of malignant large bowel obstruction (MBO) and to identify risk factors affecting operative mortality.

Methods: Data were prospectively collected from 1046 patients with MBO by 294 surgeons in 148 UK hospitals during a 12-month period from April 1998. A predictive model of in-hospital mortality was developed using a 3-level Bayesian logistic regression analysis.

Results: The median age of patients was 73 years (interquartile range 64-80). Of the 989 patients having surgery, 91.7% underwent bowel resection with an overall mortality of 15.7%. The multilevel model used the following independent risk factors to predict mortality: age (odds ratio [OR] 1.85 per 10 year increase), American Society of Anesthesiologists grade (OR for American Society of Anesthesiologists grade I versus II,III,IV-V = 3.3,11.7,22.2), Dukes' staging (OR for Dukes' A versus B,C,D = 2.0, 2.1, 6.0), and mode of surgery (OR for scheduled versus urgent, emergency = 1.6, 2.3). A significant interhospital variability in operative mortality was evident with increasing age (variance = 0.004, SE = 0.001, P < 0.001). No detectable caseload effect was demonstrated between specialist colorectal and other general surgeons.

Conclusions: Using prognostic models, it was possible to develop a risk-stratification index that accurately predicted survival in patients presenting with malignant large bowel obstruction. The methodology and model for risk adjusted survival can set the reference point for more accurate and reliable comparative analysis and be used as an adjunct to the process of informed consent.

FIGURE 1. Hierarchical logistic regression equation used for calculating the probability of death in patients undergoing surgery for malignant large bowel obstruction.

FIGURE 2. Variation of in-hospital operative mortality (plotted on the logit scale) with age across scheduled and urgent procedures. Each line represents individual prediction curves for each hospital. The median regression line (bold black) is superimposed on the prediction curves for individual hospitals.

FIGURE 3. Variation of operative mortality by age, ASA grade and operative urgency. The regression lines have been adjusted for age, ASA, operative urgency using a 3-level hierarchical logistic regression model. Sc, scheduled procedure; Ur, urgent procedure, Em, emergency procedure.

FIGURE 4. Bar charts comparing the observed and predicted in-hospital operative mortality by type of operative procedure. 95% confidence intervals are displayed around the observed outcome. (Hosmer-Lemeshow statistic: 3.21, 5df, P = 0.668).