Prognostic Modeling of Preoperative Risk Factors of Pouch Failure

BACKGROUND: The prospect of pouch failure needs to be considered when evaluating the management strategy for patients who may be candidates for an ileo anal pouch. An ability to predict the likelihood and timing of failure preoperatively may influence surgical decision making. OBJECTIVE: The aim of this study was to define a preoperative prognostic model for ileoanal pouch failure. DESIGN: A novel random forest methodology was used to evaluate the prognostic significance of 21 preoperative potential risk factors for pouch failure. A forest of 3000 random survival trees was grown to estimate pouch failure for each patient and to identify important risk factors that maximize survival prediction. SETTINGS: This study took place at a tertiary referral department at a major academic medical center. PATIENTS: Patients undergoing an ileoanal pouch at this institution between 1983 and 2008 were included. MAIN OUTCOME MEASURES: The primary outcome measured was pouch survival. RESULTS: Between 1983 and 2008, 3754 patients underwent ileoanal pouch. Type of resection (total proctocolectomy vs completion proctectomy), type of anastomosis (stapled vs mucosectomy), patient diagnosis (mucosal ulcerative colitis and others vs Crohn's disease) and diagnosis of diabetes had the strongest effect on pouch survival. Predicted survival was worse for completion proctectomy (HR, 1.44; 95% CI, 1.08-1.93), Crohn's disease (HR, 2.37; 95% CI, 1.48-3.79), handsewn anastomosis (HR, 1.72; 95% CI, 1.23-2.42), and diabetes (HR, 2.31; 95% CI, 1.25-4.24). Pouch survival was worse for the oldest group of patients. LIMITATIONS: This study was limited by its retrospective nature. CONCLUSION: Random forest techniques applied to a large number of patients undergoing the ileoanal pouch identify factors associated with pouch failure. Attention directed at these factors may improve outcomes for these patients.