AM Arozullah, VA Chicago Health Care System, Westside Division, Chicago, IL; WG Henderson, Hines Cooperative Studies Program Coordinating Center, Hines, IL; S Khuri, Boston VA Health Care System, West Roxbury, MA; J Daley, Partners Health Care/Massachusetts General Hospital, Boston, MA

Objectives: To determine the patient-specific and operation-specific predictors of mortality subsequent to developing postoperative pneumonia (POP).

Methods: We studied 401,480 cases undergoing major noncardiac surgery that were enrolled in the National VA Surgical Quality Improvement Program from September 1,1996 through August 31,1999. We excluded 15,649 cases with either preoperative pneumonia, ventilator dependence, do not resuscitate status, or comatose status leaving 385,831 cases for analysis. POP was defined using the Centers for Disease Control and Prevention (CDC) definition of nosocomial pneumonia. Failure to rescue was defined as POP followed by mortality within 30 days of the index operation. Patient-specific and operation-specific variables were selected as potential predictors based on separate, previously validated, models predicting POP and 30-day postoperative mortality. A base logistic regression model including all potential predictors was developed with failure to rescue following POP as the dependent variable. Those variables that were not significant predictors (p >=0.05) were sequentially deleted from the base model. The deleted variables were reintroduced at various stages of model development to reassess their contribution to the model. The final model included 25 significant predictors (p<0.05).

Results: There were 6759 cases (1.8% of total cases analyzed) that developed POP. Among these POP cases, there were 1256 deaths or cases with failure to rescue. Patient-specific predictors of failure to rescue included age > 60 years, dependent functional status, ASA class, history of alcohol use, recent weight loss, disseminated cancer, chronic steroid use, chronic obstructive pulmonary disease (COPD), stroke, impaired sensorium, central nervous system tumor, ascites, and bleeding disorder. Preoperative laboratory values that were significant predictors of failure to rescue included albumin level < 3.5 g/dl, blood urea nitrogen > 30 mg/dl, sodium <= 135 mmol/l, and increased bilirubin and serum glutamic oxaloacetic transaminase (SGOT) levels. The operation-specific predictors of failure to rescue included type of surgery (thoracic, abdominal aortic aneurysm repair, upper abdominal, and neurosurgery), emergency surgery, longer operation time, and higher number of red blood cell units given intraoperatively. The c-statistic for the final model including all significant patient-specific and operation-specific predictors was 0.887.

Conclusions: Patient-specific predictors of failure to rescue following POP relate to general health status (age, functional status, ASA class, recent weight loss), immune status (alcohol use, disseminated cancer, chronic steroid use), neurological status (stroke, impaired sensorium, central nervous system tumor), pulmonary status (COPD), and liver function (ascites, bilirubin and SGOT level). Operation-specific predictors relate to type and location of surgery, operative time, and blood transfused during surgery.

Impact: The patient-specific and operation-specific predictors outlined above may be useful in predicting the risk of failure to rescue following POP for subjects undergoing major noncardiac surgery. Studies to evaluate interventions such as preoperative nutritional supplementation, aggressive pulmonary rehabilitation, and intensive perioperative respiratory care for patients at high risk for failure to rescue should be pursued in the future.