Risk Stratification for Major Postoperative Complications in Patients Undergoing Intra-abdominal General Surgery Using Latent Class Analysis

BACKGROUND: Preoperative risk stratification is a critical element in assessing the risks and benefits of surgery. Prior work has demonstrated that intra-abdominal general surgery patients can be classified based on their comorbidities and risk factors using latent class analysis (LCA), a model-based clustering technique designed to find groups of patients that are similar with respect to characteristics entered into the model. Moreover, the latent risk classes were predictive of 30-day mortality. We evaluated the use of latent risk classes to predict the risk of major postoperative complications. METHODS: An observational, retrospective cohort of patients undergoing intra-abdominal general surgery in the 2005 to 2010 American College of Surgeons National Surgical Quality Improvement Program was obtained. Known preoperative comorbidity and risk factor data were entered into LCA models to identify the latent risk classes. Complications were defined as: acute kidney injury, acute respiratory failure, cardiac arrest, deep vein thrombosis, myocardial infarction, organ space infection, pneumonia, postoperative bleeding, pulmonary embolism, sepsis/septic shock, stroke, unplanned reintubation, and/or wound dehiscence. Relative risk regression determined the associations between the latent classes and the 30-day complication risks, with adjustments for the surgical procedure. The area under the curve (AUC) of the receiver operator characteristic curve assessed model performance. RESULTS: LCA fit a 9-class model on 466,177 observations. The composite complication risk was 18.4% but varied from 7.7% in the lowest risk class to 56.7% in the highest risk class. After adjusting for procedure, the latent risk classes were significantly associated with complications, with risk ratios (95% confidence intervals) (compared to the class with the average risk) varying from 0.56 (0.54-0.58) in the lowest risk class to 2.15 (2.11-2.20) in the highest risk class, a 4-fold difference. In models incorporating surgical procedure, latent risk class, and the American Society of Anesthesiologists Physical Status, the AUC for composite complications was 0.76 (0.76-0.76). However, for individual complications, there was heterogeneity in model performance using these variables, with AUCs ranging from 0.70 (0.69-0.71) for pulmonary embolus to 0.90 (0.90-0.90) for acute respiratory failure. CONCLUSIONS: LCA can be used to classify patients undergoing intra-abdominal general surgery based on preoperative risk factors, and the classes are independently associated with postoperative complications. However, model performance is not uniform across individual complications, resulting in variations in the utility of preoperative risk stratification tools depending on the complication evaluated.