Introduction: In the United States, where the average American can expect to undergo seven surgical operations during her lifetime, each year at least 150,000 patients die and 1.5 million develop a medical complication within 30 days after surgery. The purpose of this study is to find out the association between demographics and clinical predictors and the outcome of higher number of postoperative complications (PC) in various types of surgical procedures. Methods: Using a single-center retrospective cohort of 50,314 adult surgical patients admitted between 2000 and 2010, we fit ordinal response models (proportional odds model, partial proportional odds model, stereotype model, and multinomial logistic models). We modeled number of major PC (acute kidney injury, severe sepsis, prolonged mechanical ventilation and ICU admission, cardiovascular complications) grouped as 0, 1, 2, and 3 or more using small set of readily available preoperative variables (age group, gender, African-American ethnicity, primary insurance payer, Charlson comorbidity index, surgery type (cardiothoracic, non-cardiac general and vascular surgery, neurologic surgery, orthopedics and gynecologic surgery), emergency surgery status, and weekend admission status). Adjusted odds ratios (OR) were calculated along with their 95% confidence intervals (CI). Results: Multinomial logistic model performed the best among the four tested models. Age older than 65 (OR 1.5, 95%CI 1.4-1.6), male gender (OR 1.2, 95%CI 1.2-1.3), Charlson comorbidity index of at least 1 (OR 3.0, 95%CI 2.3-1.6), emergency surgery (OR 3.5, 95%CI 3.3-3.7), and weekend admission (OR 1.4, 95%CI 1.3-1.5) were more likely to have 3 or more number of PC compared to no PC. Cardiothoracic surgery had the highest odds of having 3 or more PC (OR 22.9, 95% CI 20.7-25.3) compared to specialty surgery group. Medicare insurance had the largest adjusted OR of any number of PC even after adjusting for age. Area under receiver operating curve (AUC) of the model was 0.72 showing a good discriminatory ability of the model considering the simplicity of the model. Conclusions: Using a set of readily available preoperative demographic and clinical variables and novel modeling technique we can predict the number of major PC after surgery. The simple score can be developed for the use in clinical practice.
Read full abstract