Background
Frailty has emerged as an important predictor of operative risk among elderly surgical patients. However, the complexity of prospective frailty scores has limited their widespread use. Our goal was to develop two frailty-based surgical risk models employing only routine preoperative data. Our hypothesis was these models could easily integrate into an electronic medical record (EMR) to predict 30-day morbidity and mortality.
Study Design
ACS-NSQIP participant use files from 2005–2010 were reviewed, and patients ≥65 years old who underwent elective lower gastrointestinal surgery were identified. Two multivariate logistic regression models were constructed and internally cross-validated. The first included simple functional data, a comorbidity index based on the Charlson Comorbidity Index, demographics, BMI, and laboratory data (albumin <3.4g/dL, hematocrit<35%, creatinine>2mg/dL). The second model contained only parameters that can directly auto-populate from an EMR: demographics, laboratory data, BMI, and ASA score. To further assess diagnostic accuracy, receiver operating characteristic (ROC) curves were constructed.
Results
76,106 patients met criteria for inclusion. 30-day mortality was seen in 2,853 patients or 3.7% of the study population. 18,436 patients (24.2%) experienced major complication. The c-statistic of the first expanded model was 0.813 for mortality and 0.629 for morbidity. The second simplified model had a c-statistic of 0.795 for mortality and 0.621 for morbidity. Both models were well calibrated per the Hosmer-Lemeshow test.
Conclusions
Our work demonstrates that routine preoperative data can approximate frailty and predict geriatric-specific surgical risk. The models’ predicative power was comparable to that of established prospective frailty scores. Our calculator could be used as a low cost simple screen for high-risk individuals who may require further evaluation or specialized services.