T.R. Bellomo1, M. Thornton1, S.K. Lella1, B. Gaston1, C. Png1, A. Gregg1, M. Eagleton1, S.D. Srivastava1, A. Dua1, N. Zacharias1 1Massachusetts General Hospital, Vascular Surgery, Boston, MA, USA
Introduction:
Peripheral arterial disease (PAD) affects over 200 million patients worldwide and 1 in 5 of those patients are claudicants in early stage disease. Severe disease is noted by the occurrence of major adverse limb events (MALE), with upwards of 50% of patients dying within one year of PAD related amputation. Open bypass procedures have historically benefitted patients, but only for patient with chronic limb ischemia when a vein is the conduit used. The two aims of this retrospective cohort study were to 1) confirm that vein is superior to non-vein conduits and to 2) create a prediction model for 30 day MALE within a population of patients with life-style limiting claudication.
Methods:
The Vascular Quality Initiative (VQI) national database was queried for all patients who underwent an infrainguinal lower extremity bypass as initial operation for their claudication between December 2004 and December 2017. Of the 71,561 procedures, 12,469 of these procedures were initial open procedures performed for patients with lifestyle limiting claudication. Our primary outcome was 30 day MALE and the imbalance of MALE in the data was mitigated by using inverse probability weights in our generalized linear models (GLM), Random Forest (RF), ridge regression (RR), Lasso, and elastic net (EN) models. The first aim utilized conduit type as an exposure controlling for demographic variables and cardiovascular risk factors identified in the BEST-CLI clinical trial. The second aim utilized the above machine learning models to identify variables for model inclusion. From these variables, a backwards elimination algorithm was used to produce a final model on 80% of the data set and validated on 20% of the dataset for accuracy of predicting outcomes.
Results:
A total of 367 patients progressed to 30 day MALE after open surgery and less than 5% of missingness was observed. Specifically, non-spliced great saphenous vein conduit was associated with decreased 30 day MALE on multiple machine learning models (Table 1). The final model of 30-day MALE after variable selection using machine learning models included the predictors age, sex, race, BMI, coronary artery disease, diabetes, pre-operative aspirin use, pre-operative smoking, congestive heart failure, prior stroke, urgency of procedure, and conduit type. Final model characteristics included an AUC of 0.65 with a specificity of 75% and sensitivity of 51%.
Conclusions:
Use of non-spliced great saphenous vein conduit was associated with decreased 30 day MALE. We developed a model to predict 30 day MALE in claudicants undergoing open infrainguinal bypass using clinical characteristics commonly available to physicians.