End-to-end proximal anastomosis is the sole modifiable variable in risk score for loss of patency of aortobifemoral...

The purpose of this study was to create a risk score for loss of aortobifemoral artery bypass (ABF) patency using preoperative, perioperative, and long-term follow-up (LTFU) variables in the Vascular Quality Initiative database.

The Vascular Quality Initiative suprainguinal arterial bypass module was queried from 2009 to 2025, and 4971 patients undergoing ABF had LTFU patency data and thus met inclusion criteria. These patients were then divided into a two-thirds testing cohort (N = 3364) and a one-third validation cohort (N = 1607), in which the risk score was evaluated. The first step was univariable analysis for the outcome of loss of patency of either or both ABF limbs after elective ABF, with χ² testing for categorical variables. The 67% testing cohort was used for this initial analysis. Demographics, socioeconomics, and comorbidities that were hypothesized to have any potential association with bypass occlusion were selected for the initial univariable analysis. Next, a multivariable Cox regression time-dependent analysis was performed for the outcome of thrombosis of either ABF limb using factors, with a univariable P value of .05 or less. Variables with a multivariable P value ≤.05 from the abovementioned regression were included in the risk score and weighted based on their respective regression beta-coefficient in a point scale. Variables with a beta-coefficient of less than .25 were assigned 1 point, and then a point was added for each rise in beta-coefficient at .25 intervals. Machine learning supplemental analysis with IBM modeler software was also performed.

Multivariable Cox regression analysis for the development of ABF thrombosis after the index operative hospitalization using significant univariable factors found multivariable significance (P < .05) and ultimate inclusion in the risk score for operative site infection after discharge [hazard ratio (HR), 1.84; P = .038], revision to achieve primary-assisted patency (open or endovascular) in follow-up (HR, 7.14; P < .001), ischemic tissue loss at initial operation (HR, 2.29; P < .001), and either femoral outflow target artery being less than 8 mm in diameter (HR, 1.59; P < .001). End-to-end proximal aortic anastomosis was protective (HR, 0.568; P = .003) as was patient not being selected for anticoagulation medication at the time of most recent LTFU (HR, 0.731; P < .001). Patients who fell into risk score bundle no. 1 (raw scores < 0) experienced the primary event in 2.1% of cases. Patients in risk score bundle no. 2 (raw scores, 0-5) had a 5% event rate, and patients in risk score bundle no. 3 (raw score > 5) experienced graft thrombosis at a 44.8% risk. There was thus statistically significant escalation in event rate with rising risk score (P < .001). Receiver operating characteristics for the risk score revealed an area under the curve value of .733. There was no significant difference in primary event rate between the testing and validation cohorts at any of the risk score bundles. The top machine learning methodology achieved an area under the curve of .914 and confirmed all the Cox regression significant multivariable factors (including end-to-end anastomosis) to be of high importance.

A validated risk score for the event of ABF occlusion after the index operative hospitalization has been developed. Performing an end-to-end proximal aortic anastomosis when not anatomically contraindicated should be considered to enhance long-term patency. Anatomic and disease pattern variables weigh most heavily on patency.