Incremental value of bilateral ankle-brachial index for ten-year heart and cerebrovascular mortality prediction: an exploratory temporal benchmark in NHANES
Abstract
Background: Ankle-brachial index (ABI) is associated with systemic vascular disease, but its incremental predictive value depends on the clinical comparator, endpoint and validation design. Objective: To test whether a bilateral ABI feature block improves a nonlaboratory clinical benchmark for ten-year death from diseases of the heart or cerebrovascular disease, and whether a limited nonlinear learner improves the same chronological evaluation. Methods: We rebuilt public NHANES 1999-2004 examination data linked to the 2019 public-use mortality file. Adults aged at least 40 years with bilateral ABI values in [0.20, 2.50] and ascertainable 120-month status formed a survey-weighted analytic subset. Models were trained on 1999-2002, tuned within development cycles, and evaluated on 2003-2004. Primary models were linear clinical logistic regression, flexible clinical logistic regression, flexible clinical plus ABI regression, and histogram gradient boosting plus ABI. Point estimates used full evaluation weights; conditional intervals used 32-replicate Fay balanced repeated replication with 15 design degrees of freedom. The study was not prospectively registered. Results: The corrected cohort contained 6,807 participants (4,617 development; 2,190 temporal evaluation) and 409 primary events, including 139 in evaluation. Evaluation AUROC was 0.834 for flexible clinical regression and 0.834 after ABI addition. The paired ABI increment was -0.0007 (95% CI -0.0047 to 0.0032); the paired Brier difference was 0.0008 (95% CI -0.0003 to 0.0018). Boosting plus ABI had AUROC 0.822; versus regression with the same predictor block, its AUROC difference was -0.0117 (95% CI -0.0379 to 0.0144). Its Brier difference was -0.000661 (95% CI -0.001322 to -7.53e-08); the upper limit was extremely close to zero. Neither primary hypothesis of improvement in both metrics was supported. Conclusions: In this corrective, exploratory chronological benchmark, bilateral ABI did not improve discrimination or Brier score beyond a matched flexible clinical model. The result does not negate ABI associations, establish equivalence, or support treatment decisions. It supports transparent evaluation of incremental information in a new cohort rather than claims of clinical utility.