Long-term outcome followed for more than 5 years after revascularization surgery for the treatment of atherosclerotic steno-occlusive disease: poor outcome prediction using machine learning and analysis of the results

Abstract

Cerebral revascularization for the treatment of atherosclerotic steno-occlusive disease (ASOD) was found to have no benefit compared with medical treatment. However, there is also criticism that with sufficiently long-term follow-up, a crossover might emerge demonstrating the advantages of surgery. Therefore, we examined the long-term outcome of cerebral revascularization performed on patients with carefully selected ASOD at our center. Patients undergoing bypass surgery for non-moyamoya ischemic disease were retrospectively identified. The inclusion criteria were symptomatic ASOD with hemodynamic insufficiency, follow-up of more than 5 years, and stroke or surgical complications during follow-up. The clinical course and radiological findings were investigated. Poor outcomes were predicted using machine learning (ML) models, and Shapley additive explanation (SHAP) values and feature importance of each model were analyzed. A total of 109 patients were included from 2007 to 2018. The 30-day risk of any stroke or death was 6.4% (7/109). The risk of ipsilateral ischemic stroke during median follow-up of 116 months was 7.3% (8/109). The SHAP values showed that previously and empirically known stroke risk factors exert a relatively consistent effect on the prediction of models. The number of lesions with stenosis > 50% (odds ratio [OR] 5.77), age (OR 1.13), and coronary artery disease (OR 5.73) were consistent risk factors for poor outcome. We demonstrated an acceptable long-term outcome of cerebral revascularization surgery for patients with hemodynamically insufficient and symptomatic ASOD. Multicenter studies are encouraged to predict poor outcomes and suitable patients with large numbers of quantitative and qualitative data.

Keywords: Atherosclerosis; Bypass; Cerebral revascularization; Ischemic stroke; Machine learning; Occlusion; Stenosis.