Validation of randomized controlled trial-derived models for the prediction of postintervention outcomes in chronic limb-threatening ischemia

• Published in: Journal of vascular surgery 2020-03, Vol.71 (3), p.869-879
• Main Authors: Wijnand, Joep G.J., van Koeverden, Ian D., Teraa, Martin, Spreen, Marlon I., Mali, Willem P.T.M., van Overhagen, Hans, Pasterkamp, Gerard, de Borst, Gert J., Conte, Michael S., Gremmels, Hendrik, Verhaar, Marianne C.
• Format: Article
• Language: English
• Subjects: Amputation-free survival; Critical limb-threatening ischemia; Prediction models; Validation
• ISSN: 0741-5214; 1097-6809
• DOI: 10.1016/j.jvs.2019.06.195

Chronic limb-threatening ischemia (CLTI) represents the most severe form of peripheral artery disease and has a large impact on quality of life, morbidity, and mortality. Interventions are aimed at improving tissue perfusion and averting amputation and secondary cardiovascular complications with an optimal risk-benefit ratio. Several prediction models regarding postprocedural outcomes in CLTI patients have been developed on the basis of randomized controlled trials to improve clinical decision-making. We aimed to determine model performance in predicting clinical outcomes in selected CLTI cohorts. This study validated the Bypass versus Angioplasty in Severe Ischaemia of the Leg (BASIL), Finland National Vascular registry (FINNVASC), and Prevention of Infrainguinal Vein Graft Failure (PREVENT III) models in data sets from a peripheral artery disease registry study (Athero-Express) and two randomized controlled trials of CLTI in The Netherlands, Rejuvenating Endothelial Progenitor Cells via Transcutaneous Intra-arterial Supplementation (JUVENTAS) and Percutaneous Transluminal Angioplasty and Drug-eluting Stents for Infrapopliteal Lesions in Critical Limb Ischemia (PADI). Receiver operating characteristic (ROC) curve analysis was used to calculate their predictive capacity. The primary outcome was amputation-free survival (AFS); secondary outcomes were all-cause mortality and amputation at 12 months after intervention. The BASIL and PREVENT III models showed predictive values regarding postintervention mortality in the JUVENTAS cohort with an area under the ROC curve (AUC) of 81% and 70%, respectively. Prediction of AFS was poor to fair (AUC, 0.60-0.71) for all models in each population, with the highest predictive value of 71% for the BASIL model in the JUVENTAS population. The FINNVASC model showed the highest predictive value regarding amputation risk in the PADI population with AUC of 78% at 12 months. In general, all models performed poor to fair in predicting mortality and amputation. Because the BASIL model performed best in predicting AFS, we propose use of the BASIL model to aid in the clinical decision-making process in CLTI. However, improvements in performance have to be made for any of these models to be of real additional value in clinical practice.