Limitations of Vascular Quality Initiative-derived models to predict the outcomes of intervention for infrapopliteal limb-threatening ischemia.

OBJECTIVE

The purpose of this study was to validate published Society for Vascular Surgery Vascular Quality Initiative (VQI) prediction models for patients with limb-threatening ischemia (LTI) undergoing open or endovascular revascularization for infrapopliteal occlusive disease.

METHODS

We sought to validate our prior VQI LTI models for major adverse limb events (MALEs) and amputation-free survival (AFS) in a VQI new cohort undergoing open or percutaneous interventions from September 2014 through August 2016. Receiver operating characteristic curves were generated including the C statistic, and the predicted vs actual outcomes were correlated. The Hosmer-Lemeshow (HL) statistic was calculated to determine goodness of fit, and the Tjur R statistic was derived to demonstrate the degree to which the observed outcomes were accurately predicted by the models.

RESULTS

Of 15,576 open infrainguinal and 34,679 percutaneous interventions collected in the VQI during the 24-month interval, 8852 and 17,124, respectively, were performed for LTI, among which 4410 and 5116 specifically targeted the infrapopliteal vessels. MALEs and AFS were identified for 400 of 927 (43.1%) and 576 of 982 (58.7%) open procedures and 197 of 855 (23.0%) and 658 of 1115 (59.0%) percutaneous procedures, respectively. For open operation, the predictive ability of the model was poor for MALEs (C = 0.59; HL = 107; R = 0.03) and only marginally better for AFS (C = 0.69; HL = 130; R = 0.10). Similarly, for endovascular intervention, the model performed poorly for MALEs (C = 0.62; HL = 183; R = 0.06) and slightly better for AFS (C = 0.68; HL = 68; R = 0.11). Breaking AFS into its component determinants, the predictive ability of the open operation model for patient survival (C = 0.77; HL = 70; R = 0.15) surpassed that for limb salvage (C = 0.64; HL = 54; R = 0.05). For endovascular interventions, the survival model (C = 0.71; HL = 94; R = 0.11) also outperformed the limb salvage model (C = 0.67; HL = 28; R = 0.07). For both types of intervention, the actual MALE rate was lower and AFS was higher than predicted by the models.

CONCLUSIONS

The ability of reported VQI-derived models to accurately predict major outcomes for infrapopliteal LTI is limited and cannot be advocated for clinical decision-making at this time. Further study would be necessary to determine whether this is due to intraoperative and postoperative variables not accounted for in our models, absence of pertinent data points from the registry, or incomplete follow-up.