Assistant Professor Weill Cornell Medicine, United States
Background: Electronic health records (EHR) are routinely collected longitudinal data and can augment clinical registries' capacity to assess longer-term outcomes after peripheral vascular interventions (PVI).
Objectives: To assess amputation following PVI using linked registry-EHR data and understand the utility and challenges in these data.
Methods: We used the Vascular Quality Initiative (VQI) registry and INSIGHT Clinical Research Network, aggregated EHR data from multiple institutions in New York City. We applied a validated indirect linkage method and linked patients undergoing device-based PVI in 4 centers in VQI and INSIGHT during 1/1/2013-11/30/2021. The final analytical cohort included those treated for femoral/popliteal occlusive disease. We assessed loss-to-follow-up (LTFU) in EHR by year of follow-up for eligible patients who did not reach the end of the study (12/31/2021) or die during that specific year. We examined predictors of LTFU after year 1 using a logistic regression. We identified 5 phenotypes with low to high risks of 1-year amputation using a nonparametric classification tree. For these phenotypes, we examined the risk of 1-year amputation using Kaplan-Meier analysis and Cox regression as well as LTFU.
Results: There were 5,115 eligible patients from the VQI, of which 4,512 (88%) were linked to the EHR. The final cohort included 1,405 patients receiving PVI for femoral/popliteal disease. The mean patient age was 71 (±11) years and 51% were male. 57% of patients had balloon angioplasty and 43% had stenting. 30% of patients had a concurrent atherectomy. Among eligible patients, LTFU was 2% in year 1, 17% in year 2, and 30% in year 5. Age ≥ 75 years, Medicaid, congestive heart failure, dialysis, and having reduced ambulation were predictors of LTFU after year 1. The 5 phenotypes based on the risk of amputation were: 1) patients with claudication (1y amputation: 2%); 2) patients with rest pain (15%); 3) patients with tissue loss without diabetes (20%); 3) patients with tissue loss and diabetes, but no dialysis (36%); and 4) patients with tissue loss, diabetes, and dialysis (61%). After adjusting for other covariates, group 5 had a 23-fold higher risk of amputation than group 1 (HR 22.8, 95% CI 10.9-47.5). Outcomes after angioplasty and stenting did not differ significantly for all phenotypes. Group 3 and 5 were more likely LTFU after year 1 than group 1 (29% vs. 12%).
Conclusions: Our study showed the feasibility and timeliness of linking registry and EHR data to examine intermediate-term amputation after PVI. However, patient LTFU in EHR poses challenges in studying very long-term outcomes. Patients with the highest risk of amputation also had the highest risk of LTFU, stressing the importance of long-term follow-up for research and clinical care.
