Clinical and economic outcomes of ProGlide compared with surgical repair of large bore arterial access

Patients were selected for this retrospective study if they underwent any of the following procedures involving large-bore arterial access site closure with ProGlide device during the enrollment period of 1 January 2013 through 24 April 2017: transcatheter aortic valve replacement/repair (TAVR), endovascular abdominal aortic aneurysm repair (EVAR), thoracic endovascular aortic repair (TEVAR) or balloon aortic valvuloplasty (BAV). For patients with multiple procedures in the EHR data, the most recent procedure date was selected as the index procedure date. Patients with the ProGlide device used for vessel repair of the index procedure, the ProGlide cohort, were compared with those for whom no suture-mediated closure device was used, the Cutdown cohort. Patients with a history of VCD use for the indicated procedures were excluded from the Cutdown cohort.

Patient characteristics were measured during a 6-month pre-index baseline period. Clinical outcomes were measured during three time periods from the index date through the length of the index procedure inpatient admission, the first 30 days post-index and the first 6 months post-index. Clinical outcomes were compared between the ProGlide and Cutdown cohorts overall and within subgroups of patients undergoing each of the four procedures: TAVR, EVAR, TEVAR and BAV.

This study utilized patient-level de-identified US EHR data from the IBM Explorys Therapeutic Dataset (Explorys; IBM, NY, USA). The Explorys Dataset contains data for approximately 55 million patients (∼15% of US population), integrated from 23 large health systems comprising approximately 360 hospitals and 330,000 providers. The data, collected from EHRs, outgoing billing and adjudicated claims from both commercial and public payers, provides a full longitudinal view of a patient's medical history across the care continuum.

All study data are fully de-identified and were accessed with protocols compliant with US patient confidentiality requirements, including the Health Insurance Portability and Accountability Act of 1996 regulations; therefore, this study was exempted from Institutional Review Board approval.

Patients with at least one record for a large-bore femoral arterial access procedure (TAVR, EVAR, TEVAR or BAV) in the inpatient setting during the enrollment period were selected for analysis. The index date was set as the most recent date for any of the four procedures. The ProGlide cohort included all patients with a ProGlide model number in the EHR medical device record on the index date and no other non-ProGlide VCD recorded on the index date. The Cutdown cohort was comprised of all patients without any VCDs recorded during the entire index procedure hospitalization record or at any other time in the study period.

Model number is not a required element within Explorys; as such, model number is reported only in a minority of large-bore access arterial procedures. As not all Explorys data contributors provide implantable medical device identifiers (i.e., model numbers) to the dataset, Cutdown patients were selected from the same subset of healthcare providers (N = 419) who reported at least one record where a ProGlide device identifier was recorded. This was done to minimize the risk that Cutdown patients were originating from data providers that simply did not provide device identifiers to the database (i.e., false negatives). The result of this criteria is a substantial drop in potential sample size from all patients receiving a large-bore access arterial procedure to just those with receiving a large-bore access arterial procedure who saw a provider who definitely provided ProGlide to at least one patient. In addition, patients with the Healthcare Common Procedure Coding System procedure code C1760: Closure device, vascular (implantable/insertable) in their EHR record on the index date were excluded from either cohort because it could indicate the use of an unspecified VCD.

Patients qualifying for the ProGlide and Cutdown cohorts from the same healthcare providers were then matched in a 1:1 ratio on the following characteristics: index procedure (TAVR, EVAR, TEVAR or BAV), age at index procedure (+/− 5 years), sex, year of index procedure (+/− 1 year), blood transfusion procedure in the 6-month pre-index period (yes/no) and diagnosis of coronary artery disease in the 6-month pre-index period (yes/no).

Patients were stratified by the procedure(s) performed on the index date based on the presence of procedure codes for TAVR, EVAR, TEVAR or BAV. Patients with more than one procedure on the index date were included in both groups.

Demographic characteristics were assessed at the index date and included age, sex, geographic region and index procedure year. Baseline clinical characteristics were measured in the 6-month pre-index period and included select comorbidities, pre-index diagnosis of clinical outcomes of interest, BMI and pre-index utilization of anticoagulants, antiplatelet medications or methicillin-resistant Staphylococcus aureus (MRSA) indicated antibiotics. BMI categories were defined based on the most recent of any BMI records in the 6-month baseline. Patients were identified as having the selected clinical comorbidity or pre-index clinical outcome of interest if they had at least one record containing an International Classification of Diseases Ninth or Tenth Revision (ICD-9 or ICD-10) diagnosis code for the condition.

The primary outcomes of interest were diagnoses of access-site complications and minor ipsilateral access-site vascular complications. These included pseudoaneurysm, arteriovenous fistula, blood transfusion, hemorrhage, hematoma, lower-extremity arterial emboli/stenosis, deep vein thrombosis and infection. Complications were identified using ICD-9 or ICD-10 diagnosis codes and did not require other criteria. For example, we did not require or capture surgical treatment for hematoma. However, we did capture blood transfusion procedures separately from a bleeding diagnosis and antibiotic use separately from an infection diagnosis.

Complications were assessed during the three post-index follow-up periods: index procedure hospital admission, first 30 days post-index and first 6 months post-index. Patients were identified as having a complication if they had at least one record containing the diagnosis code. Patients were only assessed for 30-day and 6-month post-index outcomes if they had at least one record following the end of the measurement period. Also evaluated were the postsurgery inpatient LOS and hospital readmission following discharge for the index procedure. LOS was calculated as the number of days from the index procedure date through the discharge date. Evidence of death was defined by the recorded year of death and the patients’ last medical record entry occurring during the 30-day and 6-month follow-up periods. The exact death date and reason for death were not available in the database due to privacy concerns.

Since the Explorys database does not include cost or charge data, the healthcare cost differences associated with observed clinical outcomes were estimated using administrative claims data from the IBM^® MarketScan Commercial Claims and Encounters and Medicare Supplemental and Coordination of Benefits databases (IBM Watson Health, MA, USA). The MarketScan Research Databases comprise enrollment, demographic and inpatient medical, outpatient medical and outpatient pharmacy claims information collected from over 300 large self-insured US employers and over 25 US health plans. The administrative claims do not, however, contain information on ProGlide and other medical implantable devices at the brand level. To our knowledge, there is no database that captures both medical device brand information, which was contained in the Explorys EHR data, and the patient-level cost information found in administrative claims data. Therefore, utilizing this administrative claims database to quantify the cost of clinical outcomes associated with the device, as opposed to a direct comparison of patient costs, was the best and only feasible option for an economic analysis given the available real-world data sources.

Patients undergoing the TAVR, EVAR, TEVAR or BAV in 2016 were identified. Cost differences were applied to the Explorys clinical outcomes by first calculating the average total healthcare costs for patients with and without each complication and the costs per day of the index hospitalization. As a direct cost comparison was not possible, cost differences between ProGlide and Cutdown cohorts were indirectly quantified based on inpatient LOS. The cost-per-day of an inpatient admission for TAVR, EVAR, TEVAR or BAV patients calculated in the claims data was multiplied by LOS results for the two cohorts for use in multivariate modeling which accounts for the skewed distribution of inpatient costs.

Bivariate statistics were used to compare all baseline characteristics and outcome variables between the matched ProGlide and Cutdown cohorts. Two-tailed Student's t-tests were used to test for significant differences in the distributions of continuous variables. The χ²tests were used for significance testing of categorical variables.

Generalized linear models with a γ-distribution and log-link function were employed to evaluate LOS and healthcare costs by cohort while adjusting for baseline characteristics. Covariates included in the LOS model were index procedure type, age, sex and a subset of baseline clinical characteristics (anticoagulant use, atherosclerosis, cancer, chronic respiratory disease, myocardial infarction, stroke, blood transfusion). For the healthcare costs generalized linear model, only index procedure type, age and sex were included as covariates.

During the enrollment period, a total of 44,039 patients in the Explorys database had a record of TAVR, EVAR, TEVAR or BAV. Of these, 839 patients had evidence of a ProGlide device in their procedure record and met all the selection criteria (Table 1). An additional 2569 patients, treated by the same providers, had no evidence of any VCD in their procedure record and were assigned to the manual surgical cutdown cohort. To reduce the possibility of false negatives from undercoding device usage, cutdown patients were selected only from providers who documented at least one ProGlide procedure.

Direct 1:1 matching resulted in 757 matched pairs used in the matched ProGlide versus Cutdown analyses. Baseline demographic characteristics in prematched and postmatched cohorts are summarized in Table 2. Prior to matching, patients in the Cutdown cohort were significantly younger (73.0 ± 11.7 years Cutdown vs 76.9 ± 11.3 years ProGlide; p < 0.001) and more likely to be male (69.2 vs 65.3%; p = 0.037).

Index procedure type and baseline clinical characteristics in prematched and postmatched cohorts are summarized in Table 3. The most commonly performed procedure on the index date was TAVR followed by EVAR. Prior to matching, pre-index procedure blood transfusions were more common in Cutdown cohort (5.2% Cutdown vs 3.5% ProGlide; p = 0.042). After matching, the distribution of index procedure types was 44.9% TAVR, 38.6% EVAR, 21.3% TEVAR and 3.4% BAV. Baseline average BMI and prevalence of clinical comorbidities and outcomes of interest were similar between cohorts after 1:1 matching with the exception of stroke (7.7% Cutdown vs 5.0% ProGlide; p = 0.035). Baseline medication utilization was different between groups after matching, with Cutdown patients have lower utilization of anticoagulants (17.8% Cutdown vs 44.9% ProGlide; p < 0.001) and higher utilization of MRSA antibiotics (27.3% Cutdown vs 11.0% ProGlide; p < 0.001).

During the index procedure hospitalization, patients in the ProGlide cohort were significantly less likely than matched Cutdown patients to receive a blood transfusion (35.7% Cutdown vs 9.5% ProGlide; p < 0.001) or be diagnosed with an infection (22.2% Cutdown vs 15.6% ProGlide; p < 0.001) including infections treated with MRSA-indicated antibiotics (2.6% Cutdown vs 5.0% ProGlide; p < 0.05). Other access-site complications analyzed occurred in less than 10% of patients and did not differ significantly between the matched cohorts, as seen in Figure 1A.

In the first 30 days post-index procedure, fewer patients in the ProGlide cohort received a blood transfusion (35.0% Cutdown vs 10.7% ProGlide; p < 0.001) or were diagnosed with any of the following: infection (31.2% Cutdown vs 21.6% ProGlide; p < 0.001), hematoma (4.0% Cutdown vs 2.0% ProGlide; p < 0.05), bleeding complicating a procedure (3.7% Cutdown vs 1.7% ProGlide; p < 0.05) or evidence of death (3.0% Cutdown vs 1.1% ProGlide; p < 0.05; Figure 1B). At 6 months post-index procedure, the only remaining statistically significant difference in access-site complications between the two cohorts was the likelihood of having received a blood transfusion (42.4% Cutdown vs 13.3% ProGlide; p < 0.001; Figure 1C).

When comparing the subset of matched ProGlide and Cutdown patients who underwent a TAVR procedure at the index date, results remained significant for all outcomes that were significant in the overall comparison, as shown in Table 4. Among the TAVR subset, 39.4% more patients in the Cutdown cohort required a blood transfusion (p < 0.001) and 12.9% more patients were diagnosed with an infection during the index hospitalization (p < 0.001). At 30 days post-index, patients in the Cutdown cohort were more likely to be diagnosed with a hematoma (5.1% Cutdown vs 1.3% ProGlide; p ≤ 0.01), or bleeding complicating a procedure (4.7% Cutdown vs 1.0% ProGlide; p ≤ 0.01). The proportion receiving a blood transfusion during the index hospitalization, 30 days and 6 months post-index were higher among Cutdown patients in EVAR and TEVAR subsets, while other outcomes were not statistically significant.

The mean inpatient LOS for the index procedure hospitalization was on average 4 days longer for Cutdown patients than ProGlide patients (mean: 9.0 days Cutdown vs 5.1 days ProGlide; p < 0.001; median: 7.0 days Cutdown vs 3.0 days ProGlide; Figure 2). In the 30 days postdischarge, there was no difference in hospital readmissions between cohorts (11.0% Cutdown vs 10.0% ProGlide; p = 0.55). The generalized linear model, after adjusting for baseline differences in covariate, found that inpatient LOS for the ProGlide cohort was 42.5% shorter than for the matched Cutdown cohort (odds ratio [ProGlide vs Cutdown]: 0.575; 95% CI: 0.531–0.619; p < 0.001).

To quantify the predicted inpatient stay cost difference between cohorts, mean and median LOS results obtained from the main analysis were compared. After controlling for index procedure, age and sex, LOS was a significant predictor of costs. (Figure 3; p < 0.05). With an observed 4-day difference in mean index hospitalization LOS, ProGlide was predicted to save $14,687 USD.

Studies based on EHR data and administrative claims data have several inherent limitations. First, the EHR database does not include specific information regarding arterial anatomy (i.e., artery diameter, calcification or occlusive disease) that could influence patient selection for either VCD or Cutdown. Second, we are not able to identify or exclude patients in the ProGlide cohort who had a simultaneous cutdown procedure in the contralateral groin. Third, data collected for administrative purposes may not be collected and validated with the same rigor as data collected for researcher purposes; therefore, these datasets are subject to miss-coding and under-coding which may introduce bias or measurement error. However, there is no reason to expect systematic differences in the frequency or type of data collection errors between cohorts. Fourth, with regards to the clinical outcomes analysis, any care for complications received outside of Explorys healthcare partners would not be captured in the dataset. As a result, this study may be under-reporting the frequency of access-site complications; however, there is no reason to believe this would impact one cohort more than the other. Fifth, use of a VCD is not reported in the MarketScan administrative claims database, so cost estimates were limited to an indirect comparison based on hospitalization LOS and did not adjust for device usage. Future research using a combined dataset of EMR records and insurance claims could investigate differences in direct healthcare costs between these patient cohorts. At last, as stated above, implantable device model number is not a required field in Explorys and is, in fact, only available on a minority of Explorys patients. Because of this, the Cutdown cohort was limited to patients seen by healthcare providers who had at least one ProGlide patient (i.e., contributed data to the database with an implantable device model number). To help minimize potential confounding, we then matched this subset of patients based on parameters available into the EHR (index procedure, age, sex, index year, pre-index blood transfusion and pre-index coronary artery disease), there may be unreported factors which influence provider choice that we could not control for. For example, provider type/location (radiology, cardiology, vascular surgery) are not reported consistently across data providers there for this could not be incorporated into our matching procedure despite having a potential impact on outcomes.