Surgeon proclivity for coronary endarterectomy during isolated coronary artery bypass grafting in Medicare beneficiaries^†

Abstract

OBJECTIVES

Coronary endarterectomy (CE) is an adjunct to coronary artery bypass grafting (CABG) in patients with one or more diffusely diseased coronary arteries. Although associated with increased perioperative morbidity and mortality, it remains a therapeutic strategy to potentially improve late outcomes by facilitating the revascularization of an otherwise ungraftable target.

METHODS

Medicare beneficiaries undergoing CABG from 2001 to 2019 were identified. Surgeon proclivity for endarterectomy was determined; surgeons were stratified by quintile of endarterectomy frequency. Overlap propensity score weighting risk-adjusted measured confounding variables. Risk-adjusted survival was compared between surgeons.

RESULTS

1 500 710 Medicare beneficiaries underwent isolated CABG, of whom 32 302 (2.2%) underwent concomitant CE. Surgeons were divided into never-endarterectomizers (0% frequency, 267 245 surgeries by 1839 surgeons), occasional-endarterectomizers (0–4% frequency, 1 001 310 surgeries by 2207 surgeons) and frequent-endarterectomizers (≥4% frequency, 232 155 surgeries by 756 surgeons). Beneficiaries undergoing surgery by a never-endarterectomizer had a risk-adjusted median survival of 10.05 [95% CI: 10.00, 10.09] versus 9.90 [9.86, 9.95] years in those undergoing surgery by a frequent-endarterectomizer, a difference of 1.71 [1.08, 2.37] months, P < 0.001 for risk-adjusted survival comparison. Similarly, beneficiaries undergoing surgery by an occasional-endarterectomizer had a risk-adjusted median survival of 9.94 [9.91, 9.96] versus 9.85 [9.80, 9.90] years for those undergoing surgery by a frequent-endarterectomizer, a difference of 1.05 [0.56, 1.74] months, P < 0.001 for risk-adjusted survival comparison.

CONCLUSIONS

Medicare beneficiaries undergoing CABG by never- or occasional-endarterectomizers had small early risk-adjusted survival advantages and similar late outcomes compared to those undergoing surgery by frequent-endarterectomizers. CE remains a valuable tool in selected cases; however, it may be reasonable for surgeons to adopt a never- or occasional-endarterectomy approach to CABG.

Open in new tabDownload slide

INTRODUCTION

Coronary endarterectomy (CE) is an adjunctive procedure performed during isolated coronary artery bypass grafting (CABG) in the setting of heavily diseased and/or calcified coronary arteries [1]. By debulking extensive coronary arterial atherosclerosis and calcium, CE can improve distal arterial flow and graft patency in coronary arteries that may otherwise be ungraftable.

Since its description, the utility of CE has been debated; ultimately, CE has found a role as an infrequent adjunct to CABG in patients with diffuse calcific CAD, with a prevalence between 2% and 3% during CABG [2–4]. Proponents argue that the early hazard for death associated with CE is due to patient factors (more advanced state of CAD), and is offset by improved late outcomes due to a more complete revascularization [5–8]. No adequately powered randomized controlled trial has been undertaken to investigate the treatment effect of CE, and most analyses of CE are limited by retrospective design and small sample size, limiting the strength of supporting data. A recent analysis of The Society of Thoracic Surgeons (STS) database found that CABG with CE was associated with a higher peri-operative hazard for death (compared to CABG without CE), but that 5-year outcomes were equivalent between propensity-matched patients [4].

In practice, the application of CE during CABG varies significantly among surgeons. Based on our institutional experience, we hypothesize that despite variations in the frequency of CE, surgeons observe and treat a similar spectrum of CAD, and an important driver in the decision to perform CE is the surgeon’s belief in and comfort with performing CE during CABG. Specifically, when faced with a diffusely diseased artery, certain surgeons have a preference to perform CE, while others will either bypass into a less diseased segment of the artery or avoid bypassing the diseased artery completely [1]. We thus implement an instrumental variable (IV) analysis using surgeon preference for CE as a pseudo-randomization technique and compare outcomes between surgeons based on their frequency of CE during CABG [9]. This statistical approach is applied to circumvent bias related to unobserved confounding variables (such as the exact amount of calcific disease in a particular coronary artery), to which traditional ‘treatment-type’ (CE vs no CE) methodologies are susceptible [10].

MATERIALS AND METHODS

Study design

Administrative claims from the Centers for Medicare and Medicaid Services (CMS) between 2001 and 2019 were retrospectively reviewed [11]. The institutional review board of the Baylor Scott & White Research Institute approved and performs ongoing monitoring of this study (IRB#:019-270, 23-08-2019). Informed patient consent waived due to retrospective design. Data collection and storage were performed consistent with the WMA Declaration of Taipei.

Study population

A 100% sample of Medicare beneficiary hospitalizations during which CABG was performed was identified from the MedPAR file using International Classification of Diseases–Ninth and Tenth Revision (ICD) procedural codes (Supplementary Material, Table S3). Surgeon-billed Current Procedural Technology (CPT) codes identifying CABG from the admission relevant to the index operation were identified from the Carrier Claims file to ascertain additional procedural details as described previously (Supplementary Material, Table S4) [12, 13]. Surgeries involving concomitant CE were identified using the surgeon-billed CPT code 33572. Procedural data were doubly-adjudicated: records with discordant (CPT compared to ICD codes) or missing CPT codes and/or missing hospital/surgeon identification data were excluded.

Beneficiaries missing Medicare Part A or B coverage as well as those with a Medicare Advantage (MA) plan within 1 year of the date of surgery were excluded to avoid the under-reporting of comorbidities [14]. As well, beneficiaries missing insurance status data, and those missing 9-digit zip code data or with a 9-digit zip code lacking a corresponding measure of Area Deprivation Index (ADI) were excluded. Beneficiaries undergoing concomitant cardiac surgery aside from CE or ablation of atrial fibrillation were also excluded (Supplementary Material, Tables S5 and S6). Finally, to ensure adequate assessment of CE proclivity, beneficiaries undergoing CABG by a surgeon who performed <10 annual CABGs in Medicare beneficiaries were excluded.

Demographics, patient comorbidities and social determinants of health

Demographics at the time of CABG were established from the Master Beneficiary Summary File (MBSF). Preoperative comorbidity data were obtained from the Chronic Conditions segment of the MBSF (CC-MBSF) [15]. Diagnoses were confirmed as pre-existing if the earliest documented date of diagnosis occurred prior to the date of surgery [16]. The beneficiary race was ascertained from the MBSF using the CMS imputation algorithm as opposed to self-reported values, given previously demonstrated improved sensitivity [17, 18]. Neighbourhood deprivation using the ADI metric is assigned a value from 1 to 100 (least to most deprived neighbourhoods) and has been cross-referenced to more than 69 million 9-digit zip codes [19]. National 2015 and 2019 ADI values were used to assess deprivation; 2015 values were used for surgeries performed between 2001–2015 and 2019 for those between 2016-2019.

Hospital and surgeon volume and CE frequency

Annual hospital and surgeon CABG volumes were determined using the hospital CMS certification number associated with ICD codes for CABG, and the National Provider Identifier associated with CPT codes for CABG, respectively. Surgeon frequency for CE was determined by dividing the surgeon’s Medicare CABG cases with concomitant CE by the surgeon’s total Medicare CABG cases. To ensure adequate classification of surgeon frequency for CE, surgeons who performed <10 Medicare CABG/year were excluded. Surgeons were stratified based on their rate for performing CE during CABG into those never performing CE (0% CE, NCE), those occasionally performing CE (>0 and <4%, OCE), and those frequently performing CE (≥4%, FCE). Thresholds of 0% and ≥4% were chosen given that they represented the top and bottom quintile of surgeon CE proclivity in beneficiaries undergoing CABG. Surgeon CABG volume and CE frequency in Medicare patients served as a surrogate for overall volume and CE frequency. A sensitivity analysis of surgeons within our local North Texas healthcare system demonstrated that a surgeon’s proclivity for CE was similar among their Medicare and non-Medicare patients.

Study outcome

The primary outcome of this study was all-cause mortality. Vital status and date of death were determined by the MBSF.

Statistical methods

Year-to-year trends in the use of CE were evaluated using the Cochran-Armitage test. Beneficiaries undergoing CABG with concomitant CE were first directly compared to those undergoing CABG without CE in our ‘treatment-type’ analysis, with the acknowledgment that this analysis may be subject to bias related to unmeasured confounders.

Next, outcomes were compared after stratifying beneficiaries based on their surgeon’s proclivity for CE at the time of CABG (NCE, OCE, FCE) [9]. Conceptually, this ‘surgeon-preference’ analysis assumes that variation in surgeon practice patterns (frequency of CE) results in the pseudo-randomization of a patient’s treatment status (patients and their underlying complexity of coronary artery disease are evenly distributed among surgeons with varying preferences for performing CE) [20]. These analyses assume: (1) between-provider variation in the treatment variable (certain surgeons prefer CE and others do not), (2) the assignment of patients to surgeons is unrelated to the surgeon’s preference for concomitant CE and (3) a surgeon’s use of CE is independent of alternative treatments that affect outcomes (e.g. a surgeon’s preference for arterial grafts at the time of CABG) [21, 22]. A sensitivity analysis of surgeons and cardiologists in our local North Texas healthcare system noted that referral patterns for CABG are primarily a function of (i) the surgeon on call, (ii) overall surgeon CABG volume, (iii) perceived surgeon excellence (which was not related to surgeon frequency of CE) and (iv) personal cardiologist/surgeon relationships. Cardiologists interviewed in our healthcare system could not differentiate between NCE, OCE and FCE surgeons.

Overlap propensity score weighting (OW) was used to further risk-adjust for imbalances in patient and surgical characteristics (Supplementary Methods S1) [23, 24]. Notably, although IV approaches may protect against unmeasured confounding, they are not immune from violations of the IV assumptions. Thus, OW was applied to our IV analysis to minimize bias resulting from violations of the IV assumptions [25]; more details on the weighting procedure are included in the Supplementary Methods S1. Between-group differences were measured using standardized mean differences (SMD) before and after risk adjustment [26].

Median duration of follow-up was determined using the reverse Kaplan–Meier (KM) method [27]. The primary outcome of all-cause mortality was evaluated with KM analysis in the unweighted and weighted (risk-adjusted) populations. 95% confidence intervals (CI) [are provided in brackets] for risk-adjusted KM analyses and risk-adjusted median survival estimates were obtained using a non-parametric bootstrap procedure [28].

Data management was conducted using SAS (SAS Institute, Cary, NC). Statistical analyses were conducted with Stata/MP 18.0 (Stata Corp, College Station, TX).

RESULTS

1 500 710 Medicare beneficiaries with suitable data for analysis underwent CABG from 2001 to 2019, of whom 32 302 (2.2%) underwent concomitant CE (Fig. 1). The prevalence of CE remained stable throughout the duration of the study: 2.0–2.4%(Cochran Armitage Z-statistic 1.02, P = 0.31). Preoperative characteristics between beneficiaries undergoing concomitant CE and those undergoing CABG without CE are shown in Supplementary Material, Table S1.

Total Medicare CABG volume and frequency of CE were determined for all 4802 surgeons (Fig. 2). Surgeons were stratified based on proclivity for CE during CABG into NCE (0% frequency, 267 245 surgeries by 1839 surgeons), OCE (0–4% frequency, 1 001 310 surgeries by 2207 surgeons) and FCE (≥4% frequency, 232 155 surgeries by 756 surgeons) surgeons (Fig. 3). Although beneficiaries undergoing CABG by NCE, OCE and FCE surgeons were largely similar, we observed that FCE CABG recipients were more likely to be of Hispanic ethnicity, from the U.S. Health and Human Services ‘Dallas region’ (U.S. states of Texas, New Mexico, Oklahoma, Louisiana, Arkansas), and live in slightly more disadvantaged neighbourhoods than NCE or OCE recipients. Certain U.S. training institutions were more likely to produce FCE surgeons, while others were more likely to produce NCE surgeons (Supplementary Material, Fig. S1).

The prevalence of pre-existing comorbidities and operative urgency were similar across all 3 groups. NCE surgeons had a lower median annual CABG volume than OCE or FCE surgeons (Table 1); however, the overall correlation between surgeon volume and CE frequency was weak (Pearson correlation coefficient 0.0155, P = 0.284, Fig. 2). This discordance is because lower volume surgeons necessarily have a smaller pool of CABG cases through which to identify an instance of concomitant CE, and thus more variability in the calculation of surgeon CE frequency. Finally, the FCE group was more likely to have a redo operation and to have a slightly higher number of venous grafts. Following risk adjustment with OW, patient demographics and co-morbidities were well-balanced among groups (Supplementary Material, Table S2).

Unadjusted survival

Median follow-up was 12.57 [12.55–12.59] years. The total number of deaths (events) was 814 999 in 1 468 408 CABG recipients without CE (55.5%) and 19 558 in 32 302 recipients with concomitant CE (60.6%). Unadjusted KM survival curves comparing beneficiaries who did and did not undergo CE during CABG are shown for reference (Supplementary Material, Fig. S2), as are unadjusted survival curves for beneficiaries undergoing surgery by either an NCE, OCE or FCE (Supplementary Material, Fig. S3).

Risk-adjusted survival

In the ‘treatment-type’ analysis, beneficiaries who did not undergo CE had a risk-adjusted median survival of 9.82 [9.80–9.84] versus 8.89 [8.79–9.01] years in those with concomitant CE, a difference of 11.3 [9.9–12.4] months, P < 0.001 for risk-adjusted survival comparison (Fig. 4). In the ‘surgeon-preference’ analysis, beneficiaries undergoing CABG by NCE surgeons had a risk-adjusted median survival of 10.05 [10.00, 10.09] versus 9.90 [9.86, 9.95] years for FCE surgeons, a difference of 1.71 [1.08, 2.37] months, P < 0.001 for comparison of survival curves (Fig. 5). Similarly, beneficiaries undergoing CABG by OCE surgeons had a risk-adjusted median survival of 9.94 [9.91, 9.96] versus 9.85 [9.80, 9.90] years for FCE surgeons, a difference of 1.05 [0.56, 1.74] months, P < 0.001 for comparison of survival curves (Fig. 6). Sensitivity analyses varying the cut-point defining FCE surgeons yielded similar findings (Supplementary Material, Fig. S4).

DISCUSSION

This study analyzed survival in 1 500 710 Medicare beneficiaries undergoing CABG, stratifying CABG recipients based on surgeon preference for CE. Our analysis was notable for 4 major findings. First, the prevalence of CE during CABG was 2.2% and did not change appreciably over the 19-year duration of the study. Second, in our ‘treatment-type’ analysis, concomitant CE during CABG was associated with a substantially higher risk-adjusted hazard for death; undoubtedly, unmeasured confounding variables such as the severity of CAD contributed to this finding. Third, we noted both minimal correlation and substantial variability between surgeon CABG volume and frequency of CE. Fourth, early risk-adjusted survival was mildly improved in beneficiaries undergoing CABG by OCE and NCE surgeons compared to FCE surgeons, with no difference in late survival.

The prevalence of CE during CABG in our study (2.2%) is concordant with other analyses (2–3%), which also note a stable prevalence of CE over time [4, 8]. As well, the risk-adjusted early hazard for death noted in beneficiaries undergoing CE during CABG is in concordance with prior studies (Fig. 3) [4–7, 29, 30]. However, in contrast to a recent STS analysis, which found that survival between matched CE and non-CE CABG recipients converged by 5 years post-operatively, we found that risk-adjusted survival remained lower in CE recipients, even at 19 years post-operatively [4]. This represents an important difference between these 2 analyses that warrants future study/scrutiny.

To our knowledge, this analysis provides the first comprehensive description of the distribution of surgeon CE frequency in the US. Interestingly, over a third of all surgeons never performed a CE during CABG (at least in the Medicare population analyzed). The differences in CE prevalence by US geography warrant explanation. Although regional variation in surgeon frequency of CE may reflect geographical differences in CAD severity, we hypothesize that these differences may be better explained by the impact a surgeon’s training program has both on their frequency of CE (familiarity with and belief in its use) and on the location of their surgical practice (Supplementary Material, Fig. S1 and Table S2) [31–33]. We hypothesize our findings imply that not all surgeons are convinced of the clinical utility of CE, and that some surgeons choose to bypass a less diseased segment of a diffusely diseased target or simply avoid bypassing diffusely diseased targets altogether, acknowledging that although this may represent a less complete revascularization, it may nevertheless decrease peri-operative risk.

We found that beneficiaries undergoing CABG by FCE surgeons were subject to an absolute increase in mortality of roughly 1% by 90 days postoperatively compared to OCE or NCE surgeons, a survival disadvantage then slowly dissipated over time with near-equivalent late survival (Figs 5 and 6). While hypothesis generating, these findings suggest that surgeons who decide to adopt a never or occasional approach to CE in order to avoid its early hazard for death have some justification for their approach, given that the less complete revascularization offered by this approach does not appear to translate into worse late outcomes (at least in Medicare beneficiaries).

Limitations

This study is limited by its retrospective, observational design, selection bias related to our study’s analysis of only Medicare beneficiaries with a traditional (non-advantage) plan, our use of administrative coding data as a surrogate for discrete clinical variables and the possibility for measurement bias related to miscoding. Granular clinical details, such as CAD complexity and medication adherence, are thus not available. Surgeon preference for CE was inferred based on the relative frequency at which individual surgeons performed CE in Medicare beneficiaries, and whether some surgeons more (or less) frequently performed CE in younger non-Medicare patients is unknown. Using surgeon preference as an IV requires certain assumptions, e.g. that the distribution of CAD treated by cardiac surgeons is evenly distributed across NCE, OCE and FCE surgeons. Although this assumption is supported by the authors’ clinical experience within a group of over 50 surgeons that operated in a large North Texas healthcare system from 2001 to 2019, we are unable to assess referral patterns within Medicare data and recognize the possibility that biases in referral patterns based on CAD severity and surgeon utilization of CE could exist. Importantly, our use of overlap weighting following the implementation of surgeon preference as an IV is protective against potential violations of the IV assumptions. Finally, because the results of this study are drawn from Medicare beneficiaries, these findings may not apply to patients younger than 65 years of age.

CONCLUSION

Our analysis of 1.5 million CABGs in Medicare beneficiaries noted a stable prevalence of CE (2.2%) and corroborated the results of prior studies that CE during CABG is associated with an increased early risk-adjusted hazard for death, findings that are likely related to unmeasured confounding variables such as the severity of CAD. Substantial differences were noted in surgeon proclivity for CE. Beneficiaries undergoing CABG by NCE and OCE surgeons benefitted from a clinically small early survival advantage compared to FCE surgeons, with similar late outcomes. In the absence of adequately powered randomized controlled trials to describe the treatment effect attributable to CE, these data, while hypothesis-generating, suggest it may be reasonable for surgeons to adopt a never- or occasional-endarterectomy approach to CABG.

ACKNOWLEDGMENTS

We are especially grateful to Dr Alessandro Gasparini (Red Door Analytics) for his substantial contributions regarding our statistical methodologies.

SUPPLEMENTARY MATERIAL

Supplementary material is available at EJCTS online.

FUNDING

Data acquisition and efforts of J.B.E., K.A.M. and J.K.B. were supported by a philanthropic gift of Satish and Yasmin Gupta to Baylor Scott & White–The Heart Hospital Plano.

Conflict of interest: none declared.

DATA AVAILABILITY

Datasets derived from sources in the public domain: MEDPAR; https://www.cms.gov/data-research/statistics-trends-and-reports/medicare-fee-for-service-parts-a-b/medpar.

Author contributions

John B. Eisenga, MD, MS: Investigation; Visualization; Writing—original draft; Writing—review & editing. Kyle A. McCullough: Writing—original draft; Writing—review & editing. Austin Kluis: Writing—review & editing. Jasjit K. Banwait: Data curation; Investigation; Methodology; Writing—original draft; Writing—review & editing. Sarah Hale: Project administration; Supervision; Writing—original draft; Writing—review & editing. Michael J. Mack: Supervision; Writing—original draft; Writing—review & editing. J. Michael DiMaio: Supervision; Validation; Writing—original draft; Writing—review & editing. Justin M. Schaffer: Conceptualization; Formal analysis; Investigation; Methodology; Visualization; Writing—original draft

Reviewer information

European Journal of Cardio-Thoracic Surgery thanks Ari Mennander, Samuel Heuts, Tim Berger and the other anonymous reviewers for their contribution to the peer review process of this article.

REFERENCES

ABBREVIATIONS

 
• ADI

  Area deprivation index

 
• CABG

  Coronary artery bypass grafting

 
• CAD

  Coronary artery disease

 
• CC-MBSF

  Chronic conditions segment of the master beneficiary summary file

 
• CE

  Coronary endarterectomy

 
• CI

  Confidence intervals

 
• CMS

  Centers for Medicare and Medicaid Services

 
• CPT

  Current procedural terminology

 
• FCE

  Frequent coronary endarterectomizer

 
• ICD

  International classification of diseases, ninth revision and tenth revision

 
• IV

  Instrumental variable

 
• KM

  Kaplan-Meier

 
• MBSF

  Master beneficiary summary file

 
• NCE

  Never coronary endarterectomizer

 
• OCE

  Occasional coronary endarterectomizer

 
• OW

  Overlap propensity score weighting

 
• SMD

  Standardized mean difference

 
• STS

  Society of Thoracic Surgeons

Author notes