Coronary revascularization outcomes in relation to skilled nursing facility use following hospital discharge

Abstract Background Observational analyses comparing coronary artery bypass surgery (CABG) and percutaneous coronary intervention (PCI) among elderly or frail patients are likely biased by treatment selection. PCI is typically chosen for frail patients, while CABG is more common for patients with good recovery potential. Hypothesis We hypothesized that skilled nursing facility (SNF) use after revascularization is a measure of relative frailty associated with outcomes following coronary revascularization. Methods We used a 20 percent sample of Medicare beneficiaries aged 65 years or older who received inpatient PCI or CABG between 2007–2014. Key explanatory variables were the revascularization strategy and SNF use after revascularization. We used Cox regression to evaluate death and repeat revascularization within one year and logistic regression to evaluate SNF use and 30‐day readmissions/death. Results CABG patients were 25.1 percentage points [95% confidence interval: 24.7, 25.5] more likely to use SNF following revascularization than inpatient PCI patients. SNF use was associated with a higher death rate (hazard ratio (HR): 3.19 [3.02, 3.37]) and a 16.2 percentage point (15.5, 16.9) increase in 30‐day readmissions/death. Among patients with SNF use, CABG was associated with a decrease in 30‐day readmissions/death compared to PCI. Conclusions While CABG was associated with higher rates of SNF use and 30‐day readmission/death overall, CABG was associated with significantly lower rates of 30‐day readmissions/death among patients with SNF use. The findings suggest that caution is needed in treatment selection for patients at high‐risk for SNF use and that selection of inpatient PCI over CABG may be associated with frailty and worse outcomes for some patients.


| INTRODUCTION
The share of percutaneous coronary intervention (PCI) relative to coronary artery bypass surgery (CABG) that is performed in elderly and frail patients has been growing in recent years. 1 This pattern is driven in part by dramatic increases in the share of the population that are elderly. 2,3 PCI, especially in an inpatient setting, may be more suitable for frail patients with limited ability to recovery from major surgery, whereas CABG may be more suitable for patients seeking long-term durability of revascularization. There has been intense interest in using observational data to compare the effectiveness of these two procedures, but such comparisons have important limitations.
The decision between CABG and PCI involves key tradeoffs that may have different implications for patients who are elderly or frail.
CABG is associated with lower rates of mortality in observational research [4][5][6] and a reduced risk of repeat revascularization in randomized controlled trials (RCTs). [7][8][9][10][11][12][13] The survival benefits may be more pronounced among elderly patients undergoing revascularization. 6 The main advantage of PCI is that less time is required for recovery following the procedure, 14 and it is associated with a lower risk of needing post-acute care services including skilled nursing facility (SNF) use. 1 The risk of SNF use among elderly patients is especially important because preoperative frailty is a key predictor for needing SNF care following discharge. While other factors including postoperative complications and prolonged ICU and hospitalization stays are also important predictors, 15,16 these factors are also associated with greater frailty. 14,17 Additionally, PCI is associated with a lower risk of stroke. 7 While some physicians may favor CABG because it is viewed as more durable than PCI and has a reduced risk of repeat revascularization, others physicians may view the reduced risk of morbidity and SNF use associated with PCI as offsetting these advantages. The current guidelines support both treatment alternatives as reasonable options for patients with left main and multivessel coronary artery disease, but do not address the association of likely needing SNF care after discharge with prognosis among patients selected for coronary revascularization. 18 Despite prior research, gaps exist in our understanding of how the evidence applies to elderly or frail patients. Patients who are very elderly (85 and older) or frail are typically excluded or underrepresented in in randomized trials comparing CABG to PCI. 7-13 Additionally, randomized trials do not typically report the use of SNF services after revascularization. [7][8][9][10][11][12][13] Given the importance of these outcomes for the elderly and frail population, more research in this area would help support selection of patients for treatment.
In this study, we addressed these gaps by using a large data source of elderly and frail patients. We sought to compare the utilization of SNF based on initial revascularization strategy and patient comorbidities. We then evaluated the association of treatment and SNF use with key outcomes including all-cause mortality, repeat revascularization, and 30-day readmissions or death. In particular, we assessed whether the association of treatment with outcomes was different among patients who did or did not have SNF use after discharge from inpatient revascularization.

| METHODS
We derived the study cohort from a 20 percent random sample of Medicare beneficiaries 65 and older who had simultaneous coverage of Medicare Parts A, B, and D for at least 1 month between 2007-2014.
We identified all patients who had PCI or CABG as part of an inpatient admission. We limited our sample to patients who had Medicare feefor-service (Parts A and B) during the month of coronary revascularization since these patients have full claims data. We also excluded patients who had PCI performed as an outpatient procedure because these patients tend to be healthier and not likely to require SNF services. Lastly, we excluded patients who were already in a nursing home before revascularization because such patients were already frail and may be more likely to use SNF after revascularization. All data came from Medicare claims and enrollment records. We received Institutional Review Board Approval from the University of North Carolina at Chapel Hill. We did not obtain informed consent from patients because of the de-identified nature of this secondary data source.
We focused on three outcomes following live discharge after revascularization: time to all-cause death, time to first repeat revascularization, and 30-day readmissions or death. We measured all-cause death using the Centers for Medicare & Medicaid (CMS) date of death variable. We identified the date of repeat revascularization and readmissions using claims data. The time to all-cause death and repeat revascularization were measured starting at the discharge date for the index hospitalization during which the patient received an initial revascularization.
The key explanatory variables were whether the patient used Medicare inpatient post-acute care services within 30 days following discharge and the initial revascularization strategy (PCI or CABG). The inpatient post-acute care services included stays in long-term care hospitals (LTAC), inpatient rehabilitation facilities (IRF), and SNF. The vast majority of post-acute care utilization was for SNF (82.5%) and so we refer to this dichotomous explanatory variable as 'SNF' although it also encompasses LTAC and IRF. In addition, we controlled for additional clinical and demographic factors that may be associated with outcomes: age, race/ethnicity, admitting diagnoses, multi-vessel vs. single-vessel PCI, the Charlson Comorbidity Index, Medicare/Medicaid dual eligibility, hemodialysis status, and discharge year.
We performed the analyses using Stata version 16 (StataCorp, College Station, TX). We first evaluated SNF use as an intermediate outcome using logistic regression and then included SNF use as an explanatory variable in all subsequent regressions. We did not attempt to adjust for the fact that SNF use technically occurs after revascularization choice, so all estimation results pertain to associations rather than causal effects. We evaluated time-to-death and time-to-revascularization using Cox proportional hazards regression for deaths and for revascularizations for up to 365 days following discharge. In addition to these regression models, we also evaluated time-to-death between 30 and 365 days by excluding patients who died within the first 30 days. We performed this additional regression because SNF use was measured within the first 30 days. Since patients had to survive long enough to receive SNF at some point in the first 30 days, this measure could have led to immortal time bias. 19 By focusing on patients who survived the first 30 days, we address the possibility of such bias. We evaluated 30-day readmissions or death using logistic regression.
We evaluated the Cox proportional hazards assumptions using the Schoenfeld residuals test. 20 We found that the null hypothesis of proportional hazards was rejected for the 0-365 days death model, the 30-365 days death model, and the repeat revascularization model (p < .001). To address the violation of the proportional hazards assumption, we performed sensitivity analyses using extended Cox models. We divided the days into four time periods (0-90 or  (Figure 1(A)) illustrates the wide gap in time-to-death between patients with and without SNF use for both CABG and PCI. The Kaplan-Meier curves for repeat revascularization (Figure 1(B)) show that the incidence of repeat revascularization was similar for CABG patients regardless of SNF use. The incidence of repeat revascularization was higher for PCI patients compared to CABG patients, but the incidence was similar for patients with and without SNF use.
The adjusted outcomes were similar to the unadjusted outcomes (  Figure 1A). The HRs for PCI and CABG patients with SNF compared to PCI patients without SNF were greater in magnitude for days 0-90 and the magnitude steadily declined to the period for days 270-365. The findings were similar for the 30-365 days model (Supplemental Figure 1B). This pattern suggests that the association between SNF use and death was stronger for time periods closer to the date of revascularization.
For repeat revascularization, the results were more stable over the time periods (Supplemental Figure 1C). The rate of repeat revascularization for PCI with SNF use was slightly lower and the rate for CABG with SNF was much lower compared to PCI without SNF across all four periods.

| DISCUSSION
In a large sample of patients receiving inpatient coronary revascularization, SNF use was more common following CABG compared to PCI.
However, among patients with SNF utilization, mortality was significantly higher among patients selected for PCI. These findings were present despite adjustment for age, sex, multi-vessel vs. single-vessel PCI and medical comorbidities, suggesting that unmeasured confounders are associated with SNF use and subsequent mortality, rather than revascularization itself. In addition, SNF use overall was associated with lower rates of revascularization despite higher mortality, suggesting that patients transferred to SNF were less likely to be offered repeat procedures and were likely treated more conservatively due to higher burden of comorbidities.
We found that SNF use was strongly associated with an increased all-cause mortality and 30-day readmissions. While it is not surprising that patients with SNF use had worse outcomes since these patients are more likely to be in worse health, have experienced postoperative complications and have had prolonged stays, this association was significant despite controlling for comorbidity burden and age. The strong association with all-cause mortality also remained after excluding patients who died within the first 30 days (the same period when SNF use was measured). We also found that the association was strongest in the first period after revascularization and weaker in later periods, suggesting that post-discharge SNF utilization is a particularly strong predictor of early mortality.
These results add important context to the tradeoffs that exist when selecting a revascularization strategy for elderly and frail patients. Like previous studies, we found that patients who received CABG had a slightly lower all-cause mortality 4-6 and a lower rate of repeat revascularization compared to patients who received inpatient PCI. [7][8][9][10][11][12][13] We also confirmed that patients who received CABG were more likely to have SNF use. 1  CABG was associated with a decrease in mortality and the percentage of patients with readmission or mortality compared to PCI. However, the proportion of CABG patients who required SNF use was much higher. Additionally, PCI inpatients with SNF use also had a higher comorbidity burden and were older than CABG patients with SNF use. These results suggest that the small subset of patients who received PCI and required SNF use were a group that was at especially high risk for poor outcomes. This is particularly important given recent findings of randomized clinical trials demonstrating the effectiveness of medical therapy alone as an initial treatment strategy for patients with stable coronary artery disease. 23,24 For patients with advanced frailty likely to require SNF utilization and associated poor prognosis, clinicians may be more likely to consider an initial trial of medical therapy alone rather than invasive evaluation.
These findings also have important implications with respect to treatment selection for revascularization patients who are elderly and frail who remain candidates for invasive evaluation. First, the findings support the concerns about the risk of CABG use in this frail and elderly population. CABG patients were much more likely to require SNF use despite being younger and having a lower average comorbidity burden. The results suggest greater caution may be warranted when selecting treatment for patients at high risk for SNF use. Second, the results suggest that there is selection in favor of PCI among patients with extreme frailty and more advanced age at baseline. Inpatient PCI patients were older, had a higher comorbidity burden, and had a worse prognosis conditional on having SNF use. This selection of patients with extreme frailty to receive inpatient PCI may be reasonable since PCI is a minimally invasive procedure whereas CABG is an invasive surgery that was associated with a greater risk of postrevascularization SNF use. However, the poor outcomes among the subset of inpatients with PCI and SNF use may reflect a need for greater consideration for whether patients with extreme frailty are good candidates for revascularization. Additionally, such selection presents a challenge for observational research that compares outcomes between PCI and CABG. The patients who receive inpatient PCI may be more frail in unobservable ways and have a worse prognosis prior to revascularization. This issue may partially explain the survival advantage for CABG compared to PCI that was found in this study and in prior observational research. [4][5][6] Third, current results from clinical trials are unlikely to determine the optimal revascularization strategy for elderly and frail patients. Physicians are unlikely to allow this complex treatment selection to be left to chance, and improved ability to analyze observational data will remain an important method of evaluating the comparative effectiveness of revascularization strategies.
The strengths of this cohort include the large sample size and that it is nationally representative of elderly Medicare fee-for-service beneficiaries.
The study also has several limitations. First, we excluded Medicare Advantage beneficiaries and our results may not generalize to these patients.
Second, as an observational study, the associations between receipt of CABG or inpatient PCI and outcomes should not be interpreted as causal.
A randomized controlled trial would be necessary to evaluate the comparative effectiveness of inpatient PCI and CABG in this population. Third, SNF use is an indicator associated with functional status, postoperative complications, and prolong stays that occurs after revascularization and did not cause the observed associations with outcomes. Fourth, while we were able to control for comorbidity burden and admission diagnoses, we were unable to control for clinical fa//ctors that are not available in claims data such as severity of CAD or anatomic variants. 25 In summary, among a large cohort of elderly and frail Medicare fee-for-service beneficiaries receiving revascularization, we found that CABG was associated with a lower rate of death and repeat revascularization, but a higher risk of readmissions. We also found that SNF use was more likely among CABG patients and that SNF use was associated with poor outcomes, particularly among patients receiving inpatient PCI. These observations highlight the important limitations in conducting observational research in the comparative effectiveness of CABG compared to inpatient PCI. The results suggest that caution is needed in treatment selection for patients at high-risk for SNF use and that selection of inpatient PCI over CABG may be associated with greater frailty and a worse overall prognosis.