Home Health Care Use and Outcomes After Coronary Artery Bypass Grafting Among Medicare Beneficiaries

The data for this study included Medicare fee-for-service administrative claim files from 2016 to 2019, including inpatient, home health, skilled nursing facility, Carrier, outpatient facility, and beneficiary summary files. Access to these data files was granted through an unfunded data use agreement (20-UFA04161) and approved by the University of Michigan Institutional Review Board (HUM00175541; approved on March 2, 2020). American Hospital Association Annual Survey and Distressed Community Index (DCI) data files were used to supplement Medicare claims with hospital and regional socioeconomic factors. The DCI is a composite measure of local community distress based on the American Community Survey (2014–2018) developed by the Economic Innovation Group, including zip code level rates of high school diploma education, poverty, employment, housing vacancy, median household income, change in employment, and change in local establishments.¹⁵ The underlying data in this study cannot be made available due to existing data use agreement restrictions.

The study sample included all Medicare fee-for-service beneficiaries aged ≥65 years who underwent CABG with or without concomitant valve procedures between July 1, 2016, and December 31, 2018, with continuous coverage for 6 months before surgery and 1 year after discharge, and were discharged alive (n=152 806; Figure S1). International Classification of Diseases, Tenth Revision, procedure codes were used to identify CABG procedures (0210*, 0211*, 0212*, and 0213*). Beneficiaries were excluded if they had a concomitant cardiovascular procedure (n=26 141), were discharged to a location other than home (n=46 667), or had missing covariate information (n=2667) for a final sample of 77 331 beneficiaries.

Home health files were used to identify any claims of HHC for the beneficiaries in the sample. The primary definition of HHC use in our study was 30-day HHC use versus no HHC use based on the presence of any HHC claims within 30 days of discharge.¹¹

Clinical outcomes evaluated in this study included 30- and 90-day postdischarge mortality, ED visits (without hospital admission), all-cause hospital readmission, and CR enrollment. Mortality was identified as having a date of death present in the Medicare beneficiary summary file that occurred within 1 year of discharge; otherwise, the beneficiary was assumed to be alive.¹⁶ ED visits were identified using outpatient facility claims with revenue center codes 040-049.¹⁷ Hospital readmissions were identified as at least 1 acute care hospitalization admission in the Medicare inpatient files.¹⁶ Enrollment in CR was defined as having at least 1 CR claim within 30 or 90 days after discharge using Current Procedural Terminology codes 93797 or 93798 and Healthcare Common Procedure Coding System codes G0423 of G0424 paired revenue center code 943 in the carrier or outpatient facility claims files.¹⁸

Beneficiary-level covariates were created from Medicare claims files and included demographic and clinical information. Demographic factors included beneficiary age, sex (male versus female), reported race/ethnicity category (White, Black, other, Asian, Hispanic, North American Native, and unknown), and Medicare/Medicaid dual-eligibility status (yes versus no). Clinical information included elective admission status (yes versus no), transfer from another acute care facility (yes versus no), and index hospitalization length of stay (in days). The degree of comorbidity burden was evaluated using the Charlson comorbidity index.¹⁹ Established claim–based algorithms were also used to identify postoperative complications, including acute kidney injury, bleeding, pneumonia, cardiac arrest, stroke, myocardial infarction, and sepsis.²⁰ The claim-based frailty index was used to ascertain the degree of frailty in beneficiaries and categorized as 0 to 0.14, 0.15 to 0.24, and 0.25 or more²¹ based on prior work.²² Beneficiary zip code was used to incorporate DCI data, including the DCI composite score quintile (first quintile, prosperous community to fifth quintile, distressed community) and census region (midwest, northeast, south, and west). Hospital characteristics from the American Hospital Association Annual Survey included bed size (≥500, 300–499, 100–299, and <100 beds), teaching status (major teaching, minor teaching, and nonteaching), and ownership status (nonprofit, for-profit, and public).

Univariate analysis described HHC use and the frequency of HHC visits within 30 and 90 days and 1 year after discharge. Bivariate analysis compared patient and hospital factors associated with HHC use with χ² tests and ANOVA used to test for statistically significant differences. Hierarchical logistic regression models were used to identify independent predictors of HHC use among patients and hospital factors associated with HHC use in bivariate analyses and included a random hospital effect. Interhospital variation in HHC use was described using hospital-level rates of HHC use among patients attributed to each hospital. The intraclass correlation coefficient and median odds ratios were estimated using the hospital random effect variance estimate from the hierarchical logistic regression model to quantify the extent to which variation in HHC use was attributed to the discharging hospital.

To compare outcomes between HHC users and nonusers, we used a propensity matching approach that included 2 steps: (1) constructing a propensity score for HHC use for everyone based on patient demographic, clinical, geographic, and hospital factors and (2) estimating the average treatment effect of HHC use on outcomes in a matched cohort of HHC users and nonusers. For the first step, a hierarchical logistic regression model was used to construct propensity scores for each beneficiary, representing the probability of the individual receiving HHC. Covariates in the model were chosen based on the significant independent predictors identified previously and also included the hospital random effect to account for clustering.²³ Box and Whisker plots were created to compare the distribution of propensity scores among HHC users and nonusers. In the second step, the propensity scores were used to match each beneficiary with HHC use to another patient without HHC use with a similar propensity score. Matching was conducted using a nearest neighbor 1:1 approach with a caliper of 0.25*SD, of the propensity score (SD, 0.23) of 0.575.²⁴ Raw and standardized absolute differences in covariates for the matched sample were estimated and displayed using a love plot to illustrate covariate balance before and after propensity matching. Conditional logistic regression was used to evaluate the average treatment effect associated with HHC use across all outcomes and to account for matching. The average treatment effect among the treated was used to estimate the effect of HHC, which avoids making inferences on the effect of HHC among beneficiaries that did not use HHC. All analyses were conducted using Stata/SE, version 18.0.

In our sample of 77 331 beneficiaries undergoing CABG who were discharged to home, 26 751 (34.6%) had any HHC use within 30 days of discharge. The mean age of our sample was 72.7±5.1 years, 60 518 (78.3%) were male, 69 226 (89.5%) were White, and 6226 (8.1%) were dually eligible for Medicare and Medicaid (Table 1). Beneficiaries using HHC tended to be older, male, White race/ethnicity, in a lower frailty category, located in the midwest and northeast regions, in more prosperous communities, from suburban geographic areas, and admitted to hospitals with more beds and major teaching hospitals. Non-HHC users were more likely to be dual eligible, transferred in from another hospital, more comorbid, from the south and west regions, from more distressed communities, located in rural and small-town locations, and admitted to smaller hospitals and nonteaching hospitals. The frequency of individual comorbidities in the Charlson comorbidity index for the overall sample and by HHC use can be found in Table S1.

Multivariable analysis identified several patient and hospital characteristics associated with HHC use after CABG (Table 2). Notably, increasing age (per year; adjusted odds ratio [aOR], 1.015 [95% CI, 1.012–1.018]; P<0.001), male sex (aOR, 1.07 [95% CI, 1.03–1.11]; P<0.001), higher claim-based frailty index category (0.25+ versus 0–0.14; aOR, 1.10 [95% CI, 1.02–1.19]; P<0.001), and suburban (versus urban) geographic location (aOR, 1.22 [95% CI, 1.16–1.28]; P<0.001) were associated with greater relative odds of HHC use adjusting for all other variables. On the other hand, Black (aOR, 0.90 [95% CI, 0.83–0.98]; P<0.001) and Hispanic (aOR, 0.81 [95% CI, 0.68–0.96]; P<0.001) race/ethnicity (versus White), dual eligibility (aOR, 0.91 [95% CI, 0.85–0.97]; P<0.001), increasing Charlson comorbidity index (per unit increase; aOR, 0.97 [95% CI, 0.96–0.98]; P<0.001), located in the south (aOR, 0.58 [95% CI, 0.56–0.61]; P<0.001) or west (aOR, 0.41 [95% CI, 0.39–0.43]; P<0.001) census regions (versus midwest), located in the more distressed communities (fifth DCI quintile versus first; aOR, 0.70 [95% CI, 0.66–0.74]; P<0.001), located in rural (aOR, 0.78 [95% CI, 0.74–0.82]; P<0.001) or small town (aOR, 0.87 [95% CI, 0.82–0.91]; P<0.001) geographic locations (versus urban location), and smaller bed size hospital (<100 versus 500+ beds; aOR, 0.44 [95% CI, 0.40–0.49]; P<0.001) were associated with lower relative odds of HHC use adjusting for all other variables.

Of the 872 hospitals with at least 20 CABG cases in our sample during the study period, the average hospital–level rate of HHC use was 35.1%±23.6% (Figure 1). The median hospital–level HHC use rate was 31.0% (interquartile range, 13.7%–54.5%) and ranged from 0% to 94.2%. After adjusting for patient- and hospital-level factors, 29.6% of the variation in HHC use was attributed to the discharging hospital (intraclass correlation coefficient, 0.296 [95% CI, 0.275–0.318]). The median odds ratio for the hospital variance estimate was 3.07 (95% CI, 2.90–3.26).

Descriptive characteristics of the propensity-matched sample can be found in Table S2. Crude outcomes for HHC users and nonusers can be found in Figure 2. Thirty-day mortality (0.7% versus 0.4%) and 90-day mortality (1.2% versus 0.9%) were higher in HHC users compared with nonusers (both P<0.001). However, HHC users had lower rates of ED visits (30 days: 11.7% versus 14.4%; 90 days: 19.9% versus 22.6%; both P<0.001) and all-cause readmissions (30 days: 7.9% versus 12.2%; 90 days: 13.9% versus 18.3%; both P<0.001). Enrollment in CR was also significantly higher for HHC users compared with nonusers at 30 (22.9% versus 18.8%; P<0.001) and 90 days (64.1% versus 52.0%; P<0.001) after discharge.

The distribution of predicted probabilities (or propensity scores) of HHC use for HHC users and nonusers can be found in Figure S2. After matching, our sample was limited to 50 580 beneficiaries with 26 751 individuals in the HHC use and nonuse groups, respectively. The raw and standardized absolute differences in covariates can be seen in Figure 3. Standardized differences in baseline characteristics did not exceed 0.10, indicating good covariate balance.²⁴ After matching, the average treatment effects of HHC use on mortality was 0.3% (95% CI, 0.1%–0.6%; P=0.005) at 30 days after discharge and 0.3% at 90 days after discharge (95% CI, 0.01%–0.5%; P=0.046; Table 3). HHC users were significantly lower for readmissions at 30 (−7.7% [95% CI, −8.9% to −6.6%]) and 90 days after discharge (−8.1% [95% CI, −9.5% to −6.8%]) and ED visits at 30 (−3.1% [95% CI, −4.1% to −2.2%]) and 90 days after discharge (−3.3% [95% CI, −4.4% to −2.2%]; all P<0.001). Enrollment in CR was significantly higher for HHC users at 30 (11.2% [95% CI, 9.9%–12.8%]) and 90 days after discharge (11.1% [95% CI, 9.7%–12.6%]; both P<0.001).

Despite reports demonstrating that HHC is the fastest-growing discharge destination among CABG recipients, few studies have examined its impact on improving transitions of care and outcomes in this patient population. Similar to our findings, a study of Medicare and commercially insured patients with CABG in Michigan found that nearly over half of patients received HHC within 90 days of discharge.²⁵ In the only other national investigation, Deo et al¹⁴ used 2012 to 2014 data from the National Readmission Database and found HHC to be associated with lower odds of 30-day readmission. Our study confirms this finding but advances our understanding of HHC use in CABG by our use of a large, national data set and a comprehensive set of outcomes, some of which are process measures and others of which are patient-centered. One explanation for the association between HHC use and lower readmission rates is that HHC supported postoperative care (ie, vital signs, medication reconciliation, surgical site, and wound care) and with ambulation and other aspects of physical recovery in the home in high-risk patients.²⁶ However, we were unable to evaluate the patterns or intensity in which HHC agency clinicians (eg, nurses and physical therapists) or other professionals (aides) provided care for these patients. The use of HHC among high-risk patients might also explain our findings that suggest an association between HHC use and increased mortality, which could be attributed in part to unmeasured confounding and might be further assessed through rigorous comparative effectiveness studies or clinical trials.

Importantly, the study results also highlighted significant disparities in HHC use after CABG across racial/ethnic and socioeconomic groups. For example, HHC use was less common among Black and Hispanic beneficiaries, individuals dually eligible for Medicare and Medicaid, and individuals located in socioeconomically distressed communities and rural areas. Given longstanding disparities in quality and outcomes for individuals undergoing CABG, as well as some reported disparities in HHC use by region and race, it will be important for future work to identify factors and downstream impact of lower use of post-CABG HHC use among these vulnerable populations.^27,28

Notably, we found that for patients discharged to home after CABG, HHC use is primarily driven by the discharging hospital compared with specific patient-level factors. One possible explanation for this is the availability of HHC services, which differs by geographic region in the United States.^29–32 Another is that the hospital’s relationship with HHC agencies may be a powerful determinant of service use. That is, while some hospitals are fully integrated with HHC agencies, others refer to a list of preferred HHC agencies, and others may not have any relationship.³³ Such factors can impact the ease with which HHC assessments in the hospital are done (ie, before discharge), HHC orders are signed by physicians overseeing the plan of care, and services are initiated in the patients’ homes. Finally, some hospitals may discharge patients to SNFs over HHC, even when HHC might be feasible. Prior work also established wide variation in SNF use after CABG.³⁴ While this study demonstrates the potentially important role that the discharging hospital plays in HHC use after CABG, future studies are needed to better understand how hospital-specific factors impact HHC referral, including the perspectives of key stakeholders in the hospital-to-home transition for patients with CABG. Stakeholders are likely to include hospital executives and leaders, staff members, and HHC agency leaders and staff.

Our findings add to the discussion around recent calls from cardiac and surgical societies that promote recovery and rehabilitation at home after undergoing cardiac surgery.³⁵ Use of HHC among high-risk patients might also explain our findings that suggest an association between HHC use and increased mortality, which could be attributed in part to unmeasured confounding and might be further assessed through rigorous comparative effectiveness studies or clinical trials.^25,36–41 For example, what are the clinical benefits and potential risks of HHC for patients with CABG? While this study provides insight into the clinical outcomes associated with HHC use, a more rigorous comparative effectiveness analysis that leverages rich clinical data or more experimental designs is needed to justify expanded use. If HHC is indeed beneficial, what is the optimal timing of HHC services? Prior studies in HF and sepsis suggest that HHC nurse visits are frontloaded or received early and often in the HHC episode, alongside early physician outpatient follow-up visits.^42,43 The optimal timing of these visits is unknown in the context of CABG. Additionally, which HHC visits for patients with CABG are most beneficial, nursing or physical therapy visits? How should HHC be enhanced to optimize cardiac surgery recovery? Future studies are needed to inform targeted areas of improvement and hospital responses to value-based reimbursement policies.

While this study provides novel findings about the use of HHC after CABG, there are limitations that should be considered, and the results should be interpreted cautiously. First, although participants were balanced on several demographic, clinical, geographic, and hospital covariates, it is likely that unmeasured confounding remained. For example, we lacked data on certain factors, which might have increased the probability of participants receiving HHC and experiencing a poor outcome, such as functional status, dementia/cognitive impairment, social support, and receipt of follow-up ambulatory visits postdischarge. Moreover, risk adjustment using claims data may be limited, and future analyses should seek to replicate the study findings with robust clinical data that can better match patients based on clinical factors. Second, the present study did not have information about the number, duration, timing, frequency, and type of HHC services, which limits the ability to characterize the HHC benefit in detail. Additionally, we did not validate the fact that patients in the present sample had a minimum exposure to HHC; thus, it is unknown if patients had 1, 2, or several HHC visits postdischarge. Finally, we were not able to determine the appropriateness and preventability of readmissions and deaths observed. It is possible that some of these were clinically appropriate.

Figures S1 and S2

Tables S1 and S2