Predicted Mortality and Cardiology Follow-up Following Heart Failure Hospitalizations Among Veterans Health Administration Patients.
Jan 6, 2025DOI: 10.1016/j.cardfail.2024.12.006Journal ArticleJCF
Written by: Rebecca L TisdaleFang CaoMegan SkyeOrly VardenyKarim SallamNeil KalwaniStephanie HsaioAnubodh S VarshneyPaul A HeidenreichAlexander T Sandhu
Heart failure (HF) is the fastest-growing cardiovascular disease in the USA, affecting an estimated 6.7 million patients (1). It is associated with 1-year mortality between 13–20% (2–4), and significant and increasing costs (5). Patients hospitalized for HF remain at high risk for recurrent HF decompensation and readmission in the immediate post-discharge period (6): one quarter of patients admitted for HF will be readmitted within 30 days (7,8), with a third of these readmissions taking place within 7 days of initial discharge (9). Earlier follow-up after HF hospitalizations has been associated with better outcomes (10–12). Hence, professional guidelines (13) suggest that post-discharge follow-up after a HF hospitalization should ideally take place within 1–2 weeks to mitigate these risks.A complementary literature shows that this post-discharge follow-up most improves HF outcomes when conducted by a cardiology specialist, as compared to a primary care provider (PCP) (5,14–16). The most likely mechanisms for the added benefit of follow-up by a cardiology specialist are titration of diuretics and expedient uptitration of guideline-directed medical therapy (GDMT). Titration of GDMT to the maximum tolerated doses (13) in the minimum tolerated time period is critical for optimizing patient outcomes (17), yet treatment inertia in GDMT titration remains too common in the outpatient setting (2,18,19). As cardiologists may be more likely than generalists to rapidly and completely up-titrate outpatient GDMT medications (2), timely follow-up with a cardiologist, in place of or in addition to a PCP, is warranted after a HF diagnosis or hospitalization.However, in the setting of limited cardiology specialist resources (20,21), follow-up within 1–2 weeks may not be feasible for all patients in many healthcare systems. Under such circumstances, those at highest risk of HF mortality and/or readmission could be rational candidates to prioritize for earliest follow-up. Yet, no existing publications explore the extent to which current post-discharge follow-up patterns after HF hospitalizations correlate with baseline clinical risk. In this study, we address this gap in the literature by examining the extent to which estimated HF mortality risk at the time of hospital discharge, as assessed by the MAGGIC Heart Failure Risk Score (22), is associated with timely follow-up after HF hospitalization in the Veterans Health Administration (VA), hypothesizing that higher-risk patients will be more likely to have timely follow-up. Because the VA is the nation’s largest integrated healthcare system and patients receiving care in the integrated VA system face lower financial barriers to accessing care than in most healthcare systems, this is an ideal population in which to explore this question.
Methods
Our study cohort was drawn from VA’s Corporate Data Warehouse, which stores Veteran demographics, clinical encounters, laboratory values, vital signs, and pharmacy dispensing records drawn from VA administrative claims and electronic health record data. This study was approved by the Stanford University Institutional Review Board.
The study cohort included Veterans hospitalized with HF in an acute care VA hospital between January 1, 2018 and September 15, 2022 and discharged on or before September 15, 2022. Those receiving hemodialysis or with a prior heart transplant or ventricular assist device were excluded from the cohort. Patients were followed through September 15, 2023. We excluded those without lookback data of at least two years to ensure adequate capture of patient data within the VA system. HF diagnosis was based on ICD-10 codes (Table S1), and we included patients with HF in the primary or secondary discharge diagnosis position. We used VA ―stop codes‖ that categorize outpatient visits to identify cardiology follow-up visits, which may have been conducted by physicians or advanced practice providers (nurse practitioners or physician assistants).
We included patient- and neighborhood level sociodemographic characteristics relevant to HF risk, including age, sex, race, ethnicity, rurality, and Center for Disease Control (CDC) Social Vulnerability Index (SVI), a measure of neighborhood-level social risk (23). In VA, race and ethnicity are over 95% self-identified. Rurality was derived from U.S. Census Bureau criteria (24); Veterans residing in the US territories (US Virgin Islands, Guam, Palau, and the Northern Marianas Islands) were excluded from the analytic cohort (<0.01%). In addition, we incorporated left ventricular ejection fraction (LVEF), systolic blood pressure, heart rate, comorbidities (diabetes and chronic obstructive pulmonary disease [COPD]), smoking status, and laboratory values (serum creatinine). In all cases, we took the value closest to discharge in order to most accurately calculate risk at the time of discharge. The LVEF was derived from VA’s natural language processing algorithm, as described elsewhere (25). We also identified HF medication prescriptions at discharge and whether the HF diagnosis was incident or preexisting.
We calculated a risk score and corresponding one-year mortality risk for each patient at the time of hospital discharge based on the MAGGIC risk model (22), which takes into account a number of clinical and sociodemographic characteristics, including age, sex, LVEF, systolic blood pressure, serum creatinine, co-morbidities, and medical therapy. We imputed missing data using multiple imputation chained equations with 20 adds. As we did not have data on New York Heart Association Class (NYHA), we assumed that patients had NYHA Class III heart failure if they were primarily hospitalized for HF and Class II if they had a secondary HF diagnosis. We assumed that those without smoking information were non-smokers.
We calculated and described sociodemographic and clinical characteristics by MAGGIC risk, using Cohen’s D to characterize effect size for continuous variables and Crámer’s V for categorical variables. For a sub-cohort of Veterans surviving for one year after discharge, we calculated the proportion of Veterans with each MAGGIC risk quintile with an outpatient follow-up visit post-hospital discharge within two weeks and one, six, and twelve months. We described the effect size of the absolute differences between these proportions with Cohen’s D statistic. We repeated this calculation with stratification by left ventricular ejection fraction (≤40% versus >40%) and across VA facilities, using ANOVA to test whether differences were statistically significant. In addition, we performed several other sensitivity analyses. First, we assumed all patients had NYHA Class III heart failure at discharge for the purposes of calculating the MAGGIC score. Second, we compared discharges that took place before versus during the COVID-19 pandemic, with March 11, 2020, the day when the World Health Organization formally declared a pandemic (26), considered the first day of the pandemic period. Third, we recategorized hospital readmissions within a given time window as visits.
We constructed a time-to-event analysis to estimate the effect of MAGGIC mortality risk on one-month cardiologist follow-up, with readmission and death as competing risks. One month was chosen for the primary analyses based on hypothesized logistical and administrative challenges with achieving cardiology follow-up of a majority of patients within two weeks of discharge, despite guidelines recommending that this is the ideal window (13).
We also performed a logistic regression analysis of MAGGIC mortality risk on one-month cardiologist follow-up visit with fixed effects for site, adjusting for calendar year, age, sex, race/ethnicity, rurality, and CDC SVI as well as presence of HFrEF diagnosis, since HFrEF prevalence is known to vary by sex and race/ethnicity in VA (27), and whether the discharge was associated with a primary hospital diagnosis of HF. We limited these analyses to patients surviving at least one month (30 days) without HF readmissions. We then repeated these analyses with two-week cardiology clinic follow-up as our outcome variable. Given that some individuals had multiple hospitalizations during the study period, models accounted for within-individual correlation of variance via clustering of standard errors on the individual. For all logistic regression analyses, we modeled MAGGIC risk with restricted cubic splines using Stata’s mkspline and marginscontplot2 packages. Due to technical complications of graphing splines with imputed data, we repeated the analyses with complete cases to depict the relationship between MAGGIC risk and one-month cardiologist follow-up; these figures were constructed among the sub-cohort of patients surviving at least one year after discharge to account for survivorship bias. We constructed similar regression models limiting the cohort to patients with de novo HF diagnoses (first HF diagnosis within 18 months prior to hospital admission, using the same criteria as the MAGGIC score) and limiting to patients with a primary hospital diagnosis of HF.
Finally, we examined the variability in proportion of HF hospitalization discharges followed by cardiologist follow-up at various intervals across facilities with 20 or more HF hospitalizations during the study period. This included construction of a mixed effects logistic regression model, adjusted for the same patient-level characteristics as the logistic regression models described above, for the purposes of calculating a median odds ratio.
As our data included only VA-based visits and admissions, readmissions outside VA hospitals were not reflected in this analysis. All analyses and statistics were performed in Stata 18 (StataCorp LLC, College Station, TX).
Results
Our study cohort included 84,348 Veterans hospitalized with HF with a total of 120,619 hospital admissions (Table 1). The sub-cohort of patients surviving for at least one year after discharge included 57,554 Veterans with a total of 79,866 hospital admissions, of which 32.1% were followed by a cardiology visit within two weeks; 49.3% within one month; 78.6% within six months; and 84.4% within one year. Patient baseline characteristics stratified by whether or not follow-up took place within these time periods are presented in Table S1.
MAGGIC risk scores among the full cohort ranged from 3–49 with a median score of 23. Quintile cut-offs were 19, 23, 26, and 30. Predicted mortality increased from 2.0% one-year mortality/5.2% three-year mortality for the low score of 3, to 81.6% one-year/98.0% three-year mortality for the high score of 47 (22). Table S2 provides risk score-predicted mortality pairings for all relevant risk scores.
Among patients surviving at least one year after discharge, the proportion of individuals with a follow-up visit increased as the time interval from discharge lengthened for all risk quintiles. In general, patients in the highest risk quintile (Quintile 5) had the highest rates of post-discharge follow up for all time intervals evaluated. Differences in proportions of admissions with a follow-up visit within a given interval across quintiles were significant, but were small (maximum Cohen’s D = −0.05, for the difference between follow-up proportions for Quintile 1 vs 5 for two-week follow-up) (Figure 1, Table S3).
Within a given risk quintile, patients with HFrEF (EF≤40%) were more likely to have follow-up than those with HFpEF (EF>40), across all follow-up intervals and risk quintiles (Table S4). Admissions for HFpEF had the same or marginally lower follow-up rates as risk quintiles increased, while admissions for HFrEF had similar follow-up rates across risk quintiles. Additionally, admissions for patients with HFrEF in the lowest risk quintile (Quintile 1) had higher rates of follow-up visits than admission for patients with HFpEF in the highest risk quintile (Quintile 5) for all intervals evaluated.
Marginal probabilities of one-month follow-up were higher for patients discharged with the highest risk, though with less than a 10% absolute difference in marginal probability for lowest- vs highest-risk discharges (Figure 2).
In our competing risk time-to-event model, we found a slightly negative association between MAGGIC score and 1-month cardiology follow-up; for each unit change in MAGGIC score, the likelihood of 1-month follow-up decreased by 0.004 (95% confidence interval [CI] −0.005 - −0.003).
In our logistic regression models, after adjusting for selected patient-level sociodemographic and clinical characteristics, neighborhood-level social risk, year, and facility among patients surviving without readmission at least one month, there were differences in probability of follow-up across sociodemographic groups (Figure 3). For a given level of HF-associated risk, female and Black Veterans were less likely to have follow-up within one month of hospital discharge (Table S5). Black Veterans were also less likely to have follow-up within two weeks of discharge, as were individuals without ethnicity listed. Patients with HFrEF were more likely than those with HFmrEF or HFpEF to have one-month or two-week follow-up, as were those with a primary hospital diagnosis of HF compared to patients with only a secondary diagnosis of HF. In addition, discharges in 2019–2022 were slightly less likely to be associated with one-month follow-up compared to the reference year of 2018. Rurality and neighborhood-level social risk according to CDC SVI were not associated with likelihood of two-week or one-month follow-up.
Results varied slightly when the cohort was limited to patients with de novo HF and/or with a primary hospital diagnosis of HF (Figure S1, Table S6). There was no longer a difference by sex when limiting to de novo or primary HF diagnoses. There was also less variability across the spectrum of HF risk when only discharges from a hospitalization with a primary hospital diagnosis of HF were included (Figure S1B). When we assumed that all patients were discharged with NYHA Class III heart failure, results were similar to our primary analysis (Figure S2), though there was less variability across MAGGIC quintiles. Follow-up proportions within 14 and 30 days were similar for discharges that took place before versus during the COVID-19 pandemic (Figure S3) as well as with VA hospital readmissions counted as visits (Figure S4).
A total of 114 facilities with at least 20 HF hospitalizations and with any follow-up visits during the study period were identified. Among discharges from these facilities, rates of two-week follow-up varied from 4% to 54% with a mean of 29%; one-month follow-up varied from 5% to 69% with a mean of 44% (Figure 4); six-month follow-up varied from 12% to 84% with a median of 70%; and one-year follow-up varied from 18% to 86% with a median of 75%. The median odds ratio was 1.82 (95% CI 1.68, 2.00), which can be interpreted as 1.8-fold variation in the probability that identical patients at two randomly-selected VA facilities would receive follow-up within one month (28).
Discussion
In this cohort study, we found that less than one-third of VA acute care hospital admissions with HF in 2018–2022 were followed by a cardiology visit within two weeks, and less than half within one month. Across Veterans of varying mortality risk, there was a minimal relationship between risk and likelihood of having follow-up within one month of discharge; even among a subgroup surviving at least one year after discharge, those with the highest risk as estimated by MAGGIC scores were only slightly more likely to have follow-up within one month of discharge. This relationship was consistent before and during the COVID-19 pandemic. We demonstrated inequities in follow-up across sociodemographic and clinical groups, with female and Black Veterans, and admissions for HFmrEF/HFpEF, having significantly lower follow-up rates. Additionally, there was substantial variability across VA facilities in rates of timely follow-up. To our knowledge, this is the first publication characterizing the relationship between risk and timeliness of cardiology follow-up in patients hospitalized with heart failure.
The overall limited and variable cardiology follow-up care after discharge from HF hospitalizations is a significant opportunity for care quality improvement, given the association of timely follow-up with lower readmission and mortality rates. Individuals may be following up with primary care providers in a shorter window, which may mitigate some risk. However, evidence of a greater risk-reduction effect with subspecialty over generalist follow-up suggests these patients may be best served by early cardiology visits after discharge (14,16).
Our results identifying at best a modest relationship between timeliness of follow-up and estimated mortality risk, particularly at the highest levels of risk, support our hypothesis that Veterans at higher risk would be more likely to have timely follow-up. This suggests a tacit acknowledgment that follow-up within two or four weeks after every HF hospitalization may not be feasible, and a corresponding triage of higher-risk patients to closer follow-up. If it is indeed infeasible for every HF hospitalization discharge to be followed by a cardiology follow-up visit within two to four weeks, it is likely appropriate to reserve far more visits in this period for patients at higher risk for near-term readmission and death. However, the association between risk and timeliness of follow-up was weaker than anticipated. This could be attributable to incomplete mental models of risk: for example, the fact that we found elevated odds of timely follow-up after hospitalizations with a primary diagnosis of HF compared to hospitalizations with a secondary HF diagnosis, even when adjusting for MAGGIC risk, suggesting higher-risk patients may be passed over for early post-HF hospitalization follow-up when HF is not the focus of the hospitalization. Making a patient’s risk level clearer to the multi-disciplinary team and targeting higher-risk patients for more intensive intervention, for example with EMR-based dashboards (29–31), could help to strengthen the relationship between risk and prioritization for follow-up.
It is possible that making risk information more transparent could also mitigate the differences we found in timeliness of follow-up between male and female Veterans, and across race and ethnicity groups, for a given level of HF-associated risk. These differences persisted even when adjusting for neighborhood-level social risk and ejection fraction—and for facility, meaning these differences reflect patterns within a given facility. Unfortunately, this mirrors findings related to timeliness of cardiology follow-up in other domains, such as delayed incident HF diagnosis in Black Veterans (32) and a lower rate of timely angiography for both women and Black patients presenting with acute MI compared to men and White patients, respectively (33). These structural inequities suggest that additional efforts to ensure access to timely specialty care are needed to adequately reach these Veteran subpopulations.
Veterans with HFrEF had much higher rates of timely follow-up in our cohort than those with HFmrEF or HFpEF for a given risk level. This may reflect perceptions that the scope of interventions benefiting from cardiology specialty care is larger for the subset of patients with reduced versus mildly reduced or preserved EF, such as more extensive GDMT (13). However, the most recent guidelines recommend consideration of several of the same categories of GDMT used in HFrEF to decrease hospitalizations in those with HFmrEF and HFpEF (13), and timely follow-up is associated with improved mortality across the EF spectrum (34)—suggesting there is indeed much that cardiology specialty follow-up can offer patients with HFmrEF or HFpEF after discharge from HF hospitalizations.
As has been described in other settings (12), the rate of one-month post-discharge follow-up varied substantially across VA facilities with at least 20 HF hospitalizations in our cohort, from less than 10% to nearly 70%. This variation could be attributable to differences in quality of care, unmeasured patient-level characteristics, or local resource availability. To the extent that limited resource availability plays a role, telehealth-based clinics like VA’s clinical resource hub program, which provides virtual care for patients based at facilities with staffing shortages, are a potential mechanism to close this gap (35,36). Leveraging the entire multidisciplinary HF team, for example with close HF pharmacist-led or nurse-led follow-up to optimize GDMT, is another evidence-based strategy to improve access to guideline-concordant care (37).
We used mortality probabilities derived from the original MAGGIC cohort (22). It is likely that HF-related mortality has changed over the decade since publication of these mortality probabilities (2,38). However, if mortality has improved across the spectrum of HF severity, these probabilities would not be expected to vary differentially by risk level. As such, even if the exact mortality probabilities have evolved since 2013, this is unlikely to substantially change the patterns and relationship demonstrated in this study. Still, our study is limited by use of predicted rather than observed mortality rates.
Due to data limitations, we were unable to specify NYHA class, a component of the MAGGIC score calculation. We therefore assigned patients with a primary hospital diagnosis of HF to Class III and those with a secondary hospital diagnosis to Class II. While we have conducted a sensitivity analysis to test the effect of categorizing all patients as Class III, that we are unable to precisely attribute NYHA class across the study cohort remains a limitation of this analysis.
Our data prevents determination of patient-specific factors that could contribute to suboptimal follow-up. We also do not have data regarding the availability of general cardiology and HF-specific providers at a given facility, which may have impacted cardiology follow-up in certain institutions and contributed to the variability in follow-up across the VA system.
Finally, this study took place in the VA healthcare system and thus within a patient population in which women and races and ethnicities besides Black and White are underrepresented compared to the general U.S. population. As such, any associations identified within these populations should be considered hypothesis-generating only with respect to other settings.
In this contemporary, large cohort of HF hospitalizations in the VA system, the rate of cardiology subspecialty follow-up visits following discharge was overall low and varied across sex, race/ethnicity, ejection fraction, and hospitals. Additionally, follow-up rates were only moderately correlated with estimated mortality risk at discharge in a sub-cohort surviving one year past discharge, and this relationship was close to null when competing events (death or rehospitalization) were accounted for in the full patient cohort. More closely matching timeliness of cardiology follow-up after discharge from a HF hospitalization with mortality risk may represent an important intervention to improve HF outcomes and optimize resource allocation, particularly in higher-risk individuals.
SV
Dr. Sunny Varshney
Cardiology · Advanced Heart Failure & TransplantationStanford MedicinesunnyStanford, CA
