The Worsening Renal Perfusion Index Predicts the Prognoses of Heart Failure Patients Treated with Sacubitril/Valsartan

Guideline-directed medication therapies (GDMTs) are evidence-based treatments for heart failure with a reduced ejection fraction (HFrEF). Along with β-blockers and sodium-glucose cotransporter 2 inhibitors (SGLT2i), the blockade of the renin-angiotensin-aldosterone system (RAAS) with angiotensin-converting enzyme inhibitors (ACEIs), angiotensin II receptor blockers (ARBs), and mineralocorticoid receptor antagonists (MRAs) has been the cornerstone of pharmacotherapy for HFrEF. As the leading angiotensin receptor-neprilysin inhibitor, sacubitril/valsartan improved patients’ prognoses with HFrEF [1] due to its mechanical and functional reverse ventricular remodeling effect [2]. Sacubitril/valsartan is used to manage heart failure in ambulatory and hemodynamically stable hospitalized patients [3‒5].

A decline in the estimated glomerular filtration rate (eGFR) is expected upon therapy with RAAS inhibitors, including sacubitril/valsartan [6]. The European Heart Failure Association has issued a consensus document stating that the drop in eGFR does not necessarily indicate a poorer prognosis or necessitate a change in therapy [7]. eGFR is determined by evaluating serum creatinine levels and is commonly used as a clinical indicator of renal function. Nevertheless, eGFR in heart failure has constraints due to various factors that impact serum creatinine, including volume status and renal hemodynamics, on top of age and muscle mass [8].

In addition to eGFR, blood urea nitrogen (BUN) has been strongly recommended to evaluate renal function in patients with heart failure [9, 10]. More importantly, BUN is one of the most compelling prognostic predictors in patients with heart failure because its elevation implies neurohormonal activation or renal hypoperfusion [11‒13]. However, the time-varying BUN data in most studies lacked [1, 14, 15], and the recommended dose adjustment of sacubitril/valsartan is based on changes in serum creatinine levels [7]. Therefore, in this pharmacoepidemiological study, our objective was not only to examine the real-world efficacy of sacubitril/valsartan but also to evaluate the prognostic impact of changes in BUN levels, together with changes in serum creatinine levels and eGFR, in patients with HFrEF who received sacubitril/valsartan. We collected patients’ demographic, drug history, laboratory, echocardiographic, and clinical data before and after the first sacubitril/valsartan prescription to identify risk factors, prognostic factors, or predictors of all-cause death or heart transplant in patients with HFrEF.

This observational cohort study was conducted at the National Taiwan University Hospital (NTUH), a tertiary care medical center, and was approved by the NTUH Institutional Review Board (201907090RINB). We identified 911 patients with sacubitril/valsartan prescriptions from March 2017 to March 2019 in the NTUH Integrated Medical Database (NTUH-iMD). The index day for each patient was defined as the date of the first prescription of sacubitril/valsartan. The follow-up period was from the index day to December 2020, or the occurrence of all-cause death or heart transplant. Thus, the primary endpoint of this study was the survival time from the index day to the event of all-cause death or heart transplant. However, as shown in the flow diagram in Figure 1, 620 patients with age <20 years, a baseline left ventricular ejection fraction (LVEF) ≥40%, lack of baseline or follow-up LVEF data during the follow-up period, sacubitril/valsartan initiated in other medical facilities, or inaccessible baseline datawere excluded from the data analysis.

The real-world data of 291 remaining patients were retrospectively collected from the NTUH electronic medical records, but the NTUH-iMD further verified the occurrence of death. Specifically, we collected real-world demographic and clinical characteristics data, drug histories of sacubitril/valsartan and concomitant medications, laboratory results, echocardiography, and hospitalizations at baseline and during the follow-up period in all recruited patients, where the “baseline” was defined as within 1 year before the index day (esp., for medications) or on the index day.

In particular, the drug histories of sacubitril/valsartan and concomitant medications, including the GDMT [16] and treatments for anemia, hyperlipidemia, and other cardiovascular conditions, respectively, were carefully extracted from the electronic medical records, where the drug history consisted of the date of prescription, the name of the drug, the dose, and the duration of a prescribed drug. Each GDMT drug was recorded in terms of the equivalent daily dose of a particular drug in each class, including valsartan for ACEIs and ARBs, carvedilol for β-blockers, spironolactone for MRAs, and oral furosemide for loop diuretics. The equivalent daily dose of valsartan was calculated using the ratio of the target dose of each ACEI or ARB [5]. In contrast, the equivalent daily doses of carvedilol, spironolactone, and oral furosemide were calculated based on the conversion tables of previous studies [17, 18].

Statistical analysis was performed using R 4.1.3 software (R Foundation for Statistical Computing, Vienna, Austria). In statistical testing, a two-sided p ≤ 0.05 was considered statistically significant. The distributional properties of continuous variables were expressed by mean ± standard deviation, categorical variables were presented by frequency and percentage (%), and survival curves of survival outcomes were estimated using the Kaplan-Meier method. In univariate analysis, the unadjusted effect of each potential risk factor, prognostic factor, or predictor of all-cause death or heart transplant was examined using the Wilcoxon rank sum test, χ² test, and log-rank test, as appropriate for the data type. Next, multivariate analysis was conducted by fitting Cox’s proportional hazards models with time-dependent covariates (the “Cox’s model”) to estimate the adjusted effects of risk factors, prognostic factors, or predictors on the hazard rate of all-cause death or heart transplant. As depicted in Figure 2 and explained in online supplementary Methods (for all online suppl. material, see https://doi.org/10.1159/000534095), we defined 158 time-dependent covariates for the following four categories of variables and considered all of them to fit Cox’s models.

• Drug histories: sacubitril/valsartan, β-blockers, ACEIs, ARBs, isosorbide mononitrate, vasodilators, MRAs, SGLT2i, digoxin, amiodarone, ivabradine, loop diuretics, thiazides, calcium channel blockers, antiplatelets, anticoagulants, erythropoietin stimulating agents, iron, vitamin B, and blood transfusions (online suppl. Table 1).

• Heart-related variables: left ventricular end-diastolic diameter (LVEDD), left ventricular end-systolic diameter, LVEF calculated by the Teicholz method (LVEF_Teich) and the Simpson’s method of disks, respectively, and the left ventricular mass index (LVMI).

• Kidney-related variables: BUN, creatinine, eGFR, BUN-to-creatinine ratio, acute kidney injury (AKI), chronic kidney disease (CKD), end-stage renal disease (ESRD), and hemoglobin level (Hb).

• Hospitalizations: All-cause and cardiovascular-related hospitalizations.

In particular, given the potentially important and unique role of BUN levels in the prognoses of fresh sacubitril/valsartan users, we defined the maximum increase in BUN levels within a prespecified period (e.g., 1 month, 3 months, 6 months, 9 months, and 12 months) as the Worsening Renal Perfusion Index in patients with HFrEF treated with sacubitril/valsartan (abbreviated as the WRP_SV Index), as explained in Figure 2a–c. Technically, we applied the following steps to derive, compute, and store these 158 time-dependent covariates (online suppl. Table S2) in our dataset for regression analysis. First, as shown in Figure 2 and online supplementary Methods, we reconstructed the original wide-form survival data from 291 patients (indexed by n) into a long-form data structure with 47 events of all-cause death or heart transplant, using the so-called counting process style of input for the studied event. Next, as demonstrated in the online supplementary long-form data, at each ordered event time of all-cause death or heart transplant, we computed the values of these 158 time-dependent covariates for each patient at risk at that event time to produce the long-form dataset of 11,423 observations (indexed by m). Finally, we fitted Cox’s models to this long-form survival data of 11,423 observations from the 291 recruited patients with all relevant time-fixed and time-dependent covariates using the coxph() function (e.g., Surv(start, stop, event)) of the survival package in R [19].

To ensure a good quality of analysis, the model-fitting techniques for (1) variable selection, (2) goodness-of-fit assessment, and (3) regression diagnostics and remedies were applied in our regression analysis. They were described in detail in online supplementary Methods [20‒22].

Among the 291 recruited patients, 47 events of all-cause death or heart transplant occurred – 11 received heart transplants and 40 died, of which 4 had received heart transplants before death. As shown in Figure 3a, the estimated 90th percentile survival time for these 291 patients was 16.8 months. In Figure 3b, the survival rates in the 100 patients with CKD at baseline were significantly worse than those of the 191 patients without CKD at baseline (log-rank test p = 0.0472). Among the 100 patients with CKD at baseline, 10 had ESRD that required routine dialysis.

In Table 1 and online supplementary Table S3, the distributions of most variables did not differ significantly between the heart failure patients who survived without a heart transplant and those who died or received a heart transplant, except for body mass index (kg/m²) (25.16 vs. 23.54, p = 0.0316), New York Heart Association (NYHA) functional classification (p = 0.0147), serum N-terminal pro-b-type natriuretic peptide concentration (pg/mL) (1,912 vs. 4,866, p < 0.0001), valvular heart disease (48.4% vs. 68.1%, p = 0.0204), dialysis (1.6% vs. 12.8%, p = 0.0006), BUN (mg/dL) (24.34 vs. 33.40, p = 0.0029), creatinine (mg/dL) (1.34 vs. 2.15, p = 0.0155), eGFR (mL/min/1.73 m²) (71.35 vs. 57.76, p = 0.0061), LVEF_Teich (%) (30.22 vs. 27.29, p = 0.0028), LVEF calculated by the Simpson’s method of disks (%) (29.92 vs. 26.99, p = 0.0042), left atrial diameter (cm) (4.49 vs. 4.77, p = 0.0384), left ventricular posterior wall thickness at end-diastole (cm) (1.08 vs. 1.02, p = 0.0173), moderate or severe mitral regurgitation (43.4% vs. 68.1%, p = 0.0033), and prescription of vitamin B₁₂ or folic acid (1.2% vs. 8.5%, p = 0.0146).

Next, in Table 2, the best final Cox’s model with time-dependent covariates was fitted to the 291 fresh users of sacubitril/valsartan to identify the risk factors, prognostic factors, and predictors of all-cause death or heart transplant with the adjusted generalized R² = 0.7359 > 0.15 and concordance = 0.9569 (SE = 0.0153) > 0.7, indicating an excellent fit. The following 14 covariates significantly increased or decreased the risk of all-cause death or heart transplant: ESRD (estimated hazard ratio [HR] = 24.73, p < 0.0001), oral loop diuretics equivalent daily dose at baseline >28.56 mg/day (HR = 7.19, p < 0.0001), nonsteroid anti-inflammatory drug (NSAID) use at baseline (HR = 255.16, p < 0.0001), coronary artery disease (CAD) without any prescription of antiplatelet at baseline (HR = 0.09, p = 0.0001), 11.75 g/dL < Hb level at baseline ≤14.87 g/dL (HR = 5.35, p = 0.0001), Hb level at event time t (g/dL) (HR = 0.6284, p < 0.0001), 45.03 mm < LVEDD at event time t ≤ 66.22 mm (HR = 8.22, p < 0.0001), average LVEF_Teich since the initiation of sacubitril/valsartan at event time t ≤ 29.95 (%) (HR = 4.03, p = 0.0006), 167.09 g/m² < average LVMI since the initiation of sacubitril/valsartan at event time t ≤ 297.41 g/m² (HR = 6.24, p = 0.0001), maximum percentage decrease in eGFR (mL/min/1.73 m²) within the past 3 months at event time t (HR = 0.96, p < 0.0001), maximum increase in BUN level (mg/dL) within the past 1 month at event time t (HR = 1.06, p = 0.0007), maximum increase in BUN level (mg/dL) within the past 9 months at event time t (HR = 1.03, p = 0.0002), sacubitril/valsartan daily dose at event time t – 30 days (mg) (HR = 0.99, p < 0.0001), and nicorandil accumulative dose >387.78 mg within the past 3 months at event time t – 30 days (HR = 55.13, p < 0.0001). The baseline was specified as within 1 year before the index day (esp., for medications) or on the index day, where the index day for each patient was defined as the date of the first prescription of sacubitril/valsartan. The variables labeled “at event time t” at the end were time-dependent covariates, where t indicated an ordered event time, as depicted in Figure 2 and defined in online supplementary Methods. Then, “at event time t – 30 days” literally meant that the latency period of a covariate’s effect on the occurrence of the studied event was at least 30 days. Lastly, the cut-off values of these continuous covariates were estimated using the p-spline smoothed effect plots of Cox’s model, as shown in online supplementary Methods and online supplementary Figures 1–8.

In particular, after adjusting for the effects of the baseline demographic and clinical characteristics (i.e., ESRD, oral loop diuretics, NSAID use, CAD without antiplatelet use, and Hb level) and the derived time-dependent covariates (i.e., the Hb level, LVEDD, LVEF_Teich, LVMI, nicorandil, and the maximum percentage decrease in eGFR), a 1 mg increase in the daily dose of sacubitril/valsartan at event time t – 30 days would significantly reduce the hazard rate of all-cause death or heart transplant to 99.15% (95% confidence interval [CI] = 98.77–99.54%, p < 0.0001) and a 1 mg/dL maximum increase of the BUN level within the past 1 month and the past 9 months at event time t would significantly increase the hazard rate of all-cause death or heart transplant to 105.75% (95% CI = 102.37–109.25%, p = 0.0007) and 102.72% (95% CI = 101.28–104.18%, p = 0.0002), respectively. As mentioned before, given the clinical importance and unique role of the maximum increase in BUN levels in the prognoses of fresh sacubitril/valsartan users, we defined the maximum increase in BUN levels within a prespecified period (e.g., 1 month, 3 months, 6 months, 9 months, and 12 months) as the Worsening Renal Perfusion Index in patients with HFrEF treated with sacubitril/valsartan (abbreviated as the WRP_SV Index), as explained in Figure 2c.

Finally, as demonstrated in online supplementary Figure 9, an R Shiny application was developed to assess the risk of all-cause death or heart transplant in patients with HFrEF who take sacubitril/valsartan according to the Cox’s model in Table 2. It is available for free access on the website of https://wantsenghsu.shinyapps.io/HF_SV_WRPindex_Model/.

We corroborated the efficacy of sacubitril/valsartan in these patients with HFrEF and believed that continued therapy with sacubitril/valsartan could reduce the risk of death or heart transplant in such patients. Specifically, after adjusting for the effects of the other covariates, sacubitril/valsartan directly affected the hazard rate of all-cause death or heart transplant with a time lag of 30 days (Table 2). On the other hand, sacubitril/valsartan could have positive or negative effects on the other time-dependent covariates in the Cox’s model, including LVEDD at event time t, average LVEF_Teich at event time t, average LVMI at event time t, the maximum percentage decrease in eGFR within the past 3 months at event time t, and maximum increases in BUN level within the past 1 month and 9 months at event time t. Thus, sacubitril/valsartan, in turn, had various indirect effects through them on the hazard rate of all-cause death or heart transplant in these patients (Table 2). In summary, the “sum” of these direct and indirect effects of sacubitril/valsartan would be its total impact on the hazard rate of all-cause death or heart transplant, indicating that sacubitril/valsartan had both direct and indirect pathways to affect the risk of all-cause death or heart transplant in heart failure patients receiving sacubitril/valsartan.

The novelty of this study is that we demonstrated that even after adjusting for the effect of a decrease in eGFR within the past 3 months at event time t, the prognostic effects of the maximum increase in BUN level within the past 1 month (acute effect) or 9 months (chronic effect) at event time t remained statistically significant. This finding revealed the additional role played by BUN in patients with HFrEF who received sacubitril/valsartan. BUN concentration is determined by the net balance of urea generation and excretion by the kidneys, of which the latter depends on the degree of urea reabsorption. Reabsorption of urea occurs with sodium and water in the proximal tubule through a passive process and is also related to water reabsorption in the distal tubules, which is regulated by an antidiuretic hormone [23]. The interplay between sodium and water reabsorption, antidiuretic hormone regulation, and urine flow rate significantly influences urea reabsorption in the kidneys [23]. Consequently, elevated BUN indicates a state of renal hypoperfusion caused by factors such as hypovolemia or reduced cardiac output. Under such conditions, BUN levels can increase independently of changes in eGFR due to increased urea reabsorption activated by sympathetic nervous and RAAS [24], which are well known to be associated with cardiovascular risks. Besides, exposure to diuretics [25] or NSAIDs [26] was reported to be associated with elevations in BUN. After adjusting for the effects of NSAID use and the equivalent daily dose of oral loop diuretics at baseline in the fitted Cox’s model (Table 2), BUN elevations are independently associated with poor outcomes in our patients with HFrEF. This finding suggests that a clinically comprehensive evaluation is necessary to examine the underlying cause of BUN elevation in each patient who receives sacubitril/valsartan. Accordingly, we defined the WRP_SV Index as the maximum increase in BUN levels within a prespecified time period in heart failure patients treated with sacubitril/valsartan in Figure 2c, which highlighted the importance of the pathophysiological conditions that result in increases in BUN levels over short and long periods in heart failure patients treated with sacubitril/valsartan.

Specifically, Figure 4 illustrates in a diagrammatic way the importance of tissue perfusion as the ultimate confluence of all pathophysiological and therapeutic factors in patients with HFrEF who received sacubitril/valsartan. Among the organs, the kidney is considered one of the most sensitive to inadequate cardiac output. The stroke volume of the left ventricle usually falls first in patients with HFrEF, leading to a diminishment in renal blood flow. Reduced effective forward flow, caused by mitral or aortic regurgitation, further compromises renal perfusion. The worsening of renal perfusion in patients with HFrEF could be further exacerbated by the aggressive use of diuretics or the administration of NSAIDs [27]. Besides, anemia affects oxygen delivery to peripheral tissues, predisposes renal function decline [28], and jeopardizes survival in patients with HFrEF [29]. RAAS blockade can also reduce Hb levels [30], so monitoring Hb levels is essential to maintain adequate oxygenation of the perfusate. Collectively, compromise in renal perfusion should be carefully evaluated by Hb levels, a decrease in eGFR, and an increase in BUN level, respectively, to help minimize renal function decline and ensure acceptable tolerance to sacubitril/valsartan in patients with HFrEF. Because BUN elevation was a potential indicator of renal hypoperfusion secondary to decreased cardiac output and increased neurohormonal activation [24, 31], we proposed that maximum increases in BUN levels within the past 1 month and 9 months might reflect the magnitude of renal hemodynamic abnormalities in patients with HFrEF during treatment with sacubitril/valsartan. Mechanistically, sacubitril/valsartan is likely to reduce renal perfusion in patients who are hypotensive or hypovolemic [32]. Clinicians need to explore optimal dosing titration strategies for sacubitril/valsartan in such patients to strike a balance between arterial underfilling and afterload reduction. For example, attention to volume status and reduction in diuretics have been reported to be critical to achieving the target dose of sacubitril/valsartan [33].

However, we discovered in Table 2 that after adjusting for the effects of the other covariates, CAD without antiplatelet prescription at baseline (HR = 0.0873, 95% CI: 0.0260−0.2931, p = 0.0001) and 11.75 g/dL < Hb level at baseline ≤14.87 g/dL (HR = 5.3471, 95% CI: 2.2691−12.6005, p = 0.0001) were associated with a decreased and increased risk of death or heart transplant, respectively. Although these unexpected findings appeared correct, they should be interpreted cautiously. In the case of patients with the comorbidity of CAD who did not receive antiplatelets at baseline, it is always possible that these individuals with CAD may be asymptomatic or have mild symptoms. As a result, their overall prognosis may be more favorable compared to patients with symptomatic or more advanced CAD. These unfamiliar results should not be deliberately interpreted out of context to mistakenly prevent patients with CAD from using antiplatelets if there are no contraindications. In the case of patients with relatively normal Hb levels at baseline, the effect of Hb levels was the change score of Hb levels between the event time t and the baseline. Since the higher the Hb level at event time t, the lower the risk of death or heart transplant, patients with relatively normal Hb levels at baseline would benefit less compared to those with abnormal Hb levels at baseline.

Finally, we acknowledged five limitations of this pharmacoepidemiological study. First, the sample size of 291 patients from a single medical center was relatively small, and these patients were homogeneous in terms of race, accessibility to healthcare, insurance coverage, etc. Future studies with larger sample sizes and more diverse patients with HFrEF are needed to verify and expand our results. Second, patients without baseline LVEF or renal function data were inevitably excluded from this study (Fig. 1). However, such an exclusion due to missing data at baseline could reduce the generalizability of this study because missing data could reflect the clinical judgments of caring physicians. Third, the lack of data on indications or reasons for sacubitril/valsartan dose titration during the treatment course made it difficult to remove or reduce the risk of confounding by indication, which could increase or decrease the estimated direct effect of sacubitril/valsartan on all-cause death or heart transplant (Table 2). Therefore, we performed a preliminary analysis of the drug histories of 291 subjects during the follow-up time. Specifically, individual plots of GDMTs, including sacubitril/valsartan, β-blockers, MRA, SGLT2i, ivabradine, digoxin, and loop diuretics, were drawn for all 291 patients – that is, a plot of the prescribed dose of a selected drug over time for each subject (online suppl. Fig. 10–16). Fourth, we proposed the WRP_SV Index to capture the prognostic impact of renal hypoperfusion in patients with HFrEF, but the patterns of renal function recovery among the heart failure patients with a sacubitril/valsartan-related decrease in kidney function were unclear. Fifth, as shown in Table 1, 60.8% of our patients (177/291) already received an ACEI/ARB before initiating sacubitril/valsartan. This fact may suggest a potential association between changes in BUN and the neprilysin inhibitor component of sacubitril/valsartan. Since no one has seriously investigated the possibility of an elevation in BUN and its association with adverse outcomes in patients with HFrEF who receive ACEI/ARB, it remains unknown whether the increases in BUN are only related to the neprilysin inhibitor component of sacubitril/valsartan. Additional research is needed to falsify our findings.

Adequate maintenance of renal perfusion and function is crucial for the continuous use of sacubitril/valsartan in patients with HFrEF. We defined the maximum increase in BUN levels within a specified period as the Worsening Renal Perfusion Index (WRP_SV Index) to capture the prognostic effect of renal hypoperfusion in patients with HFrEF. Our finding suggests that a clinically comprehensive evaluation is necessary to examine the underlying cause of acute and chronic BUN elevation in each patient with HFrEF during utilization of sacubitril/valsartan.

Dr. Fu-Chang Hu and Ms. Fang-Yu Wen at the international-Harvard (I-H) Statistical Consulting Company helped conduct the statistical analysis, interpret the results, draw Figure 2a, prepare online supplementary Materials, and statistically edit the manuscript. Wallace Academic Editing edited the English of this manuscript. Dr. Shih-Rong Wang at Min-Sheng General Hospital assisted in acquiring funding.

The Research Ethics Committee of the National Taiwan University Hospital (202001049RINB) approved this study. Because this study analyzed nonidentifiable data, the Research Ethics Committee of the National Taiwan University Hospital confirmed that no consent to participate was required and no consent for publication was required.

The authors had no conflicts of interest to disclose.

This work was supported by research grants from the National Science and Technology Council of Taiwan (111-2320-B-002-042-MY2) and Min-Sheng General Hospital (109F005-111-M).

Wan-Tseng Hsu, Chii-Ming Lee, and Tao-Min Huang contributed to the study’s conception. Tsun-Yu Yang, Chao-Kai Chang, Yi-Hsuan Lin, and Yu-Yang Cheng performed data collection and analysis. Yu-Yang Cheng developed the R Shiny application. Wan-Tseng Hsu acquired funding. Wan-Tseng Hsu wrote the manuscript. All the authors approved the final manuscript.

The data analyzed in this study were obtained from the Integrated Medical Database and electronic medical records of the National Taiwan University Hospital. The datasets generated and analyzed during the current study are not publicly available because they contain information that could compromise the privacy of research participants but are available from the corresponding author at reasonable request. Statistical analysis was performed using R 4.1.3 software (R Foundation for Statistical Computing, Vienna, Austria). The R code is available upon request.