Introduction

Between 2001 and 2014, there were 57.4 million hospitalizations for acute decompensated heart failure (ADHF) [1]. During hospitalization, patients with “persistent or worsening symptoms of heart failure requiring an escalation in intravenous therapy or initiation of mechanical and ventilatory support” are considered to have worsening heart failure (WHF) [2]. Since its conception, WHF has demonstrated a prognostic role in predicting mortality and rehospitalizations [3, 4], and has become an important clinical endpoint in randomized controlled trials [5,6,7,8,9,10,11]. However, there is a lack of standardization in its definition due to differences in the modality of rescue therapy used, the timing of WHF, and subjective assessment of symptoms present [12,13,14]. Currently, there are limited studies assessing the effects of increasing diuretic dose with or without the presence of WHF on heart failure outcomes. In this study, we introduce diuretic failure (DF), which is defined as an escalation in intravenous loop diuretic therapy after 48 h of stable dosing regardless of the presence of WHF. In doing so, we created a simplified prognostic marker in ADHF and assessed its effects on length of stay, mortality, and rehospitalizations.

Materials and methods

Study population

This was a multicenter retrospective study at a single hospital system in Pennsylvania. We identified patients with an admission and discharge diagnosis of acute decompensated heart failure (ADHF), receiving intravenous diuretics between October 1, 2017 and January 1, 2020. Patients’ baseline characteristics, medical history, medications, and laboratory values were obtained using the electronic medical record. There were 1641 patients who were eligible based on these screening criteria. The first encounter was used for patients with multiple admissions for heart failure. To eliminate confounding disease processes, 102 patients with shock, end-stage renal disease, cirrhosis, nephrotic syndrome, loculated pleural effusions, use of intravenous fluids upon admission, and vasopressor requirement were excluded. Of those 1539 patients, 150 were excluded for receiving less than 24 h of intravenous diuretics. This was to factor out patients who only received intravenous diuretics in the emergency department as well as ensure adequate therapy duration for comparison with the DF group. As a result, there were a total of 1389 patients assessed in this retrospective study.

Definitions

Initial hospital diuretic dose was the average daily intravenous diuretic dose for 48 h before escalation or de-escalation of dosing occurred. Specifically, DF was defined as an escalation of loop diuretic dosing after at least 48 h or the use of hemodialysis. Patients who had de-escalation of dosing followed by escalation were not considered to have DF. Transthoracic echocardiogram findings were reported if completed within 6 months of hospitalization. All echocardiogram findings were categorized into mild, mild to moderate, moderate, moderate to severe, and severe.

Statistical methods

Statistical analyses were conducted in R. Two-tailed p-values were used to assess statistical significance with values < 0.05 considered statistically significant. Mean and standard deviation were used for continuous variables and compared across DF and no DF groups with the Student’s T test. Absolute frequencies were used for categorical variables and compared between groups using the chi-square test. Univariate analysis was conducted on all patient baseline statistics and medical history listed in Table 3 using logistic regression. Independent predictors of DF were identified by first stratifying predictors according to their p-value in the univariate analysis, using only those with p values below 0.5. Stepwise multivariate logistic regressions in both backward and forward directions were then used for feature selection and model generation. Association of DF with a length of stay, 60-day mortality, 60-day rehospitalizations, and composite outcome of 60-day mortality and cardiac rehospitalizations, were assessed using logistic regression with and without adjustment for covariates. Kaplan–Meier estimates were used to show the effects of DF on 60-day mortality and rehospitalizations, and compared using log-rank tests.

Results

Baseline characteristics

Between October 1, 2017 to January 1, 2020 there were 1389 individuals hospitalized for ADHF. Among which 6.4% (89) of the patients developed DF. Baseline characteristics for patients with and without DF are summarized in Table 1. Patients with DF were more likely to have peripheral edema, lower systolic blood pressure, lower initial hospital diuretic dose, higher BUN, and higher baseline creatinine (p < 0.05). They also were less likely to have an ICD or a history of hyperlipidemia (p < 0.05). The most common reasons for DF included persistent overload (33%), lack of urine output (33%), and failure of shortness of breath resolution (16%) (Table 2).

Table 1 Baseline characteristics of the study cohort divided into patients with and without diuretic failure
Full size table
Table 2 Clinical characteristics resulting in DF
Full size table

Rate of events in patients with diuretic failure

Cumulative rate of 60-day mortality for patients with and without DF was 19.1 and 7.08% respectively (HR 4.522, 97.5% CI 2.01 to 10.17; p = 0.0002) (Fig. 1). There was also a significant difference in rates of 60-day rehospitalizations; 39.32% of patients with DF were readmitted in comparison to 26.92% of patients without DF (HR 1.639, 97.5% CI 1.08 to 2.5; p = 0.0214) (Fig. 2).

Fig. 1
figure 1

Cumulative rate of mortality in patients with diuretic failure through 60-days

Full size image
Fig. 2
figure 2

Cumulative rate of rehospitalizations in patients with diuretic failure through 60-days

Full size image

Predictors of diuretic failure

Significant univariate predictors of DF included markers of renal function (potassium, BUN, creatinine), markers of right ventricular dysfunction (severity of pulmonary hypertension and lack of right ventricle contractility on echocardiogram), and markers of severity of heart failure (BNP and blood pressure) (p < 0.05) (Table 3). A multivariate model was created using gender, weight, troponin, hypertension, presence of an ICD, initial hospital diuretic dose, use of intravenous bumetanide, the severity of pulmonary hypertension and tricuspid regurgitation, and beta-blocker or mineralocorticoid antagonist use. Peripheral edema was found to be statistically significant between DF and non-DF group in baseline analysis (Fig. 3), but was not included in the multivariate analysis due to the lack of its documentation in the electronic medical records: 442 patients in the cohort did not have adequate documentation of peripheral edema severity. For the variable to fit in the model, there would have been a significant reduction in the sample size. In model performance, the training set had an accuracy of 0.6924, and a c-index of 0.8006. The predictive value of the model using these variables had an accuracy of 0.6667 and a c-index of 0.7012 in the holdout set.

Table 3 Univariate and multivariate predictors of diuretic failure
Full size table
Fig. 3
figure 3

Box plot comparing DF to the severity of peripheral edema across DF groups. Y corresponds to patients with DF and N corresponds to those who do not

Full size image

Effects of diuretic failure on outcomes

Using logistic regression, patients with DF had a longer length of stay (p < 0.0001), increased rate of 60-day mortality (p < 0.0001), increased rate of 60-day rehospitalizations (p = 0.012), and an increase in composite outcome of 60-day mortality and 60-day cardiac rehospitalizations (p = 0.0004) (Table 4). After adjustment for covariates associated with DF, DF remained an independent predictor of a longer hospital stay (OR 26.56, 97.5% CI 11.11 to 63.49; p < 0.0001), 60-day mortality (OR 2.07, 97.5% CI 1.06 to 3.84; p = 0.026), 60-day rehospitalizations (OR 1.63, 97.5% CI 1.03 to 2.57; p = 0.036), and a composite outcome of 60-day mortality and cardiac rehospitalizations (OR 1.78, 97.5% CI 1.09 to 2.84; p = 0.018).

Table 4 Association between diuretic failure and outcomes
Full size table

Discussion

WHF has a prognostic role in predicting mortality and rehospitalizations in numerous post-hoc analyses  [11,12,13]. However, there is a lack of standardization in its definition due to the reliance on subjective assessments of symptoms, a variety of therapies used, and different timing parameters. DF avoids these pitfalls by assessing the dosing of a single therapy, loop diuretics, after a set time of 48 h without requiring an assessment of heart failure symptoms. This marker is simple to use and generalizable to nearly all hospitalized patients with heart failure, as loop diuretics are a standard of care for decongestion [14]. Because of its ease of use, this could be applied in real-time during a hospitalization to identify patients who could benefit from more intensive monitoring and follow-up.

The primary finding of this study is that DF has a strong association with adverse heart failure outcomes including length of stay, 60-day mortality, 60-day rehospitalizations, and a composite outcome of 60-day mortality and cardiac rehospitalizations (p < 0.05). These findings are consistent with the broader literature on WHF [3, 4] and lack of diuretic response [15, 16] as a whole.

Notable univariate predictors of DF were markers of kidney dysfunction and right ventricular dysfunction: increasing BUN, creatinine, baseline potassium, pulmonary hypertension and tricuspid regurgitation, the severity of peripheral edema, decreasing eGFR, right ventricular contractility, and initial hospital diuretic dose. Patients with markers of right ventricular dysfunction are more likely to have venous congestion, renal dysfunction, and cardiorenal syndrome [17, 18]. As a result, they may require larger doses of diuretics due to renal dysfunction, increased neurohormonal response, and nephron remodeling [19]. Interestingly, the use of beta-blockers was an important component of the univariate and multivariate model, and was associated with an increased risk of DF. The negative inotropic effects from beta-blockers could potentiate the poor response to diuretics and cause DF [19].

We examined potential confounders of patient groups with disease severity, as indicated by a set of clinical variables. There was a statistically significant difference in the presence of peripheral edema of patients with and without DF (p < 0.001). 62% of patients without DF had 1 + peripheral edema or less compared to only 33% in the DF group. The driving force behind the significant difference is primarily the absence of peripheral edema in patients without DF. Jugular venous pulse, one of the most sensitive and specific markers for elevated right and left-sided filling pressures, was not significantly different among the groups (p = 0.254) [20].

A notable limitation to our retrospective study is the limited sample size of patients experiencing DF: 89 of 1389 (6.4%). As mentioned above, the lack of integration of peripheral edema in the multivariate analysis due to the lack of documentation of it in the EHR is another limitation. Furthermore, this was a multicenter retrospective study in Pennsylvania, and its findings may not be generalizable to other populations. Additionally, DF only factored increases in loop diuretic dosing and not adjunct therapy like thiazide diuretics, mineralocorticoid antagonists.

In conclusion, DF is a simple yet potent prognostic marker for adverse heart failure outcomes including length of stay, mortality, rehospitalizations, and a composite outcome of mortality and cardiac rehospitalizations at 60-days. Markers of renal dysfunction as well as right ventricular dysfunction may help identify patients who may be at risk for DF. Given its association with poor short-term outcomes, these patients should be identified and may benefit from earlier and more intensive follow-up.