Diuretic Response in Acute Heart Failure
Diuretic Response in Acute Heart Failure
Baseline characteristics of the entire study population are presented in Supplementary material online, Table S1. Patients excluded from analysis had lower blood pressures and worse NYHA class, renal function, and outcomes (Supplementary material online, Tables S1–2). Baseline characteristics per quintile of diuretic response are presented in Table 1 .
The mean weight change on Day 4 was −2.8 ± 3.0 kg. The median diuretic dose through Day 3 was 240 mg [140–400] and 1702 (97%) patients received furosemide. The median diuretic response was −0.38 (−0.80 to −0.13) kg/40 mg furosemide. Poor responders showed strong differences in baseline characteristics, including more frequent renal impairment, diabetes, and ischaemic heart disease, but less hypertension and atrial fibrillation (all P < 0.05). Trends were similar in the placebo group and the congested subgroups (Supplementary material online, Tables S3–5).
Predictors of diuretic response are presented in Table 2 . Low systolic blood pressure, low serum potassium, high BUN, diabetes, and atherosclerotic disease were associated with poor diuretic response. Rolofylline treatment independently predicted good diuretic response (all P < 0.05). Patients on rolofylline showed a better diuretic response than those on placebo [−0.39 (−0.82−0.14) vs. −0.38 (−0.75−0.133) kg/40 mg furosemide, P = 0.018], despite excellent baseline matching (Supplementary material online, Table S6). This effect was driven by greater weight loss for rolofylline vs. placebo (3.0 ± 2.8 vs. 2.6 ± 2.9 kg, P = 0.019) as diuretic doses through Day 3 were similar [240 [140–380] vs. 240 [140–412] mg, P = n.s.] There were no interactions between any of the predictors, patient characteristics, study treatment or renal function parameters (BUN, eGFR, or serum creatinine).
Across quintiles, patients with a poor diuretic response had worse outcomes on all endpoints (Figure 1 and Table 3 ). Patterns for the placebo group alone and in patients with manifest signs of congestion (with and without inotrope use) were similar (Supplementary material online, Tables S7–9).
(Enlarge Image)
Figure 1.
Distribution of the primary composite endpoint per quintile of diuretic response. P for trend ≤0.001. Medians are presented per quintile, see Table 1 for inter-quartile range per quintile.
In Cox proportional hazards models, worse diuretic response was associated with poor outcome (all P < 0.001), and remained independently associated with a poor outcome even after multivariable adjustment ( Table 4 and Table 5 , all P < 0.05). There were no interactions between diuretic response and renal function (BUN, eGFR, and serum creatinine), study treatment, left ventricular ejection fraction, or other patient characteristics. Similar patterns were seen in the placebo and congested subsets (Supplementary material online, Table S10).
Figure 2 shows the adjusted Cox hazard function for diuretic response for the 180-day mortality endpoint, fitted using a penalized spline function. Unadjusted Kaplan–Meier survival estimates across quintiles showed consistent survival benefit for a better diuretic response (log-rank P < 0.001) (Figure 3).
(Enlarge Image)
Figure 2.
Adjusted hazard ratio for 180-day mortality for diuretic response. Adjusted for model 3 covariates (Table 4). Legend: dark blue: hazard function, fitted using a penalized spline, light blue: 95% CI; grey: density plot.
(Enlarge Image)
Figure 3.
Survival per quintile of diuretic response. Unadjusted Kaplan–Meier survival curves.
Associations between responsiveness on Days 2 and 3 and measures using i.v. doses only were examined; all showed consistent, similar patterns in baseline characteristics and outcomes, with the chosen definition presenting the largest effect size and smallest P-value in models (data not shown). Trends across quintiles of diuretic response were examined separately in patients receiving low vs. high dose furosemide (above and below the median dose of 240 mg on Days 1–3, i.e. an average of 80 mg furosemide per day), which showed improved diuretic response was consistently associated with similar differences in baseline characteristics (including low systolic blood pressure, worse renal function, diabetes, and atherosclerotic disease, all P < 0.05). The incidence of 180-day mortality, 60-day heart failure rehospitalization and 60-day death or cardiovascular or renal rehospitalization was also consistently higher across quintiles in both groups (all P for trend < 0.05). Patients on high vs. low diuretic doses did have worse 180-day and 60-day outcomes (unadjusted log-rank P < 0.001), though these differences did not persist after multivariable correction (covariates form Table 4 – Table 5 ) in survival models (all P = n.s.).
Next, we examined the effect of changes in diuretic response over time. Patients were divided into three groups, based on whether they remained in the same quintile of diuretic response or were reclassified between Day 2 and Day 4. In univariable Cox models, corrected for baseline (Day 2) diuretic response, patients with stable vs. improving diuretic response did not show any statistically significant differences in 180-day mortality or the 60-day endpoints. Patients with worsening diuretic response, however, were more likely to meet all endpoints (all P < 0.05). After multivariable correction, this only held for the 60-day outcomes [corrected for covariates in Table 4 and Table 5 ; 60-day HF rehospitalization: hazard ratio (HR) 1.48, 95% confidence interval (CI) 1.13–1.93, P = 0.004; 60-day death or renal or cardiovascular rehospitalization: HR: 1.49, 95% CI: 1.22–1.81, P < 0.001].
Analyses were performed to evaluate the added value of introducing diuretic response compared with its individual components (weight change and diuretic dose) as covariates in Cox proportional hazards models. In univariable models, diuretic response showed a greater effect size per SD than weight change and diuretic dose alone ( Table 4 – Table 5 ).
In multivariable 180-day mortality models, inclusion of diuretic response vs. its components showed similar performance, with a trend favouring diuretic response; in the full-study population, Harrell's C-index (higher is better) and AIC (lower is better)—measures for model performance and fit—were similar for both models (0.720 and 3409, respectively, for both), while the continuous NRI—a measure for reclassification improvement—slightly favoured diuretic response (0.01, 95% CI: −0.26–0.18). In patients with manifest congestion, diuretic response showed a slightly stronger trend towards an improved performance for Harrell's C-index (0.717 vs. 0.712), AIC (2464 vs. 2468), and continuous NRI (0.08, 95% CI: −0.16–0.31). Similar patterns for diuretic response vs. the components were observed for 60-day death or renal or cardiovascular rehospitalization in the full population (Harrell's C-index 0.650 vs. 0.647, AIC 6425 vs. 6432, continuous NRI 0.16, 95% CI: −0.06–0.28) and the congested subgroup (Harrell's C-index 0.651 vs. 0.646, AIC 4643 vs. 4650 continuous NRI 0.23, 95% CI: −0.11–0.37).
Diuretic response showed a better performance than its components in 60-day heart failure rehospitalization models. In the full population, the diuretic response model outperformed diuretic dose and weight change individually (C-index 0.692 vs. 0.686; AIC 3537 vs. 3550; continuous NRI 0.29, 95% CI: 0.04–0.47). These effects were also present in patients with manifest congestion (C-index 0.681 vs. 0.672; AIC 2538 vs. 2554; continuous NRI 0.35, 95% CI: 0.01–0.47).
Results
Baseline Characteristics and Identifiers of Diuretic Response
Baseline characteristics of the entire study population are presented in Supplementary material online, Table S1. Patients excluded from analysis had lower blood pressures and worse NYHA class, renal function, and outcomes (Supplementary material online, Tables S1–2). Baseline characteristics per quintile of diuretic response are presented in Table 1 .
The mean weight change on Day 4 was −2.8 ± 3.0 kg. The median diuretic dose through Day 3 was 240 mg [140–400] and 1702 (97%) patients received furosemide. The median diuretic response was −0.38 (−0.80 to −0.13) kg/40 mg furosemide. Poor responders showed strong differences in baseline characteristics, including more frequent renal impairment, diabetes, and ischaemic heart disease, but less hypertension and atrial fibrillation (all P < 0.05). Trends were similar in the placebo group and the congested subgroups (Supplementary material online, Tables S3–5).
Predictors of diuretic response are presented in Table 2 . Low systolic blood pressure, low serum potassium, high BUN, diabetes, and atherosclerotic disease were associated with poor diuretic response. Rolofylline treatment independently predicted good diuretic response (all P < 0.05). Patients on rolofylline showed a better diuretic response than those on placebo [−0.39 (−0.82−0.14) vs. −0.38 (−0.75−0.133) kg/40 mg furosemide, P = 0.018], despite excellent baseline matching (Supplementary material online, Table S6). This effect was driven by greater weight loss for rolofylline vs. placebo (3.0 ± 2.8 vs. 2.6 ± 2.9 kg, P = 0.019) as diuretic doses through Day 3 were similar [240 [140–380] vs. 240 [140–412] mg, P = n.s.] There were no interactions between any of the predictors, patient characteristics, study treatment or renal function parameters (BUN, eGFR, or serum creatinine).
Clinical, Mortality, and Rehospitalization Outcomes
Across quintiles, patients with a poor diuretic response had worse outcomes on all endpoints (Figure 1 and Table 3 ). Patterns for the placebo group alone and in patients with manifest signs of congestion (with and without inotrope use) were similar (Supplementary material online, Tables S7–9).
(Enlarge Image)
Figure 1.
Distribution of the primary composite endpoint per quintile of diuretic response. P for trend ≤0.001. Medians are presented per quintile, see Table 1 for inter-quartile range per quintile.
In Cox proportional hazards models, worse diuretic response was associated with poor outcome (all P < 0.001), and remained independently associated with a poor outcome even after multivariable adjustment ( Table 4 and Table 5 , all P < 0.05). There were no interactions between diuretic response and renal function (BUN, eGFR, and serum creatinine), study treatment, left ventricular ejection fraction, or other patient characteristics. Similar patterns were seen in the placebo and congested subsets (Supplementary material online, Table S10).
Figure 2 shows the adjusted Cox hazard function for diuretic response for the 180-day mortality endpoint, fitted using a penalized spline function. Unadjusted Kaplan–Meier survival estimates across quintiles showed consistent survival benefit for a better diuretic response (log-rank P < 0.001) (Figure 3).
(Enlarge Image)
Figure 2.
Adjusted hazard ratio for 180-day mortality for diuretic response. Adjusted for model 3 covariates (Table 4). Legend: dark blue: hazard function, fitted using a penalized spline, light blue: 95% CI; grey: density plot.
(Enlarge Image)
Figure 3.
Survival per quintile of diuretic response. Unadjusted Kaplan–Meier survival curves.
Sensitivity Analyses
Associations between responsiveness on Days 2 and 3 and measures using i.v. doses only were examined; all showed consistent, similar patterns in baseline characteristics and outcomes, with the chosen definition presenting the largest effect size and smallest P-value in models (data not shown). Trends across quintiles of diuretic response were examined separately in patients receiving low vs. high dose furosemide (above and below the median dose of 240 mg on Days 1–3, i.e. an average of 80 mg furosemide per day), which showed improved diuretic response was consistently associated with similar differences in baseline characteristics (including low systolic blood pressure, worse renal function, diabetes, and atherosclerotic disease, all P < 0.05). The incidence of 180-day mortality, 60-day heart failure rehospitalization and 60-day death or cardiovascular or renal rehospitalization was also consistently higher across quintiles in both groups (all P for trend < 0.05). Patients on high vs. low diuretic doses did have worse 180-day and 60-day outcomes (unadjusted log-rank P < 0.001), though these differences did not persist after multivariable correction (covariates form Table 4 – Table 5 ) in survival models (all P = n.s.).
Next, we examined the effect of changes in diuretic response over time. Patients were divided into three groups, based on whether they remained in the same quintile of diuretic response or were reclassified between Day 2 and Day 4. In univariable Cox models, corrected for baseline (Day 2) diuretic response, patients with stable vs. improving diuretic response did not show any statistically significant differences in 180-day mortality or the 60-day endpoints. Patients with worsening diuretic response, however, were more likely to meet all endpoints (all P < 0.05). After multivariable correction, this only held for the 60-day outcomes [corrected for covariates in Table 4 and Table 5 ; 60-day HF rehospitalization: hazard ratio (HR) 1.48, 95% confidence interval (CI) 1.13–1.93, P = 0.004; 60-day death or renal or cardiovascular rehospitalization: HR: 1.49, 95% CI: 1.22–1.81, P < 0.001].
Diuretic Response vs. Weight Change and Diuretic Dose
Analyses were performed to evaluate the added value of introducing diuretic response compared with its individual components (weight change and diuretic dose) as covariates in Cox proportional hazards models. In univariable models, diuretic response showed a greater effect size per SD than weight change and diuretic dose alone ( Table 4 – Table 5 ).
In multivariable 180-day mortality models, inclusion of diuretic response vs. its components showed similar performance, with a trend favouring diuretic response; in the full-study population, Harrell's C-index (higher is better) and AIC (lower is better)—measures for model performance and fit—were similar for both models (0.720 and 3409, respectively, for both), while the continuous NRI—a measure for reclassification improvement—slightly favoured diuretic response (0.01, 95% CI: −0.26–0.18). In patients with manifest congestion, diuretic response showed a slightly stronger trend towards an improved performance for Harrell's C-index (0.717 vs. 0.712), AIC (2464 vs. 2468), and continuous NRI (0.08, 95% CI: −0.16–0.31). Similar patterns for diuretic response vs. the components were observed for 60-day death or renal or cardiovascular rehospitalization in the full population (Harrell's C-index 0.650 vs. 0.647, AIC 6425 vs. 6432, continuous NRI 0.16, 95% CI: −0.06–0.28) and the congested subgroup (Harrell's C-index 0.651 vs. 0.646, AIC 4643 vs. 4650 continuous NRI 0.23, 95% CI: −0.11–0.37).
Diuretic response showed a better performance than its components in 60-day heart failure rehospitalization models. In the full population, the diuretic response model outperformed diuretic dose and weight change individually (C-index 0.692 vs. 0.686; AIC 3537 vs. 3550; continuous NRI 0.29, 95% CI: 0.04–0.47). These effects were also present in patients with manifest congestion (C-index 0.681 vs. 0.672; AIC 2538 vs. 2554; continuous NRI 0.35, 95% CI: 0.01–0.47).
