OUP user menu

Digoxin and reduction in mortality and hospitalization in heart failure: a comprehensive post hoc analysis of the DIG trial

Ali Ahmed, Michael W. Rich, Thomas E. Love, Donald M. Lloyd-Jones, Inmaculada B. Aban, Wilson S. Colucci, Kirkwood F. Adams, Mihai Gheorghiade
DOI: http://dx.doi.org/10.1093/eurheartj/ehi687 178-186 First published online: 8 December 2005


Aims To determine the effects of digoxin on all-cause mortality and heart failure (HF) hospitalizations, regardless of ejection fraction, accounting for serum digoxin concentration (SDC).

Methods and results This comprehensive post-hoc analysis of the randomized controlled Digitalis Investigation Group trial (n=7788) focuses on 5548 patients: 1687 with SDC, drawn randomly at 1 month, and 3861 placebo patients, alive at 1 month. Overall, 33% died and 31% had HF hospitalizations during a 40-month median follow-up. Compared with placebo, SDC 0.5–0.9 ng/mL was associated with lower mortality [29 vs. 33% placebo; adjusted hazard ratio (AHR), 0.77; 95% confidence interval (CI), 0.67–0.89], all-cause hospitalizations (64 vs. 67% placebo; AHR, 0.85; 95% CI, 0.78–0.92) and HF hospitalizations (23 vs. 33% placebo; AHR, 0.62; 95% CI, 0.54–0.72). SDC≥1.0 ng/mL was associated with lower HF hospitalizations (29 vs. 33% placebo; AHR, 0.68; 95% CI, 0.59–0.79), without any effect on mortality. SDC 0.5–0.9 reduced mortality in a wide spectrum of HF patients and had no interaction with ejection fraction >45% (P=0.834) or sex (P=0.917).

Conclusions Digoxin at SDC 0.5–0.9 ng/mL reduces mortality and hospitalizations in all HF patients, including those with preserved systolic function. At higher SDC, digoxin reduces HF hospitalization but has no effect on mortality or all-cause hospitalizations.

  • Digoxin
  • Heart failure
  • Preserved systolic function
  • Diastolic heart failure
  • Mortality
  • Hospitalization

See page 127 for the editorial comment on this article (doi:10.1093/eurheartj/ehi686)

Digoxin is the oldest and probably the least expensive drug for heart failure (HF).1 In the randomized controlled Digitalis Investigation Group (DIG) trial, digoxin reduced hospitalizations due to worsening HF without affecting overall mortality in HF patients with ejection fraction (EF)≤45%.2 For HF patients with EF>45%, digoxin did not affect all-cause mortality or HF hospitalizations.2 Patients with diastolic HF (clinical HF with normal or near normal EF) or HF patients with preserved systolic function (PSF) constitute over half of all HF patients.3 Yet, little is known about the effect of digoxin in these patients, for whom the benefits of angiotensin-converting enzyme (ACE)-inhibitors and beta-blockers are unproven.4 As most diastolic HF patients are elderly and women, in studying the effect of digoxin, it is important to account for serum digoxin concentration (SDC), which is an important determinant of therapeutic benefit and harmful adverse effects of digoxin.57 However, there are no data on the effect of digoxin at low SDC on outcomes in diastolic HF.

Despite the US Food and Drug Administration (FDA) approval and guideline recommendations,1,4,8 use of digoxin in HF patients is in decline.911 Recently, ACC/AHA HF guidelines downgraded the role of digoxin in HF.12 This shift in practice and policy is likely to adversely affect HF care and outcomes in the developing nations, as local physicians follow guidelines and practice patterns in the USA, Europe, and other developed nations. As most of those patients cannot afford ACE-inhibitors or beta-blockers, digoxin remains the only affordable therapy for HF. Additionally, as most of these patients also cannot afford evaluation of left ventricular function, there is a need for a drug that might be beneficial in all HF patients, regardless of EF.

In this context, we conducted the most comprehensive post hoc re-analysis of the DIG data set to date to determine the effect of digoxin, accounting for SDC, on outcomes in ambulatory men and women with chronic HF and impaired or preserved systolic function. We hypothesized that digoxin at low SDC would reduce all-cause mortality and HF hospitalization in a broad spectrum of ambulatory chronic HF patients with normal sinus rhythm.


Study design

The DIG trial was a randomized controlled trial to evaluate the effects of digoxin on mortality and hospitalization in 7788 ambulatory adults with chronic HF and normal sinus rhythm.2 Patients received four different daily doses of digoxin or matching placebo (0.125, 0.25, 0.375, and 0.50 mg) on the basis of age, sex, weight, and serum creatinine level.13 Most patients were receiving ACE-inhibitors and diuretics. Beta-blockers were not approved for HF at the time of randomization. The design and results of the DIG trial have been previously reported.2


In the DIG trial, 6800 HF patients with EF≤45% (main trial) and 988 patients with EF>45% (ancillary trial) were recruited from the USA (186 centres) and Canada (116 centres) between January 1991 and August 1993. As both trials followed very similar protocols, we combined the main and ancillary trial data sets to study the effect of digoxin in all HF patients regardless of EF. SDC was measured in a random subset of original DIG participants, and investigators were blinded to the results of the SDC. Investigators were also discouraged from checking SDC for clinical reasons except in life-threatening emergencies.13 As SDC is an important determinant of digoxin effects,57 we restricted our analysis to patients with SDC measurements. Of the 3889 (49.9%) patients randomized to digoxin, data on SDC at 1 month after randomization based on specimens collected at least 6 h after the last dose of digoxin were available for 1687 patients. Of these, 982 had low SDC (0.5–0.9 ng/mL) and 705 had high SDC (≥1.0 ng/mL). In the DIG trial, SDC was reported to a single significant figure. We chose these cut points of SDC because of their significant association with outcomes.6,14 Of the 3899 (50.1%) patients randomized to placebo, 3861 were alive at 1 month. Our analysis focuses on these patients (n=5445).


In the DIG trial, the primary outcome was all-cause mortality at amedian follow-up of 39.7 months. Vital status of all patients wascollected up to 31 December 1995 and was 98.9% complete.15 Pre-specified secondary outcomes included mortality due to cardiovascular causes and HF and hospitalizations due to all causes, cardiovascular causes, and worsening HF.

Statistical analysis

We compared baseline characteristics of HF patients receiving placebo and those with low and high SDC using Pearson χ2 tests and one-way analysis of variance tests as appropriate to test for statistical significance. We used Kaplan–Meier analysis and log-rank tests to estimate the effects of low and high SDC on all-cause mortality and HF hospitalizations compared with placebo. We then used bivariate and multivariable Cox proportional hazards regression models to determine the association of low and high SDC with various pre-specified outcomes. In the multivariable model, separate indicators of low and high SDC were entered as independent predictor variables with placebo as the reference category. Covariates used in the multivariable model are displayed in Table 1. Proportional hazards assumptions were checked using log–minus–log scale survival plots for patients receiving placebo and the two SDC groups.

View this table:
Table 1

Baseline characteristics of patients

n (%) or mean (±SD)Placebo (n=3861)SDC 0.5–0.9 (n=982)SDC ≥1.0 (n=705)Overall P-value
Age (years)63.9 (±10.8)62.8 (±10.6)65.1 (±10.1)<0.001
Women (%)955 (24.7)217 (22.1)194 (27.5)0.037
Non-whites (%)563 (14.6)126 (12.8)86 (12.2)0.129
Body mass index (kg/m2)27.4 (±5.35)27.2 (±5.23)26.8 (±5.38)0.016
HF duration (months)29.7 (±36.4)32.2 (±38.7)34.7 (±38.6)0.002
Primary cause of HF
 Ischaemic (%)2654 (68.7)672 (68.4)507 (71.9)
 Hypertensive (%)411 (10.6)98 (10.0)53 (7.5)0.119
 Idiopathic (%)528 (13.7)148 (15.1)106 (15.0)
 Others (%)268 (6.9)64 (6.5)39 (5.5)
Comorbid conditions
 Prior myocardial infarction (%)2445 (63.3)624 (63.5)461 (65.4)0.577
 Current angina (%)1030 (26.7)262 (26.7)195 (27.7)0.860
 Hypertension (%)1826 (47.3)433 (44.1)334 (47.4)0.187
 Diabetes (%)1109 (28.7)265 (27.0)215 (30.5)0.284
 CKD (%)1795 (46.5)368 (37.5)415 (58.9)<0.001
Medications0.24 (±0.07)0.24 (±0.08)0.25 (±0.07)0.249
 Prior digoxin use (%)1674 (43.4)457 (46.5)340 (48.2)0.022
 ACE-inhibitors (%)3617 (93.7)931 (94.8)664 (94.2)0.401
 Nitrates and hydralazine (%)47 (1.2)12 (1.2)8 (1.1)0.982
 Potassium-sparing diuretics (%)318 (8.2)77 (7.8)59 (8.4)0.905
 Other diuretics (%)3017 (78.1)715 (72.8)586 (83.1)<0.001
 Potassium supplement (%)1101 (28.5)243 (24.7)193 (27.4)0.061
Dose of study medication0.243 (±0.07)0.244 (±0.07)0.249 (±0.07)0.121
Symptoms and signs
 Dyspnoea at rest (%)856 (22.2)192 (19.6)162 (23.0)0.150
 Dyspnoea on exertion (%)2926 (75.8)721 (73.4)532 (75.5)0.308
 Elevated JVP (%)513 (13.3)112 (11.4)115 (16.3)0.014
 Third heart sound (%)861 (22.3)243 (24.7)182 (25.8)0.056
 Pulmonary râles (%)632 (16.4)134 (13.6)144 (20.4)0.001
 Leg oedema (%)791 (20.5)183 (18.6)165 (23.4)0.057
NYHA functional class
 I (%)542 (14.0)173 (17.6)83 (11.8)
 II (%)2125 (55.0)536 (54.6)391 (55.5)0.002
 III (%)1125 (29.1)259 (26.4)208 (29.5)
 IV (%)69 (1.8)14 (1.4)23 (3.3)
Heart rate (b.p.m.)78.6 (±12.6)77.2 (±12.7)77.6 (±12.7)0.004
Blood pressure (mmHg)
 Systolic127.5 (±20.4)126.6 (±19.8)126.2 (±20.4)0.191
 Diastolic75.2 (±11.1)75.6 (±10.8)73.9 (±10.9)0.005
Chest radiograph findings
 Pulmonary congestion (%)535 (13.9)114 (11.6)134 (19.0)<0.001
 Cardiothoracic ratio>0.5 (%)2324 (60.2)558 (56.84)441 (62.6)0.048
Laboratory data
 Serum creatinine (mg/dL)1.28 (±0.37)1.21 (±0.32)1.37 (±0.41)<0.001
 Estimated GFR (mL/min/1.73 m2)63.4 (±24.5)67.1 (±19.41)58.0 (±19.1)<0.001
 Serum potassium (mEq/L)4.33 (±0.43)4.33 (±0.46)4.35 (±0.44)0.626
EF31.9 (±12.5)31.9 (±12.1)31.0 (±12.3)0.176

GFR, glomerular filtration rate; JVP, jugular venous pressure.

Because patients were not randomized to low or high SDC, there were significant imbalances in baseline covariates between patients receiving placebo and those with low and high SDC. To account for this selection or predisposition bias, we separately calculated propensity scores or predicted probabilities for the development of low and high SDC for each patient. The propensity score is the conditional probability of receiving an exposure, given a vector of measured covariates,16,17 and is used to adjust for selection bias in causal modelling for observational studies.18

We used non-parsimonious multivariable logistic regression models to estimate propensity scores, using the following baseline characteristics as covariates in the model: age, sex, race, body mass index, duration of HF, aetiology of HF, prior myocardial infarction, current angina, hypertension, diabetes, use of ACE-inhibitors, combined use of nitrates and hydralazine, non-potassium-sparing and potassium-sparing diuretics, potassium supplement, pre-trial use of digoxin, current symptoms of dyspnoea at rest and exertion, activity limitation, New York Heart Association (NYHA) class, heartrate, systolic and diastolic blood pressures, elevated jugularvenous pressure, pulmonary râles, third heart sound, lower extremity oedema, number of symptoms, serum potassium and creatinine, radiological evidence of cardiothoracic ratio >0.5 and pulmonary congestion, and EF. In addition, the following first-order interaction terms were included in the model: age*creatinine, sex*creatinine, race*creatinine, body mass index*creatinine, diabetes*creatinine, non-potassium-sparing diuretic use*creatinine, and pre-trial digoxin use*creatinine.

The models used to estimate propensity scores calibrated and discriminated well. However, because propensity score models are sample-specific adjusters and are not intended to be used for out-of-sample prediction or estimation of coefficients, measures of fitness and discrimination are not important for the assessment of the model's effectiveness.1922 Instead, comparing the balance of baseline covariates between treatment groups before and after adjustment (for instance, matching) is the appropriate way to check the adequacy of the propensity model. The standardized difference is a sensible summary measuring covariate balance between the two treatment groups.23,24 Expressed as a percentage, the standardized difference measures the degree of bias in the means of a covariate, in units of the pooled standard deviation. Our assessment of covariate balance after matching focused on these standardized differences.

Before matching, the mean propensity score for low SDC in the placebo group (n=3861) was 0.19894 and in the low-SDC group (n=982) was 0.21801, which yielded a standardized difference of 34.6% and t-test P-value less than 0.0001. Before matching, the mean propensity score for high SDC was 0.15033 in the placebo group (n=3861) and 0.17694 in the high-SDC group (n=705). This represents a standardized difference of 43.4% and t-test P-value less than 0.0001.

Using an SPSS macro,25 we then matched all 982 patients with low SDC and all 705 patients with high SDC separately with 982 and 705 patients (respectively) receiving placebo who had similar propensity scores (within 0.02). After matching, the mean low-SDC propensity score was 0.21803 in the placebo group (n=982) and 0.21801 in the low-SDC group (n=982), which yields a standardized difference of 0.0% and t-test P=0.994. After matching, the mean high-SDC propensity score was 0.17700 in the placebo group (n=705) and0.17694 in the high-SDC group (n=705), which yields a standardized difference of 0.1% and t-test P=0.986.

Finally, we re-estimated the associations of digoxin and outcomes in the two matched cohorts using Kaplan–Meier survival and bivariate and multivariable Cox regression analyses, with appropriate adjustments to obtain hypothesis testing results incorporating the matched pairs design using stratification.26 In the multivariable Cox regression model, we also adjusted for propensity score (raw score entered as a linear term) and covariates used for the unmatched cohort, at first separately and then combined together, in both cases incorporating the matched design by embedding it into the model.

To assess potential heterogeneity of effect, we estimated the effects of low SDC (vs. placebo) on all-cause mortality in several subgroups of patients: age, gender, race, HF aetiology, NYHA class, EF, diabetes, chronic kidney disease (CKD), defined as glomerular filtration rate <60 mL/1.73 m2 of body surface area, and ACE-inhibitor and diuretic use. We tested for first-order interactions in multivariable Cox proportional hazards models entering interaction terms between low SDC and the above covariates.

To identify baseline predictors of high SDC, we developed a multivariable logistic regression model based on patients receiving digoxin. We used initial bivariate analyses to identify potential predictors. We used χ2 tests and Wilcoxon rank sum tests as appropriate to assess associations of various covariates with high SDC. We then used logistic regression to develop a multivariable model. Age, sex, and race were forced in that model. Covariates significantly associated in the bivariate analysis (P<0.05) were then entered (forward selection: likelihood ratio) into the model. These covariates included digoxin dose, body mass index, serum creatinine, non-potassium-sparing diuretic use, NYHA class, elevated jugular venous pressure, pulmonary râles, lower extremity oedema, and radiological evidence of cardiothoracic ratio >0.5 and pulmonary congestion. The following first-order interaction terms were also entered into the model: age*creatinine, sex*creatinine, race*creatinine, and body mass index*creatinine. Digoxin dose 0.25 and >0.25 mg/day were used as dummy variables, using ≤0.125 mg/day as the reference. This model was first applied in a derivation cohort based on random 50% of patients receiving digoxin (n=844). Then, we used the predictor variables in a validation cohort consisting of the remaining patients (n=843). Both models were fit to data during all steps of the regression analyses and had comparable c-statistic (0.71 and 0.69, respectively, for the derivation and validation models). Finally, the same model was applied to all 1687 patients receiving digoxin (Hosmer and Lemeshow goodness-of-fit test χ2=13.38; P=0.10 and c-statistic 0.70). All statistical tests were evaluated using a two-tailed 95% confidence level. All data analyses were performed using SPSS for Windows version


Patient characteristics

Patients (n=5548; low SDC=982, high SDC=705, and placebo=3861) had a median age of 65 years, 24% were women, 14% were non-white, and 12% had EF>45%. Among the 1687 randomly selected patients with SDC, 260 (15%) were receiving digoxin ≤0.125 mg/day, 1237 (73%) were receiving 0.25 mg/day, and 190 (11%) were receiving >0.25 mg/day, with a median dose of 0.25 mg/day.

Baseline patient characteristics are displayed in Table 1. When compared with patients receiving placebo, those with low SDC were younger, had lower mean serum creatinine, and were less likely to have pulmonary congestion and higher NYHA class symptoms, and also less likely to receive diuretics (Table 1). In contrast, compared with placebo patients, those with high SDC were older, had higher mean serum creatinine, and were more likely to have pulmonary congestion and higher NYHA class symptoms, and also more likely to receive diuretics (Table 1).

Matching on propensity scores adequately balanced all 32 covariates included in the propensity model, as well as the eight included interaction terms. All 982 low-SDC patients were successfully matched to a placebo patient to within 0.02, and 979 pairs matched to within 0.01 point of propensity scores. All 705 high-SDC patients, likewise, were matched to a placebo patient whose propensity score was within 0.02 of the high-SDC patient, with 700 pairs actually matched to within 0.01 point. After matching, none of the 32 covariates or eight interaction terms had large or statistically significant differences between placebo and SDC patients. The standardized difference was ≤8.2% in absolute value for all covariates and interactions in both the 982 low-SDC-matched pairs and the 705 high-SDC-matched pairs, where standardized differences <10% in absolute value are usually taken to indicate adequate balance.23,24

Mortality and hospitalizations

During 40 months of median follow-up, 1854 (33%) patients died from all causes, including 1456 (26%) from cardiovascular causes, and 650 (12%) from worsening HF. Of the 3717 (67%) participants hospitalized for all causes, 2920 (53%) were due to cardiovascular causes and 1712 (31%) due to worsening HF. No patients in the digoxin group died during the first month and only two patients receiving placebo were lost to follow-up during that period.

Digoxin and mortality

Compared with 33% deaths among patients receiving placebo, 29% of those with low SDC and 42% of those with high SDC died from all causes during the follow-up. In the Kaplan–Meier analysis, compared with placebo, low and high SDC were, respectively, associated with unadjusted reduction and increase in all-cause mortality (Figure 1). Absolute risk (AR), absolute risk reduction (ARR), unadjusted and adjusted hazard ratios (HRs), 95% confidence intervals (CIs), and P-values for mortality due to all causes, cardiovascular causes, and worsening HF for patients with low and high SDC compared with placebo are displayed in Table 2. ARR was calculated separately for the two SDC groups relative to placebo patients. Low SDC was associated with a 22% reduction in unadjusted total mortality, which was essentially unchanged after multivariable adjustment for other covariates (Table 2). High SDC was associated with a 23% relative increase in unadjusted mortality, which did not persist after multivariable adjustment (Table 2).

Figure 1 Kaplan–Meier plots for cumulative risk of death due to all causes by SDC.

View this table:
Table 2

Effects of digoxin on mortality

Events/totalAR (%)ARRCrude HR (95% CI)P-valueAdjusteda HR (95% CI)P-value
All-cause mortalityPlacebo1272/386132.9Reference1Reference1Reference
SDC 0.5–0.9 ng/mL288/98229.3 3.6%0.78 (0.69–0.89)<0.00010.77 (0.67–0.89)<0.0001
SDC ≥1.0 ng/mL294/70541.7−8.8%1.23 (1.08–1.39)   0.0021.06 (0.93–1.20)   0.406
Cardiovascular mortalityPlacebo985/386125.5Reference1Reference1Reference
SDC 0.5–0.9 ng/mL237/98224.1 1.4%0.84 (0.73–0.96)   0.0130.83 (0.71–0.97)   0.019
SDC ≥1.0 ng/mL234/70533.2−7.7%1.26 (1.10–1.46)   0.0011.07 (0.93–1.24)   0.339
HF mortalityPlacebo468/386112.1Reference1Reference1Reference
SDC 0.5–0.9 ng/mL 86/982 8.8 3.30.63 (0.50–0.80)<0.00010.63 (0.49–0.82)<0.0001
SDC ≥1.0 ng/mL 96/70513.6−1.51.09 (0.87–1.36)   0.4490.87 (0.70–1.09)   0.236

aAdjusted for age, sex, race, body mass index, duration of HF, aetiology of HF, prior myocardial infarction, current angina, hypertension, diabetes, pre-trial use of digoxin, use of ACE-inhibitors, diuretics, combination of hydralazine and nitrates, current dyspnoea at rest and dyspnoea on exertion, heart rate, systolic and diastolic blood pressures, current jugular venous distension, third heart sound, pulmonary râles, lower extremity oedema, NYHA functional class, pulmonary congestion by chest X-ray, cardiothoracic ratio >0.5, estimated glomerular filtration rate, EF, and the interaction term for SDC 0.5–0.9 ng/mL and race.

Among the subset of 982 propensity score pairs matched for low SDC, in 204 of those pairs, low-SDC patients definitely died before their matched placebo patients. In contrast, in 252 of those pairs, placebo patients definitely died before the matched low-SDC patients. The order of death could not be determined in the remaining 526 pairs. The appropriate matched-samples comparison of hazard rates gave a Z-statistic of −2.248 (two-tailed P=0.025) for the low SDC (Figure 2, upper panel). When compared with matched placebo, low SDC was associated with reduction in all-cause mortality (unadjusted HR 0.81, 95% CI 0.67–0.97, P=0.025), which remained essentially unchanged after additional multivariable adjustment (adjusted HR 0.81, 95% CI 0.67–0.99, P=0.042).

Figure 2 Kaplan–Meier plots for cumulative risk of all-cause death among patients matched by propensity to develop SDCs 0.5–0.9 (upper panel) and ≥1.0 ng/mL (lower panel).

In the cohort of 705 propensity-matched pairs for high SDC, among 217 of those pairs, high-SDC patients definitely died before the matched placebo. In contrast, in 182 of those pairs, placebo patients definitely died before the matched high-SDC patients. In 306 pairs, we could not determine the order of death. The appropriate matched-samples comparison of hazard rates gave a Z-statistic of 1.752 (two-tailed P=0.080) for the high SDC vs. placebo comparison (Figure 2, lower panel). When compared with placebo, high SDC was associated with increased mortality that bordered on significance (unadjusted HR 1.19, 95% CI 0.98–1.45, P=0.080). Even though, this association remained essentially unchanged after multivariable adjustment, it lost its statistical significance (adjusted HR 1.19, 95% CI 0.95–1.48, P=0.130).

Digoxin and hospitalization

When compared with 33% of placebo patients, 23% of those with low SDC and 29% of those with high SDC were hospitalized due to HF. Kaplan–Meier plots for first hospitalization due to worsening HF are displayed in Figure 3. Low SDC was associated with reduction in the risk for all-cause, cardiovascular, and HF hospitalizations (Table 3). High SDC was associated with a reduction in HF hospitalization, but not with all-cause hospitalization (Table 3).

Figure 3 Kaplan–Meier plots for cumulative risk of hospitalization due to worsening HF by SDC.

View this table:
Table 3

Effects of digoxin on hospitalizations

Events/totalAR (%)ARR (%)Crude HR (97.5% CI)P-valueAdjusteda HR (97.5% CI)P-value
All-cause hospitalizationPlacebo2594/386167.2Reference1Reference1Reference
SDC 0.5–0.9 ng/mL 625/98263.6   3.60.82 (0.75–0.89)<0.00010.85 (0.78–0.92)<0.0001
SDC ≥1.0 ng/mL 498/70570.6−3.41.05 (0.95–1.15)   0.3330.95 (0.87–1.05)   0.331
Cardiovascular hospitalizationPlacebo2059/386153.3Reference1Reference1Reference
SDC 0.5–0.9 ng/mL 471/98248.0   4.50.77 (0.70–0.86)<0.00010.79 (0.72–0.88)<0.0001
SDC ≥1.0 ng/mL 390/70555.3−2.01.01 (0.90–1.12)   0.8910.91 (0.82–1.01)   0.086
HF hospitalizationPlacebo1279/386133.1Reference1Reference1Reference
SDC 0.5–0.9 ng/mL 229/98223.3   9.80.60 (0.52–0.69)<0.00010.62 (0.54–0.72)<0.0001
SDC ≥1.0 ng/mL 204/70528.9   4.20.81 (0.70–0.94)   0.0050.68 (0.59–0.79)<0.0001

aAdjusted for the same covariates as in Table 2.

In the propensity-matched cohort, compared with patients receiving placebo, low SDC was associated with 38% unadjusted (HR 0.62, 95% CI 0.51–0.76, P<0.0001) and multivariable-adjusted (HR 0.62, 95% CI 0.50–0.79, P<0.0001) reduction in HF hospitalization. Compared with placebo, matched high-SDC patients also had significant reduction in HF hospitalization: unadjusted HR 0.61 (95% CI 0.49–0.76, P<0.0001) and adjusted HR 0.53 (95% CI 0.40–0.69, P<0.001).

Subgroup analysis

Associations between low SDC and all-cause mortality in various subgroups of patients are displayed in Figure 4. Low SDC was associated with reduced mortality in most subgroups, with the exception of non-whites (adjusted P-value for interaction is equal to 0.006). There was no significant interaction of low SDC with EF>45% (P=0.834) or gender (P=0.917).

Figure 4 HR (95% CI) for all-cause mortality in subgroups of patients with HF with SDC 0.5–0.9 ng/dL and placebo.

Predictors of high SDC

Among patients receiving digoxin (n=1687), 41.8% had high SDC. In the derivation sample (n=844), 42.5% patients had high SDC and in the validation sample (n=843), 41.0% patients had high SDC. Age, gender, dose of digoxin, serum creatinine, use of diuretics, and pulmonary congestion were predictors of high SDC in both derivation and validation cohorts. Median predicted probability for high SDC was 0.40 (range 0.05–0.96). Unadjusted and multivariable covariate-adjusted odds ratios (ORs) and 95% CIs for high SDC based on all 1687 patients are presented in Table 4. When compared with patients receiving a daily digoxin dose of ≤0.125 mg, those receiving 0.25 mg (adjusted OR 3.17, 95% CI 2.25–4.47, P<0.0001) and >0.25 mg (adjusted OR 8.70, 95% CI 5.04–15.00, P<0.0001) had greater odds for high SDC.

View this table:
Table 4

Predictors of SDC ≥1.0 ng/mL among patients receiving digoxin (n=1687)

Unadjusted ORP-valueAdjusteda ORP-value
Age (years)1.02 (1.01–1.03)<0.00011.02 (1.01–1.03)0.002
Female sex1.34 (1.07–1.67)0.0112.21 (1.70–2.88)<0.0001
Non-white race0.94 (0.70–1.27)0.6990.84 (0.61–1.16)0.286
Body mass index0.99 (0.97–1.01)0.1490.95 (0.93–0.98)<0.0001
Serum creatinine3.45 (2.59–4.60)<0.00015.98 (4.21–8.50)<0.0001
Use of diuretic1.84 (1.44–2.34)<0.00011.56 (1.21–2.02)0.001
Pulmonary congestion by chest X-ray1.79 (1.36–2.34)0.0011.66 (1.25–2.22)0.001
Digoxin 0.25 mg/day1.09 (0.88–1.36)0.4313.17 (2.25–4.47)<0.0001
Digoxin >0.25 mg/day1.12 (0.83–1.52)0.4738.70 (5.04–15.00)<0.0001

aAdjusted for the same covariates as in Table 2.


The results of our study demonstrate that digoxin reduces HF hospitalization in a wide spectrum of ambulatory chronic HF patients in normal sinus rhythm receiving diuretics and ACE-inhibitors. In addition, digoxin reduces all-cause mortality and all-cause hospitalization at low SDC. Our data suggest that digoxin can play an important role in HF care, regardless of EF or gender. In the developed nations, where diuretics and ACE-inhibitors are the most commonly used regimen for HF therapy, digoxin can play an important role in reducing morbidity and mortality and in decreasing healthcare costs. In the developing nations, where most HF patients do not have a left ventricular function evaluation and cannot afford ACE-inhibitors and beta-blockers, digoxin is the only available drug that can play a critical role in reducing mortality and morbidity.

The beneficial effects of digoxin observed by us are likely due to haemodynamic and neurohormonal properties of digoxin. Digitalis is the only inotropic agent that does not increase mortality and improves neurohormonal profile in HF.28,29 The effects of digoxin have long been known to be dependent on SDC.1,5,30,31 However, ours is the first study to demonstrate a beneficial effect of low SDC in all HF patients, including those with diastolic HF. As risk of hospitalization is relatively high in HF patients with PSF, digoxin can play an important role in reducing morbidity and healthcare costs. Furthermore, digoxin reduces mortality and all-cause hospitalization if care is taken to achieve an SDC between 0.5 and 0.9 ng/mL.

Digitalis is the oldest and probably the least expensive drug for the management of HF. Yet, despite FDA approval and guideline recommendations, there has been a recent decline in the use of digoxin in HF patients.911,32 The recent ACC/AHA HF guidelines have also downgraded the role of digoxin in HF care, although promoting expensive device-based therapies with selective indications.12 This change in policy and practice regarding digoxin use appears to be due in part to lack of mortality benefit,31,33 in particular, in those with PSF.4 Despite recent evidence to the contrary,14,34 a widely publicized earlier study questioning the safety of digoxin in women35 also likely played a role. Finally, availability of newer neurohormonal antagonists such as beta-blockers and aldosterone antagonists,4 lack of industry sponsorship for digoxin, and aggressive promotion of expensive devices with selective indications10,11 have further slighted the role of digoxin in HF care. However, our data demonstrate that in ambulatory men and women with chronic stable HF and impaired or preserved systolic function, use of digoxin was associated with >30% reduction in HF hospitalization regardless of SDC and significant reduction in mortality and hospitalizations from all causes at low SDC. This is important as HF is one of the major causes of hospitalization, especially among older adults, and HF hospitalizations are expensive and associated with poor patient outcomes.

Our findings support a more expanded role of digoxin in HF. Digoxin should be used in systolic HF patients with or without atrial fibrillation who continue to remain symptomatic despite therapy with ACE-inhibitors or angiotensin receptor blockers (ARBs), and beta-blockers, or for those who cannot afford or tolerate these drugs. First, specifically, digoxin should be used for relief of HF symptoms before considering aggressive therapy with high-dose non-potassium-sparing diuretics, as these drugs may increase mortality and hospitalizations.36,37 Secondly, digoxin should be considered before initiating an aldosterone antagonist as these drugs often cause hyperkalaemia.38 Thirdly, digoxin should be prescribed before adding an ARB to an ACE-inhibitor. In the Val-HeFT and CHARM-added trials, addition of ARBs, which are expensive, did not result in reduction in mortality and was associated with increased adverse effects.32,39 Finally, digoxin should be tried before considering cardiac resynchronization therapy (CRT), which is invasive and costly.10,11,40 In addition, CRT is reserved for systolic HF patients with wide QRS complex, and response to CRT is often unpredictable.

The role of digoxin in the management of HF in the developing nations deserves special consideration. Digoxin, along with diuretics, forms the basis of HF therapy in most of the developing countries. However, clinicians in those countries often follow guidelines and practice patterns in the USA and Europe. As the use of digitalis glycosides is being discouraged in the developed nations,41 clinicians in the developing countries are also likely to decrease the use of digitalis. This could have the untoward consequence of increasing mortality and morbidity in millions of HF patients worldwide. This is important, as most HF patients in the developing countries cannot afford expensive therapy with ACE-inhibitors and beta-blockers. In addition, our data suggest that digoxin is beneficial regardless of EF. Thus, digoxin can play a very special role in HF care in the developing nations, as most HF patients there cannot afford an echocardiographic evaluation of left ventricular function.

Correlates of high SDC identified in our study can be used to reduce the risk of high SDC in situations where SDC cannot be measured. On the basis of our observations, we suggest that for young men with clinically stable HF (no pulmonary congestion by chest X-ray and not receiving diuretics) and normal renal function, a daily digoxin dose of ≤0.25 mg would be therapeutic. In contrast, for HF patients who are elderly, women, have pulmonary congestion or receiving diuretics, or have renal impairment, a daily dose of 0.125 mg would probably be more appropriate. For patients with multiple risk factors for high SDC, such as an elderly woman with CKD or a woman with pulmonary congestion receiving a non-potassium-sparing diuretic, digoxin should be started at a daily dose of 0.0625 mg. When possible, SDC should be monitored to guide therapy for these patients.

To the best of our knowledge, this is the most comprehensive and in-depth analysis of the DIG data set based on pre-defined outcomes. DIG was the largest multi-centre randomized controlled trial of digoxin in patients with HF with the longest follow-up. Because most of the studies in the literature were based on subgroups of DIG patients,14,35,42 did not account for SDC,35 were of shorter duration, or used intermediate outcomes,6,4346 our findings provide the strongest evidence of a significant benefit of digoxin in a wide spectrum of HF patients.

HF patients in the DIG trial were not receiving beta-blockersor aldosterone antagonists. A post hoc analysis of the US Carvedilol Trial's data demonstrated that digoxin use was associated with a reduction in the combined endpoint of death or hospitalization from all causes in patients both receiving (relative risk 0.64, 95% CI 0.52–0.79) and not receiving (relative risk 0.82, 95% CI 0.71–0.99) carvedilol.47 In the RALES trial, the survival benefit of spironolactone, an aldosterone antagonist, was only significant in patients receiving digoxin.48 Therefore, there is no reason to withhold digoxin from systolic HF patients receiving beta-blockers or aldosterone antagonists. As there is currently no evidence of survival benefit from beta-blockers or aldosterone antagonists in HF patients with PSF, digoxin could be considered in preference to those drugs in this population.

Several limitations of our study must be acknowledged. The results of this study are based on post hoc analysis and patients were not randomized to a certain SDC. Despite our efforts to balance baseline covariate distribution by propensity score matching, bias due to unmeasured covariates cannot be ruled out as a possible explanation of our results. Although we can never know whether such a hidden bias exists, careful sensitivity analysis permits us to quantify the size of hidden bias that would be required to invalidate our conclusions. Our sensitivity analysis demonstrates that for an unmeasured binary covariate (unrelated to covariates in our propensity model) to explain away our results, the unmeasured covariate would need to increase the odds of a patient developing low SDC by at least 48% and would also need to be a near-perfect predictor of all-cause mortality, suggesting that these results are at least somewhat resistant to hidden bias.

About 22% of DIG participants randomized to placebo were receiving open labelled digoxin, and ∼20% of those randomized to digoxin were not receiving the drug.2 In our study, all patients in the treatment group were receiving digoxin at 1month follow-up, as indicated by the presence of a detectable serum digoxin level. However, digoxin may have been subsequently discontinued in some patients, and some of the placebo patients may have been started on digoxin during follow-up. However, as our analysis starts at 30 days after randomization, we were not able to estimate an early effect of digoxin.

Results of our study are based on predominantly white, male, and relatively younger patients with mild to moderate HF and normal sinus rhythm. Our finding of a significant interaction between low SDC and race might be due to chance and lacks biological basis. Subgroup analyses without biological basis can be misleading. For example, in the MERIT-HF trial, survival benefit of extended release metoprolol was only observed in HF patients in Europe and not in the USA.49

In summary, our comprehensive post hoc analysis of data from the DIG trial indicates that at low doses, which is likely to achieve low SDC (<0.9 ng/mL), digoxin reduces mortality and hospitalization in a broad spectrum of ambulatory men and women with chronic HF and preserved or impaired systolic function. Digoxin should be used in systolic HF patients who are symptomatic despite ACE-inhibitors or ARB and beta-blocker therapy. In those patients, use of digoxin should be considered before CRT, high-dose diuretics, aldosterone antagonists, or adding an ARB to an ACE-inhibitor.


The DIG study was conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with the Digitalis Investigation Group (DIG) Investigators. This manuscript has been reviewed by NHLBI for scientific content and consistency of data interpretation with previous DIG publications and significant comments have been incorporated prior to submission for publication.

The authors wish to thank Jerome L. Fleg, NHLBI, for his critical review of the manuscript. The authors also wish to thank Raynald Levesque, Aon Consulting, Montreal, Canada for his generous assistance with the SPSS macro and other programs.

A.A. is supported by a National Institute on Aging, National Institutes of Health grant 1-K23-AG19211-01.

A.A. conceived the study hypothesis and design and wrote the paper in collaboration with M.W.R., M.G., and T.E.L. A.A. did the biostatistical analyses with assistance from I.B.A., T.E.L., and D.M.L.-J. All authors analysed and interpreted the data, participated in critical revision of the paper for important intellectual content, and approved the final version of the article. A.A. and I.B.A. had full access to the data.

Conflict of interest: none declared.


  • The authors wish to dedicate this article to the memories of Thomas W. Smith, MD (1936–1997) and Richard Gorlin, MD (1926–1997) who played a crucial role in enhancing our understanding of digoxin in heart failure and in the planning and conduct of the randomized controlled DIG trial.

    An oral abstract based on this manuscript was presented at the 2005 American Heart Association Scientific Sessions in Dallas, Texas on 15 November.


View Abstract