We studied a consecutive cohort of 2604 patients with a primary diagnosis of acute heart failure admitted to the cardiology ward of a tertiary care university hospital between January 2004 and December 2016. Acute heart failure was defined according to clinical practice guidelines.10 Patients with acute decompensated heart failure resulting from new-onset or chronic heart failure were included. Of the 2604 patients in the registry, 1467 were excluded because a) they did not have a diagnosis of atrial fibrillation (n = 1262), b) they did not have a diagnosis of valvular atrial fibrillation according to the recommendations of the European Cardiology Society3 (n = 157), or c) they were being treated with an anticoagulant other than a VKA (n = 48). The final sample thus consisted of 1137 patients.

A wide range of variables, including clinical, physical examination, biochemical, electrocardiographic, and echocardiographic variables, together with details of concomitant treatments, were recorded during the initial hospital stay. Treatment was individualized at the discretion of the attending cardiologist and in accordance with clinical practice guidelines applicable over the study period.

The study was performed in accordance with the principles set forth in the Declaration of Helsinki and was approved by ethics committee of the hospital.

INR was measured during the first medical contact in the emergency department. A value of 2 to 3 was considered to be within the optimal therapeutic range. Patients were classified into 3 categories depending on their INR value: a) a therapeutic category (INR 2-3), b) a subtherapeutic category (INR < 2), and c) a supratherapeutic category (INR > 3).

Survival after hospital discharge was checked by reviewing the patients’ electronic medical records. The researchers responsible for assigning the events did not have access to the patients’ INR data. The primary endpoint was the association between INR and all-cause mortality during follow-up. The secondary endpoint was the association between INR and cause-specific mortality, categorized as a) cardiovascular vs noncardiovascular death) and b) death due to heart failure vs death due a cause other than heart failure. Cardiovascular death was defined using the criteria recommended by the American Heart Association and included sudden death attributable to heart failure, acute myocardial infarction, stroke, vascular bleeding, or peripheral artery disease, and death due to an unknown cause. In all other cases, the cause of death was considered to be noncardiovascular.11

Continuous variables are expressed as means ± standard deviation or, when nonnormally distributed, as medians and interquartile range. Discrete variables are expressed as percentages. The baseline characteristics of patients in the INR categories were compared using analysis of variance. The Kruskal-Wallis test was used for nonparametric variables.

The proportional hazards assumption was tested for all the study variables in the 3 INR categories using scaled Schoenfeld residuals and log-log curves. A different method of analysis was used depending on whether the assumption of proportional hazards was met or not. In the first case, a Royston-Parmar model was used and in the second case, differences in restricted mean survival time (RMST) were calculated.12 RMST is an alternative to the hazard ratio for situations in which the assumption of proportional hazards does not hold. RMST differences represent the years of life lost associated with a given study variable.12,13 The variables for which IRN did not meet the proportional hazards assumption were all-cause mortality, cardiovascular death, and death due to heart failure. The assumption was met for noncardiovascular death and death due to a cause other than heart failure. The maximum follow-up time in the RMST analysis was set at 5 years. Results are expressed as the difference in number of years up to the event.

All the variables in Table 1 were evaluated for their prognostic value. Backward elimination was used to simplify the final multivariate model. The optimal polynomial for the continuous variables was calculated during stepwise selection to ensure linearity with the event. Cumulative incidence curves were generated to analyze cause-specific mortality and differences were evaluated using the Gray test.14 RMST adapted to competing risks was used in the multivariate analysis of cause-specific mortality.

A 2-tailed P value of less than .05 was considered significant for all analyses. The statistical analyses were performed by MedStats Consulting (Reading, Pennsylvania, United States) using STATA 15.1 (StataCorp LP; College Station, Texas, United States).

The mean ± standard deviation age of the patients was 74 ± 10 years; 581 patients (51.1%) were women and 568 (50.2%) had already been hospitalized for heart failure. Most of the patients (n = 927, 81%) had an INR outside the therapeutic range. There were 210 patients (18.5%) in the therapeutic category (INR, 2-3), 660 (58.0%) in the subtherapeutic category (INR < 2), and 267 (23.5%) in the supratherapeutic category (INR > 3).

The patients’ baseline characteristics are shown by INR group in Table 1. INR values outside the therapeutic range, in particular, subtherapeutic values, were associated with various clinical, biochemical, and echocardiographic variables that are typically associated with poor prognosis in heart failure. Notably, there were no differences between the 3 groups for biochemical liver function parameters. Supratherapeutic INR values were more common in patients with preserved left ventricular ejection fraction than with reduced left ventricular ejection fraction (Table 1).

Over a median follow-up period of 2.15 (0.71-4.29) years, 495 patients (43.5%) died. Crude mortality rates were higher in patients with INR values outside the therapeutic range (Table 2). According to the Kaplan-Meier survival analysis, thus, the risk of all-cause mortality was higher in patients an INR value lower than 2 and in particular in those with an INR value above 3 (log rank test, P = .014) (Figure 1).

Figure 1.

Kaplan-Meier all-cause mortality curves by international normalized ratio (INR) category.

(0.11MB).

The multivariate analysis of RMST differences confirmed an independent association between INR on admission to hospital and an increased risk of all-cause mortality: RMSTINR<2 = –0.50 years (95% confidence interval [95%CI], –0.77 to –0.23) (P < .001) and RMSTINR>3 = –0.40 years (95%CI, –0.70 to –0.11) (P = .007) (Table 3). The Harrell C statistic was 0.771. The survival curves adjusted for all-cause mortality are shown by INR category in Figure 2. These curves were estimated from the multivariate RMST model and include the interaction between INR and time. On comparison of patients in the subtherapeutic and therapeutic categories, the strength of the association between subtherapeutic values and risk of all-cause mortality decreased over time and lost its significance in the second year. The association in the supratherapeutic group, by contrast, remained relatively constant over time but was only significant from the third month up to the third year (Figure 3). The modeling of time-dependent hazard ratios is presented in Table 1 of the supplementary material. When INR was established as a continuous variable in the multivariate model, both subtherapeutic and supratherapeutic INR values were found to be associated with an increased risk of all-cause mortality (P < .001) (Figure of the supplementary material).

Figure 2.

Adjusted survival curves for all-cause mortality by international normalized ratio (INR) category. RMST, restricted mean survival time.

(0.11MB).

Figure 3.

Estimated all-cause mortality risk according to time. A, INR < 2 vs. 2–3. B, INR > 3 vs. 2–3. INR, international normalized ratio.

(0.11MB).

Of the 495 deaths recorded, 378 (76.4%) were attributable to cardiovascular causes. The crude rates for the specific causes of death by INR category are shown in Table 2.

The cumulative risk of cardiovascular death was higher in patients with nontherapeutic INR values (Gray test, P = .063) and was particularly high in those with values above 3 (Figure 4). After the multivariate adjustment, however, although both subtherapeutic and supratherapeutic INR values were associated with a higher risk of cardiovascular death, the association was somewhat stronger for subtherapeutic values (differences in RMST at 5 years: RMSTINR<2 = –0.39 years (95%CI, –0.68 to –0.11); P = .007, and RMSTINR>3 = –0.26 years (95%CI, –0.57 to 0.05) years; P = .098) (Table 3).

Figure 4.

Cumulative indicence. A: cardiovascular death. B: noncardiovascular death. C: death due to heart failure. D: death due to a cause other than heart failure. INR, international normalized ratio.

(0.35MB).

The survival curves for the secondary endpoints are shown in Figure 5. The association between the risk of cardiovascular death and subtherapeutic INR values was stronger in the early follow-up period, as is also reflected in the time-dependent hazard ratios (Table 1 of the supplementary material). Compared with INR values in the therapeutic range, supratherapeutic values were associated with a sustained increased risk, but this did not reach statistical significance.

Figure 5.

Adjusted survival curves. A: cardiovascular death. B: noncardiovascular death. C: death due to heart failure. D: death due to a cause other than heart failure. INR, international normalized ratio; RMST, restricted mean survival time.

(0.22MB).

The crude rates for cause-specific mortality were low (Table 2 of the supplementary material). No significant differences were observed for death attributable to ischemic stroke, acute myocardial infarction, hemorrhagic stroke, or sudden death, although there were a higher number of deaths due to hemorrhagic stroke and sudden deaths in patients with INR values above 3. The cause of death was unknown in 84 patients (16%).

In total, 117 deaths (23.6%) did not have a cardiovascular cause. The incidence of noncardiovascular death was higher in the subtherapeutic and supratherapeutic categories (Table 2), but the differences with patients in the therapeutic category were not significant (Gray test, P = .379) (Figure 4).

Because the assumption of proportional hazards was met for this criterion, the prognostic values of the different INR categories are presented as single hazard ratios in Table 4. The results confirm the lack of association between INR and the risk of noncardiovascular death identified in the RMST analysis (Table 3).

The number of deaths due to bleeding was very low (Table 2 of the supplementary material) and no differences were observed in cumulative risk between the 3 INR categories (Gray test, P = .949).

A total of 230 deaths were attributed to heart failure (46.5%). Although the crude incidence rate was higher in patients with nontherapeutic INR values (Table 2), no significant differences were observed for cumulative risk between the 3 categories (Gray test, P = .866) (Figure 4).

The category associated with the highest risk of mortality due to heart failure was the subtherapeutic category, which was mostly attributable to early events. In the multivariate RMST analysis, only patients with subtherapeutic INR values had an increased risk of death due to heart failure: RMSTINR<2 = –0.31 years (95%CI, –0.57 to –0.05) (P = .018) (Table 3 and Figure 5). The time-dependent hazard ratios (Table 1 of the supplementary material) showed a pronounced and significant increase in risk in the subtherapeutic category, but this decreased rapidly and lost its significance after the third month of follow-up.

There were marked differences between the 3 INR groups in terms of the cumulative incidence of death due to causes other than heart failure, with the highest rate observed in the supratherapeutic category (Gray test, P = .018) (Figure 4).

In the multivariate analysis (Table 3), patients with supratherapeutic INR values had a considerably increased risk of death due to a cause other than heart failure compared with patients with values in the therapeutic range: RMSTINR>3 = –0.48 (95%CI, –0.81 to –0.16) (P = .004) (Figure 5). The assumption of proportional hazards was also met for this criterion and the prognostic values of the different INR categories are thus shown as a single hazard ratio in Table 4, supporting the results of the RMST analysis.

This study has several limitations. First, as an observational, single-center study, its results may have been influenced by several biases and aspects inherent to daily practice. Second, we were missing data on time in therapeutic range on admission and on subsequent INR levels, meaning that we have no information on level of anticoagulant therapy control beyond the acute episode. Third, assignment of specific causes of death in observational studies is complicated and has evident limitations.11 In this regard, the low incidence of deaths attributable to certain cardiovascular events, together with the presence of deaths due to unknown causes, makes it difficult to draw any firm conclusions about specific causes of death attributable to INR. Finally, INR was not measured at the same time as other variables (eg, echocardiographic variables).