We retrospectively studied a cohort of 3302 consecutive patients admitted for AHF to the Cardiology Department of a third-level teaching center between January 2007 and June 2018. AHF was diagnosed by trained cardiologists according to the definition proposed by the guidelines applicable at the time of patients’ inclusion. On admission, all patients had symptoms or signs attributable to congestion. We included both patients with new onset AHF and decompensated chronic HF. We excluded patients with a missing CA125 value and those who underwent surgical valve replacement, transcatheter valvular intervention during the index admission, or transfer to heart transplant (n=71). The final study sample included 3231 patients.

Left ventricular ejection fraction (LVEF) was assessed by 2-dimensional echocardiography during the index hospitalization. Medical treatment was individualized based on the established guidelines.

Follow-up was limited to 6 months. Patients’ follow-up was censored if death or cardiac transplant (n=2) occurred within this period. The study was approved by an institutional review committee (Hospital Clínico Universitario, Valencia), and patients gave written informed consent. The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a priori approval by the institution's human research committee.

The Biology Study to Tailored Treatment in Chronic Heart Failure (BIOSTAT-CHF) was a multicenter, multinational, prospective, observational study that included 2516 patients with new onset or worsening signs or symptoms of HF from 69 centers in 11 European countries from December 2010 to December 2012.15 Median follow-up was 21 [interquartile range (IQR) 15-27] months. Patients were included after presentation with either new onset or worsening HF, defined as LVEF ≤ 40% and/or brain natriuretic peptide (BNP)> 400 pg/mL or N-terminal pro-B-type natriuretic peptide (NT-proBNP)> 2000 pg/mL. Patients were either seen in the outpatient setting or the hospital. The present study included 1583 hospitalized patients with available CA125 assessments. All patients enrolled in BIOSTAT-CHF gave their written informed consent to participate in the study, and BIOSTAT-CHF was conducted in accordance with the Declaration of Helsinki, national ethics, and legal requirements, as well as relevant European Union legislation. The study was also approved by national and local ethics committees. The characteristics of the BIOSTAT-CHF cohort have been described elsewhere.15

The receiver operating characteristic curve (ROC) was estimated for both endpoints at 1 month. To determine the optimal cutoff values of CA125 aimed at identifying low-risk patients, sensitivity, specificity, negative predictive value (NPV), and positive predictive values were estimated along the continuum of the biomarker (diagt command). From a range of cutoffs considered to be predictive of low risk, those with an NPV of 90% or greater, sensitivity of 85% or greater, and those that classified at least 10% of patients as low risk were further analyzed in multivariate survival analyses. In this latter scenario, the best-cutoff was the one with the best performance using the Akaike information criterion (AIC) and the Bayesian information criterion (BIC). Moreover, the prognostic adequacy of the cutoff deemed to be optimal was also tested at 6 months of follow-up. Sensitivity, NPV, AIC, and BIC for the preselected CA125 values are presented in figure 1. The best cutoff was compared with prior reported cutoffs in the literature (traditional cutoff suggested by most of the assays: 35 U/mL, and prognostic cutoffs used in prior studies: 65 U/mL).6–11 For this comparison, patients were classified according to the following CA125 categories: C1, <23 U/mL; C2, 23-34.9 U/mL; C3, 35-64.9 U/mL; and C4, ≥ 65 U/mL.

Figure 1.

Performance of different CA125 cutoffs for predicting 1-month adverse events. A: all-cause mortality. B: all-cause mortality/HF readmission. AIC, akaike information criterion; BIC, Bayesian information criterion; CA125, carbohydrate antigen 125; HF, heart failure; NPV, negative predictive value.

(0.29MB).

Employing a flexible parametric regression model–Royston-Parmar model–we determined the independent prognostic effect of CA125 categories with the 2 clinical outcomes. To hold a linearity assumption with the outcome when modelled as continuous exposure, the best fractional polynomial transformation was chosen based on AIC/BIC criteria. In the derivation cohort, regression estimates from the mortality model were adjusted by age, sex, prior admission for AHF, New York Heart Association (NYHA) before admission, etiology, atrial fibrillation, heart rate, systolic blood pressure, hemoglobin, blood urea nitrogen, NT-proBNP, LVEF, use of beta-blockers during hospital stay, and furosemide equivalent dose on admission. For the composite endpoint, the Charlson comorbidity index and severe tricuspid regurgitation were added to the mortality set of covariates. Under the same multivariate scenario, the prognostic adequacy of the different CA125 cutoffs was also tested.

A similar modelling strategy was applied to the validation cohort, except that these regression estimates were only adjusted with an outcome-specific risk score (BIOSTAT-CHF risk score).16 The BIOSTAT-CHF risk score for mortality included age, blood urea nitrogen, NT-proBNP, serum hemoglobin, and the use of beta-blockers. The BIOSTAT-CHF risk score for the composite endpoint included age, previous HF-hospitalization, peripheral edema, systolic blood pressure, NT-proBNP, hemoglobin, high-density lipoprotein, sodium, and use of beta-blockers. In a sensitivity analysis on the derivation cohort, risk estimates were also adjusted by the appropriate BIOSTAT-CHF risk score for both endpoints.16 Estimates of risk across sex, LVEF (≥ 50% and <50%), estimated glomerular filtration rate (> 60mL/min/1.73 m2 vs ≤ 60mL/min/1.73 m2), and NT-proBNP (above or equal to the median vs below median) were also calculated. All estimates were presented as hazard ratios (HR) with 95% confidence intervals (95%CI). Harrell c-statistics was used as the metric for the model's discrimination performance and presented in figure legends.

We set a 2-sided P-value of <.05 as the threshold for statistical significance. Stata 15.1 (Stata Statistical Software, Release 15 [2017]; StataCorp LP, College Station, United States) was used for the main analysis.

At 1 month of follow-up, 137 (4.2%) deaths and 328 (10.2%) composite endpoints were registered. Ninety-eight patients (3.0%) died during hospital stay. Crude rates of both endpoints increased when moving from lower to higher CA125 categories (table 1). Kaplan-Meier curves showed a progressive separation of curves throughout the entire follow-up with a tangible separation of curves within the first month for both endpoints and especially for those with CA125 <23 U/mL (figure 2A,B). After a multivariate adjustment, and compared with CA125 ≥ 23 U/mL, those with CA125 <23 U/mL showed an 80% reduction in the risk of death (HR, 0.20; 95%CI, 0.08-0.50; P <.001) and a 37% reduction in the combined endpoint (HR, 0.63; 95%CI, 0.45-0.89; P=.009). A similar risk reduction was also found when compared with patients with CA125 between 23 and 35 U/mL (HR, 0.20; 95%CI, 0.07-0.53; P=.001; and HR, 0.52; 95%CI, 0.34-0.79; P=.002 for death and death/HF readmission, respectively). Survival curves among the prespecified subgroups are presented in figure 1 and figure 2 of the supplementary data. Multivariate analyses revealed a nondifferential association across sex (male vs female), LVEF (≥ 50% vs <50%), NT-proBNP categories [above or equal to median (3480 pg/mL) vs below median], and estimated glomerular filtration rate (> 60mL/min/1.73 m2 vs ≤ 60mL/min/1.73 m2) (figure 3).

Figure 2.

Kaplan-Meier survival curves across CA125 categories. A: all-cause mortality. B: all-cause mortality/HF readmission. CA125, carbohydrate antigen 125; HF, heart failure.

(0.25MB).

Figure 3.

CA125 <23 U/mL and risk of adverse events. Multivariate risk estimates. Subgroup analyses. A: all-cause mortality. B: all-cause mortality/HF readmission. 95%CI, 95% confidence interval; AHF, acute heart failure; CA125, carbohydrate antigen 125; eGFR, estimated glomerular filtration rate; HF, heart failure; HR: hazard ratio; LVEF, left ventricle ejection fraction; NT-proBNP, N-terminal pro-B-type natriuretic peptide. For all patients, estimates of risks for mortality were adjusted by age, sex, prior admission for AHF, previous New York Heart Association class before admission, etiology, atrial fibrillation, heart rate, systolic blood pressure, hemoglobin, blood urea nitrogen, NT-proBNP, LVEF, use of beta-blockers during hospitalization, and furosemide equivalent dose on admission. For the composite endpoint, the Charlson comorbidity index and severe tricuspid regurgitation were added to the mortality set of covariates. Harrell's c-statistics for multivariate models for mortality and the composite endpoints were 0.81 and 0.71, respectively.

(0.25MB).

A sensitivity analysis, adjusting for the BIOSTAT-CHF risk score, also revealed the same pattern of risk for 1-month endpoints. Patients with CA125 <23 U/mL remained associated with a similar lower risk (HR1-monthdeath, 0.17; 95%CI, 0.07-0.41; P <.001; and HR1-monthdeath/HFreadmission, 0.43; 95%CI, 0.31-0.59; P <.001).

At the 6-month follow-up, we registered 424 (13.1%) deaths and 863 (26.7%) combined endpoints. At this time point, the differences indicating a lower risk in patients with CA125 <23 U/mL were of lower magnitude but remained significant for both endpoints (figure 2A,B). Regarding these secondary endpoints, adjusted risk estimates confirmed that CA125 <23 U/mL identified a subgroup of patients with a 51% lower risk of death (HR, 0.49; 95%CI, 0.35-0.69; P <.001) and 22% of death/HF readmission (HR, 0.78; 95%CI, 0.64-0.94; P=.010).

At 1 month of follow-up, 42 (2.6%) and 120 (7.6%) patients died and experienced the combined endpoint, respectively. Fifty-one (3.2%) patients died during hospitalization. At the 6-month follow-up, 184 (11.6%) patients died, and 371 (23.4%) died or were readmitted for HF. The NPV and sensitivity of CA125 <23 U/mL for 1-month mortality were 98.6% and 85.7%, respectively. The same parameters for 1-month death/HF readmission were 96.6% and 88.3%, respectively. Table 2 of the supplementary data shows the crude rates for 1- and 6-month endpoints across CA125 categories.

Patients with CA125 <23 U/mL had the lowest adverse event rates with differences already present since the first month (figure 4A,B). After multivariate adjustment for the BIOSTAT-CHF risk score, patients with CA125 <23 U/mL continued to show a lower risk when compared with those with CA125> 23 U/mL for 1-month death/HF readmission (HR, 0.54; 95%CI, 0.32-0.91; P=.021). The adjusted risk estimates for 1-month mortality did not reach statistical significance (HR, 0.61; 95%CI, 0.26-1.47; P=.267). At 6 months, CA125 <23 U/mL was associated with a lower risk for both endpoints (HR6-monthdeath, 0.59; 95%CI, 0.39-0.90; P=.014; and HR6-monthdeath/HFreadmission, 0.65; 95%CI, 0.49-0.86; P=.002).

Figure 4.

Kaplan-Meier survival curves. BIOSTAT-CHF international cohort. A: all-cause mortality. B: all-cause mortality/HF readmission. HF, heart failure.

(0.27MB).