C₂HEST score predicts clinical outcomes in heart failure with preserved ejection fraction: a secondary analysis of the TOPCAT trial

We acquired the dataset of the TOPCAT trial (phase III, randomized, double-blind, placebo-controlled) from the National Heart, Lung, and Blood Institute. The dataset of this trial was obtained from the National Heart, Lung, and Blood Institute (NHLBI) by applying to Biologic Specimen and Data Repository Information Coordinating Center (BIOLINCC, https://​biolincc.​nhlbi.​nih.​gov/​). Our study was approved by the Medical Ethical Committee of the First Affiliated Hospital, Sun Yat-sen University. The TOPCAT investigators were not involved in our current study.

The study design of the TOPCAT trial had been reported previously [11]. A total of 3445 patients from the Americas (USA, Canada, Brazil, and Argentina) and Russia/Georgia, with an age of ≥ 50 years, a left ventricular ejection fraction (LVEF) of ≥ 45%, a serum potassium level of < 5.0 mmol/L, and a history of HF hospitalization within the previous 12 months or elevated brain natriuretic peptide level within 60 days before randomization were enrolled. Exclusion criteria included severe systemic illness with a life expectancy of less than 3 years, severe renal dysfunction, and specific coexisting conditions.

In this study, we included 2202 patients who were without AF at baseline including a history of AF or AF confirmed on an electrocardiogram (ECG) at enrollment. Variables of patient characteristics at baseline were retrieved from the dataset to calculate the C₂HEST score with a total of 6 individual components including coronary artery disease (1 point), chronic obstructive pulmonary disease (COPD, 1 point each), hypertension (1 point), elderly (age ≥ 75 years, 2 points), systolic HF (2 points), and hyperthyroidism (1 point). These risk factors were determined based on a combination of medical record review and interview at baseline visits. Of note, since there were no data on “coronary artery disease” in the TOPCAT trial dataset, we modified the coronary artery disease criterion using a history of myocardial infarction (MI, as 1 point). Besides, since our studied population was HFpEF patients with merely subtle abnormalities in systolic function [12], the item “systolic HF” in the C₂HEST score received no point in our current study. The included patients were classified into three risk strata according to C₂HEST score: the low-risk of 0 to 1 point, the medium-risk of 2 to 3 points, and the high-risk of ≥ 4 points [7].

Participants were followed up every 4 months during their first year on the study, and every 6 months thereafter, to monitor the events. The first onset of AF during follow-up was the observed endpoint, which was defined as an irregular rhythm with no discernible P-waves confirmed by a physician as AF after ECGs or rhythm strips were adjudicated by a critical event committee. Besides, we also studied the outcomes of stroke, all-cause death, cardiovascular death, any hospitalization, and HF hospitalization, definitions of which were previously described [11]. Data on participants who did not have an event of time-to-event outcomes were censored at the date of last available follow-up information for clinical events.

Continuous variables were presented as mean ± standard deviation and compared using the unpaired Student’s t tests (following normal distribution). Categorical variables were presented as proportions and compared using the chi-square test or Fisher’s test as appropriate. Kaplan-Meier curves with the log-rank test were plotted to display the differences of AF, stroke, all-cause death, cardiovascular death, any hospitalization, and HF hospitalization according to the risk strata of the C₂HEST score, and death was censored in non-death outcomes. A Cox proportional hazard model was used to explore the association of the C₂HEST score, its components, and its risk strata with all-cause death, and competing risk regression models for cumulative incidence were used when outcomes were incident AF, stroke, cardiovascular death, any hospitalization, and HF hospitalization, and results were reported with hazard ratios (HRs) and confidence intervals (CIs). Death was the competing risk in models concerning incident AF, stroke, any hospitalization, and HF hospitalization, and non-cardiovascular death was the competing risk for cardiovascular death. The adjusted model included various variables, including gender, treatment arm, diabetes mellitus, smoke or ever smoke, body mass index, heart rate, diastolic blood pressure, and estimated glomerular filtration rate. The score was evaluated through the time-dependent (at follow-up at 5 years) area under the curve (AUC) for the receiver operating characteristic (ROC) curve for predicting incident AF and other outcomes. Time-dependent ROC was performed following the methods introduced by Blanche et al. [13]. In brief, it defines cases and controls by subjects with events and censored, and then uses the inverse probability of censoring weighting (IPCW) approach to calculated time-dependent AUC. Blanche et al. also introduced two ways to define controls: (i) a control is defined as a subject i that is free of any event, and (ii) a control is defined as a subject i that is not a case, and we used definition (i) in the present study.

In a sensitivity analysis, we repeated the above-mentioned analyses by replacing “systolic HF” with “HF.” As reported previously [14, 15], we also used “thyroid disease” to replace “hyperthyroidism” for another sensitivity analysis. In subgroup analysis, we divided subgroups by region of participants (the Americas versus Russia/Georgia) because of concerns about the quality of enrollment as previously reported [16]. Besides, subgroup analyses divided by gender (males versus females) and treatment arm (spironolactone versus placebo) were also performed. All statistical analyses were performed with R version 4.0 (with packages tableone, survival, cmprsk, and timeROC). A two-tailed P value of < 0.05 was considered statistically significant.

Baseline patient characteristics were summarized in Table 1. The low-risk C₂HEST score stratum was the youngest, with the lowest prevalence of males, and had the highest mean heart rate, diastolic blood pressure, body mass index, and estimated glomerular filtration rate. These patients in the low-risk stratum had the lowest prevalence of MI, stroke, COPD, hypertension, dyslipidemia, and peripheral artery disease history. Baseline echocardiographic characteristics were summarized in Additional file 1: Table S1, but only a small proportion of enrolled patients had echocardiography data available. Patients in the high-risk stratum had the biggest mean maximal left atrial anterior-posterior diameters.

Among 2202 patients included in our study, 130 (5.9%) incident AF events were recorded during a median follow-up time of 3.07 years. The incidence rates of AF across the C₂HEST scores were presented in Additional file 1: Table S2. Overall, the average incidence rate of AF was 1.79 per 100 person-years in HFpEF patients. Baseline characteristics of HFpEF patients with or without incident AF were presented in Additional file 1: Table S3.

The associations of individual components in the C₂HEST score with incident AF are presented in Additional file 1: Table S4. In the competing risk regression models, age ≥ 75 years old (HR 3.21, 95% CI 2.28–4.53) and hyperthyroidism (HR 5.31, 95% CI 2.16–13.10) were independently associated with an increased risk of incident AF. When analyzed as a continuous variable, a 1-point increase in the C₂HEST score was associated with a 50% increased risk of incident AF (HR 1.50, 95% CI 1.29–1.75; Table 2). When patients were divided into three risk strata, the Kaplan-Meier curves showed a graded increased risk for incident AF (log-rank P < 0.001, Fig. 1). Compared with patients in low-risk stratum, those in medium-risk stratum (HR 2.00, 95% CI 1.34–2.99) or high-risk stratum (HR 3.32, 95% CI 1.93–5.71) showed increased risks of incident AF (Table 2).

Among the studied patients, the average incidence rates of stroke, all-cause death, cardiovascular death, any hospitalization, and HF hospitalization were 0.70, 3.81, 2.42, 15.50, and 3.32 per 100 person-years in HFpEF patients, respectively. The associations of individual components in the C₂HEST score with outcomes were presented in Additional file 1: Table S4. In Cox proportional hazard and competing risk regression models, previous MI history was significantly associated with higher risks of all-cause death (HR 1.32, 95% CI 1.03–1.68), cardiovascular death (HR 1.39, 95% CI 1.03–1.88), and any hospitalization (HR 1.22, 95% CI 1.06–1.40); COPD history was significantly associated with higher risks of any hospitalization (HR 2.09, 95% CI 1.75–2.50) and HF hospitalization (HR 2.26, 95% CI 1.64–3.11); age ≥ 75 years old was significantly associated with higher risks of all-cause death (HR 1.99, 95% CI 1.56–2.53), cardiovascular death (HR 1.50, 95% CI 1.09–2.06), any hospitalization (HR 1.60, 95% CI 1.38–1.85), and HF hospitalization (HR 1.63, 95% CI 1.24–2.13); thyroid disease history was significantly associated with higher risks of any hospitalization (HR 1.55, 95% CI 1.30–1.85) and HF hospitalization (HR 1.53, 95% CI 1.09–2.14); and hyperthyroidism was significantly associated with higher risk of HF hospitalization (HR 4.08, 95% CI 2.01–8.30).

When analyzed as a continuous variable, per 1-point increase in the C₂HEST score was associated with increased risks of all-cause death (HR 1.20, 95% CI 1.08–1.33), cardiovascular death (HR 1.15, 95% CI 1.00–1.33), any hospitalization (HR 1.22, 95% CI 1.14–1.29), and HF hospitalization (HR 1.14, 95% CI 1.01–1.29), but not stroke (HR 1.24, 95% CI 1.00–1.54) (Table 2). The cumulative incidences of these outcomes in different risk strata were shown in Fig. 1, and the differences in all-cause death, cardiovascular death, any hospitalization, and HF hospitalization risks among different risk strata were of statistical significance (all P < 0.001). In Cox proportional hazard and competing risk regression models, compared with patients in low-risk stratum, those in medium-risk stratum had greater risks of all-cause death (HR 1.39, 95% CI 1.07–1.81) and any hospitalization (HR 1.50, 95% CI 1.30–1.73), but not stroke, cardiovascular death, or HF hospitalization, whereas those in high-risk stratum showed increased risks of all-cause death (HR 1.98, 95% CI 1.37–2.86), cardiovascular death (HR 1.89, 95% CI 1.16–3.09), any hospitalization (HR 1.87, 95% CI 1.47–2.37), and HF hospitalization (HR 1.71, 95% CI 1.10–2.64), except for stroke (Table 2).

As shown in Fig. 2, the time-dependent AUC for the C₂HEST score in predicting incident AF (0.694 [95% CI 0.640–0.748]) was higher than other studied outcomes including stroke (0.644 [95% CI 0.565–0.727], all-cause death (0.624 [95% CI 0.583–0.664]), cardiovascular death (0.612 [95% CI 0.564–0.660]), any hospitalization (0.638 [95% CI 0.606–0.670]), or HF hospitalization (0.621 [95% CI 0.577–0.665]).

The results of the sensitivity analysis were summarized in Additional file 1: Table S5. When we used “HF” as a scoring item instead of “systolic HF” or replacing “hyperthyroidism” with “thyroid disease,” the results were similar with primary analyses as mentioned above. Performances of the C₂HEST score on incident AF and other outcomes were similar with primary analyses (Additional file 1: Fig. S1–S2).

The results of subgroup analyses were summarized in Additional file 1: Table S6. In the subgroup analysis based on region, we observed no significant difference between participants from the Americas and Russia/Georgia regarding the outcomes of AF, stroke, and HF hospitalization (all P for interaction > 0.05). The C₂HEST score was shown to be predictive for all-cause death and cardiovascular death only in participants from Russia/Georgia, but not in those from the Americas. The predictive value of the C₂HEST score was significantly greater in those from Russia/Georgia than in those from the Americas (all P_interaction < 0.05). For all outcomes of interest, there were no significant differences in subgroup analyses based on sex (males versus females) or treatment arm (spironolactone versus placebo) (all P_interaction > 0.05).

We acknowledged several limitations in this study. First, the regional variation of the TOPCAT population might affect results as shown in subgroup analysis, along with our slight modification of the C₂HEST score criteria to use what was available in the TOPCAT dataset (e.g., using MI to stand for coronary artery disease). Second, our results were based on the data of a retrospective analysis of the TOPCAT trial and it is possible that healthier patients were selected and the unmeasured confounders were not found, which might influence the validity of our findings. Third, incident AF as the outcome event was collected in the follow-up every 4 or 6 months. Coupled with some missed visits, there could be a certain degree of underestimation of incident AF. In addition, some participants with paroxysmal AF might not be aware of it and not recorded when they received ECG examination at baseline, and therefore, the follow-up outcome in our study might be recurrent AF.