Development and Validation of a Nomogram for Predicting Mortality in Patients with Atrial Fibrillation and Acute Coronary Syndrome Who Underwent Percutaneous Coronary Intervention in a Chinese Multicenter Cohort.

Background: This study is aimed at to establish an effective prognostic nomogram for patients with atrial fibrillation (AF) and acute coronary syndrome (ACS) underwent percutaneous coronary intervention (PCI).
Methods: The nomogram was based on a retrospective study of 977 patients with AF and ACS who underwent PCI who were admitted to any of the 11 tertiary hospitals in the Beijing area between 2009 and 2015. The predictive accuracy and discriminative ability of the nomogram were determined by a concordance index (C-index) and calibration curve and were compared using current risk scores such as GRACE, CRUSADE, CHA2DS2-VASc, and HAS-BLED. The results were validated using bootstrap resampling and a retrospective cohort study of 409 patients enrolled in Fuwai Hospital at the same institution.
Results: Independent factors derived from multivariable analysis of the primary cohort to predict all-cause mortality were age, pattern of ACS, red blood cell distribution width, N-terminal proBNP, and serum creatinine, all of which were assembled into the nomogram. The calibration curve for the probability of recurrence showed that the nomogram-based predictions were in good agreement with actual observations. The C-index of the nomogram for predicting mortality was 0.764 (95% CI, 0.718-0.810), which was statistically higher than the C-index values for the current risk scores (from 0.573 to 0.681). In the validation cohort, the C-index of the nomogram for predicting all-cause death was 0.706 (95% CI 0.601-0.811), with no significant differences compared with GRACE and CRUSADE, but better than that of CHA2DS2-VASc and HAS-BLED. Conclusions: The nomogram has good prognostic prediction for patients with AF and ACS who underwent PCI.

1. Introduction

Atrial fibrillation (AF) and acute coronary syndrome (ACS) often coexist. Atrial fibrillation complicates acute myocardial infarction (AMI) with an incidence between 6 and 21% [1]. Patients with a history of AF commonly underwent percutaneous coronary intervention (PCI), which varied in incidence by institution of 2.5% to 18.4% [2]. Patients with ACS, complicated by AF, had poorer short-term and long-term clinical outcomes, especially in the elderly [3]. In China, the prevalence of AF increased 20-fold from 2001 to 2012. The lifetime risk of AF was approximately one in five among Chinese adults, and it increased with advancing age [4]. In the China Acute Myocardial Infarction (CAMI) registry, 740 (3.0%) patients were recorded with AF during hospitalization, and the in-hospital mortality was significantly higher in patients with AF than those without AF [5].

Risk assessment plays a major role in the management of patients with AF and ACS. Several scores are already widely used clinically, including the Global Registry of Acute Coronary Events (GRACE) [6] and the Can Rapid Risk Stratification of Unstable Angina Patients Suppress Adverse Outcomes with Early Implementation of the American College of Cardiology/American Heart Association Guidelines (CRUSADE); these have shown good predictive values for both short-term and long-term mortality in ACS [7]. In AF, the CHA2DS2-VASc score [8] is used to estimate thromboembolic risk. The Hypertension, Abnormal Renal/Liver Function, Stroke, Bleeding History or Predisposition, Labile International Normalized Ratio, Elderly, Drugs/Alcohol Concomitantly (HAS-BLED) score [9] is recommended for bleeding risk prediction.

However, at present, there is no ideal risk score in this special population of AF combined with ACS. By creating an intuitive graph of a statistical predictive model, nomograms are reliable tools to quantify risk and have demonstrated advantages over the traditional staging systems used to predict patient outcomes. This study is aimed at establishing a prognostic nomogram for patients with AF and ACS who underwent percutaneous coronary intervention (PCI) in a Chinese cohort and compared with the current risk score systems.

2. Methods

This retrospective, multicenter study included patients with AF and ACS who were admitted in any of the 11 tertiary hospitals in Beijing between December 2009 and July 2015. The inclusion criteria were as follows: (1) patients with a diagnosis of ACS and ongoing PCI during the index hospital stay, including unstable angina (UA), non–ST-segment elevation myocardial infarction (NSTEMI), ST-segment elevation myocardial infarction (STEMI), and history of AF (paroxysmal, persistent, or permanent) or ongoing AF during the index hospital stay and (2) patients who provided the informed consent to participate. The exclusion criteria were as follows: (1) patients who died in the hospital; (2) patients with coronary artery bypass graft; and (3) patients with missing risk scores data. Among them, cases from Fuwai Hospital (about 30%) served as the validation cohort. Individual patient management decisions were decided by the interventional cardiologist and/or the treating clinical cardiologist. All demographic and clinical characteristics were obtained by screening hospitalization reports through the computerized system of the institution.

The validated risk scores, such as GRACE, CRUSADE, CHA2DS2-VASc, and HAS-BLED, were also calculated at the same time based on the definitions used in their validation cohorts. However, in the HAS-BLED score, the labile international normalized ratio (INR) could not be assessed and was thus omitted. This method has been used in other retrospective studies among patients with AF [10-12]. Therefore, a maximum of 8 points was used for “modified HAS-BLED” analysis in this study. The primary endpoint was all-cause mortality. All patients were followed up through telephone or face-to-face interviews from March 2016 to June 2016.

Data were analyzed using SPSS version 25.0 (SPSS Inc., Chicago, IL, USA). Categorical variables were grouped based on clinical findings, and decisions on the groups were made before modeling. Continuous variables were expressed as means and standard deviations, which were then compared using the Mann–Whitney test. Categorical variables were expressed as frequencies and percentages and compared using Fisher's exact test. Cox regression analysis was used for multivariate analyses. A nomogram was formulated based on the results of multivariable Cox regression analysis and by using the rms package in R version 3.6.2 (http://www.r-project.org/). The performance of the nomogram was measured using the concordance index (C-index). Bootstraps with 1,000 resamples were used for these activities. During the external validation of the nomogram, the total points of each patient in the validation cohort were calculated according to the established nomogram, and the C-index and calibration curve were derived based on the regression analysis. The C-statistics for risk scores were compared using the nonparametric test developed by DeLong et al. [13] (MedCalc version 12.3.0; MedCalc Software; Mariakerke, Belgium). A P value <0.05 was considered statistically significant.

3. Results

A total of 1386 patients completed follow-up, and 977 patients were in the primary cohort, while 409 patients from Fuwai Hospital were included in the validation cohort. The characteristics of the patients in the primary and validation cohorts are listed in Table 1. In the primary cohort, 139 patients (14.2%) died, with a median follow-up duration of 37.89 months. In the validation cohort, 29 patients (7.1%) died, with a median follow-up duration of 41.94 months. The primary cohort had higher age, serum creatinine, history of diabetes mellitus, and all-cause mortality than the validation cohort.

Table 1

Patient demographics and clinical characteristics.

Demographic or characteristic	Primary cohort (n = 977)	Validation cohort (n = 409)	P
Male (n%)	662 (67.8%)	300 (73.3)	0.039
Age (n%)			<0.001
<65year	319 (32.7%)	187 (45.7%)
65-74year	373 (38.2)	145 (35.5%)
≥75year	285 (29.2%)	77 (18.8%)
History of hypertension (n%)	721 (73.8%)	352 (86.1%)	<0.001
History of diabetes mellitus (n%)	326 (33,4%)	259 (63.3%)	<0.001
Current smoker (n%)	439 (44.9%)	310 (75.8%)	<0.001
Initial systolic blood pressure (mmHg, mean ± SD)	129.55 ± 20.3	129.13 ± 18.33	0.198
Pattern of ACS (n%)			0.006
Unstable angina pectoris	631 (64.6%)	300 (73.3%)
NSTEMI	189 (19.3%)	57 (13.9%)
STEMI	157 (16.1%)	52 (12.7%)
Pattern of AF			<0.001
Paroxysmal	751 (76.9%)	356 (87%)
Persistent	186 (19%)	47 (11.5%)
Permanent	40 (4.1%)	6 (1.5%)
WBC (∗109/L, mean ± SD)	7.39 ± 2.79	7.46 ± 2.01	<0.001
Hemoglobin (g/dl, mean ± SD)	135.78 ± 19.28	134.76 ± 17.47	0.029
RDW (%, mean ± SD)	13.45 ± 1.24	11.95 ± 1.45	<0.001
Platelets (∗109/L, mean ± SD)	193.77 ± 56.61	197.28 ± 59.07	0.270
Glucose (mmol/L, mean ± SD)	6.89 ± 2.83	6.15 ± 2.14	<0.001
Serum albumin (g/L, mean ± SD)	39.52 ± 4.64	40.93 ± 3.44	<0.001
LDL-C (mmol/L, mean ± SD)	2.42 ± 0.81	2.47 ± 0.86	0.098
NT-proBNP (n%)			<0.001
<300 pg/mL	159 (16.3%)	12 (2.9%)
300-1800 pg/mL	459 (47%)	188 (46%)
1800-18000 pg/mL	338 (34.6%)	203 (49.6%)
>18000 pg/mL	21 (2.1%)	6 (1.5%)
Serum creatinine (umol/L)	91.18±37.06	84.71 ± 23.76	<0.001
GRACE	126.88 ± 30.58	116.21 ± 29.12	0.211
CRUSADE	34.94 ± 14.53	30.45 ± 12.74	<0.001
CHA2DS2-VASc	3.50 ± 1.83	2.95 ± 1.61	<0.001
HAS-BLED	1.95 ± 0.98	1.71 ± 0.89	0.929
Follow-up time (month, mean ± SD)	37.88 ± 18.59	41.94 ± 18.82	0.507
All-cause death	139 (14.2%)	29 (7.1%)	<0.001

In the primary cohort, we entered variables with P values of <0.10 on univariate analysis into the multivariate model according to Cox analysis. With regard to all-cause death, 11 variables yielded from the univariate analysis entered into the subsequent multivariate Cox analysis, including age, initial systolic blood pressure, pattern of ACS, WBC, hemoglobin, and RDW (Table 2). In the multivariate analysis, we found that age and pattern of ACS, RDW, N-terminal proBNP (NT-proBNP), and serum creatinine levels were independently associated with prognosis, and we established a nomogram for this risk model (Figure 1). The prediction model showed a C-index of 0.764 (95% CI 0.718–0.810) (Figure 2), which was significantly superior to that of traditional risk scores (P < 0.05) such as GRACE, CRUSADE, CHA2DS2-VASc, and HAS-BLED (Table 3). In addition, the calibration curve demonstrated good concordance between the predicted and actual outcomes (Figure 3).

Table 2

Univariable analysis and Cox proportional hazards regression analysis.

Variable	Univariable analysis, P	Multivariable analysis			Selected factors for building the model
		Hazard ratio	95% CI	P	Hazard ratio	95% CI	P
Sex	0.112
Age	< .001	1.237	0.982-1.558	0.072	1.257	1.008-1.568	0.042
History of hypertension	0.672
History of DM	0.219
Current smoker	0.681
Initial SBP	0.007	0.948	0.669-1.342	0.762
Pattern of ACS	0.001	1.259	1.002-1.582	0.048	1.351	1.095-1.666	0.005
Pattern of AF	0.353
WBC	<0.001	1.427	0.939-2.174	0.098
Hemoglobin	<0.001	0.878	0.611-1.262	0.483
RDW	<0.001	1.206	1.097-1.326	<0.001	1.23	1.124-1.346	<0.001
Platelets	0.004	0.973	0.547-1.731	0.926
Glucose	<0.001	1.459	0.916-2.322	0.112
Serum albumin	<0.001	0.971	0.699-1.349	0.861
LDL-C	0.484
NT-proBNP	<0.001	1.799	1.383-2.342	0.000	1.823	1.403-2.369	<0.001
Creatinine	<0.001	1.005	1.002-1.008	0.000	1.005	1.002-1.008	<0.001

Figure 1

Establishment of a nomogram risk model for prediction all-cause mortality in patients with atrial fibrillation and acute coronary syndrome who underwent percutaneous coronary intervention. To use the nomogram, an individual patient's value is located on each variable axis, and a line is drawn upward to determine the number of points received for each variable value. The sum of these numbers is located on the Total Points axis, and a line is drawn downward to the survival axes to determine the likelihood of 1-, 2-, or 3-year survival.

Figure 2

ROC curve of the GRACE, CRUSADE, CHA2DS2VASc, HAS-BLED, and nomogram to predict all-cause mortality in primary cohort.

Table 3

C-statistics of the GRACE, CRUSADE, CHA2DS2VASc, HAS-BLED, and nomogram to predict all-cause mortality and comparisons of the predictive accuracy of the risk scores for all-cause mortality by DeLong test in primary cohort.

Risk model	All-cause mortality
	C-statistic	95% confidence interval	Z	P value
Nomogram	0.764	0.718-0.810	vs.
GRACE	0.642	0.591-0.692	4.911	<0.001
CRUSADE	0.681	0.633-0.729	3.951	<0.001
CHA2DS2VASC	0.627	0.577-0.677	5.326	<0.001
HAS-BLED	0.573	0.520-0.626	7.060	<0.001

Figure 3

The calibration curve for predicting patient outcome at (a) 1 year, (b) 2 years, and (c) 3 years in the primary cohort, and at (d) 1 year, (e) 2 years, and (f) 3 years in the validation cohort. Nomogram-predicted probability of overall outcome is plotted on the x-axis; actual overall outcome is plotted on the y-axis.

In the validation cohort, the C-index of the nomogram for predicting all-cause death was 0.706 (95% CI 0.601-0.811) (Figure 4). There were no significant differences in the C-indices compared with those of GRACE or CRUSADE, but it was significantly better than those of CHA2DS2-VASc and HAS-BLED (P < 0.05) (Table 4). A calibration curve showed good agreement between prediction and observation in the probability of 1–3 year survival (Figure 3).

Figure 4

ROC curve of the GRACE, CRUSADE, CHA2DS2VASc, HAS-BLED, and nomogram to predict all-cause mortality in validation cohort.

Table 4

C-statistics of the GRACE, CRUSADE, CHA2DS2VASc, HAS-BLED, and nomogram to predict all-cause mortality and comparisons of the predictive accuracy of the risk scores for all-cause mortality by DeLong test in validation cohort.

Risk model	All-cause mortality
	C-statistic	95% confidence interval	Z	P value
Nomogram	0.706	0.601-0.811	vs.
GRACE	0.721	0.606-0.835	0.250	0.803
CRUSADE	0.703	0.590-0.815	0.065	0.949
CHA2DS2VASC	0.621	0.520-0.723	1.346	0.017
HAS-BLED	0.608	0.513-0.704	1.656	0.009

4. Discussion

Patients with AF and ACS undergoing PCI have poor short-and long-term mortality. In the ACS or AF population, several validated risk scores, including GRACE, CRUSADE, CHA2DS2-VASc, and HAS-BLED, showed good predictive values for poor clinical outcome. However, in this specific group of patients, their predictive values are unknown.

The GRACE score is the most extensively investigated score for ACS. In 2003, the GRACE registry group published a risk score for in-hospital mortality in patients with ACS and later for 6-month mortality [14]. Its predictive value was further validated over 6 months on 5-year follow-up [15]. In our primary and validation cohorts, we found that it had moderate value for predicting long-term all-cause mortality, with the C-statistics of 0.642 and 0.721, respectively.

The CHA2DS2-VASc is a simple risk stratification schema to estimate thromboembolic risk in patients with nonvalvular AF. Kim et al. [16] evaluated 15,681 patients with ACS and showed that the CHA2DS2-VASc score was an important predictor of long-term mortality. In our study, the C-statistics of the CHA2DS2-VASc score for predicting long-term all-cause mortality were 0.621 and 0.627, respectively.

Meanwhile, bleeding complications have emerged as a major adverse clinical outcome in the management of ACS and AF because they are associated with poor outcomes [17]. The CRUSADE score was developed in a broad population of community-treated patients with NSTEMI, and recently, it has also been validated in STEMI, demonstrating good performance in this scenario [18]. Therefore, this score is the preferred method for bleeding risk stratification in the current clinical practice guidelines. In our cohort, we found that they had a moderate value for predicting long-term all-cause mortality, with the C-statistics of 0.681 and 0.703.

The HAS-BLED score was initially used to assess the risk of bleeding in patients with AF receiving anticoagulation therapy. Recently, in patients without AF after PCI, Capodanno et al. [19] and Konishi et al. [20] concluded that the HAS-BLED score could be used to predict the clinical outcomes at 3 years of follow-up. In our study, we found that they had poor value for predicting long-term all-cause mortality, with the HAS-BLED score having C-statistics of 0.573 and 0.0.608.

Cox multivariate analysis showed that five factors including age, pattern of ACS, NT-proBNP, RDW, and serum creatinine were finally included in the nomogram model. NT-proBNP has been shown to be of prognostic value not only in patients with chronic heart failure but also in patients with stable coronary artery disease and those with ACS, similar to the grading indicators of heart function in other risk scores. Serum creatinine was also included in GRACE and CRUSADE scores. Compared with other existing scores, the factor of RDW is novel. RDW is a parameter of circulating erythrocytes measured using a hematology analyzer. Recent studies have shown that RDW, as an easy and cheap biomarker, is associated with clinical outcomes in patients with acute coronary syndrome [21], heart failure [22], and atrial fibrillation [23]. A meta-analysis by Abrahan et al. [24] of 13 trials involving 10,410 patients showed that a low RDW was associated with a statistically significant lower all-cause or CV mortality in ACS (RR 0.35, (95% CI 0.30-0.40), P < 0.00001, I2 = 53%), a finding that was consistent both in the short and long term follow-up. In the two-year follow-up of an Israeli cohort of adults with atrial fibrillation including 69,412 patients, Saliba et al. [25] showed the cumulative all-cause mortality rate increased across the RDW quartiles: 9.8%, 13.6%, 18.8%, and 28.5%, respectively. The hazard ratio (HR) for mortality was 1.82 in the highest RDW quartile compared to the lowest quartile after adjustment for other factors. The study showed that RDW was independently associated with the risk of all-cause mortality in patients with atrial fibrillation. In our primary cohort, RDW was also associated with the risk of all-cause mortality and contributed prominently in the nomogram model.

To our best knowledge, the nomogram model was firstly established in a Chinese population with atrial fibrillation and ACS which underwent PCI. It only contained five indicators, including age, pattern of ACS, NT-proBNP, RDW, and serum creatinine, which were simple and commonly used in clinical practice. Samaras et al. [26] established a new risk score and validated in 887 patients with AF and found most important predictors of death included both cardiac biomarkers and clinical information, such as NT-proBNP, high-sensitivity troponin-T (hs-TnT), kidney impairment, and age. Similarly, Cai et al. [27] constructed an internally validated nomogram containing 5 baseline predictors in AF patients before cardiac resynchronization therapy, including NT-proBNP, history of syncope, and previous pulmonary hypertension. Compared with several traditional validated risk scores, in the modeling cohort, its predictive performance is superior to the existing scoring, and it is equivalent to GRACE and CRUSADE in the validation cohort and higher than other scoring systems, including CHA2DS2-VASc and HAS-BLED. However, Morrone et al. [28] reported that both CHA2DS2-VASc and HAS-BLED scores predicted mortality similarly in anticoagulated patients with AF. Meanwhile, Jaakkola et al. [29] reported that these two scores predicted the type of intracranial complication in patients with AF only at very high risk levels. Therefore, it is suggested that alternative risk scores are required to predict mortality of different subpopulation of AF patients.

The main limitation of this study is its retrospective and observational nature. Furthermore, the risk scores were calculated on a post hoc basis. Comprehensive information on liver function or labile INR was not available; thus, they were omitted in the calculation of the modified HAS-BLED score, which may have diminished the value of using HAS-BLED in this population. Although this study is a multicenter study, it is confined to tertiary hospitals in Beijing. The follow-up of the study was mainly from outpatient and telephone follow-up on patients admitted in the hospital between December 2009 and July 2015. Though we obtained 1386 available information, there were still many patients that could not complete the follow up, which may lead to a decrease in data reliability. The verification cohort of this study was used to collect single-center data over the same period and was not strictly an external verification cohort. It is better to collect a prospective cohort for further verification.

5. Conclusion

We developed and validated nomograms predicting long-term all-cause death in patients with ACS and AF in a Chinese cohort. The proposed nomogram is simple and commonly used; in this study, it provided significantly better discrimination than the current risk scores. To generalize the use of this nomogram, validation with data from other areas or prospective cohorts is required.