Prediction of Cardiac Death in Patients with Hypertrophic Cardiomyopathy Using Plasma Adipokine Levels

Abstract

Backgrounds and Aims

Hypertrophic cardiomyopathy (HCM) causes cardiac death through both sudden cardiac death (SCD) and death due to heart failure (HF). Although adipokines lead to adverse cardiac remodeling in HCM, the prognostic value of plasma adipokines in HCM remains unknown. We aimed to predict cardiac death in patients with HCM using plasma adipokines.

Methods and Results

We performed a multicenter prospective cohort study of patients with HCM. The outcome was cardiac death including heart transplant, death due to HF, and SCD. With data from 1 institution (training set), a prediction model was developed using random forest classification algorithm based on 10 plasma adipokines. The performance of the prediction model adjusted for 8 clinical parameters was examined in samples from another institution (test set). Time-to-event analysis was performed in the test set to compare the rate of outcome events between the low-risk and high-risk groups determined by the prediction model. In total, 389 (267 in the training set; 122 in the test set) patients with HCM were included. During the median follow-up of 2.7 years, 21 patients experienced the outcome event. The area under the covariates-adjusted receiver-operating characteristics curve was 0.89 (95% confidence interval [CI] 0.71–0.99) in the test set. revealed the high-risk group had a significantly higher risk of cardiac death (hazard ratio 17.8, 95% CI 2.1–148.3, P = 0.008).

Conclusion

The present multicenter prospective study demonstrated that a panel of plasma adipokines predicts cardiac death in patients with HCM.

INTRODUCTION

Hypertrophic cardiomyopathy (HCM) is one of the most common genetic cardiac diseases. HCM affects 1 in 200–500 people all over the world [1], and is one of the leading causes of sudden cardiac death (SCD) in the young [2]. HCM can cause advanced heart failure (HF) and death due to HF, which is rescued only by heart transplant [1]. Current guidelines propose prediction models for SCD of patients with HCM using age, clinical history, and echocardiographic findings. However, the accuracy of prediction remains suboptimal [3–5]. Although some circulating biomarkers, such as N-terminal pro-brain natriuretic peptide [6] and troponin T [7, 8], are known to be associated with prognosis in HCM, the current prediction models for SCD do not utilize any circulating biomarkers [9]. Furthermore, the prediction model for fatal adverse cardiac events other than SCD, such as death due to HF, has not been established in patients with HCM. Development of a more accurate prognostic model with a small panel of circulating biomarkers is warranted to better inform patients and family members and to enable frequent surveillance and early referral to specialists.

Adipokines are a group of bioactive molecules known to be secreted mainly from adipose tissue and affect the whole body’s homeostasis through the bloodstream [10, 11]. Some adipokines are related to cardiovascular events, especially among people with obesity [12–14]. Adipokines such as adiponectin [15], leptin [15], and resistin [16], are known to have cardiovascular effects by mediating inflammation and cardiac fibrosis. Prior studies reported that levels of certain adipokines are elevated in patients with HCM compared with healthy controls [9, 17–19]. Especially, some inflammatory adipokines, such as resistin [17] and apelin [9], are associated with myocardial fibrosis and scarring in HCM. Therefore, these adipokines may be associated with disease progression in HCM, possibly by provoking systemic inflammation and myocardial fibrosis. However, the value of adipokines to predict cardiac death in patients with HCM remains unknown. Thus, in this multicenter study, we prospectively tested whether a small set of plasma-circulating adipokines can predict cardiac death in patients with HCM. We also aimed to identify patients with HCM who are at high risk of cardiac death using plasma adipokine levels.

METHODS

Study Design and Sample

We performed a prospective cohort study on adults (aged ≥ 18 years) with a diagnosis of HCM. Participants were enrolled from the HCM programs at Columbia University Irving Medical Center (CUIMC) (New York, NY) and Massachusetts General Hospital (MGH) (Boston, MA) between April 2008 and February 2021, and were consecutively included in this study if phlebotomy for plasma adipokine measurement was performed. The diagnosis of HCM was established by echocardiographic evidence of left ventricular (LV) hypertrophy (maximum LV wall thickness ≥ 15 mm) that was out of proportion to systemic loading conditions. We excluded patients with HCM phenocopies such as Fabry disease or cardiac amyloidosis by performing additional testing (e.g., genetic testing, cardiac magnetic resonance [CMR] imaging, technetium-99m pyrophosphate scintigraphy imaging, and heart biopsy) when needed. For patients with a family history of HCM, LV wall thickness ≥ 13 mm was considered diagnostic of HCM [5, 20].

The baseline characteristics were collected at the time of phlebotomy to determine plasma adipokine levels. The parameters of echocardiography and CMR imaging were obtained from these tests performed closest to the time of the phlebotomy. The training set to derive the prediction model consisted of patients from MGH. The independent test set for validation was based on patient data from CUIMC. The Mass General Brigham Institutional Review Board and that of CUIMC approved the study protocol and all participants provided written informed consent.

Blood Sample Processing and Plasma Adipokine Level Measurement

Venous blood specimens were drawn at the time of an outpatient clinic visit. Samples were collected in K₂EDTA-treated tubes and centrifuged for 10 min at 3100 rpm. Immediately after aliquoting, the supernatant plasma was frozen at −80 degrees Celsius [21].

Plasma levels were determined on 10 adipokines that were previously reported to be associated with adverse events in patients with cardiovascular diseases [22–24] using the SomaScan assay (SOMALogic, Inc., Boulder, CO) [21, 25, 26]. This assay is highly multiplexed, sensitive, quantitative, and reproducible, with the lower limit of quantification of 100 fM and median coefficient of variance of < 5% for human proteins [25, 27]. We performed a thorough literature search using the terms “adipokine” and “hypertrophic cardiomyopathy” on PubMed in English on November 27^th, 2023. We found 21 papers based on this search and included all adipokines discussed in these 21 papers in the present study. The list of 10 adipokines included resistin, adiponectin, leptin, C1q/tumor necrosis factor-related protein (CRTP)-1, CRTP-3, CRTP-9, visfatin, apelin, omentin-1, and chemerin [22–24].

Outcome Measures

The primary outcome of cardiac death was defined as a composite of heart transplant, death due to HF, and SCD with or without successful resuscitation [21]. Successful resuscitation refers to return of spontaneous circulation with external shocks or internal shocks from implantable cardioverter-defibrillator (ICD). The event rate was expressed as number of events per 100 person-years. Events were independently adjudicated by two attending cardiologists. The first one reviewed the medical record, and if needed, performed direct interviews with the treating physicians and/or patients. The second one also reviewed the medical record. Any disagreements were resolved through discussion.

Univariable Analysis

Continuous variables were presented as mean ± standard deviation if normally distributed and as median [interquartile range] if not normally distributed. To compare clinical characteristics between patients with and without cardiac death, the unpaired Student’s t-test was used for normally distributed continuous variables and the Mann-Whitney-Wilcoxon test for other continuous and ordinal variables. Statistical significance was declared if the 2-sided P value was < 0.05.

Development of 10-Adipokine Model to Predict Cardiac Death

Random forest model was used to establish the prediction model for cardiac death using the aforementioned 10 adipokines. Five-fold cross-validation method was applied for the training set using the R caret package. In hyperparameter tuning, Gini impurity was selected as the split rule. Three patterns of mtry were attempted (i.e., 2, 6, and 10) and subsequently mtry 2 was selected. The minimum node size was set as 1. The predictive ability of the model in the independent test set was measured to test the performance of the model developed from the training set. After calculating the receiver-operating-characteristic (ROC) curve in the test set, the ROC curve was adjusted for clinical factors that were included in the previously-reported prediction model for cardiac death from a large-scale meta-analysis using the R ROCnReg package [28, 29]. The following clinical factors were included: age, New York Heart Association (NYHA) functional class, family history of SCD, syncope, atrial fibrillation (AF), non-sustained ventricular tachycardia (NSVT), maximum LV wall thickness, and LV outflow tract gradient at rest [28, 29]. Covariate-adjusted ROC (AROC) curve is an average of each covariate-specific ROC curve where the averaging is made with respect to the distribution of the covariates. AROC curve takes covariate information into account [30, 31]. The area under the AROC curve (AUC) was calculated as an indicator of the model’s predictive ability.

Time-to-first-event Analysis to Identify Patients at High Risk of Cardiac Death

Using the cutoff value of the predicted probability (ranging from 0% to 100%) for cardiac death determined by the nearest point to the top-left corner of the original ROC curve, the test set was divided into a low-risk group and a high-risk group. In this time-to-first-event analysis, logrank test was used to compare the Kaplan-Meier survival estimates for patients in the low- and high-risk groups. Cox proportional hazards model was performed to examine the longitudinal association of the group classification in the test set with cardiac death.

RESULTS

The present study included 389 patients with HCM. The training set to develop the prediction model was comprised of 267 patients from MGH, of whom 14 patients subsequently had cardiac death. The independent test set for validation consisted of 122 patients from CUIMC, including 7 patients who subsequently experienced cardiac death.

Baseline characteristics are presented in Table 1. Patients with cardiac death were more likely to have HF symptoms with NYHA class 3 or 4, prior AF, and prior sustained ventricular tachycardia or ventricular fibrillation. They were also likely to be taking loop diuretics, anticoagulation, and amiodarone. In echocardiographic evaluation, patients with cardiac death had a larger LV end-diastolic diameter and intraventricular septal thickness as well as a greater degree of mitral regurgitation, whereas they had lower LV ejection fraction. With exercise stress tests, patients with cardiac death had lower metabolic equivalents and peak heart rate during stress. Finally, in terms of CMR characteristics, they had larger LV end-systolic and diastolic volumes. Other demographic and clinical characteristics were similar between the two groups.

Table 1.

Baseline clinical characteristics of the study sample

		Patients with cardiac death (n = 21)	Patients without cardiac death (n = 368)	P value
Demographics
	Age (years)	63 ± 17	59 ± 16	0.30
	Male	14 (67)	236 (64)	0.99
	Race/Ethnicity			0.69
	European ancestry	17 (85)	286 (79)
	African American	0 (0)	25 (7)
	Asian	0 (0)	8 (2)
	Native American	0 (0)	1 (0.3)
	Unidentified	3 (15)	41 (11)
	Height (m)	1.70 ± 0.1	1.70 ± 0.1	0.92
	Weight (kg)	85 ± 25	87 ± 20	0.68
	BMI (kg/m2)	29.0 ± 6.1	30.0 ± 5.8	0.43
	SBP (mmHg)	125 ± 15	124 ± 16	0.78
	DBP (mmHg)	73 ± 9	74 ± 9	0.55
Past medical history
	Hypertension	9 (43)	199 (54)	0.40
	NYHA functional class ≥3	5 (24)	31 (8)	0.048
	Prior AF	12 (57)	96 (26)	0.005
	Prior sustained VT/VF	7 (33)	14 (4)	<0.001
	Prior NSVT	7 (33)	76 (21)	0.27
	Prior syncope	6 (29)	76 (21)	0.56
	Prior stroke	2 (10)	20 (5)	0.76
Family history
	Family history of SCD	0 (0)	36 (10)	0.26
	Family history of HCM	6 (29)	83 (23)	0.72
Medications
	β-blocker	19 (91)	240 (65)	0.06
	Non-dihydropyridine calcium channel blocker	7 (33)	71 (19)	0.20
	Loop diuretic	8 (38)	37 (10)	<0.001
	Aspirin	8 (38)	137 (37)	>0.99
	Anticoagulation	10 (48)	75 (20)	0.008
	Thiazide	2 (10)	35 (10)	>0.99
	ACE inhibitor	3 (14)	34 (9)	0.70
	ARB	3 (14)	54 (15)	>0.99
	Potassium spearing diuretic	3 (14)	22 (6)	0.29
	Clonidine	1 (5)	7 (2)	0.91
	Statin	14 (67)	185 (50)	0.22
	Digoxin	0 (0)	2 (0.5)	>0.99
	Disopyramide	4 (19)	23 (6)	0.07
	Amiodarone	5 (24)	9 (2)	<0.001
Genetic testing (n = 240)		n = 13	n = 227
	Pathogenic or likely pathogenic	5 (42)	49 (24)	0.30
Prior procedures
	Prior myectomy	4 (19)	68 (19)	>0.99
	Prior ASA	1 (5)	35 (10)	0.73
	Prior ICD	13 (62)	94 (26)	0.001
	Prior permanent pacemaker	3 (14)	38 (10)	0.83
Echocardiographic characteristics
	LVDd (mm)	45 ± 9	43 ± 6	0.11
	LVDs (mm)	30 ± 12	26 ± 5	0.006
	Maximum wall thickness (mm)	19 ± 4	18 ± 4	0.10
	IVST (mm)	18 ± 4	16 ± 4	0.046
	LVPWT (mm)	13 ± 3	12 ± 2	0.052
	Left atrial diameter (mm)	44 ± 8	42 ± 7	0.19
	LV ejection fraction (%)	60 ± 20	69 ± 9	<0.001
	LV outflow tract gradient at rest (mmHg)	6.5 [0–50]	10 [0–41]	0.93
	LV outflow tract gradient with Valsalva (mmHg)	40 [0–89]	16 [0–70]	0.29
	Mitral valve SAM	8 (42)	163 (45)	>0.99
	Degree of mitral regurgitation*	2.5 [1.5–3.0]	2.0 [1.0–2.5]	0.02
Exercise stress test characteristics (n = 290)		n = 12	n = 278
	METs achieved with stress test	6.2 ± 3.8	9.5 ± 3.8	0.01
	Peak HR during stress	115 ± 43	138 ± 27	0.005
	Peak SBP during stress (mmHg)	165 ± 38	168 ± 29	0.76
	Peak DBP during stress (mmHg)	78 ± 21	80 ± 13	0.60
CMR characteristics (n = 231)		n = 10	n = 221
	LV mass (g)	188 ± 61	167 ± 59	0.37
	LV end-diastolic volume (mL)	169 ± 68	144 ± 41	0.07
	LV end-systolic volume (mL)	78 ± 72	52 ± 26	0.009
	Stroke volume (mL)	85 ± 21	91 ± 23	0.40
	Late gadolinium enhancement	6 (60)	128 (58)	>0.99

Data are given as the mean ± SD, or n (%), or as the median [interquartile range].

ACE, angiotensin-converting-enzyme; AF, atrial fibrillation; ARB, angiotensin II receptor blocker; ASA, alcohol septal ablation; BMI, body mass index; BSA, body surface area; CMR, cardiac magnetic resonance imaging; DBP, diastolic blood pressure; HCM, hypertrophic cardiomyopathy; HR, heart rate; ICD, implantable cardioverter-defibrillator; IVST, intraventricular septum thickness; LV, left ventricle; LVDd, left ventricular end-diastolic diameter; LVDs, left ventricular end-systolic diameter; LVPWT, left ventricular posterior wall thickness; MET, metabolic equivalent; NSVT, non-sustained ventricular tachycardia; NYHA, New York Heart Association; SAM, systolic anterior motion; SBP, systolic blood pressure; SCD, sudden cardiac death; VT/VF, ventricular tachycardia or ventricular fibrillation

^*Degree of mitral regurgitation was converted to numerical values according to the following rule: none = 0, trace = 1, trace to mild = 1.5, mild = 2, mild to moderate = 2.5, moderate = 3, moderate to severe = 3.5, severe = 4

As shown in Table 2, the primary outcome event of cardiac death was observed in 21 patients (5.4%) during the study period (median 2.7 years, interquartile range 1.9–5.1 years). This number is in agreement with a previous meta-analysis on cardiac death in patients with HCM reporting that the 3-year event rate was 5.7% (95% CI, 4.4–6.9%) [28]. The incidence rate per patient-year was 1.6%.

Table 2.

Primary outcome events observed during the study period

		n = 389
Events, n (%)
	Heart transplant	5 (1.3)
	Death due to heart failure	7 (1.8)
	Sudden cardiac death	9 (2.3)
Total patients with events, n (%)		21 (5.4)
Follow up time (years)		2.7 [1.9–5.1]
Incidence rate per patient-year (%)		1.6

Data were expressed as numbers (percentages) or median [interquartile range]

The random forest model to predict cardiac death derived from the 10 plasma adipokine levels in the training set exhibited high predictive ability in the test set after the adjustment for clinical covariates (AUC 0.89, 95% confidence interval [CI] 0.71–0.99; Fig. 1). The calibration plot and confusion matrix of the random forest model in the test set is displayed in Supplemental Fig. 1 and Supplemental Table 1. The cutoff value of the predicted probability for cardiac death at the nearest point to the top-left corner of the original ROC curve was 7.7%. With the cutoff value, the sensitivity was 0.86 (95% CI 0.42–1.00) and the specificity was 0.76 (95% CI 0.67–0.83) in the test set. Fig. 2 displays the relative importance of the 10 adipokines in the prediction model. Resistin, leptin, and apelin were the first, second, and third most important proteins in the prediction model.

Fig. 1.

Covariate-adjusted receiver-operating-characteristic curve of the model using plasma adipokine levels to predict cardiac death in patients with hypertrophic cardiomyopathy in the test set. Covariates are age, New York Heart Association functional class, family history of sudden cardiac death, syncope, atrial fibrillation, non-sustained ventricular tachycardia, maximum left ventricular wall thickness, and left ventricular outflow tract gradient at rest. AUC, area under the receiver-operating-characteristic curve; CI, confidence interval; ROC, receiver-operating-characteristic

Fig. 2.

Relative importance of adipokines in the model to predict cardiac death in patients with hypertrophic cardiomyopathy. CRTP, C1q/tumor necrosis factor-related protein

Patients in the test set were categorized to low-risk group (n = 88) or high-risk group (n = 34) according to the cutoff value of the predicted probability. Fig. 3 shows the time to the first outcome event in the test set stratified by the group allocation. The time-to-event analysis showed a significant difference in the rate of cardiac death between the low-risk group and the high-risk group (P_logrank < 0.001). Compared with the low-risk group, the Cox-proportional hazards model demonstrated that the high-risk group had a significantly higher rate of developing cardiac death, with a hazard ratio (HR) of 17.8 (95% CI 2.1–148.3, P = 0.008; Table 3). In this model, the proportional hazard assumption was not violated (Schoenfeld individual test P = 0.52, Supplemental Fig. 1).

Fig. 3.

Kaplan-Meier curves for time to cardiac death according to the risk group classified using plasma adipokine levels in the test set.

Table 3.

Hazard ratio for time to first outcome event in the test set

Groups	Patients with event, n (%)	Incidence rate, per 100 patient-years (95% CI)	Hazard ratio (95% CI)	P value
Low-risk group (n = 88)	1 (1.14)	0.34 (0.0087–1.9)	Reference	Reference
High-risk group (n = 34)	6 (17.6)	7.16 (2.63–15.6)	17.8 (2.1–148.3)	0.008

CI, confidence interval

DISCUSSION

Summary of Findings

In the present multicenter prospective cohort study of 389 patients with HCM, the 10-adipokine model derived from the training set, with adjustment for known clinical predictive factors, exhibited high predictive ability to predict cardiac death in the test set for validation. After dividing the test set into low- and high-risk groups according to the prediction model, the high-risk group had a significantly higher rate of developing cardiac death. The present study serves as the first to use plasma adipokine levels to predict cardiac death in HCM.

Results in Context

LV hypertrophy and fibrosis are the mainstays of HCM pathogenesis. They lead to cardiac death in patients with HCM through various pathways. Hypertrophy causes diastolic dysfunction and resultant HF. Fibrosis can lead to fatal tachyarrhythmias and advanced HF [1, 32]. Although the molecular mechanisms that lead to hypertrophy and fibrosis of HCM are largely unknown, prior studies reported that systemic inflammation was associated with myocardial fibrosis and diastolic dysfunction in patients with HCM [18, 33, 34].

Adipokines, derived from adipose tissue, are known to be associated with the induction of myocardial inflammation, hypertrophy, and fibrosis by stress signaling pathways caused by cardiac fibroblasts and macrophages [13, 35]. Leptin is a well-known pro-inflammatory adipokine that increases the production of tumor necrosis factors and interleukin-6 [12, 15, 22]. Leptin is related with myocardial fibrosis, which leads to fatal arrhythmia or cardiac death [36]. Resistin and visfatin are also known as pro-inflammatory adipokines [14, 22]. On the other hand, adiponectin is known as an anti-inflammatory adipokine with cardio-protective effects [15, 22–24]. Adiponectin protects against pathological cardiac remodeling and ischemia injuries by reducing myocardial oxidative stress, suppressing inflammation, and improving energy supply [37]. Omentin-1, chemerin, apelin, and CRTPs are also known as anti-inflammatory adipokines [15, 22–24]. The present prediction model was developed using 10 adipokines that are known to be associated with cardiovascular diseases based on the a priori knowledge [12, 14, 15, 22–24].

Some studies investigated the relationship between adipokines and HCM. Plasma leptin and resistin levels have been reported to be altered in patients with HCM [38]. Gene polymorphism of resistin was associated with myocardial fibrosis among patients with HCM [17, 39]. Adiponectin was associated with LV systolic dysfunction in patients with HCM [9, 33, 34]. Apelin was correlated with late gadolinium enhancement in CMR [9, 40]. Omentin-1 was reported to be dysregulated in HCM compared to healthy control and associated with adverse cardiac events among patients with HCM [9, 19]. These prior studies collectively suggest the association of adipokines with HCM disease progression. However, no prior studies have prospectively examined the prognostic value of a set of multiple adipokines in HCM. In this context, the present study adds to the body of knowledge by demonstrating for the first time that a panel of plasma adipokines can predict cardiac death in patients with HCM independent from known clinical risk factors [28].

Clinical Significance of Identifying High-risk Patients with HCM

The identification of patients with HCM at high risk of cardiac death is essential for the timely implementation of appropriate interventions and medications to prevent SCD and death due to HF. Young patients with HCM, especially those younger than 40 years old, have a high risk of SCD [32, 41]. Indeed, HCM is the most common cause of SCD in the young, especially among competitive athletes [2]. ICD for patients at high risk of fatal tachyarrhythmia would appropriately abort SCD [42]. Although the risk of SCD declines with age, the risk for HF increases, becoming the most prevalent HCM-related morbidity by mid-to-late adulthood [32, 41]. Referral to advanced HF/transplant team before they become severely symptomatic may improve HF symptoms and finally prevent death due to HF [43]. Furthermore, the newest class of medications, cardiac myosin inhibitors, turned out to be efficacious in preventing LV remodeling [44–46] and relieving HF symptoms [47, 48] in symptomatic patients with HCM. Thus, it is clinically critical to identify high-risk patients with HCM as they would likely receive the greatest benefit from these interventions and/or this new class of medications.

Also, improvement in the prediction of cardiac death enables us to increase the frequency of monitoring for high-risk patients and to focus on the management of reversible risk factors such as AF [49] and obesity [50, 51]. Since AF can result in death due to HF, stroke, and severe functional disability in patients with HCM [49], the burden of AF should be decreased and anticoagulation therapy initiated in a timely manner before patients develop these devastating consequences. Similarly, obesity is associated with the progression of HF symptoms in patients with HCM [50, 51]. Indeed, our prior study have demonstrated that bariatric surgery can prevent HF exacerbations [52]. Therefore, a timely referral to weight loss programs may improve HF symptoms and play a role in the prevention of death due to HF.

Finally, a better prediction system for HCM-related mortality is necessary to inform the patient and family how the disease might affect their longevity. Nevertheless, other than our machine learning model based on clinical parameters [29], there are no risk stratification systems to predict cardiac death in HCM to date, and the current models only predict SCD with modest accuracy [3]. The current study serves as the first to exhibit the ability of plasma adipokines to predict cardiac death among patients with HCM.

Strengths of the Present Study

The present study is a prospective, multicenter cohort study including the largest number of patients with HCM who underwent the measurement of plasma adipokines. The prospective method improves the quality of outcome adjudication, and the multicenter nature of the present study enhances generalizability. Also, our model demonstrated a high predictive ability which was reproducible in the independent test set for validation, thereby further buttressing the generalizability of the inferences. In addition, our prediction model was adjusted for known clinical variables of predictive value. The same clinical variables were used in a prior study in which machine learning models were developed to predict cardiac death in patients with HCM [29]. Finally, the level of plasma adipokines can be measured non-invasively, quickly, and at low cost using enzyme-linked immunosorbent assay method. Therefore, our prediction model with the small panel of plasma adipokines has a high potential for application in clinical settings, such as shared decision-making with patients about the necessity of frequent monitoring and the initiation of additional medications.

Potential Limitations

There are several potential limitations in the current study. First, the study was performed at two high-volume tertiary care centers, both of which serve as referral centers for patients with advanced HCM. The study environment may lead to selection bias and limit the generalizability of our findings to patients with less advanced diseases or those treated at smaller centers. Second, misclassification of the outcome was possible, although the prospective study design and rigorous event adjudication with multiple attending cardiologists minimize the possibility. Third, the possibility of informative censoring could not be excluded. Fourth, the variable importance plot is susceptible to the initial seed of random forest models and may display slightly different values depending on the seed. Fifth, plasma samples have only been collected at a single time point. Repeated measurements could be informative. Sixth, since the plasma samples were taken in outpatient settings, the results may be different if these samples were obtained when patients were hospitalized for cardiac events. Seventh, by the nature of the machine learning-based model, the cutoff threshold for each adipokine could not be calculated. Eighth, analysis of adipose samples would enhance the findings from this study, however, adipose tissue was not available. Finally, the number of events observed in the current study was not very large and the internal validity might be limited. The study participants were predominantly white. Future studies with larger and more diverse samples are warranted.

CONCLUSIONS

The present study demonstrated the ability of a plasma adipokine-based model to predict future cardiac death in patients with HCM. The prediction model from the training set performed well in predicting cardiac death in the independent test set from another institution, exhibiting a high level of reproducibility and generalizability. The high-risk patients determined by the plasma adipokine-based model showed a significantly worse prognosis than low-risk patients in the test set. The present study serves as the first step to develop of a better prediction model for identifying patients with HCM who are at high risk of cardiac death. The prediction model may help physicians identify patients with HCM who are at high-risk of cardiac death, which in turn would serve as the first step to facilitate more frequent surveillance and improve prognosis.

Highlights:

• Plasma adipokine levels are known to be associated with cardiac remodeling in HCM

• The prognostic value of plasma adipokines in HCM is unknown

• A prediction model for cardiac death in HCM was developed using ten plasma adipokines

• This model will help to specify patients with HCM at risk of future cardiac death

Nonstandard Abbreviations and Acronyms:

AF

atrial fibrillation

AUC

area under the receiver-operating-characteristic curve

CMR

cardiac magnetic resonance

CRTP

C1q/tumor necrosis factor-related protein

HCM

hypertrophic cardiomyopathy

ICD

implantable cardioverter-defibrillator

LV

left ventricular

NSVT

non-sustained ventricular tachycardia

NYHA

New York Heart Association

ROC

receiver-operating-characteristic

SCD

sudden cardiac death