A mathematical model using routine serum biomarkers for predicting of pulmonary hypertension in patients with left heart disease

Background: Pulmonary hypertension (PH) insidiously occurs in patients with left heart disease (LHD), but its diagnosis is often delayed due to the lack of specic symptoms. This observational study aimed to identify a biomarker panel for the noninvasive detection of PH in LHD patients. Methods: Patients with LHD were consecutively recruited and analyzed for correlations between the pulmonary artery systolic pressure (PASP) and levels of routine serum biomarkers, and 100 age-matched subjects served as healthy controls. The diagnostic accuracy of biomarkers was assessed by receiver operating characteristic (ROC) curve analysis. Forward stepwise binary logistic regression was performed to determine the optimal combination of biomarkers for predicting PH. Results: A total of 426 LHD patients were divided into the LHD group (n=216, PASP <35 mmHg) and PH-LHD group (n=210, PASP ≥ 35 mmHg). The ranges of routinely examined biomarkers were signicantly different between the two groups. The levels of biomarkers, including brain natriuretic peptide (BNP), total bilirubin (TB), direct bilirubin (DB), indirect bilirubin (IB), red cell distribution width (RDW), uric acid (UA), and cystatin c (Cys-C), were higher, while high-density lipoprotein (HDL) levels were lower in the PH-LHD group than in the LHD and healthy control groups. Positive correlations were found between the PASP and levels of BNP, Cys-C, UA, bilirubin, and RDW, while a negative correlation was found between the PASP and HDL level. A mini program based on the formula “P=1/(1+exp(6,314-0.898×[BNP]/1000-0.146×[DB]-0.318×[RDW]))” was developed using forward stepwise binary regression for calculating the probability of PH. The combination of BNP, DB, and RDW (cutoff, 0.256) demonstrated a better predictive value than did individual biomarkers, with a sensitivity of 0.822 and specicity increased relative to those of individual biomarkers. Our study indicated that a multiple biomarker approach is helpful for the screening and identication of PH in LHD patients. Moreover, using the formula generated by the bivariate logistic regression analysis, we developed a mini program to calculate the predicted probability of PH based on BNP, DB and RDW, which might be a simple tool to help recognize and screen PH in LHD patients in clinical practice.

Circulating biomarkers have been proposed as potentially noninvasive and objective diagnostic and prognostic markers, as well as markers of response to therapy. In contrast to echo, biomarkers re ect a disease-associated molecular change in bodily tissues and uids [3]. Serum biomarkers involving endothelial, in ammatory, and right ventricular (RV) dysfunction, as well as metabolic mediators, have been studied in pulmonary arterial hypertension (PAH) [4,5]. However, PH has a highly complex etiology. Unlike idiopathic pulmonary arterial hypertension (IPAH), which originates from a genetic or family background [6], PH-LHD usually develops in LHD in response to a passive backward transmission of lling pressures [7]. The biomarker characteristics of PH-LHD remain largely unknown. To nd a reliable marker that could help to simply screen PH in LHD patients using routinely performed blood tests, in this study, we investigated the difference between routinely examined serum biomarkers in LHD patients with and without PH and analyzed their predictive value for PH in LHD patients.

Patients and study design LDH patients of the cardiovascular department of the Second A liated Hospital of Chongqing Medical
University and Army University Xinqiao Hospital were consecutively studied from January 2014 to July 2019. According to the guidelines on the classi cation of PH, eligible LHD subjects included patients with heart failure with reduced ejection fraction (HFrEF), heart failure with preserved ejection fraction (HFpEF), and cardiac valve disease [8,9]. The etiology of heart failure includes coronary artery disease, hypertension and cardiomyopathy, and there is no evidence for connective tissue disease, congenital heart disease, diabetes, or lung disease to be included in the etiology of heart failure. Cardiac valve disease includes diseases affecting the aortic and mitral valves. Patients in PH groups 1, 3, 4, and 5 were excluded. The pulmonary artery systolic pressure (PASP) was examined by echocardiography on the day of admission. We nally enrolled 426 cases with LHD. Based on the PASP, the subjects were further divided into the LHD control group (PASP < 35 mmHg) and the PH-LHD group (PASP ≥ 35 mmHg) based on the Guidelines for the Echocardiographic Assessment of the Right Heart in Adults [10]. Another 100 age-and sex-matched subjects served as healthy controls. After informed consent was obtained, peripheral venous blood was drawn from all the enrolled subjects for routine clinical examination. The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki and has been approved by the institutional ethical committee of the Second A liated Hospital of Chongqing Medical University, Xinqiao Hospital of Army University and Shenzhen University General Hospital.

Clinical measurements
The levels of uric acid (UA), creatinine, bilirubin, cystatin C (Cys-C), and lipoproteins (high-density lipoprotein [HDL] and low-density lipoprotein [LDL]) were assayed using a Beckman Coulter AU5800 system (USA). The red cell distribution width (RDW) was assayed in a routine blood analysis using a Sysmex XN-20 analyzer (Sysmex, Japan). The levels of brain natriuretic peptide (BNP) and troponin I (TnI) were assayed by a ReLIA Analyzer SSJ-2 using the reagents provided by the manufacturer (ReLIA Biotechnologies, USA).
Echocardiographic recordings were obtained in all subjects using a GE VIVID7 (GE Healthcare, Norway) color Doppler ultrasound system. Data were stored on a DVD and analyzed by two independent investigators. The simpli ed Bernoulli equation describes the relationship between tricuspid regurgitation velocity and the peak pressure gradient of tricuspid regurgitation, where the peak pressure gradient of tricuspid regurgitation = 4 × (tricuspid regurgitation velocity). The PASP was calculated as the tricuspid regurgitation pressure gradient plus the estimated right atrial pressure. The left ventricular ejection fraction (LVEF) was calculated using the Teichholz method and the modi ed biplane Simpson method, as previously described [11].
Blood pressure (BP) and heart rate (HR) were measured with an Omron HEM-6200 monitor (Japan).

Statistical analysis
The statistical analysis was performed using SPSS software (IBM Corp.,version 20.0). All continuous variables are summarized as medians and interquartile ranges, and all categorical variables are expressed as proportions. Signi cant differences in the biomarker levels between the groups were determined using a Kruskal-Wallis nonparametric test followed by pairwise multiple comparisons. Correlations between the PASP and biomarker levels were assessed by calculating Spearman correlation coe cients. The diagnostic accuracy of serum biomarker levels was assessed by receiver operating characteristic (ROC) curve analysis. Forward stepwise binary logistic regression analyses were performed to determine the optimal combination of biomarkers for diagnosing PH-LHD. For the regression analyses, the liberal entrance criterion was 0.05 and the liberal exit criterion was 0.10; p < 0.05 was considered statistically signi cant.

Clinical characteristics
The study sample comprised 100 healthy control subjects, 216 LHD control patients, and 210 PH-LHD patients. The average age, sex, and BMI were comparable between the three groups. The proportions of LHD in the patient groups were as follows: coronary artery disease (CAD, 44.91%), hypertension (14.35%), cardiomyopathy (21.30%), and cardiac valve disease (19.44%) in the LHD group; and CAD (28.74%), hypertension (12.57%), cardiomyopathy (27.05%), and cardiac valve disease (31.66%) in the PH-LHD group. The median PASP was 26.00 mmHg (interquartile range, [IQR], 10.00 to 33.00 mmHg) in the healthy controls, 26.00 mmHg (IQR, 10.00 to 34.00 mmHg) in the LHD control group, and 62.00 mmHg (IQR, 55.00 to 70.25 mmHg) in the PH-LHD group. The pulmonary artery diameter (PAD) was 22.00 mm (interquartile range [IQR], 21.00 to 23.80 mm) in the healthy controls, 23.00 mm (IQR, 22.00 to 25.00 mm) in the LHD control group, and 26.00 mm (IQR, 24.00 to 29.00 mm) in the PH-LHD group. Compared with those in the LHD control and healthy control groups, the PASP and PAD were signi cantly increased, the left atrium and ventricle were enlarged, and cardiac function was decreased in the PH-LHD group. Furthermore, compared with those in the LHD group, the percentage of patients with New York Heart

Biomarker levels
The levels of biomarkers were signi cantly different between the LHD control and PH-LHD groups. Levels of biomarkers, including BNP, total bilirubin (TB), direct bilirubin (DB), indirect bilirubin (IB), RDW, UA, and Cys-C, were signi cantly higher in the PH-LHD group than in the LHD group. The level of HDL cholesterol was signi cantly lower in the PH-LHD group than in the LHD group. No differences were found in the TnI levels between the three groups ( Table 2). Results are expressed as medians (interquartile ranges).

Correlation between biomarkers and PASP
The levels of the biomarkers BNP, TB, DB, IB, RDW, UA, and Cys-C, but not HDL, were positively correlated with the PASP. No correlation was found between TnI and PASP (Fig. 2).

ROC curve analysis
The diagnostic accuracies of each biochemistry marker determined by the ROC curve analysis are shown in Table 3 (Fig. 3, Table 3).

Discussion
The potential of biomarker combinations is currently of considerable interest in the diagnosis of PH. Our study showed that the levels of routinely examined serum biomarkers were signi cantly different between the LHD control and PH-LHD patients. The correlation analysis showed positive correlations between the PASP and levels of BNP, Cys-C, UA, bilirubin, and RDW, while a negative correlation was found between PASP and HDL. The combination of BNP, bilirubin, and RDW was found to exhibit the best diagnostic accuracy for PH in LHD patients.
LHD involving HFpEF, HFrEF, and heart valves is the most frequent cause of PH. PH-LHD belongs to group 2 PH, with the following hemodynamic characteristics: mean pulmonary arterial pressure (mPAP) > 20 mmHg and pulmonary capillary wedge pressure (PCWP) > 15 mmHg [8,9]. PH insidiously occurs in LHD, and its diagnosis is often delayed due to the lack of speci c symptoms. RHC is the gold standard in the diagnosis of PH, but it is often declined by LHD patients due to its invasive nature. Transthoracic echocardiography remains a widely available tool in cardiology clinics worldwide. In our study, we consecutively enrolled 426 LHD patients and divided them into the LHD group and the PH-LHD group. With regard to etiology, we found that the ratio between cardiomyopathy and valve disease prevalence was higher than that between CAD and hypertension prevalence in the PH-LHD group. However, the ratio between CAD and hypertension prevalence was higher than that between cardiomyopathy and valve disease prevalence in the LHD control group. Similar to previous reports [7,12], our study showed that compared with that in the LHD control group, cardiac function, assessed by the NYHA grade and LVEF, was signi cantly decreased in the PH-LHD group. Unlike LHD control patients, patients with PH-LHD often exhibited enlargement and remodeling of the right atrium and right ventricle. Our study indicated that patients with LHD due to cardiomyopathy and valve disease, especially those with severe cardiac function damage, are prone to PH.
Circulating biomarkers have been investigated as potentially noninvasive and objective measures for the diagnosis and prognosis of diseases in daily clinical practice. In our study, we found that the levels of routinely examined biomarkers were signi cantly different between the PH-LHD and LHD control groups. The levels of BNP, TB, DB, IB, RDW, UA, and Cys-C were signi cantly higher and the HDL level was lower in the PH-LHD group than in the LHD group.
BNP belongs to the natriuretic peptide family and has potent vasodilator, hypertrophic, and proin ammatory properties. Increasing evidence has shown that circulating levels of BNP are correlated with the mPAP and pulmonary vascular resistance (PVR) in patients with PAH [13,14]. Our study showed that BNP was positively correlated with PASP. BNP exerts myriad cardiovascular effects [15]. The sensitivity of BNP alone (cutoff, 200.00 pg/mL) in predicting PH was high (0.874), but its speci city was low (0.630).
Patients with right heart failure often exhibit elevated bilirubin levels because of congestion and low perfusion. Takeda et al. [16] enrolled 37 patients with PAH and followed them for a median of 635 days. They found that elevated serum bilirubin was a risk factor for death. In our study, the levels of IB and DB were signi cantly higher in the PH-LHD group than in the LHD control group. Both IB and DB were positively correlated with PASP.
UA is the metabolite of purine nucleotide, and an elevated UA level represents a higher activity of oxidative metabolism. Recent evidence suggests that UA inhibits acetylcholine-mediated vasodilation by acting on the vascular endothelium [17]. In isolated porcine pulmonary artery segments, UA reduced nitric oxide levels in pulmonary arterial endothelial cells and inhibited acetylcholine-induced vasodilation [18]. Previous studies have shown that serum UA levels are associated with the severity of IPAH and ventricular dysfunction [19,20]. Cys-C, a novel marker of renal function, predicts left heart failure and cardiovascular mortality. Fenster et al. [21] found that the Cys-C level accurately correlates with RV pressure, function, and morphology. In our study, although the levels of UA and Cys-C were within normal ranges in both the LHD control and PH-LHD groups, the levels were signi cantly different between the groups. Serum levels of UA and Cys-C were signi cantly higher in the PH-LHD group than in the LHD group. Additionally, both UA and Cys-C were positively correlated with PASP.
The HDL level is a strong, independent, inverse predictor of cardiovascular disease, and it is not in uenced by statins or diet. Previous studies by Zhao et al. [22] and Heresi et al. [23] showed that serum HDL cholesterol levels were signi cantly lower in patients with IPAH than in control individuals, and the levels were decreased in proportion to the severity of the World Health Organization functional class. In our study, the HDL level in the PH-LHD group was signi cantly lower than that in the LHD control group and was negatively correlated with the PASP. HDL alone demonstrated a sensitivity of 0.729 and speci city of 0.526 in predicting PH in LHD patients.
The RDW, which is measured as part of the full blood count in standard hospital analyses and quanti es the variability in red blood cell size, had been shown to be valuable for assessing survival and prognosis in cardiovascular disease patients. A study by Rhodes et al. [24] demonstrated that the RDW outperforms the BNP level in predicting survival in IPAH patients. In our study, the RDW was signi cantly higher in the PH-LHD group than in the LHD control group and was positively correlated with the PASP. The RDW alone demonstrated a sensitivity of 0.690 and speci city of 0.689 in predicting PH in LHD patients.
The ideal serum biomarker for use in PH remains elusive despite recent extensive work. A diseasespeci c, clinically useful, and noninvasive biomarker for PH does not exist, partly due to the heterogeneity and complex etiology of PH. Our study showed that the ranges of routinely examined biomarkers, including BNP, bilirubin, UA, Cys-C, and RDW, although not speci c, were signi cantly different. Using a bivariate logistic regression model, we found that the combination of BNP, DB, and RDW demonstrated a higher AUC and Youden index, and both the sensitivity and speci city increased relative to those of individual biomarkers. Our study indicated that a multiple biomarker approach is helpful for the screening and identi cation of PH in LHD patients. Moreover, using the formula generated by the bivariate logistic regression analysis, we developed a mini program to calculate the predicted probability of PH based on BNP, DB and RDW, which might be a simple tool to help recognize and screen PH in LHD patients in clinical practice.
There are several limitations to this study. One limitation is the inaccuracy of the noninvasive methods in assessing pulmonary artery pressure. However, although more accurate, invasive RHC measurements cannot be obtained in healthy and LHD controls due to ethical considerations. Noninvasive echocardiography is widely used in clinical practice. Moreover, according to the guidelines for PH diagnosis, the diagnostic accuracy of PH by echocardiography is close to that by RHC, especially when PASP > 50 mmHg. The second limitation is the limited sample number. An expanded random sample across all genders and ages may be more representative. We also need a large cohort to verify the prediction value for PH-LHD using the mathematical model.

Conclusion
LT and PJ contributed equally to this work and share the rst authorship. QZ and XW contributed equally to this work and share the corresponding authorship. QZ and XW contributed to the study concept and design. Material preparation and data collection and analysis were performed by PJ and LT. The draft of the manuscript was written by XW. All authors read and approved the nal manuscript.