Predictive value of monocytes for coronary heart disease in Chinese adults: a population-based cohort study

Abstract

Objectives

The development of simple tools to identify individuals at high risk of coronary heart disease (CHD) would enable rapid implementation of preventive measures. This study was designed to construct predictive models and scoring systems for CHD using monocyte count and its ratio to high-density lipoprotein cholesterol (HDL-C) (MHR).

Design

Population-based prospective cohort study.

Setting

The Shanghai Suburban Adult Cohort and Biobank (SSACB).

Participants and outcome measures

This prospective study included 44 013 CHD-free participants of the SSACB. The Songjiang subcohort served as the training set, in which three predictive models and corresponding scoring systems were developed with monocyte count or MHR using stepwise Cox regression. The models and algorithms were tested internally using 10-fold cross-validation and externally in the Jiading subcohort. Discriminations were assessed based on area under the curve (AUC) values, while calibrations were evaluated using the Hosmer-Lemeshow goodness-of-fit test.

Results

During a mean follow-up period of 4.8 years, 883 CHD events occurred, with an incidence of 415.7/100 000. Monocyte count and MHR were significantly associated with the risk of CHD. The constructed model incorporating monocyte count (Model 2) achieved AUC values of 0.746 (0.726, 0.766) for 4-year CHD prediction in the training set, 0.746 (0.690, 0.796) in the cross-validation, and 0.717 (0.674, 0.761) in the external validation, comparable to the models including HDL-C (model 1) or MHR (model 3). Calibration plots demonstrated good agreement between predicted and actual probabilities. Similar results were observed for the corresponding scoring algorithms.

Conclusions

The monocyte-based model is a simple, low-cost and well-calibrated risk-stratification tool for CHD. However, the declined discrimination in external validation indicates limited generalisability. Prospective multicentre validation and recalibration are therefore warranted before clinical adoption.

STRENGTHS AND LIMITATIONS OF THIS STUDY

• This study was based on a large-scale prospective cohort study with a well-documented and standardised protocol for participant recruitment, data collection and biochemistry assays, with strict quality control ensuring high-quality data in this analysis.

• The incident coronary heart disease events were obtained from the complete and accurate clinical data and the disease reporting systems, which minimised the potential recall bias and reporting bias on the outcome.

• The study participants were all natives in Shanghai, which may limit the generalisability of the models and scoring systems in other populations.

• There may exist unmeasured predictors that should be included in the models.

• We just focused on monocyte count and monocyte count to high-density lipoprotein cholesterol ratio, but not other biomarkers for chronic inflammation that would improve the performance of the models and scoring systems.

Introduction

Coronary heart disease (CHD) is a leading cause of death globally.¹ In China, the incidence and mortality of CHD have increased during the past decades, resulting in 11.4 million prevalent cases in 2019.^{2 3} The heavy disease burden of CHD highlights the importance of prevention and control of the disease.

CHD events are often triggered by acute rupture and/or erosion of unstable atherosclerotic plaque.⁴,⁶ It is well-recognised that atherosclerosis is a common pathophysiological process in humans,⁷ which is induced by abnormal lipid metabolism and chronic inflammation.⁸,¹⁰ Monocytes play an important role in chronic inflammation and involve the metabolisms of lipids. After binding to adhesion molecules and migrating to the vascular endothelium, monocytes mature into macrophages,¹⁰ which engulf oxidised low-density lipoprotein cholesterol (LDL-C) and become foam cells.¹¹ Monocyte count has been treated as an index for chronic inflammation in arterial occlusive diseases¹² and was shown to be positively associated with atherosclerosis and cardiovascular events.¹³,¹⁷ High-density lipoprotein cholesterol (HDL-C), sometimes called ‘good’ cholesterol, functions as a reversal factor during the process of atherosclerosis and elicits a protective effect against oxidation of LDL-C.¹⁸ It could transport excess cholesterol from macrophages and reduce lipid accumulation¹⁹ and was also found to suppress the proliferation of stem cells, reduce monocytes and directly act on monocytes to inhibit the process of atherosclerosis.^{20 21} Therefore, monocyte count to HDL-C ratio (MHR) was developed to demonstrate the balance of serum lipids and chronic inflammation comprehensively.^{22 23}

Despite the fact that chronic inflammation and abnormal lipid metabolism is of equal importance in the development of CHD, the biomarkers for chronic inflammation have not been included in any established predictive models or risk scoring systems for CHD,²⁴,²⁷ whereas serum lipids (mainly HDL-C) have been used to predict the risk of CHD in multiple populations.^{28 29} MHR has never been used to predict the incidence of CHD in general populations either, although its application in clinical practice to predict the recurrence of cardiovascular events among patients with CHD.³⁰,³³

In this study, we explored the associations of monocyte count and MHR with the risk of CHD in Chinese populations based on a prospective cohort study, and established predictive models and risk scoring systems of CHD using the two biomarkers for their low cost to measure and cost-effective nature.

Materials and methods

Study design and participants

This study was based on the Shanghai Suburban Adult Cohort and Biobank (SSACB),³⁴ a prospective cohort study including the Songjiang subcohort recruited during June 2016 and December 2017 and the Jiading subcohort enrolled during July to September of 2017. Both subcohorts followed a similar multistage stratified clustered sampling to enrol study participants. For the Songjiang subcohort, four streets and towns were selected in convenience from a total of 17 urban streets and rural towns in Songjiang district of Shanghai by geographic locations and economic levels. Then, one-third of communities and villages in each street or town were randomly selected, resulting in a total of 47 communities and villages included. All people aged 20–74 years and living in the selected communities or villages for at least 5 years were recruited for the study. After excluding pregnant or breastfeeding women, individuals with cognitive or psychiatric disorders and those suffering from serious illness that precluded completion of the baseline survey, a total of 36 404 subjects were finally recruited. For the Jiading subcohort, a total of 10 042 eligible participants were randomly selected from 53 communities or villages that belong to 3 of 10 streets and towns in Jiading district of Shanghai.

In this study, we further excluded 2063 prevalent CHD cases at baseline (1489 from the Songjiang and 574 from the Jiading subcohorts). We also excluded 370 subjects (347 from the Songjiang and 23 from the Jiading subcohorts) due to incomplete biochemical measurement data. As shown in online supplemental figure 1, a total of 44 013 eligible subjects were included in this analysis. The Songjiang subcohort (34 568 subjects) was regarded as a training set to construct and cross-validate predictive models, while the Jiading subcohort (9445 subjects) was treated as an external validation set.

Data collection at baseline

Baseline data were collected through in-person interviews, physical examinations and biochemical assays. The survey data were collected using a structured questionnaire by trained healthcare staff, which included demographic characteristics (age, sex, educational level, residential area, etc, previous diseases (diabetes mellitus, hypertension, CHD, stroke, dyslipidaemia, etc), CHD in first-degree relatives and lifestyle factors (tobacco use, alcohol consumption, regular exercise, etc). Tobacco use was defined as smoking at least one cigarette per day for more than 6 months, while alcohol consumption referred to drinking at least three times per week for more than 6 months. Physical activity was assessed using the short form of the International Physical Activity Questionnaire, in which regular exercise referred to engaging leisure-time activities per week, with at least 10 min each time.

At the interview, physical examinations were performed for each participant by licensed physicians in local community healthcare centres. All subjects were asked to keep light clothing and bare feet when measuring height, weight, waist circumference (WC) and hip circumference. Body mass index (BMI) was calculated as weight (kilograms) divided by height (metres) squared using measured values. Blood pressure was measured on two arms twice after a rest for at least 5 min using a digital sphygmomanometer. Prevalent hypertension referred to those diagnosed with hypertension by physicians according to the 1999 WHO criteria, using antihypertension agents, or with measured systolic blood pressure (SBP) higher than 140 mm Hg and/or diastolic blood pressure (DBP) higher than 90 mm Hg. Considering the inconvenient measurement of blood glucose levels, we defined the prevalent diabetes as those diagnosed with type 2 diabetes by physicians according to the WHO 1999 criteria, or those using hypoglycaemic agents.

All participants were asked to fast overnight for at least 8 hours before donating blood samples. Blood samples were shipped to the Di-An Medical Laboratory Center (Occupational health and safety management system certification: 15/21S0537R00) for testing within 6 hours. Total cholesterol (TC), triglyceride (TG), HDL-C and LDL-C were measured using enzyme colourimetry (Roche COBAS C501 automatic biochemical analyser). A routine blood test was conducted using electrical impedance method (SYSMEX XS-500i, SYSMEX XN-1000).³⁵ MHR was calculated as monocyte count divided by measured HDL-C level.

Follow-up survey

All participants were followed up until 31 January 2022 for incident CHD and all-cause deaths through a record linkage with the local and municipal medical record systems and the Vital statistics, minimising the potential recall bias and reporting bias on the outcome. The incident CHD referred to the first diagnosed coronary atherosclerosis, angina and myocardial infarction (MI), which were obtained from the hospital information systems as International Classification of Disease (ICD) code of I20–I25 according to the ICD 10th revision.^{36 37} Person-years were calculated from the date of recruitment to the date of CHD diagnosis, death, loss to follow-up or the end of follow-up (31 January 2022), whichever came first.

Statistical analysis

Continuous variables were presented as mean±SD or median (IQR) and were compared across groups using t-tests or Wilcoxon tests. Categorical variables were described as count (percentage) and compared using χ² tests. Restricted cubic spline was applied to fit the potential nonlinear relationship of monocyte count and MHR with the risk of CHD.

Prior to the model development, univariate analyses were performed in the training set to select potential risk predictors for CHD from the variables collected, which included age, sex, residential area (urban/rural), current smoking, current drinking, prevalent diabetes, family history of CHD, HDL-C, TC, monocyte count, MHR, WC, blood pressure and use of antihypertension agents. The statistically significant variables (p values <0.05) in the univariate analyses were included in the multivariate analysis. A stepwise Cox regression model was used to develop the final models with Alpha-to-Enter and Alpha-to-Remove significance levels of 0.05. Model 1 was constructed using the significant conventional risk factors including HDL-C but not monocyte count and MHR. We further replaced the variable ‘HDL-C’ in model 1 with ‘monocyte count’ to construct model 2, and replaced with ‘MHR’ to construct model 3, aimed to evaluate the risk predictive values of monocyte count alone and jointly with HDL-C. Potential interactions between age and other variables were also considered.

The coefficient for each risk predictor was transferred into a point value, with each point equivalent to the risk of CHD associated with each 1-year increase in age (ie,the coefficient of age).³⁸ The risk score was then created for each subject by summing up the point values for all predictors in the final model.

The discrimination of the developed models and corresponding scoring algorithms was evaluated using areas under the receiver operating characteristic curves (AUCs) and their 95% CIs calculated with perturbation resampling for 100 repetitions,³⁹ while the calibrations, the agreement of predicted 5-year and 4-year risks with observed outcomes, were assessed using Hosmer-Lemeshow goodness-of-fit test⁴⁰ visualised via calibration plots. Model calibration-in-the-large was further assessed with Spiegelhalter’s Z-test. Net reclassification improvement (NRI) was calculated to quantify the incremental predictive performance between models, with bootstrap resampling used to derive its 95% CI.

The performance of the models and the scoring algorithms was tested internally using 10-fold cross-validation technique and externally in the Jiading subcohort.³⁹ As all the subjects in the Jiading subcohort were followed up less than 5 years, we evaluate the performance of the models and the scoring algorithms in predicting 4-year risk of CHD in the external testing set. To demonstrate the calibrations of three scoring algorithms by the risk of CHD, we further classified the participants in the training and validation sets into three categories according to their predicted risk of CHD (low risk: <5.0%, moderate risk: 5.0%–7.4%, high risk: ≥7.5%), which were clinically meaningful cut points.²⁴ Calibration was evaluated using the χ² statistic to quantify the agreement between predicted probabilities and observed event rates.

All analyses were performed using SAS V.9.4 and R 4.2.1. Two-sided p<0.05 was considered statistically significant.

Patient and public involvement

This study was based on SSACB, and the endpoints were derived through a record linkage with the local and municipal medical record systems and the Vital statistics. Therefore, the patients and the public were not involved in the research design, conduct, reporting or dissemination plans.

Results

Demographic characteristics of study participants

After a median follow-up of 4.8 years (IQR: 4.6–5.4), a total of 726 incident CHD cases were identified from the Songjiang subcohort, with an incidence of 427.6 per 100 000 person-years. In the Jiading subcohort, a total of 157 events were obtained during a median follow-up of 4.6 years (IQR: 4.5–4.6), with an incidence of 370.9 per 100 000 person-years.

In both the Songjiang and the Jiading subcohorts, the incident CHD cases were older, less educated, more likely to be diagnosed with diabetes, hypertension, stroke, dyslipidaemia and other chronic non-communicable diseases, and had a higher SBP, BMI, WC, weight, monocyte count and MHR compared with the non-CHD subjects (all p values <0.05) (table 1). The significant differences between the CHD cases and the non-cases in residential area, smoking, alcohol drinking, TG, LDL-C, HDL-C and DBP were observed only in the Songjiang subcohort. No significant difference was observed regarding regular exercise and family history of CHD in both subcohorts (all p values >0.05).

Table 1. Demographic characteristics of the study participants.

	The Songjiang subcohort		P value	The Jiading subcohort		P value
	CHD cases (n=726)	Non-cases (n=33 842)		CHD cases (n=157)	Non-cases (n=9288)
Demographics
Age (years)	63.3±7.2	55.8±11.3	<0.001	62.1±8.0	56.2±10.5	<0.001
Sex			<0.001			0.005
Men	356 (49.0)	13 675 (40.4)		80 (51.0)	3700 (39.8)
Women	370 (51.0)	20 167 (59.6)		77 (49.0)	5588 (60.2)
Educational level			<0.001			0.023
Primary school or below	495 (68.2)	15 409 (45.5)		56 (35.7)	2497 (26.9)
Junior high school	188 (25.9)	12 149 (35.9)		74 (47.1)	4554 (49.0)
Senior high school or above	43 (5.9)	6284 (18.6)		27 (17.2)	2237 (24.1)
Residential area			<0.001			1.000
Urban	205 (28.2)	14 507 (42.8)		0 (0.0)	0 (0.0)
Rural	521 (71.8)	19 335 (57.1)		157 (100.0)	9288 (100.0)
Lifestyle factors
Current smoking	172 (23.7)	6739 (19.9)	0.012	44 (28.0)	2174 (23.4)	0.176
Current alcohol drinking	118 (16.3)	4216 (12.5)	0.002	25 (15.9)	1074 (11.6)	0.091
Regular exercise	246 (34.1)	10 741 (31.8)	0.199	66 (42.0)	4059 (43.7)	0.677
Disease history
Diabetes mellitus	131 (18.0)	2555 (7.6)	<0.001	26 (16.6)	739 (8.0)	<0.001
Hypertension	539 (74.2)	17 742 (52.4)	<0.001	116 (73.9)	5172 (55.7)	<0.001
Stroke	51 (7.0)	909 (2.7)	<0.001	20 (12.7)	429 (4.6)	<0.001
Dyslipidaemia	123 (16.9)	3238 (9.6)	<0.001	33 (21.0)	1240 (13.4)	0.005
Family history of CHD	41 (5.7)	1803 (5.3）	0.705	17 (10.8)	915 (9.9)	0.684
Medications
Anti-hypertensive agents	361 (49.7)	8827 (26.1)	<0.001	95 (60.5)	3330 (35.9)	<0.001
Biochemistry assays
TC (mmol/L)	4.99 (4.26, 5.62)	4.88 (4.31, 5.49)	0.070	4.84 (4.35, 5.42)	4.79 (4.23, 5.42)	0.535
TG (mmol/L)	1.49 (1.05, 2.11)	1.34 (0.97, 1.91)	<0.001	1.56 (1.18, 2.24)	1.54 (1.10, 2.20)	0.325
HDL-C (mmol/L)	1.35 (1.09, 1.55)	1.40 (1.17, 1.62)	<0.001	1.31 (1.11, 1.48)	1.35 (1.14, 1.57)	0.177
LDL-C (mmol/L)	2.87 (2.23, 3.39)	2.74 (2.23, 3.27)	0.025	2.71 (2.20, 3.19)	2.61 (2.12, 3.14)	0.408
Monocyte count (109/L)	0.38 (0.30, 0.48)	0.35 (0.28, 0.43)	<0.001	0.41 (0.31, 0.49)	0.37 (0.30, 0.46)	0.010
MHR	0.29 (0.21, 0.40)	0.25 (0.19, 0.34)	<0.001	0.33 (0.22, 0.44)	0.28 (0.21, 0.38)	0.006
Body measurements
SBP (mm Hg)	139.1±19.6	133.3±19.4	<0.001	136.4±19.4	131.0±19.0	0.001
DBP (mm Hg)	82.0±10.3	80.0±10.5	<0.001	78.8±10.2	78.5±10.9	0.727
Weight (kg)	63.9±10.4	62.4±10.5	<0.001	65.7±10.6	62.0±10.7	<0.001
Height (cm)	159.1±8.1	160.0±8.0	<0.001	161.6±8.4	161.4±8.3	0.726
BMI (kg/m2)	25.2±3.4	24.4±3.3	<0.001	25.1±3.3	23.8±3.3	<0.001
WC (cm)	84.8±9.2	81.5±9.4	<0.001	86.8±9.3	82.2±9.3	<0.001

^*Data presented as mean±SD or median (IQR) for continuous variables, and count (%) for categorical variables.

BMI, body mass index; CHD, coronary heart disease; DBP, diastolic blood pressure; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; MHR, monocyte count to HDL ratio; SBP, systolic blood pressure; TC, total cholesterol; TG, triglyceride; WC, waist circumference.

Monocyte count and MHR with the risk of CHD

After adjusting for age and sex, both monocyte count and MHR were positively associated with the risk of CHD in the Songjiang subcohort (p for trends <0.001). As illustrated in figure 1, the risk of CHD increased in a linear fashion with rising monocyte count (p for non-linear test=0.749) and MHR level (p for non-linear test=0.064). The two biomarkers also performed well alone in predicting the risk of CHD in the population, with an AUC of 0.614 (0.593, 0.635) for monocyte count and 0.621 (0.601, 0.642) for MHR in predicting 5-year risk of CHD. Monocyte count and MHR level were also found to increase with elevated risk stratification of the subjects either classified by the China-PAR score or by the FRS (all p values for trend <0.001) (online supplemental table 1).

Figure 1. Relationship of monocyte count and MHR with the incidence of CHD and performance of the two indices in predicting the risk of CHD. (A) RCS curve of monocyte count with the risk of CHD, adjusting for age and sex; (B) RCS curve of MHR with the risk of CHD, adjusting for age and sex; (C) ROC curve for monocyte count alone in predicting incident CHD; (D) ROC curve for MHR alone in predicting incident CHD. AUC, area under the curve; CHD, coronary heart disease; HDL-C, high-density lipoprotein cholesterol; MHR, monocyte count to HDL-C ratio; RCS, restricted cubic spline; ROC, receiver operating characteristic.

Construction and validation of CHD predictive models

As shown in table 2, age, sex, residential area, current smoking, current drinking, history of diabetes, WC, HDL-C, blood pressure, monocyte count and MHR were significantly associated with the risk of CHD in univariate analysis and were used to develop the three predictive models using multivariable Cox regression. In model 1, age, residential area, history of diabetes, HDL-C and blood pressure were included as potential risk predictors. In model 2, WC was additionally retained and HDL-C was replaced by monocyte count. In model 3, MHR was included in instead of HDL-C based on model 1. There was no interaction between age and other variables.

Table 2. Cox regression analysis for CHD-related factors in the training set.

Risk factors	Univariate analysis		Model 1		Model 2		Model 3
	β (95% CI)	P value	β (95% CI)	P values	β (95% CI)	P values	β (95% CI)	P values
Age (years)	0.09 (0.08 to 0.10)	<0.001	0.08 (0.07 to 0.09)	<0.001	0.07 (0.06 to 0.08)	<0.001	0.08 (0.07 to 0.09)	<0.001
Sex
Women	ref		–	–	–	–	–	–
Men	0.36 (0.21to 0.50)	<0.001	–	–	–	–	–	–
Residential area
Rural	ref		ref		ref		ref
Urban	−0.62 (-0.78 to 0.46)	<0.001	−0.47 (-0.63 to 0.30)	<0.001	−0.46 (-0.63 to 0.29)	<0.001	−0.47 (-0.64 to 0.31)	<0.001
Current smoking	0.22 (0.05 to 0.39)	0.011	–	–	–	–	–	–
Current drinking	0.32 (0.12 to 0.51)	0.002	–	–	–	–	–	–
Diagnosis of diabetes	0.99 (0.81 to 1.18)	<0.001	0.62 (0.42 to 0.81)	<0.001	0.63 (0.43 to 0.82)	<0.001	0.60 (0.41 to 0.79)	<0.001
Family history of CHD	0.05 (-0.27 to 0.36)	0.761	–	–	–	–	–	–
HDL-C (mmol/L)	−0.49 (-0.71 to 0.27)	<0.001	−0.48 (-0.69 to 0.26)	<0.001	–	–	–	–
TC (mmol/L)	0.07 (-0.01 to 0.15)	0.060	–	–	–	–	–	–
WC (cm)	0.04 (0.03 to 0.04)	<0.001	–	–	0.01 (0.00 to 0.02)	0.026	–	–
Blood pressure*
Normal
Untreated	ref	<0.001	ref		ref		ref
Use of antihypertensive agents	1.21 (0.99 to 1.42)	<0.001	0.50 (0.28 to 0.72)	<0.001	0.47 (0.25 to 0.69)	<0.001	0.49 (0.27 to 0.71)	<0.001
High
Untreated	0.50 (0.29 to 0.71)	<0.001	0.21 (0.00 to 0.43)	0.049	0.19 (-0.02 to 0.41)	0.076	0.21 (0.00 to 0.43)	0.049
Use of antihypertensive agents	1.27 (1.07 to 1.46)	<0.001	0.61 (0.41 to 0.81)	<0.001	0.56 (0.36 to 0.77)	<0.001	0.60 (0.40 to 0.81)	<0.001
Monocyte count	1.56 (1.13 to 2.00)	<0.001	–	–	0.95 (0.44 to 1.46)	<0.001	–	–
MHR	1.41 (1.10 to 1.73)	<0.001	–	–	–	–	1.13 (0.78 to 1.48)	<0.001

^*Normal: systolic blood pressure (SBP) ≤140 mm Hg and diastolic blood pressure (DBP) ≤90 mm Hg; High: SBP >140 mm Hg and/or DBP >90 mm Hg.1

CHD, coronary heart disease; HDL-C, high-density lipoprotein cholesterol; MHR, monocyte to HDL-C ratio; TC, total cholesterol; WC, waist circumference.

The established models were comparable in discriminatory ability. The AUC in predicting 5-year risk of CHD was 0.802 (0.784, 0.820) for model 1, 0.802 (0.784, 0.819) for model 2 and 0.806 (0.788, 0.824) for model 3 in the training set, and were 0.798 (0.758, 0.841), 0.799 (0.750, 0.842) and 0.801 (0.759, 0.846), respectively, in the cross-validation. The AUCs of the three models in predicting 4-year risk of CHD decreased to 0.746 (0.725, 0.767), 0.746 (0.726, 0.766) and 0.749 (0.728, 0.770) in the training set, and to 0.745 (0.691, 0.795), 0.746 (0.690, 0.796) and 0.750 (0.691, 0.799) in the cross-validation, respectively (figure 2). In the external validation set, the AUC in predicting 4-year risk of CHD was much lower, with an AUC of 0.713 (0.670, 0.756) for model 1, 0.717 (0.674, 0.761) for model 2 and 0.712 (0.668, 0.757) for model 3.

Figure 2. Discrimination of established predictive models in the training set and validation set. Model 1, including age, residential area, prevalent diabetes, blood pressure, use of antihypertensive agents and HDL-C; Model 2, replacing HDL-C with monocyte count and WC based on model 1; Model 3, replacing HDL-C with MHR based on model 1. AUC, area under the curve; HDL-C, high-density lipoprotein cholesterol; MHR, monocyte count to HDL-C ratio; WC, waist circumference.

In the training cohort, the three models exhibited excellent distributional calibration, with predicted probabilities closely aligning to observed event rates (Hosmer-Lemeshow χ² test and Spiegelhalter’s Z test, both p>0.05). In the external validation cohort, however, calibration deteriorated: the distributions were left-skewed, indicating systematic overestimation of event probabilities (both tests, p<0.05 for each model) (figure 3).

Figure 3. Calibration plots for established prediction models. Model 1, including age, residential area, prevalent diabetes, blood pressure, use of antihypertensive agents and HDL-C; Model 2, replacing HDL-C with monocyte count and WC based on model 1. Model 3, replacing HDL-C with MHR based on model 1. P10–P100 representing decile groups by the risk of CHD. CHD, coronary heart disease; HDL-C, high-density lipoprotein cholesterol; MHR, monocyte count to HDL-C ratio; WC, waist circumference.

In the training cohort, models 2 and 3 achieved 4-year risk prediction comparable to model 1, with NRI (95% CI) of –1.1% (–3.6% to 1.2%) and 0.3% (–2.0% to 2.9%), respectively. For 5-year risk, both models showed modest improvements over model 1 (NRI (95% CI): 2.1% (–0.9% to 5.1%) and 2.2% (0.0% to 4.5%)). In the external validation set, models 2 and 3 significantly outperformed model 1 for 4-year risk prediction (NRI (95% CI): 4.9% (–0.4% to 10.6%) and 5.6% (0.4% to 10.9%)).

Development and validation of the risk scoring systems

Three risk scoring systems were developed based on corresponding models. Online supplemental table 2 shows the scoring algorithms used to calculate point values in the training set. The scores ranged from 0 to 84 for the scoring system 1, 0 to 88 for the scoring system 2, and 0 to 84 for the scoring system 3. The scoring system 3, including MHR, displayed slightly better discrimination, with an AUC of 0.805 (0.789, 0.820) vs 0.801 (0.786, 0.816) for the system 1 and 0.802 (0.786, 0.818) for the system 2 in predicting 5-year risk of CHD, and an AUC of 0.748 (0.729, 0.768) vs 0.745 (0.726, 0.764) and 0.745 (0.725, 0.765), respectively, for 4-year risk (online supplemental figure 2). The scoring algorithms displayed lower discriminative ability in the external validation set, with an AUC of 0.710 (0.667, 0.754) for system 1, 0.718 (0.675, 0.761) for system 2, and 0.712 (0.668, 0.756) for system 3.

Online supplemental tables 3 and 4 show good calibration of the three scoring algorithms in the Songjiang subcohort and in the moderate-risk and high-risk groups in the Jiading subcohort (all p values >0.05) for 4-year or 5-year CHD risk prediction. However, all the scoring systems appeared to overestimate the risk of CHD in the low-risk group of the Jiading subcohort (p values <0.05).

Discussion

In this prospective cohort study, monocyte count and MHR demonstrate substantial predictive values for the risk of CHD in Chinese populations. Specifically, (1) monocyte count and MHR were significantly associated with the risk of CHD in a linear fashion and had the potential to be independent predictors for CHD events and (2) the models including monocyte count or MHR demonstrated comparable performance in predicting the risk of CHD with the conventional model including HDL-C. Based on its simplicity to compute, low cost to measure and good discrimination and calibration, the risk scoring system including monocyte count is a promising risk assessment tool in clinical and public health practice for the purpose of risk stratification and primary prevention of CHD.

Multiple predictive models have been established as economical and convenient tools to assess the risk of CHD in general populations. The most common predictors in established models included sex, age, serum lipids (HDL-C, TC or LDL-C), blood pressure and current smoking.²⁴,²⁷ In addition to these conventional risk factors, the FRS included prevalent diabetes as an important predictor,²⁵ the cardiovascular disease (CVD) risk algorithm (QRISK) developed in UK populations incorporated family history of CVD and anti-hypertensive medication,²⁷ the most recent Pooled Cohort Equations for atherosclerosis CVD (ASCVD) considered race heterogeneity,²⁶ and the China-PAR, a risk scoring system for ASCVD in Chinese populations, included WC, region (north/south China), and residential area (urban/rural).²⁴

In this study, we constructed a predictive model for CHD (model 1) including age, residential area, prevalent diabetes, blood pressure, antihypertensive medications and HDL-C. Evidently, several conventional risk factors above-mentioned were not incorporated in model 1. For example, smoking, a well-recognised risk factor for CHD including in almost all established predictive models,^{24 41} was not incorporated in our established models. We did not observe a significant difference smoking rate between the CHD cases and the non-cases, nor a significant association of smoking with the risk of the disease. This may be partly explained by the much lower smoking rate in our subjects than other populations in China.⁴² Family history of CHD, another predictor incorporated in several models but not the FRS,^{24 27 41} was not included in model 1. It is plausible, as family history represents similar genetic susceptibility and shared risk exposures, which might vary greatly across populations. Notably, the use of antihypertension medications appeared a risk predictor for CHD in our models, which is different from other models. It is possible that the patients treated with antihypertensive medications experience worse health and thus have a higher risk of CHD. A study based on the Multi-Ethnic Study of Atherosclerosis also reported a 26% higher risk of CHD associated with use of antihypertensive agents.⁴¹

Although this study identified several novel risk predictors, the newly derived scoring algorithms achieved only moderate performance in the training cohort, with discrimination and calibration comparable to those of established risk scores. In the derivation dataset, the AUCs of scores 1–3 (0.801, 0.802 and 0.805, respectively) were similar to those of the FRS (0.74 for men, 0.77 for women)²⁵ and China-PAR (0.794 for men, 0.811 for women).²⁴ As expected, discrimination declined in the external testing set, reflecting the different baseline characteristics of the validation population; such attenuation is routinely observed when scores are transported beyond their derivation cohort.²⁴ In the external validation, the monocyte-augmented and MHR-augmented models were modestly outperformed by established tools: AUCs were 0.710, 0.718 and 0.712 for scores 1–3, vs 0.767 (men) and 0.788 (women) for QRISK3,²⁷ and 0.809 (men) and 0.829 (women) for China-PAR.²⁴ As external validation remains the definitive test of generalisability, the marked decline in performance highlights the inherent limitations of our scoring systems: overfitting and unmeasured predictors.

In this study, we evaluated the predictive values of monocyte count and MHR, which is distinct from previous studies.²⁴,²⁷ Chronic inflammation and lipid accumulation are the main causes of atherosclerosis⁸,¹⁰ and may be involved in the initiation and progression of CHD. Monocyte count and MHR, the two indices for chronic inflammation and/or lipid accumulation, have been independent predictors for in-stent restenosis in premature CHD,³⁰ major adverse cardiovascular events in patients undergoing coronary angiography,³¹ and all-cause death in patients with stroke, transient ischaemic attack or acute MI.^{43 44} Their associations with the risk of CHD were mainly cross-sectional. A study based on the NHANES 1999–2002 observed a doubled risk of peripheral arterial disease in the highest vs the lowest quartile of monocyte count.⁴⁵ Another analysis based on the NHANES 2009–2018 demonstrated a positive association of MHR with CHD in both men and women.⁴⁶ A survey in rural areas of northern China demonstrated 39.5% additional risk of prevalent CHD for each SD increase in MHR.⁴⁷ Only one cohort study reported a 1.15-fold elevated risk of CHD for per 100 cells/mm³ increase of monocytes in French male employees.¹⁶ Consistent with the study, our results based on a prospective cohort study provide additional evidence for the predictive value of monocyte count and MHR for CHD. As the model incorporating monocyte count demonstrated better calibration and just slightly lower discrimination than the conventional model including HDL-C, our results indicate that the easily-measured biomarker can be used to predict CHD instead of HDL-C, a serum lipid easily influenced by diets and usually assayed after overnight fasting.⁴⁸

This study was based on a large-scale prospective cohort study with a well-documented and standardised protocol for participant recruitment, data collection and biochemistry assays.³⁴ The strict quality control ensured high quality data in this analysis. Moreover, the incident CHD events were obtained from local and municipal medical record systems. The complete and accurate clinical data and the disease reporting systems have minimised the potential recall bias and reporting bias on the outcome.

There are several limitations in this study. First, the study participants were all natives of Songjiang and Jiading districts of Shanghai, which may limit the generalisability of the models and scoring systems in other populations. Second, there may exist unmeasured predictors that should be included in the models. In this study, however, we selected predictors from almost all conventional risk factors for CHD through multivariate analysis, which greatly mitigated our concern on the possibility. Finally, we just focused on monocyte count and MHR, but not other biomarkers for chronic inflammation that would improve the performance of the models and scoring systems such as IL-6 and CRP.^{49 50} Considering that the study aimed to develop simple scoring algorithms for convenient utility in general populations, the constructed models and scoring systems based on monocyte count or MHR performed well enough and appeared quite practical in the real world.

Conclusions

In summary, monocyte count offers an inexpensive, readily accessible biomarker for CHD risk stratification. Although the derived score is suitable for broad clinical and public-health deployment, external validation reveals a marked performance drop, indicating limited generalisability. Prospective, multicentre validation and recalibration are therefore mandatory before clinical adoption.

The funders play no role in the study design, data collection, analysis, interpretation, report writing and article publication.

Data availability statement

Data are available on reasonable request.