Biomarkers in plasma and extracellular vesicles for early detection of asymptomatic ischemic heart disease: a retrospective case-control study

Abstract

Background

Asymptomatic ischemic heart disease (aIHD) often precedes acute coronary syndrome (ACS). Early detection of aIHD with evidence-based treatment may reduce the risk of ACS and sudden cardiac death. Current risk prediction modalities may not fully capture at-risk patients. We aimed to explore whether plasma and extracellular vesicle (EV) proteins could help predict aIHD in at-risk individuals.

Methods

We performed a pilot case-control study in asymptomatic individuals with a coronary artery calcium (CAC) score of >300 who underwent stress-perfusion cardiac magnetic resonance (CMR) imaging in three Dutch hospitals between May 2019 and September 2020. In total, 44 participants demonstrated the presence of aIHD (myocardial ischemia or infarction), and 43 did not. Plasma and EV protein concentrations were measured using the Olink Cardiovascular III panel (92 proteins). Differential expression was used to identify potential biomarkers with correction for multiple testing using the Benjamini-Hochberg method. Forward and backward logistic regression analysis was performed to identify independent clinical and proteomic predictors for aIHD. The trial was registered at ClinicalTrials.gov (NCT04680338; date registered 2020-12-22).

Results

Baseline patient characteristics between patients with and without aIHD were similar; only hypertension was more prevalent in patients with aIHD (P=0.02). After Benjamini-Hochberg correction for multiple testing, no proteins were differentially expressed in groups with and without ischemia. However, in logistic regression models, higher plasma aminopeptidase-N [AP-N, odds ratio (OR) 8.74, 1.12–68.37, P=0.04], retinoic acid receptor responder protein 2 (RARRES2, OR 12.64, 1.88–85.15, P=0.009), and chitinase-3-like protein (CHI3L1, OR 0.54, 0.30–0.96, P=0.04) were independently associated with aIHD. After determining threshold values with the Youden index, the sensitivity for aIHD was 0.84 for AP-N, 0.93 for RARRES2, and 0.93 for CHI3L1. Positive predictive value was 0.62, 0.62, and 0.52, respectively. In an EV model, AP-N was able to predict the presence of aIHD (OR 7.24, 1.34–39.14, P=0.02). In a combined plasma and EV model, RARRES2 in plasma (OR 12.77,1.76–92.73, P=0.01) and AP-N in EVs (OR 13.27, 2.04–86.37, P=0.007) remained independently associated with aIHD. Established clinical risk scores [Systematic Coronary Risk Evaluation (SCORE), Framingham Heart Study (FHS)] did not discriminate aIHD (area under the curve <0.60).

Conclusions

In this exploratory study, selected protein biomarkers showed potential to help identify aIHD in asymptomatic individuals with a high CAC score. These preliminary findings support further validation of AP-N and RARRES2 as gatekeepers for non-invasive risk stratification.

Highlight box.

Key findings

• In asymptomatic individuals with a high coronary artery calcium score (>300), selected plasma and extracellular vesicle (EV) protein biomarkers were associated with the presence of ischemic heart disease as confirmed with stress-perfusion cardiac magnetic resonance scanning.

• AP-N and RARRES2 showed the strongest associations.

What is known and what is new?

• Asymptomatic ischemic heart disease (aIHD) often precedes an acute coronary syndrome or sudden cardiac death, yet established clinical risk factors have limited ability to detect subclinical ischemia in at-risk individuals.

• This exploratory study shows protein biomarkers have potential as a gatekeeper test to select those individuals with likely ischemia.

What is the implication, and what should change now?

• Our exploratory findings warrant further validation in larger cohorts to confirm their association with aIHD.

• In the end, a biomarker-based screening approach may improve non-invasive screening for myocardial ischemia in asymptomatic individuals.

Introduction

Ischemic heart disease (IHD) is the world’s leading cause of mortality (1). IHD is usually caused by obstructive coronary artery disease causing a spectrum of syndromes. This spectrum ranges from stable anginal symptoms, acute coronary syndrome (ACS) and sudden cardiac death (SCD), to ischemic heart failure (2). IHD may also be subclinical, with the presence of obstructive coronary plaques but without clinical symptoms. The prevalence of asymptomatic ischemic heart disease (aIHD) differs according to specific patient groups, ranging from 1% to 40% across general to high-risk populations (3-5). aIHD may be pathologically advanced, and the first clinical presentation is often an ACS or SCD, both having high mortality and morbidity rates (6). Early detection of aIHD may lead to lifestyle, medical, or revascularization interventions, which are associated with improved outcomes (7). This underlines the need for effective early detection and treatment strategies.

Current early detection tools able to predict a cardiovascular event in the general population include the Systematic Coronary Risk Evaluation 2 (SCORE2) and Framingham Heart Study (FHS) Risk Score (8,9). In spite of their accessibility and easy usage, problems with such risk scores are underestimation of event rates, ailure to consider comorbidities, prediction of 10-year risk only, and underrepresentation of minorities (10-12). Moreover, the abundance of risk prediction models and their similar performance limit general clinical use (13). Coronary artery calcium (CAC) scoring through computed tomography (CT) scanning is an accurate screening tool to exclude the presence of coronary atherosclerosis (14). However, it is uncertain what the best follow-up action is in asymptomatic individuals with elevated CAC scores, especially in primary care (15,16). Ideally, a biomarker that is easy to determine would be used to accurately guide therapy to prevent ACS and SCD or, as an in-between step, to identify the need for more expensive diagnostic testing for early detection of aIHD, such as cardiac magnetic resonance (CMR) scanning to assess its presence. Currently available biomarkers, such as troponin and N-terminal prohormone of brain natriuretic peptide (NT-proBNP), have only limited additional value in the prediction of IHD (17).

In addition to plasma concentrations of biomarkers, concentrations can also be measured in extracellular vesicles (EVs). These EVs are lipid-bound particles that bud off from cell membranes into the circulation and include apoptotic bodies, exosomes and microvesicles, which contain proteins, lipids, and nuclear material. They participate in intercellular communication, both locally and distantly, and through their contents exert a variety of effects on target cells (18,19). Specifically in obstructive coronary artery disease, EVs may affect the formation of atherosclerotic plaques through influence on leukocyte and low-density lipoprotein (LDL) infiltration in the vascular wall, as well as smooth muscle cell migration to the intima and proliferation of these cells (20). It has indeed been shown that circulating EV levels are increased in cardiovascular disease, including ACS, and potentially serve as source of diagnostic markers, as well as therapeutic sources (18,20,21).

The predictive ability of experimental biomarkers for detection of aIHD has not been investigated. Moreover, EV biomarkers in aIHD have barely been studied. Therefore, in this pilot study, the aim is to evaluate the diagnostic performance of experimental biomarkers to predict aIHD and to evaluate the added value of EVs as a biomarker source. We present this article in accordance with the STROBE reporting checklist (available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-1-678/rc).

Methods

Study design and patient population

The present research was a retrospective case-control pilot study embedded in the EARLY-SYNERGY trial. The design and rationale of EARLY-SYNERGY have been described before (22). In brief, EARLY-SYNERGY is a multicenter, randomized controlled trial aiming to investigate the yield of stress-perfusion CMR scanning for aIHD diagnosis and whether early diagnosis combined with evidence-based treatment improves outcomes. Asymptomatic individuals with a CAC score of >300, as demonstrated on CT performed within two other ongoing studies (ROBINSCA and ImaLife), were recruited in EARLY-SYNERGY (23,24). Individuals were then randomized 1:1 to additional stress-perfusion CMR scanning or no additional imaging. The study was conducted in three Dutch hospitals [University Medical Center Groningen, Gelre Ziekenhuis (Apeldoorn), and Nij Smellinghe (Drachten)] with patients included for this substudy between May 2019 and September 2020. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of University Medical Center Groningen in Groningen, The Netherlands (No. METC 2018/114). All subjects gave written informed consent prior to study procedures.

For the present pilot study, we randomly included 88 individuals from the CMR arm of the EARLY-SYNERGY trial, of whom 44 demonstrated aIHD on CMR investigation (cases), and 44 did not (controls). See Figure S1. For this exploratory pilot study, the sample size of 88 was determined by the capacity of a single assay plate, comprising 44 cases with aIHD and 44 controls without aIHD, matched on sex and age within one year. The CMR acquisition protocol is summarized in the Supplementary materials. Ischemia was deemed present if at least 10% of the left ventricle was ischemic or if there had been infarction (25).

EV isolation

The isolation was performed as described in previous publications (26,27). In brief, a subset of EVs co-precipitate with LDL particles, which allows separation. Magnetic beads were therefore added (nanomag-D plain for LDL). For the sequential isolation of the EV-LDL subpopulation, dextran sulphate (DS, 0.05%, MP Biomedicals, Irvine, CA, USA) was used in combination with Manganese II Chloride (MnCl₂, 0.05 M, Sigma-Aldrich, St. Louis, MO, USA). Subsequently, a handheld magnet was used to collect the precipitate. After removal of the supernatant, the remaining pellet containing the EV subpopulation was dissolved in Roche lysis buffer. After centrifugation, the supernatant was used to study the protein levels. Characterization of EVs isolated with this protocol is described in two studies (28,29). Normalization was performed on plasma volume. To get easy access to these data, an EV-TRACK ID was created: EV200044, in which the data is structured in a uniform way, including EV-specific markers.

Protein biomarker data

Venous plasma samples were collected right before stress-perfusion CMR imaging. The samples were anticoagulated with ethylenediaminetetraacetic acid (EDTA) and stored. To investigate new biomarkers, the Cardiovascular Disease III of the Olink Multiplex platform (Olink Proteomics AB, Uppsala, Sweden) was used for a batch-wise analysis. This is a validated, predefined multiplex panel of cardiovascular and inflammation-related proteins, providing broad yet disease-relevant coverage appropriate for an exploratory analysis of aIHD without requiring a priori selection of individual proteins. The panel is a reagent kit measuring 92 human proteins simultaneously. The assay is based on proximity extension assay technology. Two oligonucleotide-labelled antibodies are bound to the target protein in the EDTA sample upon which a new polymerase chain reaction (PCR) target is made. The target is quantified through real-time PCR. Protein levels are expressed in Normalized Protein eXpression (NPX). This unit is derived from the cycle threshold (Ct) values as found during the PCR analysis and is pre-processed. A high NPX value means a low Ct value and therefore higher concentration. It is presented on a log₂ scale. A 1 NPX difference then means a doubling in protein concentration. Importantly, NPX is a unitless quantity, and due to the transformation, can take on negative values (30).

Statistical analyses

Normality was checked visually using Q-Q plots. Normally distributed data are presented as mean ± standard deviation. Skewed data are presented as median with interquartile range (IQR). Discrete variables are presented as frequencies and percentages. To compare baseline characteristics between the groups, Student’s t-tests were used for normally distributed continuous variables, Mann-Whitney U tests for skewed continuous variables, and chi-squared and Fisher’s exact tests were used for categorical variables. Plasma and EV concentrations were correlated using Pearson’s r coefficients.

We used R (version 4.2.1) and the R package Linear models for microarray analysis (limma, version 3.52.2) to perform differential expression analysis on the 92 proteins in the Olink panel. We made volcano plots using the R package EnhancedVolcano (version 1.14.0) to visualize differentially expressed proteins. Proteins were deemed differentially expressed with a log fold change (logFC) of ≥0.1, and we used the Benjamini-Hochberg method to correct for false discovery rate and multiple testing. We would like to note that Olink data are expressed as NPX values on a log₂ scale, on which effect sizes for circulating plasma proteins are inherently small; absolute logFC thresholds derived from other platforms are therefore not directly applicable. In addition, logFC was not used as a standalone selection criterion but in combination with statistical significance, so the threshold served as a permissive effect-size filter rather than the primary basis for protein selection. Given the exploratory, hypothesis-generating nature of this pilot study, we deliberately chose a sensitive threshold to avoid prematurely excluding potentially relevant candidate proteins. We performed differential expression analysis of the Olink markers in both the plasma samples and the EV samples.

Univariate and multivariate logistic regression analyses were performed to study the association between biomarker concentrations, clinical characteristics and presence of aIHD. Variables with a P value less than 0.1 were included in multivariate regression analysis using a backward stepwise approach. Given the exploratory nature of the study, no correction for false discovery was made for this analysis. The resulting models were further analyzed in a receiver operating characteristic (ROC) curve analysis, with report of the areas under the curve (AUCs). The Akaike information criteria (AICs) of the models were reported. Optimal thresholds were determined using the Youden index. The performance of the best model was compared to the SCORE and FHS Score. We note that this comparison is conceptually imperfect, as SCORE and FHS are designed to predict future cardiovascular event risk in apparently healthy individuals, whereas our model was developed to detect current IHD. This comparison was therefore not intended as a formal head-to-head evaluation but rather as an exploratory reference point to contextualize the diagnostic performance of the novel biomarker model against widely used and clinically familiar risk scores.

To further assess the net clinical benefit of promising markers as screening markers for IHD, decision curve analysis (DCA) was performed using the dca command in Stata. Predicted probabilities for each marker individually and in combination were derived from logistic regression models and used as input for DCA. Net benefit was evaluated across a range of threshold probabilities from 0 to 1, reflecting the range of probability cutoffs at which a clinician might decide to refer or further investigate a patient for aIHD. Results were compared against the default reference strategies of treating all or treating none.

A two-tailed P value <0.05 was considered significant. Laboratory values that were below the lower limit of detection were imputed by replacing the missing value by the lower limit of detection divided by √2 (31). Complete case analysis was used. Statistical analyses were performed in Stata (StataSE version 17.0, StataCorp LP, College Station, TX, USA).

Results

Study population

In total, 87 participants were included, of which 44 had CMR-confirmed aIHD (cases), and 43 did not (controls). One participant was excluded because of aberrant PCR signals and therefore did not pass quality control. Analysis of individual characteristics, including the SCORE risk factors, showed that the prevalence of hypertension was higher in aIHD patients (cases), while the other classical cardiovascular risk factors and laboratory values were similar between the groups (Table 1).

Table 1. Patient characteristics.

Characteristics	Controls (n=43)	Cases (n=44)	P value
Sex (male)	38 (88)	39 (89)	0.97
Age (years)	70.4±6.3	70.5±6.1	0.91
Hypertension	12 (28)	23 (52)	0.02
Diabetes mellitus	5 (12)	6 (14)	0.78
BMI (kg/m2)	26.2±3.5	26.2±2.8	0.97
Smoking status			0.26
Never	11 (30)	13 (31)
Former	24 (65)	22 (52)
Current	2 (5)	7 (17)
Family history of CVD	19 (44)	22 (50)	0.59
SCORE (% 10-year risk)	3.7 (2.6, 6.2)	4.1 (2.4, 6.5)	0.44
FHS score (% 10-year risk)	35.5±4.7	37.0±4.4	0.17
eGFR (mL/min/1.73 m2)	82.0±18.8	82.0±17.2	1.00
CRP (mg/L)	0.3 (0.3, 1.0)	1.0 (0.3, 2.5)	0.07
Troponin I (ng/L)	3.0 (2.1, 5.0)	4.5 (3.0, 9.0)	0.07
CK-MB (µg/L)	1.9 (1.2, 3.1)	1.9 (1.1, 3.4)	0.85
NT-proBNP (pg/mL)	93.0 (46.0, 157.0)	99.0 (66.0, 169.5)	0.37
Total cholesterol (mmol/L)	4.7±1.1	4.9±1.2	0.37
Triglycerides (mmol/L)	1.5 (1.1, 2.1)	1.5 (1.3, 2.0)	0.59
HDL-cholesterol (mmol/L)	1.3 (1.1, 1.6)	1.2 (1.1, 1.4)	0.34
LDL-cholesterol (mmol/L)	2.3 (1.8, 3.1)	2.6 (2.0, 3.7)	0.17
Ischemic heart disease
Ischemia		20 (45.5)
Infarction		15 (34.1)
Ischemia + infarction		9 (20.5)

Data are presented as n (%), mean ± standard deviation, or median (interquartile range). Hypertension was defined as self-reported or medication use, and positive family history was defined as cardiovascular disease in first-degree relatives. BMI, body mass index; CK-MB, creatine kinase myocardial band; CRP, C-reactive protein; CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate; FHS, Framingham Heart Study; HDL, high-density lipoprotein; LDL, low-density lipoprotein; NT-proBNP, N-terminal prohormone of brain natriuretic peptide; SCORE, Systematic Coronary Risk Evaluation.

Protein concentrations

In plasma, 3 proteins showed higher concentrations in patients with aIHD: low-density lipoprotein receptor (LDL-R), aminopeptidase-N (AP-N) and retinoic acid receptor responder protein 2 (RARRES2). In EVs, concentrations of AP-N and RARRES2 were also higher in aIHD patients, as well as concentrations of interleukin-2 receptor subunit alpha (IL2-RA), trefoil factor 3 (TFF3), E-selectin (SELE), C-X-C motif chemokine 16 (CXCL16), tumor necrosis factor ligand superfamily member 13B (TNFSF13B), tissue-type plasminogen activator (t-PA) and platelet endothelial cell adhesion molecule 1 (PECAM1). However, after correction for multiple testing, no concentration differences remained significant (Figure 1). The full differential expression results are shown in Tables S1,S2 and Figure S2. There was no general correlation between protein concentrations in plasma and EVs (Figure S3). The full results of the differential expression analysis and principal component plots are shown in Appendix 1.

Figure 1.

Volcano plots of the differentially expression analysis in plasma (A) and extracellular vesicles (B). P values were corrected using the Benjamini-Hochberg method.

Prediction of aIHD presence

After multivariate adjustment, higher plasma concentrations of AP-N [odds ratio (OR) 8.74, 95% confidence interval (CI): 1.12 to 68.37, P=0.04] and RARRES2 (OR 12.64, 95% CI: 1.88 to 85.15, P=0.009) were independently associated with the presence of aIHD, as well as lower concentrations of chitinase-3-like protein 1 (CHI3L1) (OR 0.54, 95% CI: 0.30 to 0.96, P=0.04). For biomarkers in EVs, the AP-N concentration was independently associated with the presence of aIHD (OR 7.24, 95% CI: 1.34 to 39.14, P=0.02). AP-N is thus elevated in both plasma and EVs. None of the clinical and laboratory variables was associated with the presence of aIHD. In a plasma and EV combined model, plasma RARRES2 (OR 12.77, 95% CI: 1.76 to 92.73, P=0.01) and EV AP-N (OR 13.27, 95% CI: 2.04 to 86.37, P=0.007) remained independently associated with aIHD (Table 2).

Table 2. Plasma and EV predictors of aIHD presence derived from multivariate logistic regression.

Protein	OR	95% CI	P value
Plasma model
AP-N	8.74	1.12–68.37	0.04
RARRES2	12.64	1.88–85.15	0.009
CHI3L1	0.54	0.30–0.96	0.04
EV model
AP-N	7.24	1.34–39.14	0.02
Combined model
RARRES2	12.77	1.76–92.73	0.01
AP-N (EV)	13.27	2.04–86.37	0.007

aIHD, asymptomatic ischemic heart disease; CI, confidence interval; EV, extracellular vesicles; OR, odds ratio.

The AUCs of the plasma, EV, and combined models were 0.76 (95% CI: 0.66, 0.86), 0.63 (95% CI: 0.51, 0.75) and 0.71 (95% CI: 0.60, 0.82), respectively. The discriminatory performance of these models did not differ significantly (P=0.14) (Figure 2A). The AICs of the plasma, EV, and combined models were 113, 118 and 113, respectively, indicating that the EV model had worse predictive quality (>2 difference). The SCORE and FHS scores were not able to discriminate aIHD patients from controls without aIHD with AUCs of 0.56 (0.43, 0.70) and 0.60 (0.46, 0.73). The discriminatory performance of the plasma model was significantly better than the clinical risk prediction models (P=0.03) (Figure 2B).

Figure 2.

Receiver operating characteristic curves for the detection of asymptomatic ischemic heart disease. (A) Plasma, EV, and combined biomarker models; (B) plasma model compared with the SCORE and FHS clinical risk scores. EV, extracellular vesicle; FHS, Framingham Heart Study; SCORE, Systematic Coronary Risk Evaluation.

The diagnostic performance of these biomarkers is reported in Table 3. Importantly, AP-N, RARRES2, and CHI3L1 (in plasma) all had good sensitivities of 0.84, 0.93 and 0.93, respectively, and positive predictive values (PPVs) of 0.62, 0.62 and 0.52, respectively.

Table 3. Diagnostic performance of novel biomarkers.

Parameter	AP-N	RARRES2	CHI3L1	AP-N (EV)
Cutoff point	5.0	10.75	5.07	0.85
Youden index	0.31	0.35	0.05	0.24
Sensitivity	0.84	0.93	0.93	0.61
Specificity	0.47	0.42	0.12	0.63
PPV	0.62	0.62	0.52	0.63
NPV	0.74	0.85	0.62	0.61
AUC at cutoff	0.65	0.68	0.52	0.62

AUC, area under the curve; EV, extracellular vesicle; NPV, negative predictive value; PPV, positive predictive value.

DCA demonstrated net clinical benefit for AP-N, RARRES2, and the combined model across threshold probabilities ranging from 0 to approximately 0.50, all outperforming both the ‘treat all’ and ‘treat none’ strategies across a clinically relevant range of decision thresholds (Figure 3). The combined model showed the highest net benefit across this range, particularly at threshold probabilities between 0.30 and 0.50, suggesting that the combination of both markers provides greater clinical utility than either marker alone. RARRES2 showed the lowest net benefit of the three, falling below zero at threshold probabilities above approximately 0.55. These findings support the potential utility of AP-N and RARRES2 as adjunctive screening tools for IHD at threshold probabilities consistent with a gatekeeper role, where the clinical cost of missing a case outweighs the cost of a false positive referral.

Figure 3.

Decision curve analysis for AP-N, RARRES2, and the combined model in the detection of ischemic heart disease. Net benefit is plotted against threshold probability. DCA, decision curve analysis; IHD, ischemic heart disease.

Discussion

In this pilot case-control study, we showed that, in contrast to classical cardiovascular risk factors, higher levels of AP-N and RARRES2, as well as lower levels of CHI3L1, were associated with aIHD in an at-risk asymptomatic population (32,33).

AP-N (CD13) is a protein involved in the degradation of several peptides, including angiotensin III. Notably, it may play a role in angiogenesis and promote cholesterol crystallization. It is also a receptor for a coronavirus and human cytomegalovirus (34). Moreover, it is associated with arterial hypertension (35).

RARRES2 (chemerin) is an adipokine involved in adipocyte differentiation and lipid metabolism, with both pro- and anti-inflammatory effects (36). It has been linked to diabetes, metabolic syndrome, and heart failure (37,38).

CHI3L1 is a lectin that does not have chitinase activity. It is involved in inflammatory processes through T-helper cell 2 and interleukin-13-induced inflammation. Lower levels of CHI3L1 may be associated with hypertension and some evidence exists that it is a marker for early-onset atherosclerosis (39,40).

Of note, all three proteins with predictive ability in this exploratory study are involved in inflammatory or lipid-related processes. This is in line with current knowledge that chronic inflammation is a key driver in the pathophysiology of atherosclerotic heart disease, contributing to endothelial dysfunction, plaque formation, and plaque instability (41). The convergence of inflammatory and lipid-related pathways across three independently identified proteins suggests these mechanisms may reflect distinct but complementary aspects of the atherosclerotic process, potentially acting synergistically in IHD pathogenesis. Different anti-inflammatory therapies (colchicine, IL-1-blocker) have shown reduction of adverse endpoints in patients with cardiovascular disease (42), further supporting the biological relevance of inflammatory pathway activation in this context. Whether the proteins identified in this study are causally involved in IHD or represent epiphenomena of underlying disease activity remains to be established in future mechanistic and validation studies.

When evaluating the diagnostic performance of the individual biomarkers, it became evident that AP-N and RARRES2, in particular, demonstrate strong sensitivity and negative predictive values, despite showing lower specificity and PPVs. However, for a test designed for primary care settings, this performance is quite valuable. The primary role of these biomarkers is to exclude patients unlikely to have aIHD, effectively functioning as a gatekeeper. In this context, the high sensitivity and negative predictive value are especially crucial, given that ACS and/or SCD can often be the first clinical manifestation of aIHD. For a relatively inexpensive test, this level of performance makes it an excellent tool for early detection and ruling out the presence of aIHD. Further strengthening our findings, DCA demonstrated net clinical benefit for both individual markers and the combined model at threshold probabilities up to approximately 0.50, supporting their potential utility as adjunctive screening tools in a gatekeeper role. At these threshold probabilities, the clinical cost of missing an IHD case outweighs that of a false positive referral, making the relatively low specificity of both markers acceptable.

We found that our plasma model and combined model were able to predict the presence of aIHD with reasonable discriminatory ability with AUCs in the 0.7–0.8 range. Of note, none of the clinical or laboratory values, including the marker of cardiac injury troponin-T, were able to predict aIHD in our population. This emphasizes that asymptomatic myocardial ischemia is difficult to predict in an at-risk population. However, because of the limited sample size of our study, troponin and C-reactive protein levels may potentially have a predictive value for aIHD in larger cohorts, as demonstrated by Griffiths et al. (43).

Out of the classical Framingham cardiovascular risk factors, we showed that only the prevalence of hypertension was significantly higher in ischemic patients. Other established risk factors did not show significant differences. Moreover, both the SCORE and FHS risk score were unable to predict aIHD. Although we acknowledge that these two models were built for different endpoints (cardiovascular death and cardiovascular events, respectively), it is interesting that these widely used risk scores are entirely unable to predict the presence of aIHD. This enhances the hypothesis that silent myocardial ischemia is easily overlooked and hard to diagnose, specifically in primary care (43).

To the best of our knowledge, this is the first study comparing plasma to EV concentrations in association with aIHD. We found no generalizable correlation between plasma and EV levels, signaling that EVs are a potential biomarker source independent of biomarker concentration in plasma. Indeed, we found more proteins with predictive ability in EVs than in plasma in univariate regression analyses. In contrast, the EV model showed inferior performance for aIHD prediction. This apparent discrepancy reflects the distinction between univariate associations and combined model performance. A larger number of individually predictive proteins does not necessarily translate into superior performance of a combined model. In the EV compartment, the candidate proteins were substantially intercorrelated and therefore largely conveyed overlapping information, so that combining them added limited discriminatory value beyond the strongest individual marker. In contrast, the plasma proteins contributed more complementary information, yielding better combined discrimination despite fewer individually significant markers. In addition, EV proteomic measurements are subject to greater technical variability than plasma, and the small sample size of this exploratory pilot increases the susceptibility of multivariable model selection to instability. We recommend investigating EVs as a source for biomarkers in larger populations and different diseases.

Although it is known that silent myocardial ischemia is an independent predictor of adverse clinical outcome, it is not fully elucidated what the best management is (32). Indeed, the ISCHEMIA trial demonstrated that an initial invasive revascularization approach is not superior to optimal medical therapy (44). Nowadays, it is therefore recommended to treat patients with proven ischemia with adequate secondary prevention measures including lifestyle advice, antiplatelet and lipid-lowering therapy, and treatment of hypertension and diabetes if present (32). Similarly, secondary prevention measures after established silent ACS are also associated with better outcomes (2,45).

Limitations of this study include the sample size and retrospective design. The sample size of 87 participants (44 cases, 43 controls) was small relative to the number of proteins evaluated (n=92), resulting in insufficient statistical power to detect significant differential expression after Benjamini-Hochberg correction. Consequently, the logistic regression and diagnostic performance analyses were not corrected for multiple testing, which increases the risk of false positive findings. These results should therefore be interpreted with caution and are best regarded as hypothesis-generating. Also, performing a meaningful network analysis was not possible due to the limited statistical power. The study’s strength lies in the evaluation of a large number of investigated biomarkers.

In the future, a large validation cohort is necessary to confirm our findings. Additionally, we suggest performing a network analysis to unravel pathways involved in the development of aIHD. In the current study, we did not have enough power to construct a meaningful network analysis. Moreover, further studies should focus on the optimal treatment of aIHD and reduction of major adverse events, such as ACS and SCD.

Conclusions

Predicting aIHD in an at-risk population remains challenging. In this exploratory study, we identified three proteins (AP-N, RARRES2, and CHI3L1) with diagnostic potential, while none of the classic clinical characteristics were able to predict aIHD in a population with elevated CAC scores. Notably, AP-N and RARRES2 show high sensitivity and negative predictive value, making them potentially valuable as gatekeepers in early detection. These biomarkers could play a role in identifying individuals who require further testing and/or intensified therapies. Further studies and validations in different populations are needed to determine the clinical role for these markers.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of University Medical Center Groningen in Groningen, The Netherlands (No. METC 2018/114). All subjects gave written informed consent prior to study procedures.

Data Sharing Statement

Available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-1-678/dss

cdt-16-04-62-dss.pdf ^{(68.9KB, pdf)}

DOI: 10.21037/cdt-2025-1-678