Secondary Logo

Journal Logo

Clinical Science Aspects

STEMI, Cardiogenic Shock, and Mortality in Patients Admitted for Acute Angiography: Associations and Predictions from Plasma Proteome Data

Debrabant, Birgit; Halekoh, Ulrich; Soerensen, Mette∗,†,‡; Møller, Jacob Eifer§,||; Hassager, Christian||; Frydland, Martin||; Palstrøm, Nicolai; Hjelmborg, Jacob; Beck, Hans Christian∗,†; Rasmussen, Lars Melholt∗,†

Author Information
doi: 10.1097/SHK.0000000000001595

Abstract

INTRODUCTION

Despite considerable improvements in prevention, diagnosis, and treatment strategies over the last many decades, acute myocardial infarction (AMI) remains one of the major causes of mortality and morbidity worldwide. AMI type 1 patients presenting with ST elevation myocardial infarction (STEMI) is characterized by among other complete obstruction of blood flow and fast intervention is essential to restore coronary perfusion and to avoid myocardial damage and cell death. Especially for STEMI patients developing cardiogenic shock (CS) mortality remains high, i.e., more than 50% (1), and CS remains one of the most common causes of hospital mortality after AMI (2). Still, studies identifying biomarkers predicting STEMI or CS in the preshock phase have been scarce.

Based on a priori hypothesis-driven studies of individual candidate proteins, numerous proteins have been reported to associate with the diagnosis or prognosis of AMI, STEMI, or CS over the past many decades. However, only a few proteins have proven to be clinically applicable, foremost the cardiac muscle proteins troponin T and I and the creatinine kinase (CK) isoforms CK1/CK-BB and CK2/CK-MB, reflecting damage to the myocardium, applied primarily for AMI and STEMI diagnosis. For many of the remaining proposed protein biomarkers, their shortcomings relate to lack of organ specificity, lack of ability to discriminate AMI outcomes and their severity, or problems concerning the timing of the measurement in relation to the progression of disease (3). In addition to studies of candidate proteins, a few studies have applied a hypothesis-free proteome-wide approach. Alonso-Orgaz et al. (4) in 2014 identified 708 proteins in the thrombus of 20 STEMI patients by mass spectrometry-based proteomics. Moreover, Vélez et al. in 2014 and 2016 studied the membrane micro vesicles from 25 STEMI patients and 23 controls (5), respectively, the intracoronary platelets and the peripheral arterial platelets in 10 STEMI patients (6). Differential expression was observed for 25 proteins in the former study and for 15 proteins in the latter study.

The large majority of the studies published to date have investigated statistical associations between individual proteins and the outcome, while prediction studies have been scarcer and have been limited to a few candidate protein studies. Hence, the more well-established prediction models for AMI, mortality after AMI, STEMI, or CS have been exclusively based on general clinical risk factors (e.g., (7)). It is important to note that association modeling and prediction modeling have two different goals; association modeling investigates the presence of statistical dependencies between the exposure (e.g., a clinical risk factor or a protein) and the outcome (e.g., onset of a disease) having potential causal explanations in mind, while prediction modeling aims at forecasting the outcome for new observations (e.g., onset of a disease) based on the information contained in the variables of the model (e.g., clinical risk factors or proteins) (8). Hence, for clinical use of protein markers, the ability of the proteins to predict for instance an AMI outcome is important.

The aims of the present study were via proteome analysis to identify proteins associated with the occurrence of definitive STEMI and CS, as well as all-cause mortality after admission, and to examine the ability of the proteins combined to predict these three outcomes in suspected STEMI patients. Thereby, we aimed to identify protein-based predictors and their general potential for application at an early time point after admission. We included only very basic standard clinical measurements (e.g., systolic blood pressure and heart rate (9)) and did not include measurements, which required additional laboratory tests beyond determining the protein levels. Hence, we do not focus on whether inclusion of further biomarkers from routine laboratory tests might be beneficial to the predictions. Our study was performed by investigating 497 patients from the “Prediction and risk assessment in patients admitted to acute coronary angiography and development of cardiogenic shock study” (PREDICT-CS) (10), who had been admitted to Odense University Hospital for acute coronary angiography with suspected STEMI and for whom Nanoscale liquid chromatography coupled to tandem mass spectrometry (nano-LC-MS-MS) investigations of more than 700 plasma proteins were performed (11).

PATIENTS AND METHODS

Study population and proteome data

The PREDICT-CS study was conducted in year 2015–2016 and included 2,247 patients admitted to Odense University Hospital (OUH), Denmark or Copenhagen University Hospital, Rigshospitalet, Denmark for acute coronary angiography (CAG) due to suspected ST elevation myocardial infarction (STEMI) (10). The present study included 550 random PREDICT-CS participants admitted to OUH and for whom proteome data were available (11), see below. In the present study, we applied the third universal definition for myocardial infarction (12), as this was valid at the time of conducting the PREDICT-CS study. From ambulance files and patient files it was possible to extract the clinical characteristics relevant for the present study for 497 patients (Table 1), which consisted of 381 consecutive patients with definitive STEMI (76.7%), while 116 (23.3%) patients, who did not have definitive STEMI, were also included. Thirty-five out of the 497 patients (7.0%) experienced cardiogenic shock (CS), of whom almost all (N = 33) had an STEMI diagnosis. Out of the 497 patients, 51 (10.3%) died from any cause within 365 days after admission; the median survival time was 8 days (interquartile range of 2.5–34.5 days) for these 51 individuals. In the present study register-based all-cause mortality was investigated. The clinical characteristics of the patients, underlying the statistical analyses, can be seen in Table 1, while additional clinical information, used for imputation (see section “Preparation and imputation of data for statistical analysis”), can be found in the Supplementary Digital Content 1, http://links.lww.com/SHK/B81. Just prior to the coronary angiography procedure a standard blood sample was taken, together with an additional EDTA collection tube from which plasma was obtained after centrifugation. These plasma samples were used for proteomic analysis: all details regarding sample preparation, labeling with mass tags, fractionation of tagged peptides, nano LC-MS/MS analysis, and protein identification can be found in Beck et al. (11). Protein measurements are given as ratios of the actual amount of protein in a plasma sample to the average amount in a calibrator-plasma-pool generated from patients, who did not receive heparin, as described in detail by Beck et al. (11). Proteome data are available via ProteomeXchange with identifier PXD008468.

Table 1 - Clinical characteristics of the study population
Study population
No. individuals 497
STEMI diagnosis (%) 381 (76.7)
CS diagnosis (%) 35 (7.0)
Death within 365 days after admission (%) 51 (10.26)
Males (%) 354 (71.2)
Mean age (SD) 64.44 (13.96)
Heparin treatment before hospitalization (%) 361 (72.6)
Mean systolic blood pressure at hospitalization (SD) 126.27 (27.44)
Mean heart rate at hospitalization (SD) 80.12 (20.21)
Note: No. indicates number of; SD, standard deviation; STEMI, ST elevation myocardial infarction; CS, cardiogenic shock.

The local ethics committee for Region Hovedstaden (Capital Region of Denmark) approved the study (H-2-2014-110) and it was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from all study participants or patients next of kin plus the patient's general practitioner in accordance with Danish legislation.

Preparation and imputation of data for statistical analysis

Measurements for 723 proteins were available initially. Outlying values, i.e., protein measurements > 80, were set to be missing values. Proteins with a call rate < 50% were excluded. For the remaining 273 proteins, missing values were imputed based on observations from other proteins. The number of imputations was set to five. Finally, for each protein, measurements were standardized using the observed standard deviation and mean (average over the different imputations). Similarly, multiple imputation was used to produce five imputations for missing clinical data. The details regarding the data preparation and imputation can be found in the Supplementary Digital Content 1, http://links.lww.com/SHK/B81, which elaborates on the methods of this study (see the text, as well as the Supplementary Table 1, http://links.lww.com/SHK/B81 with the additional clinical information used for imputation).

Statistical analyses

A detailed description of the statistical methods is given in the Supplementary Digital Content 1, http://links.lww.com/SHK/B81. For each of the three outcomes, i.e., STEMI diagnosis, CS diagnosis, and time to death (as defined by survival for 365 days from admission), two types of analysis were performed: a univariate association analysis to investigate the association of the individual proteins to each outcome, and predictive modeling for exploring the ability of the proteins to predict each outcome. Both analyses relied on regression modeling, i.e., logistic regression for analysis of definitive STEMI or CS and Cox regression for analysis of time to death. Before including proteins into the analyses, standard models containing age, sex, heparin treatment, systolic blood pressure, and heart rate were estimated and model fit was checked. As the focus of this manuscript lied on the potential of protein-based predictors, the predefined set of clinical variables included into our standard models was on purpose restricted to the most basic of those variables available at an early time-point. For overview, the results of simple logistic/Cox regression analyses for each clinical covariate to the three outcomes can be found in the Supplementary Digital Content 2, Supplementary Table 1, http://links.lww.com/SHK/B82.

Association between the three outcomes and individual proteins

Using each of the five sets of imputed proteins, the associations of single proteins to any of the three outcomes in the suspected STEMI patients were investigated by extending the standard model to include a single protein. Results from the five analyses were combined according to Rubin (13) yielding overall effect measures. Correction for multiple testing was conducted by the Benjamini-Hochberg method (14) obtaining false discovery rate (FDR) corrected P values. A cutoff of 0.05 was used in order to declare a protein to be a finding.

Predictive modeling for predicting STEMI diagnosis, cardiogenic shock, or death

In short, a multivariable prediction model was built for each of the three outcomes by extending the standard model using all imputed proteins as additional potential predictors. Since the number of potential predictors is relatively large in relation to the available observations, one can expect that most predictors do not contribute to a stable prediction. In the present study, the penalized regression “lasso”(least absolute shrinkage and selection operator) (15) adapted for the outcome being either binary or a survival time was employed for selection of predictors. To improve stability of the prediction models, the lasso regression was combined with the variable selection procedure introduced by Meinshausen and Bühlmann (16) resulting in a final prediction model for each of the three outcomes. Prediction performance was measured in terms of area under the curve (AUC). As the prediction performance often is overly optimistic due to overfitting occurring when model selection/fitting as well as performance evaluation are based on the same parts of data, we calculated both a raw (apparent) AUC as well as an optimism corrected AUC (17), the latter using a cross validation approach.

See the Supplementary Digital Content 1, http://links.lww.com/SHK/B81, for more details on the selection of prediction models, validation of performance as well as the handling of the protein imputations in relation to this. All computations were performed in R version 3.5.3.

RESULTS

The clinical characteristics of the patients are listed in Table 1; of the 497 patients, 381 (76.7%) had definitive STEMI, 35 (7.0%) experienced CS, and 51 (10.3%) of the patients died within the follow-up of 365 days after admission.

Association analyses of single proteins and the three outcomes in the suspected STEMI patients

Supplementary Tables 2–4, http://links.lww.com/SHK/B82 display all the results of the association analyses of single proteins (see tables, Supplementary Digital Content 2, http://links.lww.com/SHK/B82, which elaborates on the results of this study).

The association analysis of CS did not show any proteins to pass correction for multiple testing. Analyses of definitive STEMI and all-cause mortality, on the other hand, revealed 4 proteins (see Table 2) and 29 proteins (see Table 3), respectively. The four proteins identified for definitive STEMI were all associated with an increased risk of definitive STEMI (odds ratio (OR) >1), while 21 of the 29 proteins identified with respect to all-cause mortality associated with an increased risk of dying (hazard ratio (HR) > 1) and 8 associated with a decreased risk of dying (HR<1).

Table 2 - Proteins associated to definitive STEMI after correction for multiple testing
Accession number Protein name OR (95% CI) P FDR
P02775 Platelet basic protein 1.8 (1.32,2.47) 2.40E-04 4.00E-02
P10720 Platelet factor 4 variant 1.67 (1.26,2.21) 4.00E-04 4.00E-02
P02776 Platelet factor 4 1.71 (1.27,2.30) 4.40E-04 4.00E-02
P10909_2 Isoform 2 of Clusterin 1.62 (1.23,2.14) 6.50E-04 4.40E-02
Notes: OR indicates odds ratio; CI, 95% confidence interval; P, nominal P value; FDR, false discovery rate.

Table 3 - Proteins associated to all-cause mortality after correction for multiple testing
Accession number Protein name HR (95% CI) P FDR
P62805 Histone H4 1.52 (1.30,1.78) 2.40E-07 4.80E-05
P84243 Histone H3.3 1.5 (1.28,1.75) 3.50E-07 4.80E-05
P06899 Histone H2B type 1-J 1.44 (1.24,1.66) 8.80E-07 8.00E-05
P12259 Coagulation factor V 0.27 (0.16,0.46) 1.80E-06 1.20E-04
P31431 Syndecan-4 1.41 (1.22,1.64) 3.00E-06 1.60E-04
P08697 Alpha-2-antiplasmin 1.86 (1.39,2.47) 8.50E-05 3.40E-03
P0C0S5 Histone H2A.Z 1.4 (1.18,1.65) 8.80E-05 3.40E-03
P00450 Ceruloplasmin 1.63 (1.28,2.08) 1.10E-04 3.60E-03
Q99879 Histone H2B type 1-M 1.36 (1.16,1.59) 1.20E-04 3.70E-03
O75531 Barrier-to-autointegration factor 1.46 (1.20,1.77) 1.60E-04 3.70E-03
P04908 Histone H2A type 1-B/E 1.33 (1.15,1.55) 1.50E-04 3.70E-03
Q14956 Transmembrane glycoprotein NMB 1.61 (1.26,2.05) 1.50E-04 3.70E-03
P01023 Alpha-2-macroglobulin 1.49 (1.20,1.85) 2.80E-04 5.40E-03
P16104 Histone H2A.x 1.29 (1.13,1.48) 2.80E-04 5.40E-03
P62979 Ubiquitin-40S ribosomal protein S27a 1.37 (1.14,1.65) 6.70E-04 1.20E-02
P01024 Complement C3 1.43 (1.16,1.75) 6.90E-04 1.20E-02
O00585 C-C motif chemokine 21 0.32 (0.16,0.65) 1.50E-03 2.20E-02
P02748 Complement component C9 1.35 (1.12,1.61) 1.40E-03 2.20E-02
O75367 Core histone macro-H2A.1 1.33 (1.11,1.59) 1.80E-03 2.50E-02
Q5TEC6 Histone H3 1.41 (1.14,1.74) 1.80E-03 2.50E-02
P06858 Lipoprotein lipase 0.38 (0.21,0.71) 2.60E-03 3.20E-02
P12956 X-ray repair cross-complementing protein 6 1.4 (1.14,1.73) 2.50E-03 3.20E-02
Q04695 Keratin, type I cytoskeletal 17 0.46 (0.28,0.76) 2.80E-03 3.30E-02
Q13609 Deoxyribonuclease gamma 0.38 (0.20,0.72) 3.00E-03 3.40E-02
P10646 Tissue factor pathway inhibitor 0.45 (0.27,0.77) 3.40E-03 3.70E-02
Q9HCB6 Spondin-1 0.46 (0.28,0.78) 3.90E-03 4.10E-02
P55072 Transitional endoplasmic reticulum ATPase 1.3 (1.08,1.55) 4.30E-03 4.30E-02
P07093_3 Isoform 3 of Glia-derived nexin 0.49 (0.30,0.80) 4.60E-03 4.50E-02
Q16777 Histone H2A type 2-C 1.29 (1.08,1.54) 4.90E-03 4.60E-02
Notes: HR indicates hazard ration; CI, 95% confidence interval; P, nominal P value; FDR, false discovery rate.

Protein prediction models for predicting the three outcomes in the suspected STEMI patients

Prediction models were constructed for each of the three outcomes. The standard predictions models (excluding proteins) are shown in Supplementary Table 5 (see Supplementary Digital Content 2, http://links.lww.com/SHK/B82).

The prediction model for definitive STEMI held five proteins (see Table 4): galectin-3-binding protein, polymeric immunoglobulin receptor, apolipoprotein B-100, Ig alpha-2 chain C region, and carboxypeptidase N catalytic chain, the latter of which was associated with a decreased risk of definitive STEMI (OR < 1), while the remainder were associated with an increased risk of definitive STEMI (OR>1). The model yielded an apparent AUC of 0.74. The corresponding optimism corrected AUC was 0.62. The standard protein-free model reached an apparent AUC of 0.67 and an optimism corrected AUC of 0.65. The observed decline in the optimism corrected AUC related to the inclusion of proteins implies that proteins do not improve prediction performance for definitive STEMI in our sample.

Table 4 - Prediction model for definitive STEMI
Protein name OR (95% CI) P
Age 1.01 (1.00, 1.03) 1.2e-01
Male sex 1.57 (0.96, 2.57) 7.5e-02
Systolic blood pressure at hospitalization 1.00 (0.99, 1.01) 9.1e-01
Heart rate at hospitalization 0.98 (0.97, 0.99) 4.0e-03
Heparin treatment before hospitalization 2.91 (1.79, 4.73) 1.7e-05
Q08380 Galectin-3-binding protein 1.45 (1.12, 1.88) 4.5e-03
P01833 Polymeric immunoglobulin receptor 1.11 (0.85, 1.44) 4.5e-01
P04114 Apolipoprotein B-100 1.46 (1.04, 2.03) 2.7e-02
P01877 Ig alpha-2 chain C region 1.32 (1.00, 1.75) 5.1e-02
P15169 Carboxypeptidase N catalytic chain 0.64 (0.50, 0.81) 2.0e-04
Notes: OR indicates odds ratios; CI, 95% confidence interval; P, nominal P value.

The prediction model for CS contained two proteins (see Table 5): thrombospondin-4 and hemoglobin subunit beta, associated with decreased risk of CS (OR < 1) and increased risk of CS (OR>1), respectively. The model yielded an apparent AUC of 0.94, together with an optimism corrected AUC of 0.92. The standard protein-free model reached an apparent AUC of 0.93 and an optimism corrected AUC of 0.91.

Table 5 - Prediction model for cardiogenic shock
Protein name OR (95% CI) P
Age 1.03 (0.99, 1.07) 1.4e-01
Male sex 2.93 (0.71, 12.03) 1.4e-01
Systolic blood pressure at hospitalization 0.92 (0.89, 0.94) 9.8e-11
Heart rate at hospitalization 1.04 (1.01, 1.06) 1.7e-03
Heparin treatment before hospitalization 0.98 (0.36, 2.70) 9.7e-01
P35443 Thrombospondin-4 0.38 (0.20, 0.71) 2.5e-03
P68871 Hemoglobin subunit beta 2.60 (1.57, 4.33) 2.3e-04
Notes: OR indicates odds ratios; CI, 95% confidence interval; P, nominal P value.

The prediction model for all-cause mortality comprised five proteins (see Table 6): alpha-2-antiplasmin, Ig lambda chain V-III region LOI, Syndecan-4, Transmembrane glycoprotein NMB, and target of Nesh-SH3, of which the latter was associated with a decreased risk of death (HR < 1), while the remainder were associated with an increased risk of death (HR>1). The model had an apparent AUC of 0.90, whereas the AUC corrected for optimism was with 0.89. The standard protein-free model reached an apparent AUC of 0.84 and the optimism corrected AUC was 0.83.

Table 6 - Prediction model for all-cause mortality after admission
Protein name OR (95% CI) P
Age 0.82 (0.72,0.93) 0.0028
Age2 1.00 (1.00,1.00) 0.0002
Male sex 2.56 (1.16,5.64) 0.0198
Systolic blood pressure at hospitalization 0.97 (0.96,0.98) 2.0e-05
Heart rate at hospitalization 1.03 (1.01,1.04) 0.0001
Heparin treatment before hospitalization 1.39 (0.63,3.06) 0.4129
P08697 Alpha-2-antiplasmin 1.59 (1.07,2.35) 0.0237
P80748 Ig lambda chain V-III region LOI 1.15 (0.94,1.42) 0.1830
D3YTG3 Target of Nesh-SH3 0.62 (0.42,0.91) 0.0165
P31431 Syndecan-4 1.33 (1.10,1.62) 0.0044
Q14956 Transmembrane glycoprotein NMB 1.59 (1.20,2.10) 0.0012
Notes: HR indicates hazard ratio; CI, 95% confidence interval; P, nominal P value.

The observed increase in the optimism corrected AUCs related to the inclusion of proteins implies that proteins do improve prediction performance for CS and mortality.

DISCUSSION

In this study, we applied proteome-wide data obtained for plasma samples from 497 patients admitted to the hospital for acute coronary angiography with suspected STEMI. The proteome-wide data was used for association studies of definitive STEMI, CS, and death after suspected STEMI, and for investigation of the predictive ability of the proteins combined for any of the three outcomes. Association analyses were conducted by regression modeling, while prediction modeling was performed by LASSO regression models and evaluated by AUC. Overall, several proteins were found to associate to definitive STEMI and all-cause mortality, yet not to CS, while adding proteins to the prediction models increased their performance for CS and all-cause mortality, yet not for definitive STEMI.

Of the four proteins found to associate to definitive STEMI (see Table 2), clusterin (a chaperon acting under different stress conditions, including stress-induced aggregation of blood plasma proteins) has been reported to associate to, among many other clinical characteristics, heart failure (18). The platelet basic protein (a platelet-derived growth factor involved in various cellular processes) and the platelet factor 4 variant (an inhibitor of angiogenesis) have among others been linked to coronary heart disease (19). Of the 29 proteins associated with all-cause mortality (see Table 3), four proteins relate to blood coagulation; coagulation factor V (F5), the three serine protease inhibitors glia-derived nexin, alpha-2-antiplamin (alpha-2-AP) and tissue factor pathway inhibitor (TFPI), known for example from coagulopathies like Factor V deficiency (F5), hemorrhagic diathesis (alpha-2-AP), and haemophilia (TFPI), respectively. Furthermore, 11 of the 29 proteins have been linked to heart or cardiovascular related diseases; for instance, three proteins have been reported connected to heart failure: syndecan-4 (a transmembrane heparan sulfate-carrying glycoprotein involved in intracellular signaling) (20) ceruloplasmin (a metalloprotein-binding copper in plasma) (21), as well as alpha-2-macroglobulin (alpha-2-M) (a protease inhibitor inhibiting among others thrombin) (22), while syndecan-4 and alpha-2-AP also have been linked to myocardial infarction (23, 24). Another example is lipoprotein lipase (LPL), which has been suggested linked to atherosclerosis and coronary heart disease (25). LPL is a key lipoprotein lipase of triglyceride metabolism, known from among others hyper-lipidemia characterized by high levels of cholesterol and triglyceride in the blood. Interestingly, 10 out of the 29 proteins associated with all-cause mortality were histone proteins, components of the nucleosomes of the chromosomes. All 10 histone proteins associated with an increased risk of death (HR>1). Interestingly, several studies have shown that increased amounts of circulating histones are associated with cell- and nuclear-lysis in severe sepsis in humans and experimental animals (26). In our study it is likely that a similar situation is present and that high levels of plasma histones may reflect a high degree of myocardial cell damage. In line with this notion, it is noteworthy that many of these 10 histone proteins were also found as the most nominal significant proteins for CS, all indicating an increased risk of CS (OR>1). Finally, of specific relevance for the phenotypes under study here, the TPF1 and the cell adhesion protein spondin-1, both found to associate to mortality, were reported among the 29 proteins found to change in level from the acute to the stable phase in the 48 STEMI patients studied by Kulasingam et al. (27). In this study a predefined panel of 92 cardiovascular biomarkers were investigated.

For clinical use the ability of the proteins to predict the three outcomes is important, but corresponding studies for comparison of our results are scarce. We constructed prediction models for the three outcomes and used AUC values for evaluating the prediction performance of a model. Considering the AUC values on the present models, inclusion of proteins improved the performance (as compared with the protein-free standard model) for all-cause mortality and to a lesser extent for CS, but not for definitive STEMI. To predict the presence of definitive STEMI in our setting seemed to be challenging, i.e., the optimism corrected AUCs were very modest: 0.65 for the standard protein-free model and 0.62 for full model. This indicates that the definitive STEMI outcome contains a major amount of unspecified variation, which could not be explained by proteins in our dataset. For CS and all-cause mortality the corresponding values were 0.91 (standard protein-free model) and 0.92 (full model), and 0.83 (standard protein-free model) and 0.89 (full model), reflecting good performance. Several well-established prediction models for AMI, all including clinical information, have been developed based on large registries and achieve moderate to excellent discrimination (validation AUC>0.7). For instance, the GRACE risk score (28), with an AUC of 0.75, predicts 6-months mortality, while the TIMI Risk Score (9), with an AUC of 0.79, predicts 30-days mortality of STEMI and NSTEMI patients. Compared with these established models, as well as additional models, the performance of the mortality-prediction model developed in the present study (optimism-corrected AUC = 0.89) is highest, suggesting that the selected proteins could add valuable information to the prediction of mortality after suspected STEMI. Prediction models for CS have also been developed, for example the ORBI score (7) with an AUC of 0.80 and the shock index of (29) with an AUC of 0.82. Again, the model for CS derived in the present study achieved a prediction performance that was superior at an excellent level (optimism-corrected AUC = 0.92). Finally, several of the proteins of the CS and mortality prediction models have been reported in association studies of heart-related traits and/or are known to be involved in biological processes of relevance to heart physiology. Of the five definitive STEMI predictor proteins, the apolipoprotein B-100 (a key apolipoprotein component of low lipoprotein particles) is known to be a key driver of the atherogenic process and its concentration in blood is strongly associated with all atherosclerotic events, including STEMI (30). Moreover, the carboxypeptidase N is a plasma metalloprotease among other factors involved in protection against vasoactive peptides. Vasoactive peptides affect blood vessels and consequently the blood pressure and heart rate (genecards.org). Finally, the galectin-3-binding protein (a beta-galactoside-binding protein among others implicated in modulating cell–cell and cell–matrix interactions) has been reported to associate with heart failure (23), while its interaction partner galectin-3 has been suggested as a biomarker for heart failure by the American College of Cardiology Foundation; however, its clinical appliance and the exact explanation behind this association are debated by others (3). The thrombospondin-4 (an adhesive glycoprotein that mediates cell-to-cell and cell-to-matrix interactions) of the CS predictor is known to be cardio-protective via both extra- and intracellular functions in the heart and blood vessels. Extracellularly it interacts with structural extracellular matrix proteins, whereby it is involved in remodeling processes and intracellularly it mediates endoplasmic reticulum stress response. The expression of thrombospondin-4 is known to increase during myocardial infarction, myocardial pressure overload and hypertension (31). As mentioned above, the alpha-2-antiplasmin and the syndecan-4 of the mortality predictor have been linked to heart disease-related outcomes (20, 24).

It might appear intriguing that the predictor models hold proteins, which had not been found to pass correction for multiple testing in the association analyses. This is however not unusual, since the prediction models are firstly multivariable models (i.e., exploit other potential and conditional structures) and are secondly not selected on the basis of significance thresholds. As such, the association model attempts to investigate the relationship between the exposure (here a single protein) and the outcome (here, e.g., all-cause mortality after admission), thereby approximating an explanatory causal connection between the two (although causality can generally not be deduced from observational studies alone). The prediction model, on the other hand, aims at forecasting outcome for new observations (here for instance death after admission) based on the information contained in the variables of the model (in this case sets of proteins). The prediction model does, hence, not seek a potential causal connection between the exposure and the outcome.

In the present study, we did not consider the potential variation in time from onset of symptoms to blood sampling (and hence proteome data), since our models only included very simple clinical covariates apart from proteins. Moreover, in a practical setting, information about time of symptom onset is often imprecise as it is self-reported information. However, we cannot exclude that differences in the time span might affect the results obtained regarding association and predictive abilities and future studies are warranted for detailed investigation of this aspect.

Finally, in the present study, we aimed to apply proteome data for identifying novel protein predictors for definitive STEMI, CS, and mortality in patients admitted for acute coronary angiography due to suspected STEMI. Consequently, we did not focus on known biomarkers of AMI, which have originally been identified in candidate protein studies, e.g., troponin. However, when investigating the troponin levels in the same patients in a post hoc analysis, troponin efficiently improved the prediction ability (AUC) of our basis model for definitive STEMI, yet not for CS and mortality (the methods and results of the post hoc analysis are summarized in the Supplementary Digital Content 1, http://links.lww.com/SHK/B81 and 2 (Supplementary Table 6, http://links.lww.com/SHK/B82), respectively). This is interesting considering that our proteome-based prediction models performed the best for CS and mortality, and the worst for definitive STEMI. This also underlines that different protein biomarkers play important roles in predicting outcomes after admission and that different protein biomarkers are useful in different settings and for different outcomes. Future studies could preferably shed light on systematic comparisons of proteome-based predictors and biomarkers known from a-priori-hypothesis-based studies.

In conclusion, the present study aimed for the first time to investigate whether proteins identified by plasma proteomics can improve the prediction of three outcomes in suspected STEMI patients; presence of definitive STEMI and CS, as well as all-cause mortality after admission. For all outcomes but definitive STEMI, adding protein data to the prediction model (beyond clinical data) resulted in improved performance. This suggests that protein biomarkers have the potential to improve existing prediction models based on clinical variables in suspected STEMI patients when it comes to the prediction of CS or mortality. Further studies are however needed to validate our results and to carefully combine previously established clinical predictors with proteomic features to build a prediction model for application in clinical practice.

REFERENCES

1. Helgestad OKL, Josiassen J, Hassager C, Jensen LO, Holmvang L, Sørensen A, Frydland M, Lassen AT, Udesen NLJ, Schmidt H, et al. Temporal trends in incidence and patient characteristics in cardiogenic shock following acute myocardial infarction from 2010 to 2017: a Danish cohort study. Eur J Heart Fail 21 (11):1370–1378, 2019.
2. Goldberg RJ, Spencer FA, Gore JM, Lessard D, Yarzebski J. Thirty-year trends (1975 to 2005) in the magnitude of, management of, and hospital death rates associated with cardiogenic shock in patients with acute myocardial infarction: a population-based perspective. Circulation 119:1211–1219, 2009.
3. Lyngbakken MN, Myhre PL, Røsjø H, Omland T. Novel biomarkers of cardiovascular disease: applications in clinical practice. Crit Rev Clin Lab Sci 56 (1):33–60, 2019.
4. Alonso-Orgaz S, Moreno-Lunaa R, Lópezb JA, Gil-Donesa F, Padialc LR, Moreud J, de la Cuestaa F, Barderasa MG. Proteomic characterization of human coronary thrombus in patients with ST-segment elevation acute myocardial infarction. J Proteomics 109:368–381, 2014.
5. Vélez P, Parguiña AF, Ocaranza-Sánchez R, Grigorian-Shamagian L, Rosa I, Alonso-Orgaz S, de la Cuesta F, Guitián E, Moreu J, Barderas MG, et al. Blood coagulation, fibrinolysis and cellular haemostasis identification of a circulating microvesicle protein network involved in ST-elevation myocardial infarction. Thrombosis Haemostasis 112 (4):716–726, 2014.
6. Vélez P, Ocaranza-Sánchez R, López-Otero D, Grigorian-Shamagian L, Rosa I, Bravo SB, González-Juanatey JR, García Á. 2D-DIGE-based proteomic analysis of intracoronary versus peripheral arterial blood platelets from acute myocardial infarction patients: upregulation of platelet activation biomarkers at the culprit site. Proteomics Clin Appl 10 (8):851–858, 2016.
7. Auffret V, Cottin Y, Leurent G, Gilard M, Beer J-C, Zabalawi A, Chague F, Filippi E, Brunet D, Hacot J-P, et al. Predicting the development of in-hospital cardiogenic shock in patients with ST-segment elevation myocardial infarction treated by primary percutaneous coronary intervention: the ORBI risk score. Eur Heart J 39:2090–2102, 2018.
8. Shmueli G. To explain or to predict? Statist Sci 25 (3):289–310, 2010.
9. Morrow DA, Antman EM, Giugliano RP, Cairns R, Charlesworth A, Murphy SA, de Lemos JA, McCabe CH, Braunwald E. A simple risk index for rapid initial triage of patients with ST-elevation myocardial infarction: an InTIME II substudy. Lancet 358 (9293):1571–1575, 2001.
10. Obling L, Frydland M, Hansen R, Moller-Helgestad OK, Lindholm MG, Holmvang L, Ravn HB, Wiberg S, Thomsen JH, Jensen LO, et al. Risk factors of late cardiogenic shock and mortality in ST-segment elevation myocardial infarction patients. Eur Heart J Acute Cardiovasc Care 7 (1):7–15, 2018.
11. Beck HC, Jensen LO, Gils C, Ilondo AMM, Frydland M, Hassager C, Møller-Helgestad OK, Møller JE, Rasmussen LM. Proteomic discovery and validation of the confounding effect of heparin administration on the analysis of candidate cardiovascular biomarkers. Clin Chem 64:1474–1484, 2018.
12. Thygesen K, Alpert JS, Jaffe AS, Simoons ML, Chaitman BR, White HD. Writing Group on the Joint ESC/ACCF/AHA/WHF Task Force for the Universal Definition of Myocardial Infarction, Thygesen K, Alpert JS, White HD, et al. Third universal definition of myocardial infarction. Eur Heart J 33 (20):2551–2567, 2012.
13. Rubin DB. Multiple Imputation for Nonresponse in Surveys. New York, Chichester, Brisbane, Toronto, Singapore: John Wiley & Sons, Inc.; 1987.
14. Benjamini Y, Hochberg Y. Controlling the false discovery rate a practical and powerful approach to multiple testing. J R Stat Soc Ser B Methodol 57:289–300, 1995.
15. Tibshirani R. Regression shrinkage and selection via the lasso. J R Stat Soc Ser B Methodol 58 (1):267–288, 1996.
16. Meinshausen N, Bühlmann P. Stability selection (with discussion). J R Stat Soc Ser B 72:417–473, 2010.
17. Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning. Second EditionBerlin: Springer Series in Statistics; 2009.
18. Koller L, Richter B, Winter MP, Sulzgruber P, Potolidis C, Liebhart F, Mörtl D, Berger R, Goliasch G, Lang I, et al. Clusterin/apolipoprotein J is independently associated with survival in patients with chronic heart failure. J Clin Lipidol 11 (1):178–184, 2017.
19. Poulakou MV, Paraskevas KI, Wilson MR, Iliopoulos DC, Tsigris C, Mikhailidis DP, Perrea D. Apolipoprotein J and leptin levels in patients with coronary heart disease. In Vivo 22 (4):537–542, 2008.
20. Bielecka-Dabrowa A, Gluba-Brzózka A, Michalska-Kasiczak M, Misztal M, Rysz J, Banach M. The multi-biomarker approach for heart failure in patients with hypertension. Int J Mol Sci 16 (5):10715–10733, 2015.
21. Xu Y, Lin H, Zhou Y, Cheng G, Xu G. Ceruloplasmin and the extent of heart failure in ischemic and nonischemic cardiomyopathy patients. Mediators Inflamm 2013:348145, 2013.
22. Yang S, Chen L, Sun S, Shah P, Yang W, Zhang B, Zhang Z, Chan DW, Kass DA, van Eyk JE, et al. Glycoproteins identified from heart failure and treatment models. Proteomics 15 (2–3):567–579, 2015.
23. Solbu MD, Kolset SO, Jenssen TG, Wilsgaard T, Løchen ML, Mathiesen EB, Melsom T, Eriksen BO, Reine TM. Gender differences in the association of syndecan-4 with myocardial infarction: The population-based Tromsø Study. Atherosclerosis 278:166–173, 2018.
24. Uitte de Willige S, Miedzak M, Carter AM, Lisman T, Rosendaal FR, Grant PJ, Philippou H, Ariëns RA. Proteolytic and genetic variation of the alpha-2-antiplasmin C-terminus in myocardial infarction. Blood 117 (24):6694–6701, 2011.
25. He PP, Jiang T, OuYang XP, Liang YQ, Zou JQ, Wang Y, Shen QQ, Liao L, Zheng XL. Lipoprotein lipase: biosynthesis, regulatory factors, and its role in atherosclerosis and other diseases. Clin Chim Acta 480:126–137, 2018.
26. García-Giménez JL, Romá-Mateo C, Carbonell N, Palacios L, Peiró-Chova L, García-López E, García-Simón M, Lahuerta R, Gimenez-Garzó C, Berenguer-Pascual E, et al. A new mass spectrometry-based method for the quantification of histones in plasma from septic shock patients. Sci Rep 7 (1):10643, 2017.
27. Kulasingam A, Hvas A-M, Grove EL, Funck KL, Kristensen SD. Detection of biomarkers using a novel proximity extension assay in patients with ST-elevation myocardial infarction. Thromb Res 172:21–28, 2018.
28. Eagle KA, Lim ML, Dabbous OH, Pieper KS, Goldberg RJ, Van de Werf F, Goodman SG, Granger CB, Steg PG, Gore JM, et al. A validated prediction model for all forms of acute coronary syndrome estimating the risk of 6-month postdischarge death in an international registry. JAMA 291 (22):2722–2733, 2004.
29. Wei Z, Bai J, Dai Q, Wu H, Qiao S, Xu B, Wang L. The value of shock index in prediction of cardiogenic shock developed during primary percutaneous coronary intervention. BMC Cardiovasc Disord 18 (1):188, 2018.
30. Bodde MC, Hermans MPJ, Jukema W, Schalij MJ, Lijfering WM, Rosendaal FR, Romijn FPHTM, Ruhaak LH, van der Laarse A, Cobbaert CM. Apolipoproteins A1, B, and apoB/apoA1 ratio are associated with first ST-segment elevation myocardial infarction but not with recurrent events during long-term follow-up. Clin Res Cardiol 108:520–538, 2019.
31. Palao T, Medzikovic L, Rippe C, Wanga S, Al-Mardini C, van Weert A, de Vos J, van der Wel NN, van Veen HA, van Bavel ET, et al. Thrombospondin-4 mediates cardiovascular remodeling in angiotensin II-induced hypertension. Cardiovasc Pathol 35:12–19, 2018.
Keywords:

Acute myocardial infarction; area under the curve; association analysis; LASSO regression modeling; prediction modeling; proteome-wide analysis; ST elevation myocardial infarction (STEMI); tandem mass spectrometry; AMI; acute myocardial infarction; AUC; area under the curve; CS; cardiogenic shock; FDR; false discovery rate; HR; hazard ratio; lasso; least absolute shrinkage and selection operator; nano-LC-MS-MS; Nanoscale liquid chromatography coupled to tandem mass spectrometry; OR; odds ratio; PREDICT-CS study; the prediction and risk assessment in patients admitted to acute coronary angiography and development of cardiogenic shock study; STEMI; ST elevation myocardial infarction

Supplemental Digital Content

Copyright © 2020 by the Shock Society