Abstract
Background
Fragments of insulin‐like growth factor binding protein 4 (IGFBP‐4) are potential new biomarkers for cardiac risk assessment. The fragments are generated on specific cleavage by pregnancy‐associated plasma protein‐A, which exerts proatherogenic activity. This study investigated the prognostic value of IGFBP‐4 fragments in patients with ST‐segment elevation myocardial infarction.
Methods and Results
We prospectively included 656 patients with ST‐segment elevation myocardial infarction treated with percutaneous coronary intervention from September 2006 to December 2008. Blood samples were drawn before percutaneous coronary intervention, and levels of intact IGFBP‐4 and N‐terminal and C‐terminal IGFBP‐4 fragments were measured by specific assays. End points were 5‐year all‐cause and cardiovascular mortality and the combined end point of major adverse cardiac events. Prognostic potential was evaluated on top of a clinical model in terms of discrimination, calibration, and reclassification analysis. During follow‐up, 166 patients experienced a major adverse cardiac event and 136 patients died, of whom 69 died from cardiovascular causes. Both IGFBP‐4 fragments were associated with all end points (P<0.001). After multivariable adjustments, both N‐terminal and C‐terminal IGFBP‐4 fragment levels remained associated with all end points, including cardiovascular mortality with hazard ratios per doubling in protein concentration of 2.54 (95% CI 1.59–4.07; P<0.001) and 2.07 (95% CI 1.41–3.04; P<0.001), respectively. Incorporation of IGFBP‐4 fragments into a clinical model with 15 risk factors improved C‐statistics and model calibration and provided incremental prognostic contribution, as assessed by net reclassification improvement and integrated discrimination improvement.
Conclusions
IGFBP‐4 fragments are associated with increased risk of all‐cause mortality, cardiovascular mortality, and major adverse cardiac events in patients with ST‐segment elevation myocardial infarction.
Keywords: biomarker, cardiovascular disease, insulin‐like growth factor binding protein 4, pregnancy‐associated plasma protein A, ST‐segment elevation myocardial infarction
Subject Categories: Myocardial Infarction, Atherosclerosis
Introduction
An important objective of cardiovascular disease (CVD) research is to identify novel biomarkers to improve stratification of patients with excessive risk. Pregnancy‐associated plasma protein‐A (PAPP‐A) has been regarded a candidate marker in CVD.1, 2 PAPP‐A is a metzincin metalloproteinase present in the circulation as well as within tissues, where it reversibly adheres to cell surfaces.3 In 2001, PAPP‐A was shown to be ubiquitously present in eroded atherosclerotic plaques, and circulating levels were elevated in patients with acute coronary syndrome.1 It was later revealed, however, that administration of unfractionated heparin to patients results in an abrupt increase in PAPP‐A concentrations, probably through displacement of cell surface–tethered PAPP‐A.4, 5 Because heparin is part of the standard initial treatment of patients with acute myocardial infarction, samples from these patients are not suitable for PAPP‐A measurements.
The proteolytic activity of PAPP‐A is directed primarily toward insulin‐like growth factor (IGF) binding protein 4 (IGFBP‐4), which is a key regulator of IGF bioactivity.6, 7, 8 On local PAPP‐A–mediated cleavage of IGFBP‐4, IGF accessibility is increased in proximity to the IGF‐1 receptor. IGF‐1 stimulates cell proliferation and promotes macrophage activation, low‐density lipoprotein uptake, and release of proinflammatory cytokines, which likely facilitate atherogenesis and plaque instability.9 The functional role of PAPP‐A in the process of plaque destabilization has been confirmed in PAPP‐A transgenic and gene knockout mouse models,10, 11 and these findings further support a proatherogenic effect of PAPP‐A–mediated IGF‐1 release.
It has been proposed that the PAPP‐A–generated IGFBP‐4 fragments may be reflective of PAPP‐A enzymatic activity. Circulating levels of total PAPP‐A correlate with levels of N‐terminal IGFBP‐4 (NT‐IGFBP‐4) and C‐terminal IGFBP‐4 (CT‐IGFBP‐4)12; therefore, IGFBP‐4 fragments released from the atheromatous plaque could reflect plaque burden and thus serve as markers of atherosclerosis.13 We recently verified that circulating IGFBP‐4 fragment levels are unaffected by heparin treatment of patients with ST‐segment elevation myocardial infarction (STEMI),5 and no other medical therapies are known to significantly impinge IGFBP‐4 fragment levels.
The aim of the present study was to examine the association between PAPP‐A–generated IGFBP‐4 fragments and risk of mortality and major adverse cardiac events (MACE) in a cohort of patients with STEMI. The value of the IGFBP‐4 fragments was compared with traditional cardiovascular risk factors in terms of discrimination, calibration, and reclassification analysis.
Materials and Methods
Patients
A total of 656 patients with STEMI treated with primary percutaneous coronary intervention at Gentofte University Hospital, Denmark, were included from September 2006 through December 2008. Details of the study population have been described previously.14 Inclusion criteria were onset of STEMI symptoms, including chest pain lasting for >30 minutes and <12 hours, persistent ST‐segment elevation of ≥2 mm in ≥2 contiguous precordial ECG leads or ≥1 mm in ≥2 contiguous limb ECG leads, and an increase in troponin I (TnI) to >0.5 μg/L. All patients were pretreated with unfractionated heparin, and glycoprotein IIb/IIIa inhibitors were administered at the discretion of the operator. Following the percutaneous coronary intervention procedure, patients received aspirin (75 mg/day), clopidogrel (75 mg/day for 12 months), cholesterol‐lowering treatment (statins), and β‐receptor antagonists. Hypertension and hypercholesterolemia were defined as use of blood pressure–lowering or cholesterol‐lowering drugs, respectively, on admission. Patients were considered to have diabetes mellitus if they received glucose‐lowering drugs on admission or had a fasting plasma glucose level ≥7 mmol/L or a nonfasting level of ≥11.1 mmol/L. Multivessel disease was defined as 2‐ or 3‐vessel disease and complex lesions such as type C lesions. Advanced echocardiography was performed in 344 patients (52%) at 1 to 3 days after admission using Vivid 7 or E9 (GE Healthcare). Left ventricular ejection fraction (LVEF) was obtained using the modified biplane Simpson method. Serum and EDTA plasma were drawn from the femoral sheath at the beginning of the percutaneous coronary intervention procedure and stored at −80°C. The study was approved by the local ethics committee and the Data Protection Agency and complied with the Declaration of Helsinki. Informed consent was obtained from all participants.
Follow‐up and End Points
Patients were prospectively followed for 5 years through the Danish Civil Registration system and the National Causes of Death Registry, which offers information from physicians on causes of death according to the International Classification of Diseases, 10th Revision (ICD‐10).15 Follow‐up information was collected on all‐cause mortality, death due to recorded cardiovascular event, and occurrence of nonfatal MACE defined as readmission due to acute myocardial infarction, ischemic stroke, or heart failure. Information on readmissions was obtained from the highly validated National Patient Registry using the ICD‐10 codes.16 Information was obtained for all patients, and all events were thoroughly validated using medical records, including laboratory measurements, hospital summaries, and operative reports.
Laboratory Measurements
Routine methods
High‐sensitivity C‐reactive protein was determined by a nephelometric assay (Dade Behring), and creatinine was assayed using standard laboratory methods. TnI levels were determined at baseline and after 6 and 12 hours using the Immulite 2500 STAT Troponin I immunoassay (Siemens Healthcare). Later determinations were performed if deemed necessary, and the highest obtained value was reported as peak TnI. Estimated glomerular filtration rate was calculated using the Modification of Diet in Renal Disease formula.17
Intact IGFBP‐4 and NT‐ and CT‐IGFBP‐4 fragments
EDTA plasma levels of IGFBP‐4, CT‐IGFBP‐4, and NT‐IGFBP‐4 were measured in duplicate using in‐house time‐resolved immunofluorometric assays based on monoclonal antibodies (mAb) and recombinant human (rh) calibrators generously provided by HyTest Ltd. The assays were performed as recently described.5 Intact IGFBP‐4 was measured using coating mAb IBP182 (4IGF4 IBP182) and detection mAb IBP144 (4IGF4EU IBP144). mAb IBP3 (4IGF4 IBP3) and mAb IBP180 (4IGF4EU IBP180) were used for the determination of the NT‐IGFBP‐4 fragment, and mAb IBP182 and mAb IBP163 (4IGF4EU IBP163) were applied for the determination of the CT‐IGFBP‐4 fragment. As calibrators, full‐length rhIGFBP‐4 (8IGF4), rhNT‐IGFBP‐4 (8NFB4), and rhCT‐IGFBP‐4 (8CIG4) were applied. In each fragment assay, one of the antibodies specifically recognized the proteolytic neoepitope generated on cleavage by PAPP‐A. Detection limits were 0.5 μg/L IGFBP‐4, 0.4 μg/L CT‐IGFBP‐4, and 0.9 μg/L NT‐IGFBP‐4. Intra‐ and interassay coefficient of variations (CVs) were <10% and <15%, respectively.
Statistical Analysis
Nonnormally distributed variables were log2‐transformed prior to statistical analyses. Frequency distribution of log2‐NT‐ and log2‐CT‐IGFBP‐4 and fitted normal curves are provided in Figure S1. A combined end point including all events was used for comparison of baseline characteristics. Patients with or without events during follow‐up were compared using the Student t test or Mann–Whitney U statistics on continuous variables, and the chi‐square test on categorical variables. The Bonferroni adjusted level of significance was reported for correlation coefficients (r value). Receiver operating characteristic (ROC) curves were used to analyze the prognostic values of intact IGFBP‐4, NT‐IGFBP‐4, CT‐IGFBP‐4, C‐reactive protein, and peak TnI. C‐statistics between models were compared by testing equality concordance using Mann–Whitney U statistics.18 Kaplan–Meier survival curves were performed for NT‐ and CT‐IGFBP‐4 quartiles, and incidence distributions were compared using the log‐rank test.
Model Building
Survival analyses were performed using Cox proportional hazards models. All candidate confounder variables and blood biomarkers (Table 1) were initially considered for the model, as were all transformations and interactions between them. In model 1, prespecified variables were included based on traditional cardiovascular risk factors (age, sex, systolic blood pressure, total cholesterol level, high‐density lipoprotein level, diabetes mellitus, and smoking status).19 Furthermore, a criterion for inclusion in model 1 was minimization of the Bayesian information criterion.20 The Bayesian information criterion is a likelihood‐based measure for which lower values indicate better model fit and in which a penalty is paid for an increased number of variables. Finally, a covariate was included if it changed the estimate measure of association by >10%. A baseline clinical model was generated including all variables selected for model 1 (age, sex, systolic blood pressure, total cholesterol level, high‐density lipoprotein, diabetes mellitus, smoking status, estimated glomerular filtration rate, hypertensive treatment, hemoglobin, previous acute myocardial infarction, multivessel coronary artery disease, complex lesion, C‐reactive protein, and peak TnI). The clinical model was then extended by the log2‐transformed NT‐ or CT‐IGFBP‐4 variable, and results were reported as hazard ratio (HR) and 95% CI. Accordingly, each unit increase in NT‐ or CT‐IGFBP‐4 on the log2‐scale corresponds to a doubling in NT‐ or CT‐IGFBP‐4, respectively. Because LVEF measurements were obtained in only 344 patients, model 1 was assessed with and without LVEF as a covariate. Body mass index, hypercholesterolemia, low‐density lipoprotein, triglyceride, symptom‐to‐balloon time, and use of glycoprotein IIb/IIIa inhibitor did not fulfill the criteria for inclusion but were included in a second model (model 2) as an extension of model 1. Hazard proportionality and linearity assumptions were checked by log‐log plots, fitted survival curves, and smoothed martingale and Schonenfeld residuals plots.21, 22, 23 Furthermore, because a large number of events occurred within the first 3 months after STEMI, survival analyses were repeated using a 3‐month end point.
Table 1.
Baseline Characteristics
| Characteristics | Total (n=656) | No Events (n=431) | Events (n=225) | P Value |
|---|---|---|---|---|
| Age, y | 63±12 | 60±11 | 69±12 | <0.001 |
| Male/female, n | 484/172 | 324/107 | 160/65 | 0.261 |
| BMI, kg/m2 | 26 (24; 29) | 26 (24; 29) | 26 (23; 29) | 0.034 |
| Diabetes mellitus, % | 9.1 | 7.7 | 12.0 | 0.067 |
| Current smoker, % | 52.7 | 54.8 | 48.9 | 0.153 |
| Hypercholesterolemia, % | 19.4 | 17.2 | 23.6 | 0.049 |
| Hypertension, % | 33.8 | 28.8 | 43.6 | <0.001 |
| Blood glucose, mmol/L | 8.3 (7.0; 9.9) | 8.1 (7.0; 9.4) | 8.7 (7.2; 11) | <0.001 |
| Hemoglobin, mmol/L | 8.7±1.0 | 8.8±0.8 | 8.5±1.1 | <0.001 |
| Total cholesterol, mmol/L | 4.8±1.1 | 4.9±1.1 | 4.6±1.2 | <0.001 |
| HDL cholesterol, mmol/L | 1.3±0.4 | 1.3±0.3 | 1.3±0.4 | 0.111 |
| LDL cholesterol, mmol/L | 2.9±1.0 | 3.0±1.0 | 2.7±1.1 | <0.001 |
| Triglycerides, mmol/L | 0.98 (0.69; 1.6) | 0.98 (0.67; 1.6) | 0.99 (0.73; 1.5) | 0.926 |
| Systolic blood pressure, mm Hg | 133±27 | 135±26 | 128±28 | 0.003 |
| eGFR, mL/min/1.73 m2 | 73±24 | 78±21 | 63±27 | <0.001 |
| Creatinine, μmol/L | 91 (78; 109) | 88 (76; 103) | 101 (84; 133) | <0.001 |
| C‐reactive protein, mg/L | 3 (1; 9) | 3 (1; 7) | 5 (2; 16) | <0.001 |
| Peak troponin I, μg/L | 90 (28; 244) | 81.3 (28; 212) | 117 (27; 281) | 0.059 |
| Left ventricular ejection fraction, %a | 45.8±9.0 | 46.9±8.6 | 42.8±9.3 | <0.001 |
| Glycoprotein IIb/IIIa inhibitor, % | 25.9 | 25.8 | 26.2 | 0.897 |
| Symptom‐to‐balloon time, minute | 195 (130; 323) | 190 (123; 310) | 230 (140; 355) | 0.051 |
| Multivessel disease, % | 27.0 | 24.6 | 31.6 | 0.057 |
| Culprit lesion, % | 0.353 | |||
| Left anterior descending artery | 47.6 | 47.1 | 48.4 | |
| Circumflex artery | 10.4 | 11.6 | 8.0 | |
| Right coronary artery | 42.0 | 41.3 | 43.6 | |
| Complex lesion, % | 50.9 | 48.3 | 56.0 | 0.060 |
| Previous myocardial infarction, % | 6.3 | 5.1 | 8.4 | 0.093 |
| Died during follow‐up, n | 136 | 0 | 136 | |
| Died due to cardiovascular event, n | 69 | 0 | 69 | |
| MACE during follow‐up, n | 166 | 0 | 166 | |
| IGFBP‐4, μg/L | 153±70 | 146±64 | 164±79 | 0.002 |
| NT‐IGFBP‐4, μg/L | 132 (99; 185) | 120 (94; 155) | 182 (118; 264) | <0.001 |
| CT‐IGFBP‐4, μg/L | 52 (34; 77) | 45 (31; 66) | 69 (46; 108) | <0.001 |
Data are mean±SD or median (25th percentile; 75th percentile). Categorical variables are indicated as numbers (n) or percentage (%) of patients. BMI indicates body mass index; CT, C‐terminal; eGFR, estimated glomerular filtration rate; HDL, high‐density lipoprotein; IGFBP‐4, insulin‐like growth factor binding protein 4; LDL, low‐density lipoprotein; MACE, major adverse cardiac event; NT, N‐terminal.
Left ventricular ejection fraction was available in only 344 patients (52%), of which 102 patients experienced an event.
Model Performances
The incremental value of adding log2‐NT‐ or log2‐CT‐IGFBP‐4 to the clinical model for predicting end points was evaluated. Discriminatory abilities of the models were compared by Harrell's C and Somers' D statistics for censored data using the “somersd” module in Stata version 13 (StataCorp LP). For calibration, the Royston modification of Nagelkerke's R 2 statistic for proportional hazards models was used to assess the explained variation by using 1000 bootstrap repetitions of the whole data set through the “str2ph” module in Stata.24 The Groennesby and Borgan extension of the Hosmer–Lemeshow goodness‐of‐fit test was applied to assess how well the predicted models fitted the observed incidence data.25 Well‐calibrated models show no significant differences in the Hosmer–Lemeshow test results. The Akaike information criterion and the Bayesian information criterion were calculated for each model. Lower values indicate a better model, but no statistical test compares the Akaike information criterion and Bayesian information criterion scores. Overall goodness‐of‐fit was evaluated by likelihood ratio test. Net reclassification improvement (NRI) and integrated discrimination improvement (IDI), as described by Pencina et al,26, 27 were used to evaluate improvements in risk predictions. NRI compares the nested addition of a new variable to a reference model, rewarding correct reclassification and penalizing incorrect reclassification. Any upward movement in patients experiencing an event implies improved classification, whereas a downward movement implies worse reclassification. The interpretation is opposite for patients who do not experience an event. NRI was computed both as category‐based NRI and continuous NRI. Category‐based NRI was assessed by comparing the predicted 5‐year risk of all‐cause mortality, cardiovascular mortality, and MACE between models across risk categories of <5%, 5% to 10%, 10% to 20%, and >20% risk. Because risk categories in STEMI have not been defined, thresholds were based on values suggested in previous studies of CVD events.28, 29 Conversely, the continuous NRI counts the direction of change in risk for every participant rather than the crossing of a threshold. The IDI is independent of category thresholds and reflects the actual extent of change and not merely the direction. NRI and IDI were obtained using 1000 bootstrap repetitions through the “incrisk” module in Stata.
Results are presented as mean±SD for normally distributed data and median (25th percentile; 75th percentile) for skewed data. C‐statistics and HRs are presented with 95% CIs. A 2‐tailed P<0.05 was considered statistically significant. Data were analyzed using Stata software.
Results
Baseline data are presented in Table 1. Median follow‐up of patients was 5 years (range 9 days to 5 years), during which 225 patients (34.3%) suffered from all‐cause mortality, cardiovascular mortality, or MACE. A total of 136 patients (20.7%) died, and among causes of deaths, cardiovascular events were responsible in 69 patients (10.5%). In a large number of patients, all‐cause mortality (n=46), cardiovascular mortality (n=41), or MACE (n=59) occurred within the first 3 months after STEMI. Because of the high‐quality Danish registration system, no patients were lost to follow‐up.
Patients with events had higher circulating levels of intact IGFBP‐4 as well as NT‐ and CT‐IGFBP‐4, and a strong positive correlation was observed between concentrations of NT‐IGFBP‐4 and CT‐IGFBP‐4 (r=0.83, P<0.001). IGFBP‐4 fragment levels correlated negatively with estimated glomerular filtration rate (NT‐IGFBP‐4: r=−0.56, P<0.001; CT‐IGFBP‐4: r=−0.50, P<0.001); therefore, increased fragment concentrations were associated with reduced kidney function.6, 30 IGFBP‐4 fragment levels were not associated with sex, diabetes mellitus, symptom‐to‐balloon time, culprit lesion location or complexity, peak TnI level, hypercholesterolemia, or use of glycoprotein IIb/IIIa inhibitors. LVEF did not significantly correlate with NT‐IGFBP‐4 (r=−0.119, P=0.179) or CT‐IGFBP‐4 (r=−0.114, P=0.219).
C‐statistics for NT‐ and CT‐IGFBP‐4 were significantly higher compared with peak TnI levels (P<0.001) and C‐reactive protein (P<0.05) for all end points (Figure 1 and Table 2). Intact IGFBP‐4 did not perform better than C‐reactive protein or peak TnI for any end point (data not shown), and thus was not subjected to further analyses. Kaplan–Meier survival curves for all‐cause mortality, cardiovascular mortality, and MACE in patients stratified by NT‐ and CT‐IGFBP‐4 quartiles are illustrated in Figure 2. The log‐rank analyses showed significantly different incidence distributions according to NT‐ and CT‐IGFBP‐4 quartiles (all P<0.001) (Table 2).
Figure 1.
Receiver operating characteristic curves for CRP, peak TnI, NT‐IGFBP‐4 (A), and CT‐IGFBP‐4 (B). The end point was cardiovascular mortality. CRP indicates C‐reactive protein; CT, C‐terminal; IGFBP‐4, insulin‐like growth factor binding protein 4; NT, N‐terminal; ROC, receiver operating characteristic; TnI, troponin I.
Table 2.
C‐Statistics and Log‐Rank Analyses of All‐Cause Mortality, Cardiovascular Mortality, and MACE
| C‐Statistics | NT‐IGFBP‐4 | CT‐IGFBP‐4 | CRP | Peak TnI |
|---|---|---|---|---|
| All‐cause mortality | 0.76 (0.72; 0.81)a , b | 0.75 (0.70; 0.80)a , b | 0.68 (0.63; 0.73) | 0.55 (0.50; 0.61) |
| Cardiovascular mortality | 0.82 (0.77; 0.87)a , b | 0.80 (0.75; 0.86)a , b | 0.69 (0.62; 0.76) | 0.57 (0.48; 0.65) |
| MACE | 0.71 (0.67; 0.76)a , b | 0.70 (0.66; 0.75)a , b | 0.59 (0.54; 0.64) | 0.54 (0.49; 0.60) |
| Log‐Rank | NT‐IGFBP‐4 Quartiles | P Value | |||
|---|---|---|---|---|---|
| 1 (31.6–100.8 μg/L) | 2 (100.9–132.6 μg/L) | 3 (132.7–184.8 μg/L) | 4 (184.9–810.1 μg/L) | ||
| Patients, n | 164 | 164 | 164 | 164 | |
| All‐cause mortality, n | 12 | 17 | 31 | 76 | <0.001 |
| Cardiovascular mortality, n | 1 | 8 | 12 | 48 | <0.001 |
| MACE, n | 16 | 32 | 35 | 83 | <0.001 |
| CT‐IGFBP‐4 Quartiles | P Value | ||||
|---|---|---|---|---|---|
| 1 (4.1–34.3 μg/L) | 2 (34.4–52.0 μg/L) | 3 (52.1–78.3 μg/L) | 4 (78.4–348.8 μg/L) | ||
| Patients, n | 164 | 164 | 164 | 164 | |
| All‐cause mortality, n | 12 | 16 | 36 | 72 | <0.001 |
| Cardiovascular mortality, n | 4 | 6 | 14 | 45 | <0.001 |
| MACE, n | 19 | 30 | 44 | 73 | <0.001 |
C‐statistics are reported individually for NT‐IGFBP‐4, CT‐IGFBP‐4, peak TnI, and CRP (mean and 95% CI). CRP, C‐reactive protein; CT, C‐terminal; IGFBP‐4, insulin‐like growth factor binding protein 4; MACE, major adverse cardiac event; NT, N‐terminal; TnI, Troponin I.
P<0.001 when compared with peak TnI.
P<0.05 when compared with CRP. Log‐rank values are reported a numbers (n) of patients.
Figure 2.
Risk of all‐cause mortality, cardiovascular mortality, and MACE in patients according to quartiles of NT‐IGFBP‐4 (A) and CT‐IGFBP‐4 (B). CT indicates C‐terminal; IGFBP‐4, insulin‐like growth factor binding protein 4; MACE, major adverse cardiac event; NT, N‐terminal.
Cox Regression and Survival
NT‐ and CT‐IGFBP‐4 were predictors of MACE and all‐cause and cardiovascular mortality as continuous variables (log2‐NT‐IGFBP‐4 and log2‐CT‐IGFBP‐4) in both univariable and multivariable analyses (Table 3). After multivariable adjustments, NT‐ and CT‐IGFBP‐4 fragment levels were associated with cardiovascular mortality with HRs per doubling in protein concentration of 2.54 (95% CI 1.59–4.07; P<0.001) and 2.07 (95% CI 1.41–3.04; P<0.001), respectively. The prognostic performances were not affected by inclusion of the covariates (model 2) that did not fulfill the criteria for inclusion in model 1. In the subgroup of patients with LVEF measurements, addition of LVEF to model 1 did not alter the prognostic performance, and NT‐ and CT‐IGFBP‐4 remained significantly associated with all end points (P<0.001).
Table 3.
Multivariable Cox Regression Analyses for All‐Cause Mortality, Cardiovascular Mortality, and MACE at 5 Years
| Model | All‐Cause Mortality | Cardiovascular Mortality | MACE | |||
|---|---|---|---|---|---|---|
| HR (95% CI) | P Value | HR (95% CI) | P Value | HR (95% CI) | P Value | |
| Log2‐NT‐IGFBP‐4 | ||||||
| All patients (n=656) | ||||||
| Univariable | 3.33 (2.69–4.13) | <0.001 | 4.36 (3.25–5.86) | <0.001 | 2.71 (2.22–3.30) | <0.001 |
| Model 1 | 2.13 (1.53–2.97) | <0.001 | 2.54 (1.59–4.07) | <0.001 | 1.97 (1.48–2.61) | <0.001 |
| Model 2 | 2.19 (1.55–3.10) | <0.001 | 2.70 (1.65–4.41) | <0.001 | 1.93 (1.45–2.58) | <0.001 |
| Subgroup with LVEF (n=344) | ||||||
| Univariable | 2.17 (1.47–3.21) | <0.001 | 4.09 (2.08–8.03) | <0.001 | 2.35 (1.74–3.18) | <0.001 |
| Model 1 | 1.75 (1.05–3.10) | 0.031 | 2.62 (0.97–7.05) | 0.057 | 1.66 (1.09–2.51) | 0.017 |
| Univariable plus LVEF | 2.12 (1.42–3.15) | <0.001 | 3.85 (1.93–7.67) | <0.001 | 2.13 (1.57–2.89) | <0.001 |
| Model 1 plus LVEF | 1.72 (1.02–3.09) | 0.042 | 2.63 (1.01–7.05) | 0.047 | 1.59 (1.03–2.45) | 0.035 |
| Log2‐CT‐IGFBP‐4 | ||||||
| All patients (n=656) | ||||||
| Univariable | 2.78 (2.27–3.40) | <0.001 | 3.69 (2.78–4.89) | <0.001 | 2.38 (1.98–2.86) | <0.001 |
| Model 1 | 1.74 (1.33–2.29) | <0.001 | 2.07 (1.41–3.04) | <0.001 | 1.70 (1.35–2.15) | <0.001 |
| Model 2 | 1.78 (1.34–2.37) | <0.001 | 2.13 (1.43–3.16) | <0.001 | 1.70 (1.34–2.16) | <0.001 |
| Subgroup with LVEF (n=344) | ||||||
| Univariable | 2.11 (1.49–2.97) | <0.001 | 4.29 (2.23–8.25) | <0.001 | 2.29 (1.73–3.03) | <0.001 |
| Model 1 | 1.79 (1.11–2.89) | 0.016 | 3.63 (1.32–9.99) | 0.012 | 1.65 (1.16–2.35) | 0.006 |
| Univariable plus LVEF | 2.06 (1.46–2.92) | <0.001 | 4.02 (2.07–7.82) | <0.001 | 2.12 (1.60–2.79) | <0.001 |
| Model 1 plus LVEF | 1.78 (1.10–2.87) | 0.019 | 3.62 (1.31–9.95) | 0.013 | 1.60 (1.11–2.29) | 0.012 |
Model 1: age, sex, estimated glomerular filtration rate, hypertension, high‐density lipoprotein, total cholesterol, systolic blood pressure, hemoglobin, current smoking, diabetes, previous acute myocardial infarction, multivessel coronary artery disease, complex lesion, C‐reactive protein, and peak troponin I. Model 2: model 1 plus body mass index, hypercholesterolemia, low‐density lipoprotein, triglyceride, symptom‐to‐balloon time, and use of glycoprotein IIb/IIIa inhibitor. Separate analyses were performed on the subgroup of patients with LVEF measurements (n=344). Results are reported as HR and 95% CI. One‐unit increase in NT‐ or CT‐IGFBP‐4 on the log2‐scale corresponds to a doubling in NT‐ or CT‐IGFBP‐4, respectively. CT indicates C‐terminal; HR, hazard ratio; IGFBP‐4, insulin‐like growth factor binding protein 4; LVEF, Left ventricular ejection fraction; MACE, major adverse cardiac event; NT, N‐terminal.
A large number of events occurred within the first 3 months after the STEMI. To investigate the association between NT‐ and CT‐IGFBP‐4 and early events, survival analyses were also performed using 3 months as the end point. In both univariable and multivariable analyses, NT‐ and CT‐IGFBP‐4 were predictors of MACE and all‐cause and cardiovascular mortality at 3 months. After multivariable adjustments (model 1), cardiovascular mortality HRs per doubling in protein concentration of NT‐ or CT‐IGFBP‐4 were 1.97 (95% CI 1.48–2.61; P<0.001) and 1.70 (95% CI 1.35–2.15; P<0.001), respectively. Adjusting for the variables in model 2 resulted in HRs of 2.71 (95% CI 1.42–5.17; P<0.001) for NT‐IGFBP‐4 and 2.07 (95% CI 1.23–3.48; P<0.001) for CT‐IGFBP‐4.
Incremental Prognostic Value of IGFBP‐4 Fragments Over Clinical Risk Factors
Table 4 shows the results of analysis using calibration, discrimination, and reclassification metrics to evaluate the incremental usefulness of IGFBP‐4 fragments over clinical risk factors for the prediction of all‐cause mortality, cardiovascular mortality, and MACE. C‐statistics for the prediction of all end points significantly increased when NT‐ or CT‐IGFBP‐4 was added to the clinical model. For the prediction of cardiovascular mortality, NT‐ and CT‐IGFBP‐4 provided significant discriminatory information with increases in C‐statistics of 2.9% (0.38; 5.5%; P=0.025) and 2.7% (0.15; 5.3%; P=0.038), respectively.
Table 4.
Performance of Models for All‐Cause Mortality, Cardiovascular Mortality, and MACE at 5 Years
| Performance Measures | All‐Cause Mortality | Cardiovascular Mortality | MACE | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Clinical Model | Clinical Model Plus log2‐NT‐IGFBP‐4 | Clinical Model Plus log2‐CT‐IGFBP‐4 | Clinical Model | Clinical Model Plus log2‐NT‐IGFBP‐4 | Clinical Model Plus log2‐CT‐IGFBP‐4 | Clinical Model | Clinical Model Plus log2‐NT‐IGFBP‐4 | Clinical Model Plus log2‐CT‐IGFBP‐4 | |
| Discrimination | |||||||||
| C‐statistic (95% CI) |
0.761 (0.720–0.801) Reference |
0.780 (0.742–0.818) P=0.026 |
0.778 (0.740–0.816) P=0.043 |
0.813 (0.765–0.861) Reference |
0.843 (0.799–0.887) P=0.025 |
0.841 (0.794–0.887) P=0.038 |
0.704 (0.664–0.745) Reference |
0.728 (0.688–0.767) P=0.018 |
0.726 (0.687–0.765) P=0.028 |
| Calibration | |||||||||
| Overall performance | |||||||||
| Nagelkerke's R 2 | 0.470 | 0.526 | 0.517 | 0.667 | 0.723 | 0.721 | 0.287 | 0.350 | 0.347 |
| Goodness of fit | |||||||||
| H‐L test (χ2) |
10.86 P=0.301 |
5.35 P=0.803 |
4.29 P=0.886 |
4.53 P=0.874 |
3.46 P=0.943 |
13.2 P=0.154 |
14.2 P=0.115 |
8.30 P=0.504 |
4.76 P=0.850 |
| AIC | 1627 | 1609 | 1612 | 799 | 785 | 786 | 2025 | 2005 | 2006 |
| BIC | 1695 | 1680 | 1684 | 866 | 857 | 858 | 2092 | 2077 | 2078 |
| Likelihood ratio | Reference | P<0.001 | P<0.001 | Reference | P<0.001 | P<0.001 | Reference | P<0.001 | P<0.001 |
| Reclassification | |||||||||
| Category NRI (95% CI) | |||||||||
| Event | Reference | 0.022 (−0.051 to 0.083) | −0.037 (−0.060 to 0.077) | Reference | 0.188 (−0.031 to 0.193) | 0.145 (−0.030 to 0.238) | Reference | 0.030 (−0.053 to 0.067) | 0.018 (−0.057 to 0.063) |
| Nonevent | Reference | 0.113 (0.025–0.188) | 0.081 (−0.009 to 0.166) | Reference | 0.095 (0.016–0.126) | 0.085 (0.010–0.119) | Reference | 0.094 (0.021–0.206) | 0.122 (0.019–0.205) |
| All | Reference | 0.136 (0.004–0.248) | 0.044 (−0.009 to 0.220) | Reference | 0.284 (0.005–0.296) | 0.230 (0.003–0.328) | Reference | 0.124 (0.003–0.239) | 0.141 (−0.007 to 0.238) |
| Continuous NRI (95% CI) | |||||||||
| Event | Reference | 0.162 (0.064–0.348) | 0.250 (0.075–0.369) | Reference | 0.159 (0.013–0.406) | 0.217 (0.070–0.457) | Reference | 0.145 (0.023–0.295) | 0.181 (0.065–0.305) |
| Nonevent | Reference | 0.196 (0.100–0.326) | 0.150 (0.042–0.269) | Reference | 0.179 (0.047–0.346) | 0.182 (0.056–0.373) | Reference | 0.176 (0.086–0.283) | 0.143 (0.045–0.237) |
| All | Reference | 0.358 (0.197–0.645) | 0.400 (0.145–0.610) | Reference | 0.338 (0.111–0.706) | 0.400 (0.168–0.765) | Reference | 0.320 (0.137–0.547) | 0.324 (0.129–0.512) |
| IDI (95% CI) | |||||||||
| Event | Reference | 0.025 (0.009–0.055) | 0.022 (0.005–0.050) | Reference | 0.027 (0.003–0.069) | 0.030 (0.007–0.071) | Reference | 0.022 (0.007–0.048) | 0.020 (0.006–0.044) |
| Nonevent | Reference | 0.007 (0.002–0.015) | 0.005 (0.001–0.013) | Reference | 0.002 (0.000–0.008) | 0.003 (0.001–0.009) | Reference | 0.006 (0.002–0.016) | 0.007 (0.002–0.015) |
| All | Reference | 0.033 (0.011–0.071) | 0.027 (0.006–0.063) | Reference | 0.029 (0.004–0.077) | 0.033 (0.008–0.079) | Reference | 0.028 (0.009–0.063) | 0.027 (0.008–0.059) |
A clinical model was generated based on variables selected for model 1 (age, sex, estimated glomerular filtration rate, hypertension, high‐density lipoprotein, total cholesterol, systolic blood pressure, hemoglobin, current smoking, diabetes, previous acute myocardial infarction, multivessel coronary artery disease, complex lesion, C‐reactive protein, and peak troponin I). The baseline model was then extended by the log2‐transformed NT‐ or CT‐IGFBP‐4 variable. For category NRI, patients were divided into risk categories (<5%, 5–10%, 10–20%, and >20%) and reclassified. AIC indicates Akaike information criterion; BIC, Bayesian information criterion; CT, C‐terminal; H‐L, Hosmer–Lemeshow; IDI, integrated discrimination improvement; IGFBP‐4, insulin‐like growth factor binding protein 4; MACE, major adverse cardiac event; NRI, net reclassification improvement; NT, N‐terminal.
Nagelkerke's R 2 was higher for models extended by NT‐ or CT‐IGFBP‐4. For nested models, R 2 increases with the strength of association, and thus higher R 2 indicates better calibration of a model. The P values for the Hosmer–Lemeshow statistics signified that model‐based estimates aligned with observed outcomes. Akaike information criterion and Bayesian information criterion were lower for all end points in models containing NT‐ or CT‐IGFBP‐4. Global goodness‐of‐fit was better in models including NT‐ or CT‐IGFBP‐4 than in the model with only established risk factors, as evaluated by likelihood ratio tests.
Category NRI, continuous NRI, and IDI were calculated to evaluate whether the addition of the IGFBP‐4 fragments to the clinical model led to any significant risk reclassification of the end points. Continuous NRI and IDI were significantly increased for all end points after inclusion of either NT‐ or CT‐IGFBP‐4. Consequently, participants were correctly reclassified both according to direction of change in risk and actual extent of change. Category‐based NRI using predefined risk categories of <5%, 5% to 10%, 10% to 20%, and >20% risk was significant for all end points after addition of NT‐IGFBP‐4 to the clinical model (95% CI did not contain zero). Of the 69 patients who died of CVD, 33.3% changed risk category when NT‐IGFBP‐4 was included. Of these, 78.3% were correctly reclassified to a higher risk category. Of the 587 patients not suffering from cardiovascular mortality, 27.6% changed risk category and 63.9% were correctly reclassified. The reclassification table of patients initially classified as having a 5‐year risk of cardiovascular mortality based on model 1 who were reclassified to a higher or lower risk category by the addition of NT‐IGFBP‐4 is illustrated in Table S1. CT‐IGFBP‐4 significantly improved risk reclassification for the prediction of cardiovascular mortality but did not yield any improvements for the prediction of all‐cause mortality or MACE (95% CI contained zero).
Discussion
The present study demonstrates that IGFBP‐4 fragments are associated with an increased risk of all‐cause mortality, cardiovascular mortality, and MACE in a cohort of patients with STEMI. The associations were significant both at 3 months and 5 years after the STEMI. Nevertheless, an association alone is insufficient to establish prognostic value, and the use of IGFBP‐4 fragments to determine the risk of a future event as a continuous variable demands cautious consideration. Various risk scores and statistical tests in the cardiovascular area have generated considerable controversy, and no clear consensus has been established. In this study, the incremental prognostic value of the addition of the IGFBP‐4 fragments to the clinical model was assessed by discrimination, calibration, and reclassification analyses. Discrimination and calibration analyses indicated that models including either NT‐ or CT‐IGFBP‐4 were more accurate than the clinical model alone. To assess the magnitude of improvement rather than testing the hypothesis that said improvement was greater than zero, we computed category‐based NRI, continuous NRI, and IDI. Importantly, neither NT‐ nor CT‐IGFBP‐4 had negative overall NRI values, which are indicative of worsened reclassification. When assessing the true discriminatory potential of a new biomarker in contrast to other risk factors, and especially when making comparisons between studies, continuous NRI may be the best metric. It yields the incremental strength of the new biomarker after accounting for correlations with variables included in the clinical model and, unlike HRs, can be compared between studies of biomarkers with different statistical distributions. Although challenging, Pencina et al27 suggested interpretations of the continuous NRI: NRI <0.2 is weak, NRI ≈0.4 is intermediate, and NRI >0.6 can be considered strong. Continuous NRI in the present study ranged between 0.33 and 0.40.
Biomarkers that mirror the degree and severity of the plaque burden are especially interesting because they may also be useful in syndromes such as unstable angina pectoris, in which peak TnI and creatinine kinase–MB levels are not elevated and ECG changes are often lacking or inconclusive. The rationale for the use of IGFBP‐4 fragments as biomarkers for cardiac risk assessment is based on the assumption that PAPP‐A is actively involved in the development of atherosclerosis.12, 13 It has been suggested, however, that IGF‐1 and PAPP‐A may instead be cardioprotective and increase as a compensatory mechanism attempting to limit plaque progression and overall atherosclerotic burden.31, 32 Speaking against this theory, apolipoprotein E–deficient mice with a deletion of the PAPP‐A gene and fed a high‐fat diet show an 80% reduction in plaque area, whereas transgenic overexpression of PAPP‐A accelerates plaque progression.10, 11, 33 Moreover, PAPP‐A substrate binding site inhibition with a neutralizing monoclonal PAPP‐A antibody results in 70% reduction in atherosclerotic plaque development in apolipoprotein E–deficient mice.34 Accordingly, PAPP‐A seems to play an unfavorable functional role in the process of plaque destabilization and may not only be a marker of plaque vulnerability but also be indicative of poor prognosis after the occurrence of a cardiovascular event.1, 4 The use of the IGFBP‐4 fragments as biomarkers is further supported by a number of observations of which a few merit particular attention. First, we recently showed that circulating levels of PAPP‐A correlate with levels of NT‐ and CT‐IGFBP‐4 fragments,12 and this was confirmed in the present study. Second, PAPP‐A–dependent proteolysis of IGFBP‐4 is very specific, and the IGFBP‐4 fragments do not undergo further modifications or truncations in the circulation.35 Third, the IGFBP‐4 fragments display great storage stability.5, 13 Fourth, treatment of patients with heparin promptly increases circulating PAPP‐A levels,4, 12 whereas the 2 IGFBP‐4 fragments are not affected. Thus, heparin may displace cell‐surface associated PAPP‐A, but the overall PAPP‐A activity toward IGFBP‐4 appears unaltered.5 To reaffirm this finding in the present study, PAPP‐A levels were measured in 38 randomly selected patients, and levels were significantly higher in all patients (>20 ng/mL) compared with non–heparin‐treated participants (<2 ng/mL).12 Finally, active PAPP‐A consists of 2 identical PAPP‐A subunits, whereas the inactive form is composed of 2 PAPP‐A subunits covalently linked to 2 inhibitory subunits of the proform of eosinophil major basic protein.7, 36 PAPP‐A assays, however, generally lack the ability to discriminate between the forms.37, 38
So far, studies of IGFBP‐4 fragments as cardiac risk markers in patients have provided conflicting results. In a study by Schulz et al,39 NT‐ and CT‐IGFBP‐4 failed to predict long‐term outcome in patients with stable CVD, whereas Postnikov et al13 showed strong associations between the fragments and short‐term cardiac events in patients with acute myocardial infarction. It has been proposed that the prognostic value might depend on whether patients present with acute or stable disease, and it is intuitive to speculate that IGFBP‐4 fragment levels merely recapitulate infarct size and the acute situation. However, we recently demonstrated that NT‐ and CT‐IGFBP‐4 levels were not altered during the acute phase of a myocardial infarction,5 and the fragments predicted cardiovascular mortality in type 1 diabetes patients without CVD at baseline during 12 years of follow‐up.12 These observations suggest that the discrepancies cannot solely be explained by acute versus stable disease or short versus long follow‐up. We speculate that the different outcomes relate to differences in study populations. It is worth mentioning that in neither this nor previous studies did we find an association between the NT‐ or CT‐IGFBP‐4 fragments and LVEF, peak TnI, location and complexity of the culprit lesion, symptom‐to‐balloon time, or the use of statins and antiplatelet agents, such as glycoprotein IIb/IIIa and cyclooxygenase inhibitors.5, 12
Although our findings are interesting and potentially clinically relevant, the specific mechanisms in STEMI patients are elusive and warrant further investigation. Furthermore, we acknowledge that the value and utility of biomarkers in STEMI patients is limited compared with other CVDs. One could argue that the IGFBP‐4 fragments may be more useful for risk stratification of patients with unstable angina or non‐STEMI. Instead, the potential purpose of a biomarker in STEMI may be to provide long‐term prognostication in patients at hospital discharge in the hope of optimizing treatment management. Considering this perspective, the present findings suggest that the prognostic value of the IGFBP‐4 fragments may be complementary to other risk factors and improve clinical decision making. However, our results cannot necessarily be extrapolated to other patient cohorts, and clinical adoption should await further validation of the ability of the biomarkers to advise therapy that improves patient outcomes.
Study Strengths and Limitations
The cohort size and the follow‐up period resulted in a large number of events, which allowed for extensive adjustment. As for study limitations, samples were collected only at baseline and not throughout the follow‐up period; therefore, we were unable to evaluate changes in variables over time. The study was entirely observational and yielded no evidence of causality. Unfortunately, LVEF measurements were not obtained for all patients and would have strengthened the analyses. Finally, comparisons with additional relevant biomarkers, such as B‐type natriuretic peptide, were not incorporated into the design and are beyond the scope of this study.
Conclusions
Elevated levels of PAPP‐A–generated IGFBP‐4 fragments are significantly associated with increased risk of all‐cause mortality, cardiovascular mortality, and MACE in patients with STEMI. Especially with regard to cardiovascular mortality, the IGFBP‐4 fragments possess incremental prognostic value beyond that of conventional clinical risk factors.
Sources of Funding
The work was supported by The Danish Heart Foundation, Copenhagen, Denmark (13‐04‐R94‐A4650‐22771 to Bjerre and 14‐R97‐A5233‐22853 to Hjortebjerg), The Danish Council for Independent Research, Copenhagen, Denmark (0602‐01195B to Pedersen), The A.P. Møller Foundation for the Advancement of Medical Science, Copenhagen, Denmark (to Frystyk), and by a research grant from The Danish Diabetes Academy supported by The Novo Nordisk Foundation, Odense, Denmark (OL8201 to Hjortebjerg). The funders had no influence on the design or conduct of the study.
Disclosures
None.
Supporting information
Table S1. Reclassification Table Comparing 5‐Year Risk of Cardiovascular Mortality Before and After Addition of N‐Terminal Insulin‐like Growth Factor Binding Protein 4 (NT‐IGFBP‐4) to the Clinical Model
Figure S1. Frequency distribution of log2‐NT‐ and log2‐CT‐IGFBP‐4 and fitted normal curves. Frequency distributions of log2‐NT‐ and log2‐CT‐IGFBP‐4 in the 431 patients without events (black circles) and the 255 patients with events (gray circles), along with the fitted normal curves (black). The vertical lines mark the mean protein levels. CT indicates C‐terminal; IGFBP‐4, insulin‐like growth factor binding protein 4; NT, N‐terminal.
Acknowledgments
The authors are grateful to HyTest Ltd. for providing antigens and antibodies for the IGFBP‐4 assays.
(J Am Heart Assoc. 2017;6:e005358. DOI: 10.1161/JAHA.116.005358.)
References
- 1. Bayes‐Genis A, Conover CA, Overgaard MT, Bailey KR, Christiansen M, Holmes DR, Virmani R, Oxvig C, Schwartz RS. Pregnancy‐associated plasma protein A as a marker of acute coronary syndromes. N Engl J Med. 2001;345:1022–1029. [DOI] [PubMed] [Google Scholar]
- 2. Cosin‐Sales J, Kaski JC, Christiansen M, Kaminski P, Oxvig C, Overgaard MT, Cole D, Holt DW. Relationship among pregnancy associated plasma protein‐A levels, clinical characteristics, and coronary artery disease extent in patients with chronic stable angina pectoris. Eur Heart J. 2005;26:2093–2098. [DOI] [PubMed] [Google Scholar]
- 3. Glerup S, Kløverpris S, Laursen LS, Dagnæs‐Hansen F, Thiel S, Conover CA, Oxvig C. Cell surface detachment of pregnancy‐associated plasma protein‐A requires the formation of intermolecular proteinase‐inhibitor disulfide bonds and glycosaminoglycan covalently bound to the inhibitor. J Biol Chem. 2007;282:1769–1778. [DOI] [PubMed] [Google Scholar]
- 4. Terkelsen CJ, Oxvig C, Nørgaard BL, Glerup S, Poulsen TS, Lassen JF, Møller HJ, Thuesen L, Falk E, Nielsen TT, Andersen HR. Temporal course of pregnancy‐associated plasma protein‐A in angioplasty‐treated ST‐elevation myocardial infarction patients and potential significance of concomitant heparin administration. Am J Cardiol. 2009;103:29–35. [DOI] [PubMed] [Google Scholar]
- 5. Hjortebjerg R, Lindberg S, Hoffmann S, Jensen JS, Oxvig C, Bjerre M, Frystyk J. PAPP‐A and IGFBP‐4 fragment levels in patients with ST‐elevation myocardial infarction treated with heparin and PCI. Clin Biochem. 2015;48:322–328. [DOI] [PubMed] [Google Scholar]
- 6. Hjortebjerg R, Flyvbjerg A, Frystyk J. Insulin growth factor binding proteins as therapeutic targets in type 2 diabetes. Expert Opin Ther Targets. 2014;18:209–224. [DOI] [PubMed] [Google Scholar]
- 7. Conover CA. Key questions and answers about pregnancy‐associated plasma protein‐A. Trends Endocrinol Metab. 2012;23:242–249. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Gude MF, Hjortebjerg R, Oxvig C, Thyø AA, Magnusson NE, Bjerre M, Pedersen SB, Frystyk J. PAPP‐A, IGFBP‐4 and IGF‐II are secreted from human adipose tissue cultures in a depot‐specific manner. Eur J Endocrinol. 2016;175:509–519. [DOI] [PubMed] [Google Scholar]
- 9. Bayes‐Genis A, Conover CA, Schwartz RS. The insulin‐like growth factor axis: a review of atherosclerosis and restenosis. Circ Res. 2000;86:125–130. [DOI] [PubMed] [Google Scholar]
- 10. Conover CA, Mason MA, Bale LK, Harrington SC, Nyegaard M, Oxvig C, Overgaard MT. Transgenic overexpression of pregnancy‐associated plasma protein‐A in murine arterial smooth muscle accelerates atherosclerotic lesion development. Am J Physiol Heart Circ Physiol. 2010;299:H284–H291. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Harrington SC, Simari RD, Conover CA. Genetic deletion of pregnancy‐associated plasma protein‐A is associated with resistance to atherosclerotic lesion development in apolipoprotein E‐deficient mice challenged with a high‐fat diet. Circ Res. 2007;100:1696–1702. [DOI] [PubMed] [Google Scholar]
- 12. Hjortebjerg R, Tarnow L, Jorsal A, Parving H‐H, Rossing P, Bjerre M, Frystyk J. IGFBP‐4 fragments as markers of cardiovascular mortality in type 1 diabetes patients with and without nephropathy. J Clin Endocrinol Metab. 2015;100:3032–3040. [DOI] [PubMed] [Google Scholar]
- 13. Postnikov AB, Smolyanova TI, Kharitonov AV, Serebryanaya DV, Kozlovsky SV, Tryshina YA, Malanicev RV, Arutyunov AG, Murakami MM, Apple FS, Katrukha AG. N‐terminal and C‐terminal fragments of IGFBP‐4 as novel biomarkers for short‐term risk assessment of major adverse cardiac events in patients presenting with ischemia. Clin Biochem. 2012;45:519–524. [DOI] [PubMed] [Google Scholar]
- 14. Lindberg S, Jensen JS, Pedersen SH, Galatius S, Goetze JP, Mogelvang R. MR‐proANP improves prediction of mortality and cardiovascular events in patients with STEMI. Eur J Prev Cardiol. 2015;22:693–700. [DOI] [PubMed] [Google Scholar]
- 15. World Health Organization . International Classification of Diseases, Tenth Revision (ICD‐10). 1992. [Google Scholar]
- 16. Madsen M, Davidsen M, Rasmussen S, Abildstrom SZ, Osler M. The validity of the diagnosis of acute myocardial infarction in routine statistics. J Clin Epidemiol. 2003;56:124–130. [DOI] [PubMed] [Google Scholar]
- 17. Vervoort G, Klein Gunnewiek JMT, Willems HL, Wetzels JFM. Effect of creatinine assay standardization on the performance of Cockcroft–Gault and MDRD formula in predicting GFR. Nephrol Dial Transplant. 2006;21:2998–2999. [DOI] [PubMed] [Google Scholar]
- 18. DeLong ER, DeLong DM, Clarke‐Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44:837–845. [PubMed] [Google Scholar]
- 19. D'Agostino RB, Vasan RS, Pencina MJ, Wolf PA, Cobain M, Massaro JM, Kannel WB. General cardiovascular risk profile for use in primary care: the Framingham Heart Study. Circulation. 2008;117:743–753. [DOI] [PubMed] [Google Scholar]
- 20. Harrell FEJ. Regression Modeling Strategies. With Applications to Linear Models, Logistic Regression, and Survival Analysis. Springer Series in Statistics. Springer International Publishing: Switzerland; 2001. [Google Scholar]
- 21. Therneau TM, Grambsch PM, Fleming TR. Martingale‐based residuals for survival models. Biometrika. 1990;77:216–218. [Google Scholar]
- 22. Schoenfeld D. Partial residuals for the proportional hazards regression model. Biometrika. 1982;69:239–241. [Google Scholar]
- 23. Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361–387. [DOI] [PubMed] [Google Scholar]
- 24. Royston P. Explained variation for survival models. Stata J. 2006;6:83–96. [Google Scholar]
- 25. Gronnesby JK, Borgan O. A method for checking regression models in survival analysis based on the risk score. Lifetime Data Anal. 1996;2:315–328. [DOI] [PubMed] [Google Scholar]
- 26. Pencina MJ, D'Agostino RB Sr, D'Agostino RB Jr, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med. 2008;27:157–172; discussion 207‐112. [DOI] [PubMed] [Google Scholar]
- 27. Pencina MJ, D'Agostino RB, Pencina KM, Janssens ACJW, Greenland P. Interpreting incremental value of markers added to risk prediction models. Am J Epidemiol. 2012;176:473–481. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Paynter NP, Chasman DI, Paré G, Buring JE, Cook NR, Miletich JP, Ridker PM. Association between a literature‐based genetic risk score and cardiovascular events in women. JAMA. 2010;303:631–637. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Ridker P, Buring JE, Rifai N, Cook NR. Development and validation of improved algorithms for the assessment of global cardiovascular risk in women: the Reynolds Risk Score. JAMA. 2007;297:611–619. [DOI] [PubMed] [Google Scholar]
- 30. Hjortebjerg R, Frystyk J. Determination of IGFs and their binding proteins. Best Pract Res Clin Endocrinol Metab. 2014;27:771–781. [DOI] [PubMed] [Google Scholar]
- 31. Conti E, Carrozza C, Capoluongo E, Volpe M, Crea F, Zuppi C, Andreotti F. Insulin‐like growth factor‐1 as a vascular protective factor. Circulation. 2004;110:2260–2265. [DOI] [PubMed] [Google Scholar]
- 32. Juul A, Scheike T, Davidsen M, Gyllenborg J, Jørgensen T. Low serum insulin‐like growth factor I is associated with increased risk of ischemic heart disease. Circulation. 2002;106:939–944. [DOI] [PubMed] [Google Scholar]
- 33. Conover CA, Bale LK, Powell DR. Inducible knock out of pregnancy‐associated plasma protein‐A gene expression in the adult mouse: effect on vascular injury response. Endocrinology. 2013;154:2734–2738. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Conover CA, Bale LK, Oxvig C. Targeted inhibition of pregnancy‐associated plasma protein‐A activity reduces atherosclerotic plaque burden in mice. J Cardiovasc Transl Res. 2016;9:77–79. [DOI] [PubMed] [Google Scholar]
- 35. Konev AA, Smolyanova TI, Kharitonov AV, Serebryanaya DV, Kozlovsky SV, Kara AN, Feygina EE, Katrukha AG, Postnikov AB. Characterization of endogenously circulating IGFBP‐4 fragments—novel biomarkers for cardiac risk assessment. Clin Biochem. 2015;48:774–780. [DOI] [PubMed] [Google Scholar]
- 36. Overgaard MT, Sørensen ES, Stachowiak D, Boldt HB, Kristensen L, Sottrup‐Jensen L, Oxvig C. Complex of pregnancy‐associated plasma protein‐A and the proform of eosinophil major basic protein. J Biol Chem. 2003;278:2106–2117. [DOI] [PubMed] [Google Scholar]
- 37. Dominguez‐Rodriguez A, Abreu‐Gonzalez P, Garcia‐Gonzalez M, Ferrer J, Vargas M. Circulating pregnancy‐associated plasma protein A is not an early marker of acute myocardial infarction. Clin Biochem. 2005;38:180–182. [DOI] [PubMed] [Google Scholar]
- 38. Schulz O, Reinicke M, Krämer J, Berghöfer G, Bensch R, Schimke I, Jaffe A. Pregnancy‐associated plasma protein A values in patients with stable cardiovascular disease: use of a new monoclonal antibody‐based assay. Clin Chim Acta. 2011;412:880–886. [DOI] [PubMed] [Google Scholar]
- 39. Schulz O, Postnikov AB, Smolyanova TI, Katrukha AG, Schimke I, Jaffe AS. Clinical differences between total PAPP‐A and measurements specific for the products of free PAPP‐A activity in patients with stable cardiovascular disease. Clin Biochem. 2014;47:177–183. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1. Reclassification Table Comparing 5‐Year Risk of Cardiovascular Mortality Before and After Addition of N‐Terminal Insulin‐like Growth Factor Binding Protein 4 (NT‐IGFBP‐4) to the Clinical Model
Figure S1. Frequency distribution of log2‐NT‐ and log2‐CT‐IGFBP‐4 and fitted normal curves. Frequency distributions of log2‐NT‐ and log2‐CT‐IGFBP‐4 in the 431 patients without events (black circles) and the 255 patients with events (gray circles), along with the fitted normal curves (black). The vertical lines mark the mean protein levels. CT indicates C‐terminal; IGFBP‐4, insulin‐like growth factor binding protein 4; NT, N‐terminal.