Multiple Cardiac Biomarkers to Improve Prediction of Cardiovascular Events: Findings from the Generation Scotland Scottish Family Health Study

BACKGROUND : Many studies have investigated whether single cardiac biomarkers improve cardiovascular risk prediction for primary prevention but whether a combined approach could further improve risk prediction is unclear. We aimed to test a sex-specific, combined cardiac biomarker approach for cardiovascular risk prediction. METHODS : In the Generation Scotland Scottish Family Health Study, N-terminal pro-B-type natriuretic peptide (NT-proBNP), growth differentiation factor-15 (GDF-15), cardiac troponin I (cTnI), cardiac troponin T (cTnT), and C-reactive protein (CRP) were measured in stored serum using automated immunoassays. Sex-specific Cox models that included SCORE2 risk factors evaluated addition of single and combined biomarkers for prediction of major adverse cardiovascular events (MACE). Combined biomarker models were compared to a baseline model that included SCORE2 risk factors. RESULTS : The study population comprised 18 383 individuals (58.9% women, median age of 48 years [25th–75th percentile, 35–58 years


Introduction
Cardiovascular risk estimation is one of the cornerstones of the primary prevention of cardiovascular disease (1,2).A biomarker-driven approach may refine cardiovascular risk estimates because disease-specific biomarkers can provide additional information about the presence and extent of asymptomatic cardiovascular disease which could help improve individual risk prediction.
Previous studies have shown that N-terminal pro-B-type natriuretic peptide (NT-proBNP), growth differentiation factor-15 (GDF-15), cardiac troponin I (cTnI), cardiac troponin T (cTnT), and C-reactive protein (CRP) predict cardiovascular events in people with established cardiovascular disease, and in the general population (3)(4)(5)(6)(7)(8)(9)(10)(11)(12).Elevations in these biomarkers reflect different underlying pathophysiological features of cardiovascular disease, including myocardial ischemia or injury (cTnI, cTnT), cardiac wall stretch or remodeling (NT-proBNP), inflammation (CRP), and generalized tissue damage .Previous studies have investigated whether single or combined cardiac biomarker approaches may improve risk prediction for primary prevention of cardiovascular disease (13)(14)(15)(16).However, whether a combined biomarker approach, using assays relevant to contemporary clinical biochemistry settings, could further improve prediction of risk in both sexes is unclear.Important sex differences are observed between the relationship of cardiac biomarkers and cardiovascular disease (17)(18)(19), and studies of large size are required for sex-specific evaluation of candidate cardiac biomarkers and their combinations.
We hypothesized that cardiac biomarkers would enhance the prediction of cardiovascular events compared to using conventional risk factors alone, and that an approach that used a combination of biomarkers would be superior to the use of any single biomarker in both women and men.Accordingly, we evaluated associations between NT-proBNP, GDF-15, cTnI, cTnT, and CRP and major adverse cardiovascular events (MACE) in the Generation Scotland Scottish Family Health Study (GS:SFHS).

Materials and Methods
Because of the sensitive nature of the data collected for this study, requests to access the data set from qualified researchers trained in human subject confidentiality protocols should be sent to the Generation Scotland management team at access@generationscotland.org.

STUDY POPULATION
The GS:SFHS is a family-based cohort that enrolled 24 090 participants between 18 and 98 years of age (20,21).Briefly, individuals between 35 and 65 years old were identified at random from participating general practices in Scotland between February 2006 and March 2011.Participants were then asked to identify one or more first-degree relatives ≥18 years old who would also be able to participate.For this study, we excluded participants with cardiovascular disease at baseline, those who had missing cardiac biomarker measurements, or who did not attend the clinical survey.As GDF-15 concentrations are temporarily substantially increased during pregnancy, we also excluded pregnant women (self-reported or when GDF-15 concentrations were >10 000 pg/mL in women ≤45 years old) (22).Participants completed a health questionnaire, and clinical characteristics were measured using a standardized protocol.The cohort is almost enitrely of White ethnicity (99%) (20) and therefore ethnicity is not further reported in this study.Study participants provided written informed consent, including linkage to their medical records.Ethical approval for the GS:SFHS study was obtained from the National Health Service Tayside Research Ethics Committee (Research Ethics Committee reference number 05/S1401/89).The study was conducted according to principles of the Declaration of Helsinki and follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.

BIOMARKER MEASUREMENTS
Serum concentrations of NT-proBNP, GDF-15, highsensitivity cTnT, and high-sensitivity CRP were measured on a Cobas e411 analyzer (Roche Diagnostics).Serum concentration of high-sensitivity cTnI was measured on an ARCHITECT i1000SR analyzer (Abbott Diagnostics).cTnI and cTnT were measured on a first thaw (measured 2016 to 2017) (3), with NT-proBNP and GDF-15 measured on a second thaw (measured 2020 to 2021), and CRP on a third thaw (measured 2021 to 2022).For NT-proBNP, GDF-15, and CRP, the limit of detection (LOD) is set to 10 pg/mL, 400 pg/mL, and 0.1 mg/L by the manufacturer, respectively.For these biomarkers we reported anything less than the LOD at LOD/2 for continuous analysis (5 pg/mL for NT-proBNP, 200 pg/mL for GDF-15, and 0.05 mg/L for CRP).For cTnT, the limit of blank (LOB) and LOD are 3 and 5 ng/L according to the manufacturer, respectively.For cTnI, the LOB and LOD are 1.2 and 1.9 ng/L, respectively (23).For the primary analysis, cTnT and cTnI concentrations below the LOB are set to the LOB value divided by 2 (cTnT, 1.5 ng/L; cTnI, 0.6 ng/L).Proportions of samples above the LOB or LOD are reported in Supplemental Table 1 in the online Data Supplement.During the conduct of this study, we participated in the National External Quality Assurance Scheme (UKNEQAS) for selected biomarkers (cTnI, cTnT, and NT-proBNP).The assays were calibrated, and quality controlled using the manufacturers' reagents.

CLINICAL OUTCOMES
We used the Information Services Division NHS record linkage for Scotland to collect nonfatal cardiovascular events and cause-specific deaths data until the end of August 2021.Information on cause of death was obtained using the NHS Central Register.Nonfatal cardiovascular events and cause-specific deaths were classified using the 10th revision of the International Classification of Diseases (ICD-10).The primary outcome was a composite end point of MACE that included nonfatal myocardial infarction (I21, I22), nonfatal stroke (I63, I64, G45), or cardiovascular death (all codes between I00 and I99).Secondary outcomes were the individual end points of myocardial infarction (nonfatal and fatal, I21, I22), ischemic stroke (nonfatal and fatal, I63, I64, G45), cardiovascular death (all codes between I00 and I99), and noncardiovascular death (other ICD-10 codes not classed as cardiovascular).
We used sex-specific Cox proportional hazard regression models to quantify the relationship between individual biomarkers and MACE.We assessed the impact of using a competing risk framework on biomarker-MACE risk associations, and concluded the differences in risk associations were so marginal that implementing a competing risk framework was not justified in this study.Adjusted sex-specific regression models included the SCORE2 risk factors (age, smoking status, systolic blood pressure, diabetes mellitus, total cholesterol, and high-density lipoprotein cholesterol) and as such did not include adjustment for body mass index (28).Biomarkers were entered in the model as continuous variables.For each biomarker, we applied log 2 transformation and examined them per 1 SD increase in the model accordingly.We bootstrapped the ratio of the hazard ratios (HRs) to compare the strength of the association of individual biomarkers with MACE in the adjusted model, using the HR for NT-proBNP as the denominator.We constructed HR plots to illustrate the relationship between biomarkers and MACE and used natural cubic splines to account for nonlinear relationships between a biomarker and MACE.The proportional hazards assumption was tested by plotting Schoenfeld residuals.
We evaluated combined biomarker approaches in relation to MACE using sex-specific Cox proportional hazard regression models.We assessed all possible combinations for NT-proBNP, GDF-15, cTnI, cTnT, and CRP and entered the biomarkers as continuous variables into the model (log 2 transformed and examined per 1 SD in the model).Similar covariates were included in the models as in the single-biomarker models.We also evaluated discrimination for each biomarker individually and in combination using the Harrell c-statistic, the integrated discrimination index (IDI), and net reclassification index (NRI, continuous and categorical).Testing every possible biomarker combination increases the number of statistical tests conducted, but allows each biomarker combination to be evaluated on the basis of incremental discriminative ability.In addition, the age-specific performance of biomarker models was evaluated for those <40 and ≥40 years old.
Secondary analyses were conducted to verify the robustness of our findings.First, we set cTnI and cTnT concentrations below the LOD at the LOD divided by 2 (rather than using LOB).Second, we evaluated discrimination of biomarker models compared to a base model using SCORE2 risk factors and socioeconomic deprivation status.Socioeconomic deprivation status was determined using the Scottish Index of Multiple Deprivation 2009 score, which is derived from participants' postal codes and compiled using 7 domains of deprivation (income, employment, education, health, access to services, crime, and housing) (29).Third, we additionally evaluated the kidney biomarker creatinine in relation to our primary outcome; we used raw creatinine rather than estimated glomerular filtration rate (eGFR) as a biomarker in order to avoid adjusting eGFR for risk factors already included in its calculation (i.e., age).And finally, we evaluated the associations between biomarkers and secondary outcomes.For completion, we also evaluated the association for all-cause death.
Familial clustering did not affect our analyses, and therefore we only present results from analyses without adjustment for clustering.Multiple imputation by chained equations was used to account for missing data for risk factors (but not missing biomarker concentrations) in the Cox regression models (10 imputed data sets).Statistical analysis was performed using R version 3.6.2.

COMBINING BIOMARKERS FOR THE PREDICTION OF CARDIOVASCULAR EVENTS
Discrimination of the base model using SCORE2 risk factors was excellent for both women and men (c-indices 0.826 and 0.795, respectively).As compared with the baseline model, GDF-15 improved the c-index by +0.010 with an IDI of 0.015 for women (Fig. 3 and online Supplemental Tables 4-7).For women, combined biomarker models that included GDF-15 together with NT-proBNP (+0.012), or GDF-15 plus cTnI (+0.013), but not CRP or cTnT, further improved the c-index.As compared with the baseline model, NT-proBNP improved the c-index by +0.005 with an IDI of 0.014 for men (Fig. 3 and Supplemental Tables 4-7).For men, combined biomarker models that included NT-proBNP together with GDF-15 (+0.007),NT-proBNP plus cTnI (+0.006), and NT-proBNP plus CRP (+0.008), but not cTnT, further improved the c-index.The greatest numerical improvement in discrimination from the base model was achieved with NT-proBNP, GDF-15, and cTnI for women (+0.014) and NT-proBNP, GDF-15, cTnI, and CRP for men (+0.010) (Fig. 3).The combined model incorporating NT-proBNP, GDF-15, and cTnI yielded an IDI of +0.033 and a continuous NRI of +0.254 in women (online Supplemental Tables 6 and 8).The combined model incorporating NT-proBNP, GDF-15, cTnI, and CRP yielded an IDI of +0.017 and a continuous NRI of +0.120 in men (online Supplemental Tables 6 and 9).Generally, cardiac biomarkers improved risk classification among cases more than non-cases.The base SCORE2 models performed better in individuals aged <40 years compared to those aged ≥40 years (c-index 0.831 vs 0.766, online Supplemental Table 10).Although discriminative performance was weaker in those aged ≥40 years, this group had the greatest improvement in the c-index with the addition of NT-proBNP and GDF-15.For NT-proBNP, the change in c-index, compared with the base model, was +0.001 and +0.008 in individuals aged <40 years and ≥40 years, respectively.For GDF-15, the change in c-index, compared with the base model, was +0.000 and +0.0010 in individuals aged <40 years and ≥40 years, respectively.Conversely, cTnI showed greatest improvement in individuals aged <40 years compared to their counterparts (change in c-index: +0.013 vs +0.006).A combined biomarker model that included NT-proBNP, GDF-15, and cTnI yielded a categorical NRI of +0.048 for individuals ≥40 years (online Supplemental Table 11).

SECONDARY ANALYSIS
Our results did not change when we set cTnI and cTnT concentrations below the LOD at the LOD/2 value (online Supplemental Table 12).We observed a similar pattern in the improvement of discrimination, relative to a base model that also included socioeconomic status, for single and combined biomarker models (online Supplemental Table 13).In addition, we evaluated the association of creatinine with primary outcome.Compared with NT-proBNP, the association with creatinine was weaker (online Supplemental Table 14).After adjustment, we found that higher creatinine was associated with MACE in women (HR 1.16 [95% CI, 1.06-1.28])but not in men (HR 1.06 [95% CI, 0.96-1.17]).When evaluating biomarkers for secondary outcome, we observed that NT-proBNP was numerically more strongly associated with myocardial infarction than either cTnT or cTnI in crude models, but similar associations were found in adjusted models (online Supplemental Table 15).NT-proBNP and cTnI were not associated with noncardiovascular death in adjusted models for women, although GDF-15, cTnT, and CRP were associated with noncardiovascular death.

Discussion
We evaluated multiple cardiac biomarkers to predict MACE in a large population-based cohort study.Our main finding was that combining cardiac biomarkers, particularly NT-proBNP, GDF-15, or cTnI, improved estimates of cardiovascular risk over a base model using traditional SCORE2 risk factors in both women and men.
Our study has several strengths.First, we used a large, contemporary population-based cohort study of >18 000 individuals with more than 10 years of followup.Second, the large number of women and men over a wide age range included in this study allowed us to conduct a sex-and age-specific analysis.Third, we were able to measure 5 candidate biomarkers for the prediction of cardiovascular risk in GS:SFHS.This enabled us to perform a systematic evaluation of combined biomarker approaches for cardiovascular risk prediction.Finally, NT-proBNP, cTnT, cTnI, and CRP were measured using assays that are commonly used in clinical biochemistry services around the world, with CRP, cTnT, and cTnT measured by high-sensitivity assays.
A number of studies have evaluated the ability of circulating cardiac biomarkers to predict cardiovascular disease in populations of presumably healthy individuals (13)(14)(15)(16)(30)(31)(32)(33)(34).Recently, Wu et al. evaluated the use of multiple circulating biomarkers in addition to established risk factors, and found that the addition of NT-proBNP and cardiac troponins refined cardiovascular risk estimates (32).Similarly, the Uppsala Longitudinal Study of Adult Men (ULSALM) study of 826 older men, using a research use only proteomics approach, reported NT-proBNP to be the biomarker most strongly associated with cardiovascular disease (14).In line with previous reports, we observed that NT-proBNP, traditionally considered a biomarker of heart failure, was strongly associated with MACE.NT-proBNP was particularly strongly additive to the risk score in those ≥40 years old, an age at which risk prediction models are more often applied in clinical practice.Given increasing interest in using NT-proBNP in some patient groups to screen for heart failure in the absence of signs and symptoms of the condition (35), these collective findings highlight the advantages of prioritizing NT-proBNP for incorporation in commonly applied cardiovascular risk scores.Although our findings are complementary, we provide additional insights by inclusion of 2 additional cardiac biomarkers, cTnT and GDF-15, because recent studies have shown both are independently associated with future cardiovascular events in the general population (3,9,36,37).Our findings show that GDF-15 should also be considered for cardiovascular risk assessment (32)  et al. previously assessed 30 candidate biomarkers in relation to cardiovascular risk prediction in a smaller study (n = 7915) and found that NT-proBNP, cTnI, and CRP when added to established risk factors improved performance when compared to a baseline model in men (13).We extended current knowledge by conducting a comprehensive sex-and age-specific analysis, and showed that NT-proBNP, combined particularly with GDF-15 or cTnI, showed greatest improvement in prediction of cardiovascular risk compared to a baseline model for both women and men and all age groups.A biomarker-driven strategy that uses NT-proBNP combined with GDF-15 or cTnI may contribute to further improvement in cardiovascular risk assessment.Sex disparities in primary prevention and treatment of cardiovascular disease exists (38,39), and using a biomarker-driven risk assessment approach may reduce the gap between women and men.In line with previous studies (17)(18)(19), we observed important sex-differences in cardiac biomarker concentrations and in their association with MACE.While for NT-proBNP, GDF-15, and CRP higher baseline concentrations were observed in women, higher cTnI and cTnT concentrations were observed in men.We also found that the association between all cardiac biomarkers and MACE was numerically stronger in women than men, but this divergence diminished after adjustment for cardiovascular risk factors.These observations are particularly important with respect to integration of cardiac biomarkers in our cardiovascular risk estimation systems.A binary approach that uses a uniform cardiac biomarker threshold will not contribute to reduce current inequalities, but rather may increase the existing gap.In previous research we showed that age and other cardiovascular risk factors like diabetes and body mass index are important modifying factors between sex, cardiac biomarkers, and clinical outcomes (22,40,41).Altogether, this indicates that an approach using sex-specific thresholds to predict cardiovascular disease in the primary care setting is also too simplistic.The digitalization of electronical health records enables the opportunity to embed cardiovascular risk estimation systems that includes cardiac biomarkers as a continuous variable together with other cardiovascular risk factors and preventative therapies for use in clinical practice.Evaluation of implementation of biomarker-driven risk assessment tools in practice is required and an important step to assess the impact of these tools on care for women and men.
Our findings suggest that cTnT, CRP, and creatinine are the weakest independent predictors in a presumably healthy population and are less useful for cardiovascular risk assessment.Although the underlying mechanisms are not well understood, cTnT seems to be more strongly associated with noncardiovascular disease like chronic kidney disease and muscular disease than  cTnI (42,43).We also found that CRP marginally improved risk prediction for men but not for women.This is in line with the findings of a large study that included 246 669 individuals who were presumed to be healthy, which showed that the change in c-index for CRP when added to a base model was +0.0077 (+0.0058 to 0.0096) for men and +0.0007 (−0.0007 to 0.0021) for women (44).Altogether, this indicates that the use of cTnT and CRP for cardiovascular risk estimation may be less incrementally beneficial than other cardiac biomarkers.It should be noted that CRP has been added to the secondary prevention SMART2 risk score (45).
In this study, we report that a model including established cardiovascular risk factors performed well with excellent discrimination and a c-index of approximately 0.8 for both women and men.This is likely at least in part due to the wide age range of our cohort.

Fig. 3 .
Fig. 3. Improved discrimination of major adverse cardiovascular events (MACE), relative to a baseline model, of single and combined biomarker models in women (A) and men (B).Baseline included SCORE2 risk factors: age, smoking status, systolic blood pressure, diabetes mellitus, total cholesterol, and high-density lipoprotein cholesterol.