Age-Related Differences in the Contribution of Systolic Blood Pressure and Biomarkers to Cardiovascular Disease Risk Prediction: The Atherosclerosis Risk in Communities (ARIC) Study

We sought to determine how biomarkers known to be associated with hypertension-induced end-organ injury complement the use of systolic blood pressure (SBP) for cardiovascular disease (CVD) risk prediction at different ages. Using data from visits 2 (1990 to 1992) and 5 (2011 to 2013) of the Atherosclerosis Risk in Communities (ARIC) study, 3 models were used to predict CVD (composite of coronary heart disease, stroke, and heart failure). Model A included traditional risk factors (TRFs) except SBP, model B-TRF plus SBP, and model C-TRF plus biomarkers (high-sensitivity troponin T [hsTnT] and N-terminal pro-B-type natriuretic peptide [NT-proBNP]). Harrel's C-statistics were used to assess risk discrimination for CVD comparing models B and A and C and B. At visit 2, the addition of SBP to TRF (model B vs model A) significantly improved the C-statistic (∆C-statistic, 95% confidence interval 0.010, 0.007 to 0.013) whereas the addition of hsTnT to TRF (model C vs model B) decreased the C-statistic (∆C-statistic -0.0038, -0.0075 to -0.0001) compared with SBP. At visit 5, the addition of SBP to TRF did not significantly improve the C-statistic (∆C-statistic 0.001, -0.002 to 0.005) whereas the addition of both hsTnT and NT-proBNP to TRF significantly improved the C-statistic compared with SBP (∆C-statistic 0.028, 0.015 to 0.041 and 0.055, 0.036 to 0.074, respectively). In summary, the incremental value of SBP for CVD risk prediction diminishes with age whereas the incremental value of hsTnT and NT-proBNP increases with age.

Prognostic Value of Ankle-Brachial Index in Prediction of Cardiovascular Events in an Asian Population with Multiple Atherosclerotic Risk Factors

We aimed to evaluate the incremental prognostic value after incorporation of the ankle-brachial index (ABI) into the 10-year pool cohort equation (PCE) risk model in patients with multiple risk factors (MRFs). A total of 4332 MRFs patients were divided into 2 groups as ABI ≤.9 or >.9. The primary outcome was hard cardiovascular events (hCVE: including cardiovascular death, myocardial infarction, or ischemic stroke) over a median follow-up of 36 months. The Cox proportional hazards survival model, C-statistic, and net reclassification indices (NRI) were used. The occurrence of the primary outcome in the ABI ≤.9 group (3.7%) was significantly greater than in the ABI > .9 group (1.3%), P < .001. ABI is an independent predictor of hCVE in addition to the variables in the standard risk model (age, gender, and smoking status). ABI modestly improved the C-index when added to the PCE risk model (PCE .70 vs ABI+PCE .74). The addition of ABI to the PCE risk model did not significantly improve the classification of patients (NRI -.029; 95% CI: -.215 to .130). Despite ABI being one of the independent predictors of hCVE, integration of ABI into the PCE model did not improve the efficacy of risk reclassification in patients with MRFs.

Plasma Amino Acid Neurotransmitters and Ischemic Stroke Prognosis: A Multicenter Prospective Study

BACKGROUND

Plasma amino acid neurotransmitter dysregulation is suggested to be implicated in the development of ischemic stroke, but its prognostic value for ischemic stroke remains controversial.

OBJECTIVE

We aimed to prospectively investigate the associations between plasma amino acid neurotransmitters levels and adverse outcomes after ischemic stroke in a large-scale multicenter cohort study.

METHODS

We measured 4 plasma amino acid neurotransmitters (glutamic acid, aspartic acid, gamma-aminobutyric acid, and glycine) among 3486 patients with ischemic stroke from 26 hospitals across China. The primary outcome is the composite outcome of death or major disability (modified Rankin Scale score ≥3) at 3 mo after ischemic stroke.

RESULTS

After multivariate adjustment, the odds ratios of death or major disability for the highest versus the lowest quartile were 2.04 (95% confidence interval [CI]: 1.60,2.59; P-trend < 0.001) for glutamic acid, 2.03 (95% CI: 1.59, 2.59; P-trend < 0.001) for aspartic acid, 1.35 (95% CI: 1.06, 1.71; P-trend = 0.016) for gamma-aminobutyric acid, and 0.54 (95% CI: 0.42, 0.69; P-trend < 0.001) for glycine. Each standard deviation increment of log-transformed glutamic acid, aspartic acid, gamma-aminobutyric acid, and glycine was associated with a 34%, 34%, and 9% increased risk, and a 23% decreased risk of death or major disability, respectively (all P < 0.05), in a linear fashion as indicated by spline regression analyses (all P for linearity < 0.05). Addition of the 4 plasma amino acid neurotransmitters to conventional risk factors significantly improved the risk reclassification, as evidenced by integrated discrimination improvement and net reclassification improvement (all P < 0.05).

CONCLUSIONS

Increased glutamic acid, aspartic acid, and gamma-aminobutyric acid and decreased glycine in plasma are associated with adverse outcomes after ischemic stroke, suggesting that plasma amino acid neurotransmitters may be potential intervention targets for improving prognosis of ischemic stroke. The CATIS trial was registered at clinicaltrials.gov (registration number: NCT01840072; URL: ===https://clinicaltrials.gov/ct2/show/NCT01840072?cond=NCT01840072&draw=2&rank=1).

The modification of individual factors on association between serum 25(OH)D and incident type 2 diabetes: Results from a prospective cohort study

Several epidemiological studies have suggested an association between low vitamin D status and increased risk for type 2 diabetes (T2D). This study aimed to explore the dose-response relationship of serum 25-hydroxyvitamin D [25(OH)D] concentrations with incident T2D and the interaction between serum 25(OH)D with individual factors on T2D risk. A total of 1,926 adults without diabetes (mean age: 52.08 ± 13.82 years; 42% men) were prospectively followed for 36 months. Cox proportional hazards model and restricted cubic spline analysis were performed to assess the association and dose-response relationship between serum 25(OH)D and T2D incidence. Both additive and multiplicative interactions were calculated between serum 25(OH)D and individual factors. The net reclassification index (NRI) was used to evaluate the improvement of risk prediction of T2D by adding serum 25(OH)D to traditional risk factors. There were 114 new T2D cases over a mean follow-up of 36 months. Serum 25(OH)D was not associated with T2D incidence, and no significant dose-response relationship was found in the total population. However, stratified analyses suggested a non-linear inverse relationship among individuals with baseline fasting plasma glucose (FPG) <5.6 mmol/L (P _overall = 0.061, P _non-linear = 0.048). And a significant multiplicative interaction was observed between serum 25(OH)D and FPG on T2D risk (P = 0.005). In addition, we found a significant additive interaction of low serum 25(OH)D with older age (RERI = 0.897, 95% CI: 0.080-1.714; AP = 0.468, 95% CI: 0.054-0.881), male (AP = 0.441, 95% CI: 0.010-0.871), and insufficient physical activity (RERI = 0.875, 95% CI: 0.204-1.545; AP = 0.575, 95% CI: 0.039-1.111) on T2D risk. Significant additive interactions were also observed between vitamin D deficiency/insufficiency with male, overweight/obesity, and insufficient physical activity on T2D risk. Moreover, adding low serum 25(OH)D to a model containing established risk factors yielded significant improvements in the risk reclassification of T2D (NRI = 0.205, 95% CI: 0.019-0.391). Our results indicated a non-linear relationship of serum 25(OH)D concentrations with T2D risk among individuals with normal FPG and additive interactions of serum 25(OH)D with gender, overweight/obesity, and physical activity on T2D risk, suggesting the importance of outdoor exercise.

Coronary computed tomographic angiography derived findings and risk score improves the allocation of lipid lowering therapy compared to clinical score

The initiation of therapy for atherosclerotic cardiovascular disease (ASVCD) is currently guided by cohort-based risk scores. Coronary computed tomographic angiography (CCTA) offers more personalised risk assessments to optimise therapy allocation. This study investigates the utility of CCTA determined coronary stenosis (both obstructive and non-obstructive plaque) to guide allocation of lipid lowering therapy. A retrospective analysis of 450 patients with CCTA performed for the assessment of chest pain at a single centre was conducted. Baseline characteristics, investigations, treatments and clinical outcomes were recorded. The allocation of lipid lowering therapy was evaluated with three models, cohort-based risk score (pooled cohort equation), a previously validated CCTA based clinical risk score (pooled cohort equation and CCTA findings) and CCTA alone (without clinical characteristics). The reclassification analysis included 266 patients. Compared to the cohort-based risk score, CCTA based clinical risk score in total reassigned 23% of patients. CCTA alone compared to the CCTA based clinical risk score correctly reassigned 23% and incorrectly reassigned 10%. When comparing the performance of CCTA alone against the cohort-based risk score, both the additive NRI of 25.8 (95% CI 4.12-37.56) and absolute NRI of 13.2 (95% CI 5.88-19.77) was significant. Revascularisation was required in 3% with a low cohort-based risk, but no patients with low risk as per CCTA alone or CCTA based clinical risk score required revascularisation The use of a CCTA based clinical risk score or CCTA alone compared to cohort-based risk scores can improve the allocation of lipid lowering therapy.

Prognostic Value of Stress Cardiac Magnetic Resonance in Patients With Known Coronary Artery Disease

OBJECTIVES

This study sought to determine whether stress cardiac magnetic resonance (CMR) provides clinically relevant risk reclassification in patients with known coronary artery disease (CAD) in a multicenter setting in the United States.

BACKGROUND

Despite improvements in medical therapy and coronary revascularization, patients with previous CAD account for a disproportionately large portion of CV events and pose a challenge for noninvasive stress testing.

METHODS

From the Stress Perfusion Imaging in the United States (SPINS) registry, we identified consecutive patients with documented CAD who were referred to stress CMR for evaluation of myocardial ischemia. The primary outcome was nonfatal myocardial infarction (MI) or cardiovascular (CV) death. Major adverse CV events (MACE) included MI/CV death, hospitalization for heart failure or unstable angina, and late unplanned coronary artery bypass graft. The prognostic association and net reclassification improvement by ischemia for MI/CV death were determined.

RESULTS

Out of 755 patients (age 64 ± 11 years, 64% male), we observed 97 MI/CV deaths and 210 MACE over a median follow-up of 5.3 years. Presence of ischemia demonstrated a significant association with MI/CV death (HR: 2.30; 95% CI: 1.54-3.44; P < 0.001) and MACE (HR: 2.24 ([95% CI: 1.69-2.95; P < 0.001). In a multivariate model adjusted for CV risk factors, ischemia maintained strong association with MI/CV death (HR: 1.84; 95% CI: 1.17-2.88; P = 0.008) and MACE (HR: 1.77; 95% CI: 1.31-2.40; P < 0.001) and reclassified 95% of patients at intermediate pretest risk (62% to low risk, 33% to high risk) with corresponding changes in the observed event rates of 1.4% and 5.3% per year for low and high post-test risk, respectively.

CONCLUSIONS

In a multicenter cohort of patients with known CAD, CMR-assessed ischemia was strongly associated with MI/CV death and reclassified patient risk beyond CV risk factors, especially in those considered to be at intermediate risk. Absence of ischemia was associated with a <2% annual rate of MI/CV death. (Stress CMR Perfusion Imaging in the United States [SPINS] Study; NCT03192891).

Simple Laboratory Test-Based Risk Scores in Coronary Catheterization: Development, Validation, and Comparison to Conventional Risk Factors

BACKGROUND

We developed and validated laboratory test-based risk scores (i.e., lab risk scores) to reclassify mortality risk among patients undergoing their first coronary catheterization.

METHODS

Patients were catheterized between 2009 and 2015 in Calgary, Alberta, Canada (n = 14 135, derivation cohort), and in Edmonton, Alberta, Canada (n = 12 143, validation cohort). Logistic regression with group LASSO (least absolute shrinkage and selection operator) penalty was used to select quintiles of the last laboratory tests (red blood cell count, mean corpuscular hemoglobin concentration, mean corpuscular hemoglobin, mean corpuscular volume, red cell distribution width, platelet count, total white blood cell count, plasma sodium, potassium, chloride, CO2, international normalized ratio, estimated glomerular filtration rate) performed <30 days before catheterization and by age and sex that were significantly associated with death ≤60  and >60 days after catheterization. Follow-up was until 2016. Risk scores were developed from significant tests, internally validated in Calgary among bootstrap samples and externally validated in Edmonton after recalibration using coefficients developed in Calgary. Interaction tests were performed, and net reclassification improvement vs conventional demographic and clinical risk factors was determined.

RESULTS

Lab risk scores were strongly associated with mortality (29-40× for top vs bottom quintile, P for trends <0.01), had good discrimination and were well calibrated in Calgary (C = 0.80-0.85, slope = 0.99-1.01) and Edmonton (C = 0.80-0.82; slope = 1.02-1.05)-similar to demographic and clinical risk factors alone. Associations were attenuated by several comorbidities; however, scores appropriately reclassified 11%-20% of deaths (both follow-up periods) and 6%-9% of survivors (>60 days) after catheterization vs demographic and clinical risk factors.

CONCLUSIONS

In 2 populations of patients undergoing their first coronary catheterization, risk scores based on simple laboratory tests were as powerful as a combination of demographic and clinical risk factors in predicting mortality. Lab risk scores should be used for patients undergoing coronary catheterization.

Prediction of Hemorrhagic Transformation After Ischemic Stroke: Development and Validation Study of a Novel Multi-biomarker Model

Objectives: We aimed to develop and validate a novel multi-biomarker model for predicting hemorrhagic transformation (HT) risk after acute ischemic stroke (AIS).

Methods: We prospectively included patients with AIS admitted within 24 h of stroke from January 1st 2016 to January 31st 2019. A panel of 17 circulating biomarkers was measured and analyzed in this cohort. We assessed the ability of individual circulating biomarkers and the combination of multiple biomarkers to predict any HT, symptomatic HT (sHT) and parenchymal hematoma (PH) after AIS. The strategy of multiple biomarkers in combination was then externally validated in an independent cohort of 288 Chinese patients.

Results: A total of 1207 patients with AIS (727 males; mean age, 67.2 ± 13.9 years) were included as a derivation cohort, of whom 179 patients (14.8%) developed HT. The final multi-biomarker model included three biomarkers [platelets, neutrophil-to-lymphocyte ratios (NLR), and high-density lipoprotein (HDL)] from different pathways, showing a good performance for predicting HT in both the derivation cohort (c statistic = 0·64, 95% CI 0·60–0·69), and validation cohort (c statistic = 0·70, 95% CI 0·58–0·82). Adding these three biomarkers simultaneously to the basic model with conventional risk factors improved the ability of HT reclassification [net reclassification improvement (NRI) 65.6%, P < 0.001], PH (NRI 64.7%, P < 0.001), and sHT (NRI 71.3%, P < 0.001).

Conclusion: This easily applied multi-biomarker model had a good performance for predicting HT in both the derivation and external validation cohorts. Incorporation of biomarkers into clinical decision making may help to identify patients at high risk of HT after AIS and warrants further consideration.

Cardiovascular Calcification as a Marker of Increased Cardiovascular Risk and a Surrogate for Subclinical Atherosclerosis: Role of Echocardiography

The risk prediction of future cardiovascular events is mainly based on conventional risk factor assessment by validated algorithms, such as the Framingham Risk Score, the Pooled Cohort Equations and the European SCORE Risk Charts. The identification of subclinical atherosclerosis has emerged as a promising tool to refine the individual cardiovascular risk identified by these models, to prognostic stratify asymptomatic individuals and to implement preventive strategies. Several imaging modalities have been proposed for the identification of subclinical organ damage, the main ones being coronary artery calcification scanning by cardiac computed tomography and the two-dimensional ultrasound evaluation of carotid arteries. In this context, echocardiography offers an assessment of cardiac calcifications at different sites, such as the mitral apparatus (including annulus, leaflets and papillary muscles), aortic valve and ascending aorta, findings that are associated with the clinical manifestation of atherosclerotic disease and are predictive of future cardiovascular events. The aim of this paper is to summarize the available evidence on clinical implications of cardiac calcification, review studies that propose semiquantitative ultrasound assessments of cardiac calcifications and evaluate the potential of ultrasound calcium scores for risk stratification and prevention of clinical events.

Incremental prognostic value of SPECT-MPI in chronic kidney disease: A reclassification analysis

BACKGROUND

Traditional cardiovascular (CV) risk factors have limited predictive value of CV mortality in patients with chronic kidney disease (CKD, creatinine clearance less than 60 mL/minute per 1.73 m2). The aim of this study was to evaluate incremental and independent prognostic value of single-photon emission computerized tomography-myocardial perfusion imaging (SPECT-MPI) across continuum of renal function.

METHODS

We retrospectively studied 11,518 (mean age, 65 ± 12 years; 52% were men) patients referred for a clinical indication of SPECT-MPI between April 2004 and May 2009. Primary end point was composite of cardiac death and non-fatal myocardial infarction (CD/MI). We examined the relationship of total perfusion defect (TPD) and CD/MI in multiple Cox regression models for CV risk factors and GFR. The incremental predictive value of TPD was examined using Harrell's c-index, net reclassification index (NRI), and integrated discrimination index (IDI).

RESULTS

Over a median follow-up of 5 years (25th to 75th percentiles, 3.0-6.5 years), 1,692 (14.5%) patients experienced CD/MI (740 MI and 1,182 CD). In a multivariable model adjusted for traditional CV risk factors and GFR, the presence of a perfusion defect was independently associated with increased risk of CD/MI (HR = 2.10; 95% CI 1.81, 2.43, p < .001). Using Cox regression, TPD improved the discriminatory ability beyond traditional CV risk factors and GFR [from AUC = 0.725, (95% CI 0.712-0.738) to 0.784, (95% CI 0.772-0.796), p < .0001]. Furthermore, TPD improves risk stratification of CKD patients over and above traditional CV risk factors and GFR [NRI = 14%, 95% CI (12%-16%, p < .001) and relative IDI = 60%, 95% CI (51%, 66%, p < .001)].

CONCLUSIONS

Across the spectrum of renal function, SPECT-MPI perfusion defects independently and incrementally reclassified patients for their risk of CD/MI, beyond traditional CV risk factors.

Reclassification calibration test for censored survival data: performance and comparison to goodness-of-fit criteria

BACKGROUND

The risk reclassification table assesses clinical performance of a biomarker in terms of movements across relevant risk categories. The Reclassification-Calibration (RC) statistic has been developed for binary outcomes, but its performance for survival data with moderate to high censoring rates has not been evaluated.

METHODS

We develop an RC statistic for survival data with higher censoring rates using the Greenwood-Nam-D'Agostino approach (RC-GND). We examine its performance characteristics and compare its performance and utility to the Hosmer-Lemeshow goodness-of-fit test under various assumptions about the censoring rate and the shape of the baseline hazard.

RESULTS

The RC-GND test was robust to high (up to 50%) censoring rates and did not exceed the targeted 5% Type I error in a variety of simulated scenarios. It achieved 80% power to detect better calibration with respect to clinical categories when an important predictor with a hazard ratio of at least 1.7 to 2.2 was added to the model, while the Hosmer-Lemeshow goodness of fit (gof) test had power of 5% in this scenario.

CONCLUSIONS

The RC-GND test should be used to test the improvement in calibration with respect to clinically-relevant risk strata. When an important predictor is omitted, the Hosmer-Lemeshow goodness-of-fit test is usually not significant, while the RC-GND test is sensitive to such an omission.

Serum Total 25-OH Vitamin D Adds Little Prognostic Value in Patients Undergoing Coronary Catheterization

BACKGROUND

Vitamin D deficiency is associated with an increased risk of cardiovascular disease; however, it is unclear whether vitamin D status should be considered in clinical risk assessments of patients with cardiovascular disease.

METHODS AND RESULTS

This study included 2975 patients who had their first serum total 25-hydroxy vitamin D (25-OH vitamin D) measurement before their first coronary catheterization in Alberta, Canada. Cox regression was used to examine associations between 25-OH vitamin D and mortality risk after adjusting for demographic and clinical risk factors. Interactions were tested using multiplicative terms, and prognostic value was assessed using measures of model discrimination, fit, calibration and net reclassification improvement. There were 401 deaths over a median of 5.8 years of follow-up. Serum total 25-OH vitamin D was inversely associated with mortality after adjusting for demographic and clinical risk factors, which was largely driven by excess risk in the bottom quintile (hazard ratio 1.84 for bottom versus top quintile, 95% CI 1.36-2.50, P for trend <0.001). Associations were weaker in the presence of several competing risk factors (e.g., advanced age; P for interactions <0.05). Adding 25-OH vitamin D to a model containing demographic and clinical risk factors yielded similar discrimination, model fit, and calibration and only modest improvements in risk reclassification (net reclassification improvement 1.9% for deaths, 2.3% for survivors).

CONCLUSIONS

Pre-catheterization, serum total 25-OH vitamin D was inversely associated with mortality risk after adjusting for established demographic and clinical risk factors. This association was attenuated by several competing risk factors. Overall, 25-OH vitamin D added little prognostic value over established risk factors; therefore, its measurement is not warranted in patients undergoing coronary catheterization.

EVALUATING RISK-PREDICTION MODELS USING DATA FROM ELECTRONIC HEALTH RECORDS

The availability of data from electronic health records facilitates the development and evaluation of risk-prediction models, but estimation of prediction accuracy could be limited by outcome misclassification, which can arise if events are not captured. We evaluate the robustness of prediction accuracy summaries, obtained from receiver operating characteristic curves and risk-reclassification methods, if events are not captured (i.e., "false negatives"). We derive estimators for sensitivity and specificity if misclassification is independent of marker values. In simulation studies, we quantify the potential for bias in prediction accuracy summaries if misclassification depends on marker values. We compare the accuracy of alternative prognostic models for 30-day all-cause hospital readmission among 4548 patients discharged from the University of Pennsylvania Health System with a primary diagnosis of heart failure. Simulation studies indicate that if misclassification depends on marker values, then the estimated accuracy improvement is also biased, but the direction of the bias depends on the direction of the association between markers and the probability of misclassification. In our application, 29% of the 1143 readmitted patients were readmitted to a hospital elsewhere in Pennsylvania, which reduced prediction accuracy. Outcome misclassification can result in erroneous conclusions regarding the accuracy of risk-prediction models.

Prognostic value of stress myocardial perfusion positron emission tomography: results from a multicenter observational registry

OBJECTIVES

The primary objective of this multicenter registry was to study the prognostic value of positron emission tomography (PET) myocardial perfusion imaging (MPI) and the improved classification of risk in a large cohort of patients with suspected or known coronary artery disease (CAD).

BACKGROUND

Limited prognostic data are available for MPI with PET.

METHODS

A total of 7,061 patients from 4 centers underwent a clinically indicated rest/stress rubidium-82 PET MPI, with a median follow-up of 2.2 years. The primary outcome of this study was cardiac death (n = 169), and the secondary outcome was all-cause death (n = 570). Net reclassification improvement (NRI) and integrated discrimination analyses were performed.

RESULTS

Risk-adjusted hazard of cardiac death increased with each 10% myocardium abnormal with mildly, moderately, or severely abnormal stress PET (hazard ratio [HR]: 2.3 [95% CI: 1.4 to 3.8; p = 0.001], HR: 4.2 [95% CI: 2.3 to 7.5; p < 0.001], and HR: 4.9 [95% CI: 2.5 to 9.6; p < 0.0001], respectively [normal MPI: referent]). Addition of percent myocardium ischemic and percent myocardium scarred to clinical information (age, female sex, body mass index, history of hypertension, diabetes, dyslipidemia, smoking, angina, beta-blocker use, prior revascularization, and resting heart rate) improved the model performance (C-statistic 0.805 [95% CI: 0.772 to 0.838] to 0.839 [95% CI: 0.809 to 0.869]) and risk reclassification for cardiac death (NRI 0.116 [95% CI: 0.021 to 0.210]), with smaller improvements in risk assessment for all-cause death.

CONCLUSIONS

In patients with known or suspected CAD, the extent and severity of ischemia and scar on PET MPI provided powerful and incremental risk estimates of cardiac death and all-cause death compared with traditional coronary risk factors.

Development and evaluation of multi-marker risk scores for clinical prognosis

Heart failure research suggests that multiple biomarkers could be combined with relevant clinical information to more accurately quantify individual risk and guide patient-specific treatment strategies. Therefore, statistical methodology is required to determine multi-marker risk scores that yield improved prognostic performance. Development of a prognostic score that combines biomarkers with clinical variables requires specification of an appropriate statistical model and is most frequently achieved using standard regression methods such as Cox regression. We demonstrate that care is needed in model specification and that maximal use of marker information requires consideration of potential non-linear effects and interactions. The derived multi-marker score can be evaluated using time-dependent receiver operating characteristic methods, or risk reclassification methods adapted for survival outcomes. We compare the performance of alternative model accuracy methods using simulations, both to evaluate power and to quantify the potential loss in accuracy associated with use of a sub-optimal regression model to develop the multi-marker score. We illustrate development and evaluation strategies using data from the Penn Heart Failure Study. Based on our results, we recommend that analysts carefully examine the functional form for component markers and consider plausible forms for effect modification to maximize the prognostic potential of a model-derived multi-marker score.