Long-Term Prognosis After Coronary Artery Calcification Testing in Asymptomatic Patients

Context Clinicians use coronary artery calcification scores to predict the risk for myocardial infarction from coronary artery disease. The score also predicts all-cause mortality, but more is known about the accuracy of its short-term predictions than its long-term predictions. Contribution The study found that the score accurately predicted all-cause mortality at 15 years in asymptomatic patients. Caution The study was limited to a single center. Implication Coronary artery calcification scores may help motivate patients with high scores to adopt healthier lifestyles and may help researchers stratify study patients more effectively. A coronary artery calcification (CAC) test is used to estimate cardiovascular prognosis and provides additive information, above and beyond traditional cardiac risk factors, to estimate important clinical outcomes (1, 2). The published data show a strong relationship between the extent of CAC and adverse clinical outcomes across diverse asymptomatic patient subgroups and population cohorts (110). However, most risk-stratification evidence on CAC involves short-term prognosis, with few registries reporting follow-up beyond 5 years (1, 11). For screening with CAC, a large proportion of tested persons is middle-aged and has the possibility for long-term survival. According to published data, CAC incidence increases with age and the potential interaction between CAC and the length of follow-up is important (1, 6, 7). Given the progressive nature of atherosclerotic disease and the prevalence of potentiating cardiac risk factors, an understanding of the long-term sequelae of low- to high-risk CAC scores may prove useful in defining the value of cardiovascular testing for patients of various age groups. Thus, the aim of this report was to describe the prognostic significance of long-term follow-up across an array of CAC scores for asymptomatic patient subgroups of younger and older women and men. Methods Study Population From 1996 to 1999, primary care physicians referred 9715 patients to 1 outpatient clinic as part of a cardiology outreach screening program in the Tricare Healthcare System. These patients did not have symptoms of coronary artery disease and were from the area surrounding Nashville, Tennessee (86% were white, 8% were African American, 4% were Hispanic, and 2% were Asian). The median annual per capita income was $33000. Each patient paid $69 out of pocket for the procedure at the time of service. All patients signed informed consent for the index CAC scan and follow-up procedures; Centennial Medical Center (Nashville, Tennessee) provided institutional approval. Deidentified data were sent to 3 participating institutions (Emory University School of Medicine, Atlanta, Georgia; Cedars-Sinai Medical Center, Los Angeles, California; and Weill Cornell Medical College, New York, New York) for analysis. Institutional review board approval was garnered for data analysis at each institution. Five-year follow-up was previously reported in a subgroup of these patients (3). Cardiac Risk Factor Data A detailed history of cardiac risk factors was ascertained at the time of testing, as previously described (3). In addition to age, a medical history of hypertension was collected and defined as having had a prescription for an antihypertensive medication or documented blood pressure 140/90 mm Hg or higher. A history of diabetes was defined as having had a prescription for an antidiabetic medication or a history of elevated blood glucose levels greater than 7 mmol/L (>126 mg/dL). Patients who received a dyslipidemic medication or those with a history of elevated cholesterol levels were classified as dyslipidemic. Patients who had a relative with a history of coronary heart disease (CHD) were considered to have a family history of CHD. CAC Image Acquisition and Interpretation All patients had CAC imaging using electron beam tomography or multislice computed tomography. The correlation between CAC measurements derived from electron beam tomography with multislice computed tomography was high (r= 0.99; n= 100), consistent with previous reports (12, 13). Prior reports have detailed our methods for measuring CAC (3, 7, 9, 11). The CAC score was calculated using the method of Agatston and colleagues (14) and grouped as 0, 1 to 10, 11 to 99, 100 to 399, 400 to 999, and 1000 or greater (3). Follow-up Methods Long-term follow-up was undertaken in a consecutive series of 9715 patients. Survival status was obtained by querying the National Death Index, a central computerized index of death record information from the National Center for Health Statistics (15). Follow-up status was ascertained through 1 May 2014. Mean follow-up for surviving patients was 14.6 years (range, 12.9 to 16.8). During this observational period, 936 patients were confirmed as dead. Statistical Analysis The CAC subgroups with binary risk factor variables were compared by using chi-square analysis. The continuous measures with binary variables, such as age, were compared by using t tests or analyses of variance statistics, where appropriate. The primary end point of this analysis was time to all-cause mortality. Univariable and multivariable Cox proportional hazards models were used to estimate the relationship and added value of cardiac risk factors and CAC scores. Unadjusted, overall survival curves were plotted by the CAC subgroups. Hazard ratios and 95% CIs were calculated from the Cox model. Adjusted models included the CAC score and cardiac risk factor variables, including hypertension, diabetes, dyslipidemia, age, sex, and family history of CHD. The proportional hazards assumption was evaluated by assessing the constancy of the parallel plotted lines in the loglog graph and tested on the basis of Schoenfeld residuals. When we included a variable representing equipment use (electron beam tomography vs. multislice computed tomography), the prognostic models did not change. We examined the goodness of fit of the multivariable models using the HosmerLemeshow test on the basis of quantiles of risk, including the cardiac risk factor model and the combined risk factor and CAC model; both were nonsignificant (P> 0.80). The net reclassification improvement (NRI) statistic was calculated, including the percentage of deaths and survivors correctly reclassified when comparing 2 models (model 1 with available cardiac risk factor variables, including age, sex, hypertension, diabetes, dyslipidemia, cigarette smoking, and a family history of CHD, and model 2 with the cardiac risk factor variables and the CAC score) using the methods of Pencina and colleagues (16, 17). Although no specific cut points for 15-year mortality are established, we evaluated 2 sets of cut points for the NRI corresponding to 7.5% to 22.5% mortality (18, 19) and alternatively used cut points of less than 10%, 10% to 19.9%, and 20% or greater (1921). From model 1, we calculated the predicted probability of 15-year mortality and then categorized a variable on the basis of quartile measurements. The quartiles of predicted mortality were compared by cardiac risk factors using chi-square analysis. Statistical analysis was done using Stata, version 13.0 (StataCorp); SAS, version 9.2 (SAS Institute); and SPSS, version 22.0 (IBM SPSS). Role of the Funding Source This study received no external funding. Results Cardiac Risk Factors and CAC Descriptive Statistics In Table 1, the differences across cardiac risk factors were reported by quartiles of 15-year predicted mortality. Patients in the lower quartiles of predicted mortality were generally middle-aged, were less often female, and had a higher prevalence of dyslipidemia and family history of CHD. By comparison, the patients in the higher quartiles of predicted mortality were older; were more often female; and had a greater prevalence of hypertension, diabetes, and smoking. Table 1. Cardiac Risk Factor Prevalence Across Quartiles of Predicted 15-y Mortality in 9715 Asymptomatic Patients* In Table 2, we report the frequency of risk factors across the CAC subgroups. Patients with higher-risk CAC scores were more often older; were less likely to be female; and had a greater prevalence and extent of cardiac risk factors, such as hypertension, diabetes, and smoking. Table 2. Cardiac Risk Factor Prevalence Across CAC Score Subgroups* Unadjusted All-Cause Mortality Overall mortality was 3%, 6%, 9%, 14%, 21%, and 28%, respectively, for CAC subgroups with scores of 0, 1 to 10, 11 to 100, 101 to 399, 400 to 999, and 1000 or greater (P<0.001). The relative hazard for all-cause death was 1.68, 2.91, 4.52, 5.53, and 6.26, respectively (P<0.001). Within each of the estimated predicted mortality quartiles (Figure), survival worsened in a generally proportional manner in the subgroups with increasing CAC scores (P<0.001 for all 4 patient subgroups). Figure. Cumulative incidence of all-cause mortality by CAC across 15-y predicted mortality quartiles. All P values are <0.001. CAC = coronary artery calcification. In Cox models adjusting for CAD risk factors, the CAC score was highly predictive of time to all-cause mortality (P<0.001). A graded or proportional relationship between risk and CAC extent was seen at 10 and 15 years of follow-up. NRI With CAC Over and Above the CAD Risk Factors For cut points ranging from less than 7.5% to 22.5% or greater, the categorical NRI was 0.21 (95% CI, 0.16 to 0.32), with 27.9% of patients who died correctly reclassified by the model with CAC added (Table 3). However, the model with CAC incorrectly reclassified 7.4% of survivors to a higher-risk category than the model with cardiovascular risk factors alone. Use of an alternative set of cut points (<10% to 20%) resulted in a slight increase in the NRI (0.239), with 34.7% of deaths correctly reclassified but 10.8% of survivors misclassified (Table 4). Table 3. NRI for Adding CAC Score to a Model With Cardiac Risk Factors, Including Age, Sex, Hypertensi