Desai et al reported the effect of coronary artery calcium (CAC) as a subclinical atherosclerosis measure on cardiovascular disease (CVD) events or other causes of death by a mean follow-up period of 7.1 years. They handled 6,581 subjects (3,095 men and 3,486 women) as the entire cohort of Multi-Ethnic Study of Atherosclerosis (MESA), and the numbers of participants who had CAC score of 0, 1 to 99, and ≥100 were 1,217, 905, and 973 in men and 2,109, 838, and 539 in women, respectively. Ahmed et al also analyzed the database of MESA and used morbidity and mortality as dependent variables with special emphasis on the advantage of low-risk lifestyles and CAC. They selected 4 lifestyle factors such as weight, physical activity, smoking, and diet as the independent variables. In contrast, the change in CAC was used as a dependent variable for the prediction of subclinical atherosclerosis.
I have a statistical concern on the outcome of the study by Desai et al. The authors conducted multivariate analyses using 2 types of statistical procedures for estimating the risk of CVD event or non-CVD mortality. About the statistical procedures in their Table 4, there was no difference in the statistical outcomes of Cox regression and the modified analysis by Fine and Gray. The investigators used age, race, systolic blood pressure, total and high-density lipoprotein cholesterol, smoking status, and diabetes for the adjustment. In contrast, study site, income, medications for hypertension and dyslipidemia, and variables of healthy lifestyles such as regular exercise, healthy diet, and weight maintenance were not used for the adjustment. The total number of adjusting variables becomes 8, including CAC. As the number of category on CAC was 3, and the authors finally used 9 independent variables for their multivariate analysis.
Concato and Feinstein reported that the number of events per independent variable in multivariate analysis should be 10 or more to keep statistical power. This means that 90 events are indispensable for their analysis. From the Table 2 by Desai et al, the number of CVD events in each CAC group was 124, 303, and 628 in men, and 112, 265, and 645 in women, respectively. In addition, the number of non-CVD deaths in each CAC group was 96, 204, and 263 in men, and 84, 164, and 254 in women, respectively. Namely, the total number of CVD events is satisfactory in both men and women. However, a subanalysis by classifying the cause of CVD event showed the limited number of independent variables for multivariate analysis, which showed a wide range of 95% confidence interval or not applicable in their Table 3. If the number of independent variables increases, more events are needed for their analysis.
Anyway, further study is needed by adjusting more independent variables in their analysis. In addition, gender gap for the risk assessment of CVD morbidity using CAC should be evaluated quantitatively. I speculate from their tables that there was a gender gap on CVD event except stroke. As the distribution of CAC in men shifts higher in score than that in women, hazard ratio of CAC on CVD events should be presented in combination with other hazard ratios of independent variables to elucidate the gender gap on the predictive ability for CVD events.