Effect of current smoking and blood pressure on cardiovascular events and mortality for workers: Number of events for multivariate analysis to keep validity

摘要

I read the article by Kondo et al, conducting a 7.5-year follow-up study in 25077 young/middle-aged male workers. They detected 114 cardiovascular disease events (CVDs) and 141 deaths during their study to investigate the effect of current smoking and blood pressure (BP) on CVD events and all-cause mortality. CVD events were composed of 34 myocardial infarctions, 73 strokes, five acute aortic dissections and two acute occlusions of a peripheral artery. SBP and DBP were categorized into five groups, and smoking status was categorized in a binary way, as a current smoker and as a noncurrent smoker. I have some concerns on their statistical procedure. Peduzzi et al. conducted simulations to evaluate the effect of events per independent variable (EPV) in Cox regression analysis and concluded that EPV value less than 10 has the risk of losing statistical validity. Although there is no gold standard of EPV to keep validity of the statistical outcome, the number of events is important for a safe risk assessment. I have some concern about how EPV was evaluated in Kondo et al.'s study. They presented hazard ratios after adjusting for age, total cholesterol, smoking status, fasting plasma glucose and annual change in BP. As they categorized BP into five groups, the number of independent variables became nine, and 90 events are needed at a minimum for their analysis. In cases of stratified analysis by smoking status, the number of independent variables becomes eight, and the study outcome for noncurrent smokers has a risk of unstable estimate. The wide range of 95% confidence interval of adjusted hazard ratio indicates loss of statistical power, and more events are needed for subanalysis stratified by smoking status.