Rate ratio for mortality by Poisson's regression analysis: Limitation on predictive ability for the individual events

摘要

Ohsawa et al. reported that middle-aged subjects with preserved glomerular filtration rate (GFR) and albuminuria had risks of all-cause, cardiovascular disease (CVD) and neoplasm-related disease (NPD) mortalities in a cohort of 5.6-year follow-up on average [1]. Patients with a history of myocardial infarction or stroke were excluded for their analysis, and the authors calculated rate ratios (RRs) using Poisson's regression analysis. During the follow-up periods, 304 deaths, including deaths by 57 CVD and by 152 NPD, were observed. I have some concerns on their statistical outcomes.First concern is the number of events in their study, which had been mentioned by the authors. They first used 10 variables for the adjustment, and 5 additional variables were used for the adjustment by using albuminuria and estimated glomerular filtration rate. Multiple regression analysis requires at least 10 individuals per independent variable [2], and in the case of CVD in their study, estimate of rate ratio becomes unstable with wide range of 95% confidence interval. To keep stable estimation of rate ratio, the authors should continue their follow-up study.