Human population studies with the exfoliated buccal micronucleus assay: Statistical and epidemiological issues

Abstract

Micronucleus (MN) assay in buccal exfoliated cells is a minimally invasive method for monitoring genetic damage in human populations. Statistical and epidemiological issues related to the design and analysis of studies based on this biomarker are addressed. A systematic review of recent literature on the buccal MN assay has been carried out to provide a state-of-the-art evaluation of how critical topics such as control for confounding, sample size and statistical power, number of cells scored, endpoint selection, and statistical modelling, are considered. In addition, a meta-analysis has been performed to estimate the impact of most common confounders on MN frequency, and to provide a baseline value of MN frequency in the control population. A total number of 63 eligible studies were included in the analysis. Age (98.4%), gender (85.7%), and smoking habit (90.5%) were the most commonly studied confounders. Univariate statistics were estimated in most studies while multivariate analysis was applied only in the 47.6%. Baseline MN frequency in controls was 1.10/1000 cells (95% confidence interval 0.70-1.72), and the relative increment in subjects exposed to genotoxic agents or affected by disease correlated with similar observations in lymphocytes (R(2)=0.74). A minimum number of 4000 cells is recommended to reduce the variability of the MN mean estimates, in contrast with the current practice of scoring only 2000 cells (81% of studies). Poisson or Negative Binomial are the preferred statistical models when more than 2000 cells are scored. Studies scoring smaller numbers of cells should consider the use of statistical models taking into account the excess of zeros, e.g., the Zero Inflated Poisson (ZIP) models. The quality of papers published on the buccal MN assay can be substantially improved, with better consideration of basic issues such as power analysis, control for confounding, choice of the statistical model, and the number of cells to be scored.