Statistical models for analysis of cytogenetic biomarkers

Bleomycin-induced chromosomal breaks (CB) and sister chromatid exchange (SCE) in peripheral blood lymphocytes have been shown to be sensitive cytological markers for susceptibility to DNA damage in patients with various types of cancer and in healthy controls. Factors such as age, sex, smoking, and alcohol consumption could affect the values of some of these biomarkers and should be considered as covariates when analyzing cytogenetic biomarkers because these factors can affect the frequency of CB and SCE.

We propose a statistical method using negative binomial (NB) distribution to evaluate the numbers of CB and SCE. In order to determine the best model to represent the frequency of CB and SCE, we compared the NB model with the widely used Poisson model and log-transformed normal model by using generalized linear models. To demonstrate the better fit of the NB model, we analyzed three different data sets from studies conducted at The University of Texas M.D. Anderson Cancer Center. The first set was a case-control study of lung cancer in a population of African Americans and Mexican Americans (286 cases and 156 controls), the second set consisted of 311 head and neck cancer patients, and the third set consisted of 105 Hodgkin's disease patients.

For CB; the estimates of the variability for Hodgkin's disease, head and neck, and lung cancers were 487.24, 502.82, and 520.15, respectively. For SCE, the estimates of the variability for Hodgkin's disease was 9777.01. For CB, the dispersion estimates under the three models (Poisson, NB, and Normal) for Hodgkin's disease, head and neck, and lung cancers were: 12.30, 1.20, 0.85; 8.94, 1.05, 0.22; and 10.10, 1.05, 0.25, respectively. For SCE (Hodgkin's disease only), the dispersion estimates under the three models (Poisson, NB, and Normal) were 30.91, 1.11, 0.10, respectively.

Our results demonstrate that the NB model provides a better interpretation and fit for the frequency of CB and SCE in different cancer types. Therefore, we recommend it as a model for the analysis of cytogenetic biomarkers.