A statistical analysis of the reliability and classification error in application of the mutagen sensitivity assay

In vitro mutagen hypersensitivity, determined with the bleomycin assay, has been found to be an independent risk factor for developing primary upper aerodigestive tract cancers, lung cancers, and second malignant tumors. The average number of chromatid breaks per cell (b/c) is derived from evaluating an arbitrarily set number of metaphases (usually 50) in each sample. The reliability of such an approach is of key importance in large-scale epidemiological studies. Because evaluating metaphases is a time-consuming task, it is desirable to know the minimum number of readings needed to reach an acceptable reliability. Statistical analyses were performed in 160 observations for which b/c values were determined by the same observer scoring 100 consecutive metaphases per sample. The b/c values were between 0.14 and 1.30 (mean, 0.61). The b/c values were separately calculated for the first and second sets of 50 metaphases. There was essentially no difference in the mean b/c values between the first and second 50 readings (difference, -0.002). The correlation between the two sets of readings was 0.72. The SE of the b/c values, based on scoring 50 metaphases, was 0.15. When scoring 100 metaphases, the SE decreased to 0.11. After evaluating the first 50 metaphases, the theoretical gain in reducing the SE was <1% with each additional reading. When 0.8 was used to dichotomize the mean b/c value into bleomycin-resistant or bleomycin-hypersensitive groups, sensitivity and specificity of reading 50 metaphases were above 75% and 95% when compared to 100 readings. A simulated case control study showed that there is a 15% attenuation in estimating the odds ratio with 50 readings. The findings suggest that the conventional method of reading 50 metaphases can yield an acceptable reliability in our setting and may be applied to other cancer epidemiology studies.