The multi-site tumour transplantation model for human endometrial carcinoma: a statistical evaluation.

We studied the effect of multi-site tumour transplantation on tumour growth by implanting varying numbers of EnCa 101 human endometrial tumours in athymic mice. One treatment group received a single tumour per mouse, another group received two tumours per mouse and a third group received four tumours per mouse. Tumour growth was sustained in all animals by implantation of oestradiol-17 beta pellets. We observed positive correlation between tumours within the same mouse, which implies that individual tumours are not statistically independent. The correlation is sufficiently large that failure to account for it in statistical design and analysis could result in studies with insufficient power and in spurious assertions of significance. Regression modelling of tumour growth curves showed that mean tumour volume per animal is not affected by the number of tumours growing on the animal; that is, the data are consistent with the null hypothesis that mean tumour volume is the same regardless of the number of tumours present. Our results therefore suggest that the use of multiple tumours per animal can increase the precision of experiments without loss of validity and at relatively little cost. However, correct and efficient analysis of the data so obtained requires more sophisticated techniques than those--such as fixed-effects analysis of variance and the two sample t-test--that assume independence of tumours.