Some Statistical Issues Pertaining to in Vitro Drug Testing with Human Tumor Colony Forming Assays

One should proceed with prudence in the interpretation and application of in vitro drug tests. Among the statistical considerations that should be borne in mind are the following:
1. Reproducibility of in vitro assays should be adequately addressed.
2. Experimental designs for in vitro assays need further development, and should take account of the underlying biology of in vitro cell growth, and drug pharmacology.
3. Sensitivity, specificity, and predictive values are commonly used summary indices of in vitro - in vivo associations, but arise from dichotomization of laboratory and clinical data with loss of information. Objective criteria are needed for in vitro outcomes; and, in vivo outcomes should be chosen to reflect survival advantage.
4. Multivariate statistical techniques that assess whether in vitro assays provide additional information for predicting clinical outcome with other available prognostic criteria can be useful. However, these analyses presuppose adherence to statistical criteria for experimental design (including adequate sample sizes) and require prospective evaluation. Extrapolation of results from one clinical setting to another should be done with caution.