Modeling minimal residual disease (MRD)-testing.

Abstract

There is considerable effort to develop more sensitive methods to detect minimal residual disease (MRD) in bone marrow and blood samples of persons with cancer. Results of MRD-testing are used to predict clinical outcome and determine if more anti-cancer therapy is needed. Mathematical models were developed to assess factors affecting sensitivity and specificity of MRD-testing at diverse cancer cell prevalences. Modeling results and predictions were compared to results of large published studies.Accuracy of MRD-testing depends on cancer cell prevalence and distribution in the blood or bone marrow of the subject, sensitivity and specificity of the MRD-test and sample size. In subjects with low cancer cell prevalences (< or = 10(-4)) results of MRD testing are likely inaccurate. Increasingly sensitive MRD-tests are only marginally useful; the major obstacle to accuracy is inadequate sampling. Increasing sensitivity of methods to detect MRD is unlikely sufficient to increase accuracy of MRD-testing. In contrast, increased sampling (size and frequency) and assigning a high cut-off value (for example, > or = 10(-3)) to declare a MRD-test positive will increase sensitivity and specificity, respectively.