Bias, benefit, or both: evaluating new glioma therapies

Despite the development of many new promising therapies for malignant glioma, virtually all randomized controlled trials testing them have proven negative. These disappointing results are largely due to complex mechanisms of treatment resistance, but increasingly there is evidence that experimental bias rather than benefit accounts for both the promising early phase I/II trial results and later phase III failures. This paper highlights the aspects of clinical trial design and outcome analysis that specifically affect interpretation of results from therapeutic trials for malignant glioma. Phase II trials of both tumor response and tumor control are subject to selection bias; the early promising results seen with interstitial brachytherapy and intraarterial chemotherapy and yet negative phase III results are examples of this. Methods for detecting selection bias include modeling techniques in which databases of patients with known outcomes are used to emulate phase III outcomes. Modeling may assist in the determination of whether a given phase II result appears to exceed that expected by selection bias alone. Such an experiment on paper is quite unlikely to replace a well-designed randomized trial; however, in this time of increasing numbers of novel therapies but shrinking resources, these techniques should find utility in selecting those therapies most suitable for testing in cooperative group randomized trials.