Measuring bias in uncontrolled brain tumor trials - to randomize or not to randomize?

Purpose: To help investigators decide if new therapies for glioma warrant definitive evaluation in randomized studies we have been developing a method for assessing the degree to which patient selection may have enhanced the results of uncontrolled treatment trials. In this study, we analyzed the impact of case selection on the survival of patients with malignant glioma receiving adjuvant stereotactic radiosurgery, a promising therapy reserved for those with small tumors and good performance status. Methods: Following published eligibility criteria we simulated the patient selection process for stereotactic radiosurgery given as a boost at the conclusion of conventional radiotherapy. Eligible patients were culled from a pre-existing clinical/imaging database of 101 consecutive conventionally-treated patients with biopsy-proven malignant glioma and known survival times. Median durations of survival and 2- and 3-year survival rates were determined for those judged eligible or ineligible for stereotactic radiosurgery, Results: Twenty-seven percent of patients were deemed eligible for stereotactic radiosurgery, eligible patients had more favorable prognostic factors and significantly longer median survival than ineligible patients (23.4 vs. 8.6 months; 2-year rate, 48% vs. 15%; 3-year rate, 30% vs, 7%); eligible patients also had a longer median survival than the entire group of unselected patients (23.4 vs. 11.4 months). Radiosurgery-eligible, conventionally-treated patients with glioblastoma multiforme and a group of radiosurgery-treated patients at a special referral center had similar median survival times (16.4 vs. 19.7 months), Conclusion: We provide additional evidence for selection bias in uncontrolled trials of stereotactic radiosurgery and by simulating the selection process accurately have detected a larger bias effect than noted previously. Judging from experience with interstitial radiation and intraarterial chemotherapy where substantial selection bias also occurred and randomized controlled trials proved disappointing, we conclude that a phase HI study of stereotactic radiosurgery for malignant glioma is unlikely to yield a positive result and may not be necessary.