Is there a selection bias in radiotherapy dose-escalation protocols?

BACKGROUND To investigate the existence of a selection bias using a virtual radiotherapy dose-escalation trial. In dose-escalation trials, normal tissue constraints generally remain constant while the tumor dose is increased. Since tumor dose and normal tissue constraints are competing demands, a point will be reached at which the tumor dose cannot be increased without exceeding normal tissue constraints. METHODS AND MATERIALS In 9 patients with non-small-cell lung cancer, the tumor dose was escalated from 66 Gy to 78 Gy in 4-Gy dose levels using intensity-modulated radiotherapy planning, while the limiting normal tissue dose constraints remained constant. Dosimetric, radiobiologic, and other planning parameters were compared at the 66-Gy dose level for patients eligible for all dose levels and for those eligible only for lower dose levels. RESULTS Seven of 9 patients were eligible for all dose levels (Group E). Two of 9 patients ("ineligible" or Group I) qualified only for lower total doses (95% confidence interval, 0.075-0.6, significant). In Group E, mean planning target volumes were smaller (132 vs. 404 cm(3), nonsignificant), monitor units per fraction were significantly lower (448 vs. 802, p = 0.0008), and the average composite score for plan quality was better than in Group I (0.012 vs. 0.068, nonsignificant). Average tumor-control probabilities were higher (0.33 vs. 0.23, nonsignificant), and normal tissue-complication probabilities were lower for Group E than for Group I. CONCLUSIONS Patients eligible for higher dose levels had significantly superior estimated outcome parameters. A method to eliminate this source of error in the interpretation of dose-escalation trials is suggested.