Abstract
PURPOSE: Phase I studies rely on investigators to accurately attribute adverse events as related or unrelated to study drug. This information is ultimately used to help establish a safe dose. Attribution in the phase I setting has not been widely studied and assessing the accuracy of attribution is complicated by the lack of a gold standard. We examined dose-toxicity relationships as a function of attribution and toxicity category to evaluate for evidence of toxicity misattribution. EXPERIMENTAL DESIGN: Individual patient records from 38 phase I studies activated between 2000 and 2010 were used. Dose was defined as a percentage of maximum dose administered on each study. Relationships between dose and patient-level toxicity were explored graphically and with logistic regression. All P values were two-sided. RESULTS: 11,909 toxicities from 1,156 patients were analyzed. Unrelated toxicity was not associated with dose (P = 0.0920 for grade ≥ 3, P = 0.4194 for grade ≥ 1), whereas related toxicity increased with dose (P < 0.0001, both grade ≥ 3 and ≥ 1). Similar results were observed across toxicity categories. In the five-tier system, toxicities attributed as "possibly," "probably," or "definitely" related were associated with dose (all P < 0.0001), whereas toxicities attributed as "unlikely" or "unrelated" were not (all P > 0.1). CONCLUSIONS: Reassuringly, we did not observe an association between unrelated toxicity rate and dose, an association that could only have been explained by physician misattribution. Our findings also confirmed our expectation that related toxicity rate increases with dose. Our analysis supports simplifying attribution to a two-tier system by collapsing "possibly," "probably," and "definitely" related.