Impact of miscoding dose-limiting toxicities in dose-escalation phase I studies.

Abstract

e13007 Background: We assess the impact on estimating the maximum tolerated dose (MTD) of miscoding dose-limiting toxicities (DLT) for the “3+3” dose-escalation scheme (SM) and the Continual Reassessment Method (CRM). Methods: Comparisons were carried out through 1,000 simulated trials under six different scenarios by varying the location of the true MTD, testing six dose levels. Both CRM and SM start at the lowest dose, and do not allow skipping dose levels. Two errors are considered: miscoding a DLT as a non-DLT (MT); or miscoding a non-DLT as DLT (MNT).The impact of these errors on accuracy, patient allocation, safety and study duration was evaluated by increasing the probability of occurrence of each error from 0 to 0.3. Results: Under no errors, CRM is up to 35% more accurate than SM in finding the true MTD. This superior accuracy remains in the presence of errors of type MT. A toxicity must be missed (coded as non-toxicity) 30% of the times to allow the SM to experiment at higher doses, and to be closer to CRM's accuracy. With accrual of 3 patients per month, SM takes on average 9.7 months to test six levels under no errors, but under MT errors the trial duration increases 20% without improved accuracy. CRM takes 9.2 months to complete with or without errors since the sample is fixed at 24 patients. SM assigns 10-20% fewer patients at the MTD compared to CRM in the presence or absence of MT errors. As MNT increases from 0 to 0.3, patient allocation at the MTD decreases from 26% to 4%, since CRM lowers experimentation at lower dose levels by de-escalating. For SM, 27% of patients are treated at dose level one (d1) compared to 67% when MNT is 0 and 0.3 respectively, resulting in 38% of trials recommending d1 as the MTD compared to 15% under no errors. Conclusions: SM is less robust to coding errors than the CRM. When, for the first cohort, a non-DLT is flagged incorrectly as a DLT, the SM stops early by not escalating beyond d1.The trial duration, sample size and accuracy are influenced by such errors when using the SM. CRM uses all available data and the dose escalation process is less sensitive to individual patient errors leading to a more accurate MTD, provided that the coding errors occur in ≤2/10 patients. No significant financial relationships to disclose.