Novel application of Kaplan-Meier methods to model tolerance for nonadherence to imatinib in patients with chronic myeloid leukemia (CML) in the ADAGIO study

7038 Background: Although adherence to imatinib treatment has been shown to be critical for attaining treatment response among patients with CML, some studies have suggested a 7.3-9.9% nonadherence tolerance margin before loss of treatment effects. We aimed to model probabilistically the margin of tolerance required to ensure treatment response among patients prescribed imatinib and the margin, if any, before treatment response is at risk. Methods: We performed a post hoc analysis of the ADAGIO study conducted in Belgium on 169 evaluable patients ( Blood 2009). Using the pill count ratio as, what in conventional survival analysis would be, the time variable, we modeled the cumulative likelihood of treatment response as a function of increasing pill count adherence. We applied Kaplan-Meier methods to model the likelihood of complete cytogenetic (CCyR), complete hematological (CHR), major molecular (MMR) and optimal (OR) (as defined by the European Leukemia Net) response as a function of 90-day pill count adherence. Kaplan-Meier methods thus estimated the tolerance for nonadherence to imatinib by calculating the 1 minus Kaplan-Meier estimate for treatment response. Results: Analyses (see Table) showed that ̃100% adherence of prescribed dose is associated with probabilities (rounded) of 0.84 for CHR, 0.83 for CCyR, 0.82 for OR, and 0.77 for MMR; compared to, 0.37 (CHR and CCyR), 0.35 (OR), and 0.39 (MMR) at 90% adherence. (of 0.7698 (MMR). Increasing the intake of imatinib from 90% to 100% of the prescribed dose increased the likelihood of the various treatment responses by 1.95 to 2.35-fold. Conclusions: Our findings challenge any previously estimated tolerance for nonadherence. There is virtually no margin for nonadherence if the objective is to optimize the likelihood of treatment response, and only a minimal margin to avoid impaired treatment response. Under such adherence, response rates similar to those in the pivotal IRIS trial can be obtained. Clinicians must assess and promote patient adherence, and patients must be perfectly adherent.[Table: see text]