Evaluating the Impact of Cross-over from Imatinib to Nilotinib on the Overall Survival in Chronic Myeloid Leukemia Using Inverse Probability of Censoring Weighting: Results from the Gimema Sustrenim Study

Introduction: In clinical studies cross-over from the control to the experimental treatment can obscure the true effect on overall survival (OS). Censoring a patient at the time of cross-over assumes their survival matches that of other patients in the same treatment arm who have not crossed over. This leads to an overestimate of the OS treatment effect. Censoring is often anti-conservative and cannot be recommended, and several proposals have been developed to better account for informative censoring. The Inverse Probability of Censoring Weighting (IPCW) method offers a statistical approach to address this issue by creating a pseudo-population that reflects the original population's characteristics without cross-over.

Methods: We launched in November 2016 an international, prospective, interventional, randomized, two arms, study to evaluate both the depth of the molecular response and the rate of TFR in newly diagnosed CP-CML patients treated with a second generation TKI (nilotinib, NIL) or with a first generation (imatinib IMA) followed by switching to NIL in absence of optimal response (defined according the ELN 2013 criteria (Clinical Trial number 02602314).

In the planned analysis of the Overall Survival control patients (initially receiving IMA) who crossed over to the experimental treatment (NIL) in case of suboptimal response were evaluated according to the ITT principle. The IPCW technique upweights the control patients to compensate for the cross over effect. This process effectively constructs a pseudo-population that maintains the same specified characteristics as the original population, yet without the cross-over. The IPCW method assumes that the created pseudo-population has a similar prognosis to the original population, relying on the 'no unmeasured confounders' assumption. Gender, age, ELTS risk score and BCR/ABL transcript at baseline were considered as potential confounders.

Results: From November 2016 to January 2021, 457 patients with newly diagnosed CP-CML patients were enrolled into the study and 448 of these (228 and 220 randomized to the NIL and IMA arms, respectively) were evaluable (mean age 54.21 years - range 19.4 – 85.9). At baseline, 185 (41%), 191 (43%) and 72 (16%) patients were classified as low, intermediate, or high-risk Sokal, respectively, while 280 (62%), 128 (29%) and 40 (9%) had a low, intermediate, or high-risk ELTS risk score. The median follow-up of the whole cohort of patients is 53.4 months. Eighty-three (38%) of the 220 patients of the IMA arm did not fulfil the ELN criteria for optimal response within the first 12 months of treatment and, according to the protocol, switched to NIL therapy.

The analysis of the OS using IPCW was conducted on 438 pts (10 patients were excluded due to missing data on potential confounders) and demonstrated that the pseudo-population without cross-over provided unbiased estimates. The hazard ratio for OS in the pseudo-population was 3.11 (95% CI: 1.25-7.77, p = 0.015) compared with 2.72 (95% CI: 1.13-6.56, p = 0.026) in the actual population, indicating a larger reduction in the risk of death with the experimental treatment.

Conclusions: The IPCW method is effective for evaluating the impact of cross-over on OS in clinical trials, as demonstrated in the GIMEMA Sustrenim study. This approach ensures that the pseudo-population reflects the original study population's characteristics without the bias introduced by cross-over, thus providing a clearer understanding of the treatment's true effect on OS. Future studies should further explore the practical application of IPCW in various clinical settings to evaluate the effect of the cross-over on the survival outcomes.