DOI: 10.1093/bjd/ljad162.020 ISSN: 0007-0963

395 Assessing imputation methods with the lebrikizumab clinical trial program

April W Armstrong, Jonathan I Silverberg, Jacob P Thyssen, Richard B Warren, Alan D   Irvine, Eric Wolf, Yuxin Ding, Yong  Lin, Sarah Reifeis, Mertixell Falques, Eric L   Simpson
  • Dermatology


Missing data occur in clinical trials and have the potential to lead to biased results. Subsequently, the analytical methods for handling the missing data are important to evaluate. In atopic dermatitis (AD) trials, missing data are not handled consistently across studies. Lebrikizumab is a monoclonal antibody that binds with high affinity and slow off-rate to interleukin (IL)-13, thereby blocking the downstream effects of IL-13 with high potency. In ADvocate1 (NCT04146363) and ADvocate2 (NCT04178967), two randomized, double-blinded, placebo-controlled Phase 3 trials evaluating the efficacy and safety of lebrikizumab monotherapy in adolescent and adult patients with moderate-to-severe AD, a combined nonresponder/multiple imputation (NRI/MI) approach, was applied on the primary and key secondary endpoints. This study aims to illustrate the NRI/MI method using individual patient examples and to present Week 16 study results from ADvocate1 and ADvocate2 with NRI/MI and single imputation methods, NRI and last-observation carried forward (LOCF). In the combined NRI/MI method, data from patients after initiating rescue medication or discontinuing treatment due to lack of efficacy were imputed with NRI, and missing data for other reasons were imputed with MI, which uses a statistical model based on all available patient data to impute missing values. The MI method considers each patient’s trajectory and leverages information from other patients within the same treatment arm to impute the missing data. With NRI alone, the cause leading to the missing data is not considered and missing data for any reason are imputed as nonresponse. With LOCF, missing data are replaced with the last available measurement; LOCF assumes that the patient response would be stable over time and does not consider the reason for missing data. We determined the amount of missing data in ADvocate1 and ADvocate2, and we assessed patient-level imputed IGA scores to illustrate NRI/MI, NRI and LOCF. With these missing data handling methods, we evaluated the percentage of patients achieving the co-primary endpoints of ADvocate1 and ADvocate2: an Investigator’s Global Assessment score of 0 or 1 [IGA (0,1); clear or almost clear] with ≥2-point improvement from baseline or 75% improvement in Eczema Area and Severity Index (EASI 75) at Week 16. With the NRI/MI method, most missing data at Week 16 were imputed with NRI (ADvocate1: 73%, ADvocate2: 82%) compared with MI (ADvocate1: 27%, ADvocate2: 18%). Example imputed IGA scores based on patient-level data will be presented to demonstrate the missing data handling methods. For NRI/MI, NRI and LOCF, respectively, the percentage of patients achieving IGA 0,1 were 12.7%, 11.3% and 12.8% for placebo in ADvocate1 and 10.8%, 9.6% and 11.0% for placebo in ADvocate2; 43.1%, 41.0% and 42.4% for lebrikizumab in ADvocate1 and 33.2%, 31.3% and 33.5% for lebrikizumab in ADvocate2. The percentage of patients achieving EASI 75 were 16.2%, 14.2% and 16.3% for placebo in ADvocate1 and 18.1%, 17.1% and 19.2% for placebo in ADvocate2; 58.8%, 56.5% and 60.1% for lebrikizumab in ADvocate1 and 52.1%, 50.2% and 55.5% for lebrikizumab in ADvocate2. When using the NRI/MI method, we determined most missing data in ADvocate1 and ADvocate2 were handled with NRI. This analysis suggests that the NRI/MI method may provide a realistic estimation of response rate in ADvocate1 and ADvocate2.

More from our Archive