(W-032) Longitudinal Joint Modeling of Modified Mayo Score and Dropout in Patients with Moderate to Severely Active Ulcerative Colitis.

Timothy Waterhouse, PhD – Research Scientist, Metrum Research Group; Michael Heathman, PhD – Research Scientist, Metrum Research Group; Yuchen Wang, PhD – Manager, Pfizer; Arnab Mukherjee, PhD – Executive Director, Pfizer; Vivek Purohit, PhD – Senior Director, Pfizer

Objectives: Develop a longitudinal joint model of modified mayo scores and dropout for patients with ulcerative colitis receiving etrasimod.

Methods: Two phase 3 studies were conducted in patients where there was an initial 12 week induction period followed by a 40 week maintenance period. After the end of the induction period there was a transient increase in the dropout rate for a period of several weeks. A longitudinal joint model was developed to characterize the effect of average etrasimod concentration at steady state on modified mayo score (MMS) and dropout where the change in the dropout was captured by the model.[1] The hazard function defining the risk of dropout was estimated with a constant baseline hazard that increased at the end of the induction period. The change in the hazard function was parameterized using a sin curve with the start time, amplitude, and duration of the change estimated. This dropout model was jointly estimated with a longitudinal ordinal regression model for the MMS components with an Imax indirect response model. Inter-individual random effects and structural model parameters were shared among the component scores. The link function to connect the dropout rate with the effect of etrasimod was modeled as a function of the baseline, placebo, and drug effects using the average of each of the estimates across the 5 MMS subscores with an additive parameterization. A Bayesian approach was used to estimate all the parameters.

Results: The model was able to adequately capture the increase in the hazard at the end of the induction period with the time for the beginning of the spike estimated to be 12.3 weeks after the beginning of treatment with a duration of 11.2 weeks. The effect of efficacy on the hazard was positive indicating that better efficacy was associated with a lower risk of dropout. The placebo onset was estimated to have a half-life of 7.6 weeks and half-life of the drug effect was 9.4 weeks. The VPCs showed that the model was able to capture the spike in the dropout rate. Each of the component scores were well estimated capturing the time course of the response during the induction period. The enriched population during the maintenance period was characterized much better after the dropout was incorporated into the model. However, the IIVs had a large CV during the maintenance period. The 95% credible intervals were all reasonable with large bulk and tail effective sample sizes indicating that the posterior distributions were well characterized. The composite endpoints calculated from the MMS subscores were adequately characterized (supported by the VPCs of the of the dropout, MMS subscores, and composite endpoints).

Conclusions: A longitudinal joint model of modified mayo score and dropout was developed using the average etrasimod concentration at steady state. The observed dropout rate, MMS subscores, and composite endpoints were adequately characterized using the joint model.

Citations: [1] A workflow for the joint modeling of longitudinal and event data in the development of therapeutics: Tools, statistical methods, and diagnostics
Tutorial in CPT:PSP, Zhudenkov et at 2021. https://doi.org/10.1002/psp4.12763