Director, pharmacometrics Critical Path Institute, United States
Disclosure(s):
Yi Zhang, PhD: No financial relationships to disclose
Objectives: Major depressive disorder (MDD) is a highly prevalent psychiatric disorder with a large socioeconomic impact, and yet its progression is still not fully understood. A recognized problem in using randomized controlled trials (RCT) for MDD is the complex inter-occurance of remission and recurrence. New rapid-acting therapies such as GABA-A receptor modulators and NMDA receptor antagonists necessitates more flexible disease progression models (DPM) to account for the response heterogeneity.
In this study we aim to develop a Bayesian DPM using the 17-item Hamilton’s Rating Scale for Depression (HAMD) as the outcome, focusing on predictive performance of the near-term (e.g. 4-8 weeks) post-treatment stage of rapid-acting therapies, given treatment stage observations (e.g. 2-4 weeks), to address the challenges brought on by short treatment cycles that are not present in standard SSRI/SNRI therapies.
Methods: We extend a previous investigation[1] by replacing the polynomial component with a sigmoid function to describe the post-treatment recurrence. The Weibull function is used to describe HAMD decay in the initial treatment. We assess different covariate structures and residual models during model selection. Compared to alternatives such as Gaussian process and general spline curve fitting, the proposed model has the advantage of clinically meaningful parameters, such as time to response, rate of response, time to recurrence and rate of recurrence, reflecting current knowledge that the progression consists of an improvement and a variable recurrence stage. HAMD percent of change from baseline (PCHG) is also used to assess goodness-of-fit.
Results: We compare the proposed Weibull/Sigmoidal model with the previously studied Weibull/linear model. It is shown that the predictions from the two models are comparable for the treatment stage. For the near-term follow-up period, the Weibull/Sigmoid model is superior for mean and Q90 while the Weibull/Linear model is better for Q05. In general the Weibull/Sigmoid model has a lower prediction RMSE. The same observation was made for PCHG. Model comparison based on Leave-one-out cross validation (LOO-CV) shows that age, sex, clinical global impression severity scale (CGI-S) and antidepressant use at baseline are clinically significant predictors. Though the new model predicts the mean of the recurrence pattern well, large credible intervals of the parameters still indicate the limitation of HAMD as an outcome.
Conclusions: As the first DPM study that combines Bayesian framework and scientific plausible function structure for rapid acting MDD therapies that also captures SoC disease progression processes. It is valuable to be a standalone model to assist understanding of the disease, and can be extended into a clinical trial simulation (CTS) tool for future trial designs.
Citations: [1] R. Gomeni and E. Merlo-Pich, Bayesian modeling and ROC analysis to predict placebo responders using clinical score measured in the initial weeks of treatment in depression trials, Br J Clin Pharmacol 63:5 595–613, 2006.
