Handling underlying discrete variables with bivariate mixed hidden Markov models in NONMEM

Area of Science:

• Statistical Modeling
• Pharmacometrics
• Chronic Disease Management

Background:

• Non-linear mixed effects models often overlook unobservable processes.
• Hidden Markov Models (HMMs) link observed data to unmeasurable states, like disease status.
• Mixed HMMs (MHMMs) incorporate stochasticity into HMMs for unobservable process variability.

Purpose of the Study:

• Develop and apply MHMMs to chronic obstructive pulmonary disease (COPD) data.
• Investigate estimation properties of MHMM parameters in NONMEM.
• Compare the power of bivariate versus univariate MHMMs for detecting drug effects.

Main Methods:

• Developed bivariate MHMMs for simulated COPD data (PROs and FEV1).
• Assessed parameter estimation with and without random/covariate effects.
• Quantified the influence of effect magnitudes on parameter precision.
• Conducted power analysis comparing bivariate and univariate MHMMs.

Main Results:

• High parameter precision observed, except for the remission-to-exacerbation transition rate variance (RRMSE > 150%).
• Parameter precision improved with higher transition probability magnitudes.
• Drug effect parameter precision enhanced with increasing parameter magnitude.
• Bivariate MHMMs required fewer subjects (25) for 80% power compared to univariate FEV1 (63) and PRO (>100) models.

Conclusions:

• Bivariate MHMMs provide enhanced statistical power for detecting drug effects in COPD.
• The model demonstrated good parameter estimation properties in NONMEM.
• Results advocate for using bivariate MHMMs when feasible for complex disease modeling.