Two-part models with stochastic processes for modelling longitudinal semicontinuous data: Computationally efficient...

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Two-part models with patient-specific stochastic processes offer flexibility for longitudinal semicontinuous data.
• Existing methods face computational challenges due to high-dimensional integrations in the marginal likelihood.
• Current non-standard procedures are computationally intensive, often relying on simulation.

Purpose of the Study:

• To develop an efficient computational method for fitting two-part models with patient-specific stochastic processes.
• To enable the use of standard maximum likelihood estimation for these complex models.
• To improve the analysis of longitudinal semicontinuous data in biostatistical research.

Main Methods:

• Transformed the marginal likelihood using properties of the multivariate normal distribution and CDF identity.
• Reduced high-dimensional integrations to efficient evaluation within a multivariate normal CDF.
• Utilized maximum likelihood estimation with the observed information matrix for parameter estimation.

Main Results:

• Demonstrated an efficient method for computing the marginal likelihood in complex two-part models.
• Successfully applied maximum likelihood estimation to obtain parameter estimates and standard errors.
• The methodology was validated on a dataset of psoriatic arthritis functional disability.

Conclusions:

• The proposed method significantly simplifies the fitting of advanced two-part models for longitudinal semicontinuous data.
• This efficient implementation allows for standard statistical inference, enhancing practical usability.
• The approach provides a valuable tool for analyzing complex health data, such as functional disability in psoriatic arthritis.