Longitudinal latent variable models given incompletely observed biomarkers and covariates.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Missing Data Methods

Background:

• Longitudinal studies, like the National Growth and Health Study, often face missing data in biomarkers and covariates.
• Conventional methods for handling missing data can lead to biased estimations in latent variable models.

Purpose of the Study:

• To develop a method for unbiased estimation in two-level latent variable models with missing longitudinal data.
• To address the over-identification issue in joint models used for missing data imputation.

Main Methods:

• Re-expressing the desired latent variable model as a joint distribution of observed and missing variables.
• Imposing constraints on the joint model to achieve a one-to-one correspondence with the target model.
• Utilizing a modified expectation-maximization algorithm for parameter estimation.

Main Results:

• The proposed constrained joint model provides unbiased estimation for latent variable models.
• The method efficiently handles missing data under the assumption of ignorable missingness.
• Over-identified joint models were shown to produce biased results.

Conclusions:

• Constrained joint modeling offers an effective strategy for analyzing longitudinal data with missingness.
• This approach improves the accuracy of latent variable modeling in health and growth studies.
• The modified expectation-maximization algorithm is suitable for estimating the constrained joint model.