A joint latent variable model approach to item reduction and validation.

Area of Science:

• Biomedical Science
• Statistical Modeling
• Health Measurement

Background:

• Biomedical research often relies on unobservable constructs (latent variables) for health states and disease severity.
• Traditional methods separate construct measurement and validation, potentially introducing bias.
• Latent variable methods are increasingly popular in biomedical research.

Purpose of the Study:

• To propose a Multiple Indicator Multiple Cause (MIMIC) latent variable model that integrates item reduction and construct validation.
• To address and account for bias inherent in traditional two-stage measurement and validation processes.
• To demonstrate the utility of joint latent variable models in biomedical applications.

Main Methods:

• Developed a Multiple Indicator Multiple Cause (MIMIC) latent variable model.
• Applied the model to data from the Physical Activity and Lymphedema clinical trial.
• Utilized self-reported Likert scale symptoms and a gold standard diagnostic measure for lymphedema.

Main Results:

• The MIMIC model successfully combined item reduction and construct validation.
• Identified one symptom as a potential candidate for removal, optimizing measurement.
• Demonstrated the model's ability to account for bias in the traditional two-stage approach.

Conclusions:

• Joint latent variable models, like MIMIC, offer advantages over traditional methods in biomedical research.
• The proposed MIMIC model provides an effective approach for measuring and validating complex health constructs.
• This methodology is applicable to various biomedical research scenarios, enhancing construct assessment.