An analytic strategy for modeling multiple-item responses: a breast cancer symptom example

Area of Science:

• Health Outcomes Research
• Psychometrics
• Biostatistics

Background:

• Item Response Theory (IRT) is increasingly used in health research to synthesize multi-item response data.
• IRT models symptom susceptibility as an interaction between person and symptom characteristics.
• This study integrates IRT within a multilevel framework for longitudinal health outcomes using self-reported symptom scales.

Purpose of the Study:

• To describe the statistical framework for incorporating IRT into a multilevel model.
• To extend this framework to analyze longitudinal health outcomes.
• To examine differences in symptom experience trajectories between older and younger breast cancer patients, controlling for care location.

Main Methods:

• Secondary analysis of data from two descriptive longitudinal studies.
• Utilized a 3-level hierarchical linear model (HLM) with 350 women with breast cancer reporting 21 symptoms over time.
• Level 1: Item responses (symptom presence/absence). Level 2: Individual trajectories of latent symptom experience. Level 3: Person-specific covariates (age, location of care).

Main Results:

• Fatigue and pain were the most frequently reported symptoms.
• Symptom experience generally improved over time for women with breast cancer.
• Neither patient age nor location of care significantly influenced the symptom experience trajectory.

Conclusions:

• Integrating IRT within an HLM framework provides significant advantages for analyzing complex health data.
• This approach enables the creation of a latent symptom experience variable usable as an outcome or covariate.
• It facilitates the examination of latent symptom experience trajectories and manages symptom nonresponse effectively.