Study Title
Sign Instability in Composite Food-Processing Exposures That Include Clinically Indicated Nutrition
Principal Investigator
Jimmy Chun Yu Louie
Affiliation
Department of Biomedical, Health and Exercise Science, School of Health Sciences, Swinburne University of Technology, Hawthorn, Victoria, Australia
Start Date
10 August 2026
End Date
Not specified
Study Objective
To examine how regression coefficients behave when a composite food-processing exposure combines discretionary ultra-processed foods with clinically indicated nutritional products that may have opposing associations with health outcomes, and to evaluate whether disaggregation alone is sufficient to resolve resulting bias.
Short Abstract
This methodological study combined analytical derivation with two Monte Carlo simulation studies to examine estimation bias arising from composite dietary exposures. A pooled coefficient fitted to an aggregated exposure was shown to depend on component contributions to exposure variance rather than energy intake. Under dietary substitution, coefficients could exceed the magnitude of both true component effects. A second simulation demonstrated that disaggregation alone did not resolve confounding by indication, with sign inversion persisting unless severity was controlled. The findings highlight the importance of identifying clinically indicated nutrition separately from discretionary intake when studying processing-based dietary exposures in clinical populations.
Study Design
Analytical derivation and Monte Carlo simulation study reported according to the ADEMP framework.
Population
Simulated datasets representing clinical cohort populations receiving medical nutrition. No human participants were recruited.
Sample Size
Simulated sample size of n = 3,000 per replication; 1,000 replications per scenario.
Inclusion Criteria
Simulated individuals contributing discretionary and clinically indicated components of a composite dietary exposure under predefined data-generating mechanisms.
Exclusion Criteria
Not applicable. No participant recruitment occurred.
Intervention/Exposure
Composite food-processing exposure consisting of a discretionary component and a clinically indicated nutrition component, such as oral nutritional supplements or enteral formulae.
Outcome Measures
Mean fitted regression coefficient; Monte Carlo standard error; empirical standard deviation; proportion of positive coefficients; frequency of sign inversion; and recovery of true component effects under alternative analytical models.
Funding Source
No funding was provided for this study.
Collaborating Institutions
None. Sole-author study conducted at Swinburne University of Technology.
Ethics Approval
Not applicable. This study used simulated data only and did not involve human participants, biological samples, identifiable data, or animal subjects.
Publication Status
Not published
Keywords
Ultra-processed foods; NOVA classification; exposure measurement; aggregation bias; confounding by indication; simulation study; dietary assessment; methodological research
Data Collection Methods
No primary data collection. Data were generated through Monte Carlo simulation using lognormal intake distributions and Gaussian copulas under predefined analytical scenarios.
Primary Data Availability
Simulation code and random seed will be made available as Supplementary File 1 accompanying the manuscript.
Contact Information
A/Prof Jimmy Chun Yu Louie
Department of Biomedical, Health and Exercise Science, Swinburne University of Technology, Hawthorn VIC 3122, Australia
Email: