Quantifying Algorithmic Affective Cost: Protocol for a Randomized Mixed-Methods Trial of Dietary Self-Monitoring, Interface Salience, and Orthorexia-Related Symptom Trajectories

By Dr Sam · 11 September 2026 · 4 sources cited

Abstract

Objective: To estimate the behavioral benefits and psychological costs of specific digital dietary self-monitoring interfaces, defining the causal effect of high-salience algorithmic feedback and tracking intensity on momentary food-related distress and orthorexia-related symptom trajectories.

Design: A 12-week, 2×2 factorial randomized controlled explanatory sequential mixed-methods trial (QUAN → qual), structured according to SPIRIT 2025 guidelines.

Participants: Adequately powered adult cohort (N=420) without active clinical eating disorders.

Primary Estimand: The intention-to-treat (ITT) average treatment effect of assigned high-salience algorithmic feedback on mean momentary food-related distress over 12 weeks.

Analysis: Factorial linear mixed-effects modeling (LMM), lagged multilevel mediation, and pattern-mixture models. Purposively sampled qualitative interviews provide explanatory mechanistic integration.

1. Introduction & Hypotheses

While dietary self-monitoring drives adherence, observational evidence robustly links it to disordered eating. Recent systematic reviews explicitly state that causal direction and interface-level mechanisms remain unresolved, highlighting the need to separate behavioral benefits from psychological harm.

This protocol tests seven prespecified hypotheses:

• H₁ (Salience - Primary): High-salience algorithmic feedback increases mean momentary food-related distress compared to neutral feedback (confirmatory ITT).
• H₂ (Tracking Intensity): Higher assigned tracking intensity increases subsequent checking behavior.
• H₃ (Interaction): The affective cost of high-salience feedback is greater under high tracking intensity (β_F×S>0).
• H₄ (Lagged Mediation): Algorithmic feedback effects on compensatory behavior are temporally mediated by acute distress.
• H₅ (Moderation): Trait perfectionism moderates the effect of algorithmic feedback on food-related distress.
• H₆ (Joint-Arm Trajectory): Participants assigned to the high-salience/high-intensity condition will exhibit greater 12-week increases in EHQ-derived orthorexia-related symptoms compared to the neutral/low-intensity condition (joint-arm contrast).
• H₇ (Interoception): High-salience/high-intensity assignment reduces self-reported interoceptive awareness (MAIA) over 12 weeks.

2. Causal Estimands & Exposure Architecture

To prevent confounding adherence with psychological cost, we formalize Algorithmic Affective Cost (AAC) as a population-level causal estimand: the ITT average treatment effect of assigned high-salience feedback on momentary distress.

Exposure is strictly partitioned:

• Assigned Exposure (A_i): Randomly allocated tracking intensity and feedback salience.
• Delivered Exposure (D_it): Telemetry-verified presentation of interface alerts.
• Received Exposure (R_it): Verified participant interaction (e.g., opening the alert).

Primary inference relies exclusively on A_i, while D_it and R_it inform exploratory dose-response models.

3. Measurement Model & Operational Definitions

Primary Outcome: Food-Related Distress

Defined operationally as a composite score of EMA-derived shame, guilt, and anxiety. Time-varying variables are decomposed to isolate within-person deviations from between-person traits:

H₄ Temporal Event Indexing:

• t: Standardized interface exposure event (logging input).
• t+1: Event-contingent EMA response completed ≤15 minutes post-exposure.
• t+2: Telemetry-verified compensatory checking behavior occurring ≤180 minutes post-EMA.

4. Experimental Interventions (2×2 Factorial)

Nutritional algorithms are identical across all arms; only interface psychology is manipulated.

5. Statistical Analysis & Power Simulation

Sample size was derived via Monte Carlo simulation targeting ≥90% power for the F×S interaction.

Monte Carlo Simulation Parameters:

Factorial LMM:

Estimands: β₃ tests H₁ (average effect of salience). β₄ tests H₃ (interaction). Time interactions (β₅,β₆,β₇) represent secondary trajectory deviations from the primary average effect.

6. Safety, Ethics, and Stopping Rules

Given the manipulation of evaluative dietary feedback, strict safety thresholds trigger immediate independent Data and Safety Monitoring Board (DSMB) review:

• EMA Signal: ≥3 consecutive food-distress composite scores >85/100.
• Clinical Signal: EAT-26 screening score escalating ≥10 points from baseline.
• Rescue: Zero-barrier, one-click referral to an eating disorder specialist integrated into the UI.

7. Mixed-Methods Integration & SPIRIT Schedule

Following quantitative analysis, 40-60 participants will be purposively sampled based on predefined trajectory criteria (e.g., top quartile of distress escalation) rather than latent class discovery. Reflexive thematic analysis will populate a QUAN → QUAL joint display to contextually explain why specific interface trajectories occurred.

SPIRIT 2025 Schedule of Assessments

8. Selected References

• Anderberg, I., Kemps, E., & Prichard, I. (2025). The link between the use of diet and fitness monitoring apps, body image and disordered eating symptomology: A systematic review. Body Image, 52, 101836.
• Chan, A. W., Boutron, I., Hopewell, S., et al. (2025). SPIRIT 2025 statement: updated guideline for protocols of randomised trials. BMJ, 389, e081477.
• Hopewell, S., Chan, A., Collins, G. S., et al. (2025). CONSORT 2025 Statement: Updated Guideline for Reporting Randomized Trials. JAMA, 333(22), 1998–2005.
• Moody, L., et al. (2025). Digital diet and fitness tracking applications and their relationship with eating disorder psychopathology: A systematic review. European Eating Disorders Review.

← All research articles · How we build & check these tools