Personalized Machine Learning Intervention to Improve Sleep Quality Using Wearable Technology in Healthy Middle-Aged Adults From Mexico City: Protocol for a Pilot Randomized Controlled Trial.

Global and national sleep dissatisfaction is widespread, motivating the study design.

• In 2019, global sleep surveys reported that 80% of adults want to improve their sleep quality.
• In 2021, 45% of adults were reported to be dissatisfied with their sleep.
• In 2025, among American adults, 37% reported sleep dissatisfaction and 38% reported not feeling energized after sleep.
• A 2016 nationally representative survey of Mexican adults (aged ≥18 years) found that 37% reported sleep problems.

The pilot RCT is designed to enroll 32 participants stratified by sex across two arms in Mexico City.

• Total planned enrollment is 32 participants, with 16 per arm.
• Participants are stratified by sex.
• All participants are healthy middle-aged adults from Mexico City.
• All participants wear Samsung Galaxy Watch 4 devices for 60 days.

The study collects 10 objective sleep variables during a 30-day baseline period before any intervention is delivered.

• Objective sleep data are collected during days 1–30 (baseline phase).
• The 10 variables are related to duration, efficiency, sleep stages, movements, cycles, and recovery metrics.
• The control group receives generic sleep hygiene education.
• The experimental group receives personalized recommendations on day 30 based on ML model outputs.

Machine learning personalization uses Shapley Additive Explanations (SHAP) analysis and recursive feature elimination to identify top predictive sleep parameters for each individual.

• Personalized recommendations are generated using Shapley Additive Explanations (SHAP) analysis.
• Recursive feature elimination is also used to identify the top predictive sleep parameters.
• Recommendations are delivered as a single personalized intervention on day 30 to the experimental group.
• The framework is described as 'manufacturer-independent.'

The primary outcome is the wearable device sleep score (scale 1–100) during days 31–60, analyzed using analysis of covariance with baseline sleep score as a single covariate.

• The sleep score is a composite device metric on a scale of 1 to 100.
• The analysis period for the primary outcome is days 31–60 (post-intervention).
• Analysis of covariance (ANCOVA) is used with the baseline sleep score as the single covariate.
• The study will compare 960 nights from the control group with 960 nights from the experimental group.

The secondary outcome is the Pittsburgh Sleep Quality Index (PSQI) global score assessed at three time points to validate objective findings with a subjective measure.

• The PSQI global score uses a scale of 0–21.
• It is assessed at baseline, day 30, and day 60.
• The PSQI serves as a subjective validation measure alongside the objective wearable data.
• The study combines 'objective device monitoring with a subjective measure (PSQI) to test whether precision targeting of individual sleep parameters outperforms generic recommendations.'

The study timeline spans from August 2024 recruitment through an expected March 2026 results availability date.

• Recruitment started in August 2024 and ended in July 2025.
• Data collection is expected to be completed by December 2025.
• Results will be available by March 2026.
• Findings will be submitted for publication within 6 months of study completion.

The pilot study is designed to establish feasibility and preliminary effect size for ML-personalized sleep interventions using consumer wearables.

• The study aims to 'establish the feasibility and preliminary effect size for ML-personalized sleep interventions using consumer wearables.'
• A secondary aim is to generate a dataset from objective data for iterative model training and analysis.
• The study also aims to correlate objective and subjective sleep quality metrics.
• The authors state the approach could 'advance sleep interventions from universal protocols toward individualized behavioral targeting' if validated.