Power analysis for personal light exposure measurements and interventions

Area of Science:

• Chronobiology
• Human Physiology
• Environmental Health

Background:

• Light exposure significantly impacts circadian rhythms, health, well-being, and performance.
• Field studies on light exposure generate complex data from wearable loggers, necessitating robust sample size determination.
• Current sample size decisions lack statistical power analysis, highlighting a need for standardized guidance.

Purpose of the Study:

• To introduce a novel hierarchical bootstrapping procedure for calculating statistical power and sample size for wearable light and optical radiation data.
• To provide a reproducible method applicable to various light exposure metrics and datasets.
• To address the challenge of determining appropriate sample sizes in complex field studies.

Main Methods:

• Developed a hierarchical bootstrapping procedure accounting for mixed-effects models and stepwise resampling.
• Utilized a dataset comprising one week of continuous light data from 13 shift-work participants in winter and summer.
• Applied the method to twelve different summary metrics derived from light and optical radiation logging data.

Main Results:

• Required sample sizes varied from 3 to over 50 depending on the metric, with metrics focusing on bright light needing fewer participants.
• Metrics related to dark time and daily patterns, such as timing of light below 10 lux and intradaily variability, required higher sample sizes (5-45).
• Geometric standard deviation and midpoint of darkest 5 hours lacked sufficient power within tested sample sizes.

Conclusions:

• The novel hierarchical bootstrapping method effectively estimates sample size for light exposure studies.
• The method is adaptable to different metrics and datasets, providing a statistical basis for sample size selection.
• This approach addresses the unique data structure of wearable light sensors, enhancing research rigor.