The multiverse of human light exposure analyses

Background: Light exposure can both improve and disrupt health and well-being. Most evidence concerning the impact of light exposure comes from well-controlled laboratory studies employing parametric light stimuli. However, personalised light exposure in the real world is time-dependent and highly variable. Consequently, we lack a nuanced understanding of the statistical properties of personalised light exposure. This knowledge gap is partially due to the following challenges: (1) How to best measure human light exposure, including the choice of measurement devices, their properties, and their placement, (2) how to best quantify and summarise time-dependent light exposure, (3) howto achieve reproducible and robust results given a high degree of analytic flexibility.Methods: Here, we focus on challenges (2) and (3). We subject the wrist-referenced light dosimetry data from the Multi-Ethnic Study of Atherosclerosis data set (MESA; Zhang et al., 2018; Chen et al., 2015; n=2155, mean recording duration 7 days) to a range of light-related metrics implemented in pyActigraphy (Hammad et al., 2021), including summary statistics over intensity (e.g., linear and log mean, median) and timing (e.g., time above threshold). We then analyse how these metrics are statistically related to each other to understand which metrics capture overlapping information. Toaddress the question of analytic flexibility, we develop a “multiverse” analysis in which we vary various parameters, including bin size. Finally, we simulate missing data of different durations to understand the sensitivity of metrics to incomplete data.Results: We hypothesise to find considerable statistical overlap in various metrics. We also characterise the analytic flexibility inherent in light dosimetry analyses using a specification curve analysis, which visualises the outcome variable as a function of various choices of analytic parameters.Conclusion: We provide a systematic and data-driven method for exploring and characterising light exposure patterns in humans