Human research · Observational

Burns AC et al., Nature Mental Health 2023

Landmark T03 #1, methods and limits (revised against the author accepted manuscript; publisher version paywalled - accepted manuscript + free supplement + nature.com public back matter read). Design: cross-sectional secondary analysis of UK Biobank (no trial registration; observational). Light measured with a wrist-worn Axivity AX3 accelerometer / APDS9007 silicon photodiode (spectral sensitivity ~470-650 nm), dominant wrist, 7 days (2013-2015); psychiatric outcomes from the online Mental Health Questionnaire completed on average 1.86 years LATER (2016-2017). Device output was converted to approximate lux and piece-wise calibrated against a Tektronix J17 photometer in a validation run on only 7 AX3 devices from the original study, tested at 34 light levels from 0 to 5,050 lux, ~8 years after the recordings (recovered r2=0.82 low / 0.81 high, attenuated/non-linear at the low end). Within-person light ICC was 0.82 (95% CI 0.81-0.83), based on n=2,988 participants who repeated actigraphy four times, on average 3.23 y after the main assessment. Daily 24-h profiles were reduced by factor analysis to 'day light' (07:30-20:30) and 'night light' (00:30-06:00), each split into four EQUAL-sized quartiles (day-night correlation rs=0.10). PRIMARY analysis: logistic (disorders) / linear (symptom scales) regression of each quartile vs the Q1 referent, with likelihood-ratio chi2 omnibus tests and three nested models (Model 3 = age, sex, ethnicity, photoperiod, employment, physical activity). An additional interaction model tested the day x night interaction using per-SD log10 light (Supplementary Tables 7-8); the paper does not call this 'secondary' - the quartile models are what it calls its primary findings/hypotheses. Results were robust across sensitivity analyses (excluding shift workers; excluding clinical subgroups; adjusting for sleep [Model 4], urbanicity [Model 5] and cardiometabolic health [Model 6]), with author-noted exceptions: after sleep adjustment (Model 4) the night-psychosis omnibus test was no longer significant (P=0.09; Q4 1.12, 1.01-1.25; Suppl Table 12), and after cardiometabolic adjustment (Model 6) the day-PTSD omnibus test was no longer significant (P=0.11; Q4 0.87, 0.78-0.98; Suppl Table 16) - the authors describe day-PTSD as 'weaker'. Bipolar Q4 was also not significant in Models 4 and 6 (Suppl Tables 12, 16). The clinical-subgroup exclusion applied to the symptom scales only (Suppl Table 11). Author-stated limitations (Discussion): cross-sectional (possible reverse causation; longitudinal needed); wrist (not ocular) light is 'a coarse estimate'; actigraphy and outcomes not simultaneous; device coverage cannot be distinguished from true darkness; the photopic APDS9007 'cannot report melanopic equivalent daylight illuminance'; lux from a late device validation. No medical advice.

Verified research card EL-LM-T03-01-b-R3