Not All Light Exposure Measures Are Equal: New Framework Aims to Fix Light Epidemiology

Light has quietly become one of the most talked-about variables in population health research. Over the past few years, a wave of large-scale studies has linked the patterns of light and darkness people experience every day to psychiatric conditions, metabolic disease, cardiovascular risk, sleep problems and even mortality. Some of these studies have drawn on satellite images of artificial brightness at night, others on wrist-worn sensors worn by hundreds of thousands of participants, and still others on simple light meters placed in bedrooms. The scale of this work is unprecedented, and it has propelled light exposure into the mainstream of environmental epidemiology as a potentially modifiable determinant of health.

But a new Debate article published in BMC Medicine by chronobiologist Manuel Spitschan and Johannes Zauner, both of the Technical University of Munich, argues that the field is standing on a measurement problem it has barely begun to confront. Their central claim is deceptively simple: not all light exposure measures are equal. Satellite-derived night-time radiance, wrist-worn illuminance readings, bedroom light levels, light measured at the cornea, retinal exposure after optical and pupillary modulation, and modelled circadian stimulus are all related, but they are not interchangeable. Each one captures a different construct, and each construct supports different kinds of scientific inference.

The crux of the argument lies in a question that sounds almost philosophical but has hard practical consequences: what light exposure construct is actually being measured, and how closely does it correspond to the biologically relevant stimulus arriving at the eye and the retina? The human circadian system does not respond to light in the environment in the abstract. It responds to photons striking a specific class of retinal cells, the intrinsically photosensitive retinal ganglion cells, which are most sensitive to a portion of the blue-cyan spectrum. The signal those cells send to the brain’s master clock depends on the spectral composition, intensity, timing and duration of light reaching the retina, which in turn depends on everything between the light source and the photoreceptor: windows, eyeglasses, eyelids, head orientation and the pupil itself.

Consider what a satellite actually measures. Satellites capture upward-directed radiance from the Earth’s surface at night, a proxy for outdoor artificial lighting in a neighbourhood. That measure says nothing about how much light enters a person’s eyes while they sit indoors, sleep behind blackout curtains, or walk down a lit street. A wrist-worn light sensor, by contrast, records illumination at the wrist, which can diverge dramatically from light at the eye depending on body posture, clothing and activity. Bedroom illuminance meters capture only one micro-environment in a 24-hour cycle. Each of these scalable proxies has enabled research at population scale that no laboratory study could ever achieve, but each carries its own uncertainty about what it truly represents.

Spitschan and Zauner do not argue that these scalable measures should be abandoned. On the contrary, they acknowledge that such tools have delivered evidence at scales impossible in laboratory or small field studies, and they have raised the visibility of light within population health research. The central challenge, as they frame it, is not whether proxies should be used, but how their meaning, uncertainty and relationship to actual ocular exposure can be specified and empirically quantified. A satellite pixel can be a perfectly legitimate exposure measure for a question about community-level outdoor lighting, and a wrist sensor can be legitimate for a question about behavioural light patterns, but conflating these constructs with retinal light exposure invites misinterpretation.

To address this, the authors propose a construct-centred framework for the next phase of light epidemiology. The framework rests on three pillars. First, researchers should explicitly define the exposure construct they intend to measure, stating precisely what physical quantity, at what anatomical location, over what time window, their instrument captures. Second, measures should be validated across measurement sites and spatial scales, so that the relationship between a proxy such as satellite radiance or wrist illuminance and the gold standard of corneal or retinal exposure is empirically characterised rather than assumed. Third, studies should pre-specify causal models that link behavioural determinants, 24-hour light-dark exposure, biological pathways, intermediate phenotypes and clinical outcomes, making the chain of inference explicit before data are collected.

The emphasis on pre-specified causal models reflects a broader concern about how light-health associations are interpreted. When a study reports that people living in brighter neighbourhoods at night have higher rates of a particular disease, the exposure is a neighbourhood-level proxy filtered through countless individual behaviours: time spent outdoors, window coverage, screen use, sleep timing, occupation and socioeconomic circumstances. Without a pre-specified model of how the measured construct relates to the biologically relevant stimulus and to potential confounders, such associations remain difficult to interpret causally. The framework is designed to move the field from broad associations towards mechanistically informed evidence that can actually support interventions.

The technical vocabulary of the field illustrates why construct clarity matters. Melanopic equivalent daylight illuminance, a metric standardised through the International Commission on Illumination, quantifies light in terms of its effectiveness for the melanopsin-containing photoreceptors that drive non-visual responses. It is a physiologically grounded quantity, but computing it requires spectral information that many wearable devices do not capture. Modelled circadian stimulus goes further, incorporating the nonlinear temporal integration properties of the circadian system. These refined constructs sit at one end of a spectrum; coarse but scalable proxies sit at the other. The authors argue that both ends have value, provided researchers are explicit about which construct they are measuring and what inferences it can legitimately support.

The paper also draws on validation work conducted under the European metrology project MeLiDos, which focuses on metrology for wearable light loggers and optical radiation dosimeters. Participant data illustrated in the article derive from previously conducted studies approved by the Medical Ethics Committee of the Technical University of Munich and by ethics committees at participating MeLiDos sites, with all participants providing written informed consent. This kind of multi-site validation, the authors contend, is exactly what the field needs more of: systematic characterisation of how well different instruments and proxies agree, and quantification of the uncertainty that propagates from measurement to inference.

The stakes extend beyond academic rigour. If light exposure is genuinely a modifiable health determinant, then eventually the field will be asked to produce actionable guidance, much as it has for sleep, diet and physical activity. Recommendations about daytime brightness, evening dimming and night-time darkness can only be as good as the exposure science underneath them. A framework that distinguishes constructs, validates proxies against ocular exposure and embeds measurement in explicit causal models is, in the authors’ view, the route to translationally useful evidence. Their stated goal is to support the integration of high-throughput epidemiology with physiologically interpretable measurement, enabling light epidemiology to mature from an era of discovery into an era of mechanism and intervention. As wearables proliferate and satellite data grow ever finer, the question of what exactly is being measured when we measure light may determine whether the field’s promise is realised.

Subject of Research: Exposure assessment and causal inference methodology in light epidemiology

Article Title: Not all light exposure measures are equal: a framework for exposure assessment and causal inference in light epidemiology

Article References: Not all light exposure measures are equal: a framework for exposure assessment and causal inference in light epidemiology. (n.d.). https://doi.org/10.1186/s12916-026-05259-z

Image Credits: AI Generated

DOI: 10.1186/s12916-026-05259-z

Keywords: light exposure, environmental epidemiology, exposure assessment, causal inference, circadian rhythms, wearable sensors, ocular light exposure, melanopic equivalent daylight illuminance, measurement validation, satellite radiance, retinal exposure, public health