Daily Bright Light Exposure Linked to 17% Lower Dementia Hazard in Study of 87,000 People

It is widely known that exposure to natural light during the day helps regulate the body clock and mood. In recent years, attention has turned to whether time spent outdoors and everyday light environments might also relate to long-term brain health and dementia risk. However, large-scale epidemiological studies that objectively measured the amount of light people receive in daily life and followed them over a long period to examine later dementia onset have been limited.

Published on June 24, 2026, in the journal General Psychiatry, the study by Nana Zheng and colleagues used data from the UK Biobank, a large British research resource, to analyze the relationship between time spent in bright daytime light and the risk of a future dementia diagnosis. The team focused on the time during which illuminance measured by a wrist-worn sensor reached 5,000 lux or more. They reported that people who exceeded a threshold of 0.70 hours (about 42 minutes) per day had a lower hazard of later developing dementia. However, this is a statistical association based on observational data. It does not prove that getting 42 minutes of bright light every day prevents dementia.

A Statistical Threshold of 0.70 Hours in Follow-Up Data from 87,000 People

The researchers analyzed 87,577 UK Biobank participants who had not been diagnosed with dementia at the time their light exposure was measured. The average age was 62.36 years, and women made up 56.98%. The median follow-up period was 8.1 years, during which 741 people (0.85%) were newly diagnosed with dementia.

The study totaled the time during which a wrist-worn sensor recorded illuminance of 5,000 lux or more. Using a threshold of 0.70 hours (42 minutes) per day, derived from exploratory statistical analysis, the researchers divided participants into two groups. In a Cox proportional hazards model adjusted for age, sex, lifestyle, health status, and other factors, the adjusted hazard ratio (HR) for the group exposed to 5,000 lux or more for over 0.70 hours per day was 0.83 compared with the group at or below 0.70 hours. The 95% confidence interval was 0.70–0.99 and the p-value was 0.036. Statistical significance was maintained after correction for multiple comparisons using the false discovery rate (FDR) (adjusted p < 0.05).

An HR of 0.83 means that, in the statistical model, the relative hazard of developing dementia was 17% lower. This does not mean the incidence rate itself fell by 17 percentage points. In the observed data, the incidence in the two groups was about 0.83% and about 0.87%, a difference of only about 0.05 percentage points. In addition, a hazard ratio is not a simple comparison of incidence over the follow-up period; it is an estimate that accounts for the passage of time and covariates. It is therefore necessary to distinguish between the relative difference in hazard and the difference in observed incidence.

What the Wrist Light Sensor Actually Measured

To interpret these results correctly, it is necessary to understand how light exposure was measured. Data were collected between 2014 and 2018. Participants wore an Axivity AX3, a device with an accelerometer and light sensor, on their dominant wrist and went about their normal lives for seven consecutive days. The device has a built-in APDS9007 silicon photodiode sensor that detects illuminance, allowing it to record both physical activity and ambient brightness.

The researchers defined "daytime" as 7:30 a.m. to 8:30 p.m. and "nighttime" as 12:30 a.m. to 6:00 a.m. For daytime, they calculated mean illuminance as well as the cumulative daily time spent at 3,000 lux or more, 5,000 lux or more, and 7,000 lux or more. The thresholds used in the analysis were a mean illuminance of 1,000 lux, 1.40 hours at 3,000 lux or more, 0.70 hours at 5,000 lux or more, and 0.45 hours at 7,000 lux or more. These were set based on exploratory analysis of the continuous relationship between light exposure and dementia onset.

The light sensor built into the AX3 is reported to respond approximately linearly in the range of 0 to 5,500 lux. Measurements based on prior research have an intraclass correlation coefficient (ICC) of 0.82 (95% confidence interval: 0.81–0.83), confirming a certain level of measurement reliability. However, illuminance measured at the wrist varies greatly depending on arm orientation and clothing sleeves. Even when the sensor records darkness, it cannot fully distinguish whether the surroundings were actually dark or whether the sensor was covered by a sleeve or similar.

Dementia onset was determined by cross-referencing primary care records, hospital admission records, and death registry records after the sensor was worn. Follow-up continued until the participant's date of death or January 25, 2024, the last date for which diagnostic records were available. Previous research has reported that dementia diagnoses obtained through UK Biobank record linkage have an overall positive predictive value of 82.50%. This means that some misclassification in the diagnostic information is possible.

The analysis adjusted for demographic factors such as age and sex, as well as socioeconomic status, environmental factors, diet, smoking, alcohol consumption, physical activity, morning/evening preference (chronotype), and social isolation. It also accounted for known dementia risk factors, including obesity, diabetes, hypertension, history of head injury, and hearing loss. Although these adjustments reduce the influence of measured factors, they cannot eliminate unmeasured confounders.

Consistent Association with Bright Daytime Light, but No Significant Difference for Nighttime Illuminance

The researchers examined the relationship between dementia onset and multiple illuminance levels, not just 5,000 lux or more.

For the group whose mean daytime illuminance exceeded 1,000 lux, the adjusted HR compared with the group at or below that level was 0.84 (95% CI: 0.71–0.99, p=0.039). For those exposed to 3,000 lux or more for over 1.40 hours per day, it was 0.82 (95% CI: 0.70–0.97, p=0.024), and for those exposed to 7,000 lux or more for over 0.45 hours, it was 0.83 (95% CI: 0.70–0.99, p=0.041). In every case, a statistical association was observed in which groups with longer exposure to bright daytime light had a lower hazard of dementia.

Nighttime light exposure, by contrast, produced a different result. The researchers divided mean illuminance between 12:30 a.m. and 6:00 a.m. into quartiles and compared the brightest fourth quartile (median 80.80 lux) with the darkest first quartile (median 0.89 lux). The adjusted HR was 0.96 (95% CI: 0.78–1.18, p=0.676). Because the confidence interval spans 1, no statistically significant association was found between nighttime illuminance and dementia onset.

The main results are summarized below.

All four daytime measures remained statistically significant after FDR correction. However, the absence of a significant association for nighttime illuminance does not prove that nighttime light exposure has no effect on health. During sleep, the wrist sensor may easily be covered by bedding or affected by sleeping posture, so the measured values may not match the amount of light actually reaching the eyes. In particular, if someone looks at a smartphone or other screen close to their face, a wrist sensor may not adequately capture that light.

To check the robustness of the results, the research team also conducted several sensitivity analyses. These included analyses treating death as a competing risk, analyses limited to data without missing values, adjustment for seasonal differences by month of wear, exclusion of shift workers and people with visual impairment, adjustment for APOE ε4 carrier status, history of depression, and time spent outdoors, and a redefinition of daytime based on day length. The association between daytime light exposure and dementia onset held under many of these conditions, but in some analyses it weakened and lost statistical significance. How firmly daytime light exposure is established as an independent risk factor therefore still requires further verification.

The Correct Interpretation of Machine Learning's "More Important Than Obesity"

Some news coverage of the study used expressions such as "lack of bright daytime light is a stronger predictor of dementia than obesity." The basis for this was the researchers' feature importance analysis using the machine learning model XGBoost.

The team built an XGBoost model using 19 features to distinguish the 741 people diagnosed with dementia from the 86,836 who were not. They used 10-fold cross-validation to tune the model's parameters and evaluated how much each feature contributed to the predictions using SHAP (SHapley Additive exPlanations) values. SHAP is a model interpretation method based on game theory that quantifies the contribution of individual variables to a prediction.

The result: the indicator "less than 0.70 hours per day of exposure to 5,000 lux or more" ranked 10th out of 19 items. It sat between low education level (9th) and alcohol consumption (11th), and ranked above indicators included in the model such as obesity, diabetes, and hypertension.

{
  "type": "bar",
  "title": "Relative SHAP rank of main features in the XGBoost model",
  "unit": "Rank (lower = greater contribution)",
  "categories": ["Age", "Education level (low)", "≥5,000 lux time (<0.70 h)", "Alcohol consumption", "≥3,000 lux time (<1.40 h)", "≥7,000 lux time (<0.45 h)", "Mean daytime illuminance (<1,000 lux)"],
  "series": [
    { "name": "Rank among 19 input variables", "values": [1, 9, 10, 11, 14, 15, 18] }
  ],
  "caption": "Ranking of predictive contribution in a classification model using data from 87,000 UK Biobank participants",
  "source": "Zheng et al. (2026) General Psychiatry"
}

However, SHAP values indicate how much each variable contributed to the predictions within the constructed model, not its importance as a cause of dementia. Even if a feature ranks high, deliberately changing it will not necessarily reduce dementia onset. A model variable's contribution to prediction must be clearly distinguished from the effect of an actual intervention.

Among other light exposure indicators, the condition of less than 1.40 hours at 3,000 lux or more ranked 14th, less than 0.45 hours at 7,000 lux or more ranked 15th, and mean daytime illuminance below 1,000 lux ranked 18th. That predictive contribution differs depending on how illuminance and time are defined is also important when interpreting this ranking. What the analysis shows is that a particular light exposure indicator carried some predictive information within the model, not that insufficient light is a more serious cause of dementia than obesity or lifestyle-related diseases.

Stronger Associations in Night Owls and APOE ε4 Carriers; Brain Structure Also Explored

The researchers also examined whether the association between daytime light exposure and dementia onset differed depending on participants' characteristics and genetic background. Interaction tests reported statistically significant interactions for nighttime light exposure level, chronotype, and carrier status of the APOE ε4 allele, which is linked to Alzheimer's disease risk. No significant interaction was found for age, sex, the time of day when light was received, or presence of depression.

In stratified analyses, among people with high nighttime light exposure, the group with more daytime light exposure had a 30–38% lower hazard of dementia. Similarly, associations of 31–41% lower hazard were observed in evening types and 19–27% lower in APOE ε4 carriers. However, these are relative differences estimated from statistical models within each group, not results from an intervention that increased light exposure. It has not been proven that bright light is effective for preventing or treating dementia in people with a particular genotype or sleep type.

The study also conducted mediation analysis on possible biological pathways linking daytime light exposure and dementia onset. Mediation analysis is a statistical method for estimating how much of the association between a factor and an outcome is explained through another variable. The team examined three candidates: rest–activity rhythms, brain structure, and serum vitamin D levels.

First, for rest–activity rhythms, they analyzed measures such as M10, the average activity level during the most active consecutive 10 hours, and L5, the average activity level during the least active consecutive 5 hours. The estimated proportion mediated by these measures was 5.70–32.99%, and was statistically significant before correction. After FDR correction for multiple comparisons, however, none remained significant.

Next, among 16,855 people with available brain imaging data, they examined how much structural brain features explained the association. The cortical surface area of the fusiform gyrus, which is involved in visual recognition of objects and faces, was estimated to statistically mediate 9.20% of the association between mean daytime illuminance and dementia onset. This result remained significant after FDR correction (p=0.018). Exploratory associations were also seen for the image intensity of the posterior corpus callosum and the area of the caudal anterior cingulate cortex, but significance was lost after FDR correction.

These results offer clues for investigating whether circadian rhythms and brain structure may be involved in the association between daytime light exposure and dementia onset. However, statistical mediation analysis does not directly prove a causal chain in which light exposure changes brain structure and thereby prevents dementia. In addition, no significant mediation effect was found for serum vitamin D levels, so at least in these data, a pathway via vitamin D could not be confirmed.

Can "42 Minutes a Day" Prevent Dementia? Remaining Measurement Limitations

The study is significant in that it objectively measured light exposure in daily life and linked it to long-term dementia onset data. However, several important constraints remain before the results could be used as general prevention guidance.

First, UK Biobank participants tend to be healthier and more socioeconomically advantaged than the general UK population. Because of this "healthy volunteer effect," the observed incidence rates and the light exposure thresholds derived from the statistical analysis may not apply directly to other populations. In particular, the specific figure of 42 minutes at 5,000 lux or more needs to be checked for reproducibility in groups with different living environments and health conditions.

Second, light exposure was measured for only seven days. Dementia develops over many years, yet what was measured here was the light environment of just one week of that time. It is unclear how well this short record represents lifestyle over the following years. In addition, some covariates were obtained up to 5.6 years before the sensor was worn, creating a time lag with the actual light exposure measurement.

Third, there are physical limits to the measuring device. A wrist-worn sensor does not directly measure the amount of light that actually reaches the retina. Light entering the eye varies with gaze direction, pupil size, and whether glasses or sunglasses are worn, so it does not necessarily match wrist illuminance. Furthermore, the AX3 light sensor cannot measure intensity by wavelength, so it cannot assess the spectral characteristics of light that matter for regulating circadian rhythms.

The timing of measurement must also be considered. The light exposure data were collected in 2014–2018, and since then the spread of LED lighting and patterns of nighttime digital device use have changed. The intensity and wavelength distribution of light people receive indoors and outdoors may have changed, so applying the results of that time directly to today's light environments calls for caution.

Most important of all, this is an observational study and cannot prove causation. Even with statistical adjustment for numerous factors such as age, lifestyle, and health status, the possibility of unmeasured confounders and reverse causation cannot be fully ruled out. For example, people who can spend long periods in bright environments of 5,000 lux or more during the day may already have been more physically and socially active and in better health. Conversely, declines in activity or changes in lifestyle that began before a dementia diagnosis may have shortened the time people spent in light.

The figure of "42 minutes a day" is a statistical threshold derived from epidemiological data, not a light exposure time that has been demonstrated to be necessary for dementia prevention. Going forward, it will be necessary to measure the amount and wavelength distribution of light reaching the retina more accurately and to test whether similar associations are reproduced in groups of different ages and social backgrounds. Randomized controlled trials (RCTs) that actually change the light environment are also needed to verify effects on cognitive function and dementia onset. All that can be said at this stage is that exposure to bright daytime light may be associated with future dementia risk; establishing it as a preventive effect will require further research.