Lifestyle and environmental exposures tracked closely with cognition in UK Biobank
In a large UK Biobank analysis, lifestyle and environmental exposures emerged as meaningful predictors of cognitive performance, with several everyday behaviors carrying measurable signal across the cohort. The work used machine learning across more than 10,000 participants and asked a simple but clinically relevant question: how much of cognition can be anticipated from modifiable habits and surroundings, and how much of that signal is already visible in the brain and body?
The answer was substantial. A model integrating twelve lifestyle and environment domains explained 23% of the variance in cognition and reached an r value of 0.48, a level described as comparable to the performance of body and brain markers. Physical activity, diet, alcohol consumption, sun exposure, and local environment stood out as the strongest predictors. In practical terms, cognition did not sit apart from daily life; it moved with the same behaviors and exposures that shape vascular, metabolic, inflammatory, and neurobiological health.
A composite brain marker captured 57.7% of the lifestyle-cognition association.
Brain structure and physiology absorbed much of the signal
The study moved beyond the familiar association between healthy behavior and sharper cognition by asking where that association was stored biologically. A composite brain marker, built from three neuroimaging modalities, accounted for 57.7% of the link between lifestyle and cognition. A composite body marker spanning nine physiological systems accounted for 47.8% of the same relationship. Those two findings point to a layered model of cognitive health: daily behavior appears to act through multiple pathways, and a large fraction of its association with cognition is already reflected in measurable brain and body status.
That pattern is biologically plausible. Physical activity can influence cerebral perfusion, synaptic plasticity, cardiometabolic risk, and sleep quality. Diet can shape insulin sensitivity, lipid metabolism, and inflammatory tone. Alcohol consumption may affect both vascular integrity and direct neurotoxicity depending on exposure patterns, while sun exposure may track with circadian regulation, outdoor activity, and broader lifestyle structure. Local environment adds another layer, likely capturing social and physical conditions that influence mobility, stress burden, and access to healthy routines. The model did not reduce cognition to a single determinant; it treated it as the output of several interacting systems.
Twelve domains, one composite cognitive signal
The analysis examined twelve domains spanning physical activity, diet, smoking, alcohol use, sleep, sexual behavior, electronic device use, and environmental exposures. Individually and in combination, these domains helped predict cognitive functioning across the cohort. Yet the most striking result was not the list of variables themselves. It was the breadth of the signal retained when lifestyle, environment, brain, and body were modeled together. Jointly, those domains captured 92.6% of age-related cognitive variation.
Age remains one of the strongest forces in cognitive change, so any model that absorbs nearly all age-related variation deserves careful attention. In this case, the result suggests that a large share of cognitive aging, at least within this dataset, aligns with markers of physiology and exposure rather than with age alone as an abstract chronological label. The implication is not that age can be “explained away,” but that age-related cognitive differences are tightly interwoven with the state of the organism and its lived environment.
Modifiable behavior still matters, but prediction is not causation
The strongest predictors—physical activity, diet, alcohol consumption, sun exposure, and local environment—are all modifiable or at least partly modifiable. That makes the findings attractive from a public health standpoint, because they point toward risk domains that can be observed before frank cognitive decline becomes obvious. Even so, the study remains computational and observational. Prediction strength does not establish that any single behavior directly causes the observed cognitive differences, and the direction of effect cannot be assumed from these associations alone.
Reverse causation is a real possibility in cognitive research. People with better cognition may be more able to maintain exercise routines, prepare healthier meals, spend more time outdoors, or live in environments that support those habits. Shared confounding is also likely, especially in a cohort as health-selected and relatively resource-rich as UK Biobank. The study design was strong for pattern recognition and mechanistic inference, but it cannot separate all behavioral causes from behavioral consequences.
Still, the converging signal across brain and body suggests that lifestyle is not merely a superficial correlate. It appears to be embedded in systems that are routinely measurable in clinical and research settings. Cerebrovascular health, neuroanatomical integrity, and multisystem physiology may act as the medium through which daily exposures influence cognitive performance over time.
Generalizability remains the central limitation
The investigators explicitly noted the need to evaluate whether these findings apply beyond UK Biobank. That caution is essential. UK Biobank participants are not a random slice of the general population, and machine-learning performance in one cohort can weaken when moved to different demographic, socioeconomic, or geographic settings. The model also depends on how exposures were measured and how cognition was operationalized, both of which influence predictive accuracy.
The study’s revised footnotes indicate that interpretation of prediction performance, commonality analyses, dietary findings, and outdoor time associations was clarified in the updated manuscript. Even with those refinements, the core message remains the same: lifestyle and environment carry a detectable cognitive signature, and much of that signature is visible in body physiology and the brain. For clinicians and researchers, the next step is not to overstate causality, but to test whether the same pattern holds in more diverse populations and whether the strongest predictive domains can be translated into useful prevention strategies.
The final constraint is practical. This was a computational study using existing UK Biobank data, and no new participant data were generated. The code is available, but the biological interpretation still depends on replication across populations with different backgrounds, exposures, and disease burdens. Until then, the findings speak most confidently to prediction rather than prescription—and to the degree to which cognition appears anchored in the everyday biology of aging rather than in the brain alone.