Changes in the brain unfolded differently among older adults with lower bone density in the spine: over roughly five years, they experienced faster deterioration in parts of the brain’s white matter and greater decline in global cognitive performance, although some of the study’s broadest brain-imaging findings did not remain statistically significant after correction for multiple comparisons.
The connection between skeletal health and brain aging has been observed in earlier research, but it has been less clear whether lower bone mineral density is associated with subsequent changes in the brain. To examine that question, researchers turned to the Multi-Ethnic Study of Atherosclerosis, a long-running prospective study in the United States.
The analysis focused on 715 participants who had measurements of vertebral volumetric bone mineral density, or vBMD, at one examination and valid cognitive or brain-imaging measurements at both that examination and a later follow-up. Their median age was 69, with an interquartile range of 65 to 75 years. The group included 397 men and 318 women.
The researchers used examination 6, conducted in 2016–2018, as the baseline for bone measurements. They then followed participants through examination 7, conducted in 2022–2024. Brain outcomes were assessed about 17–18 months after the baseline bone measurement and again at the later examination, giving roughly five years of follow-up between the two brain assessments.
The starting pool was considerably larger. Of 6,814 people originally enrolled in MESA, 3,303 had follow-up data, and 2,167 had quantifiable vertebral bone density at examination 6. Eighty-one people with a history of dementia or stroke before that examination were excluded. After additional exclusions related to missing or invalid cognitive and MRI measurements, 715 people contributed to one or more of the final analyses. The number varied by outcome, from 639 for the global cognitive composite to 408 for white matter hyperintensity measurements and 405 for fractional anisotropy.
Measuring bone from an ordinary chest CT
The bone measurement did not require a separate bone-density scan. The researchers extracted vertebral bone information from noncontrast chest CT scans.
A validated deep-learning system automatically segmented the thoracic vertebrae from T1 through T10. The researchers converted the CT measurements, expressed as Hounsfield units, into volumetric bone mineral density. Participants were included when at least eight of the 10 vertebral levels could be measured.
The brain was examined in several ways. MRI scans were acquired on 3-T Siemens systems using T1-weighted and fluid-attenuated inversion recovery sequences as well as diffusion tensor imaging.
One measure was white matter hyperintensity volume. These bright areas on FLAIR MRI were quantified with a validated deep-learning segmentation model. The other major imaging measure was fractional anisotropy, or FA, derived from diffusion tensor imaging. FA provides a measure of the organization of water diffusion within white matter and was used here as an indicator of white matter microstructural integrity.
The researchers examined total white matter as well as nine regional white matter areas, including the frontal, temporal, parietal and occipital lobes, deep white matter, the corpus callosum, the fornix and the anterior and posterior limbs of the internal capsule.
Cognition was assessed with several tests. The primary cognitive measure was a global cognitive composite made by averaging standardized scores across tests. The researchers also examined the Cognitive Abilities Screening Instrument, which measures global cognitive function, along with tests of processing speed, attention and working memory.
The statistical models accounted for a broad set of potential confounders, including age, sex, race or ethnicity, education, study site, APOE-ε4 status, body mass index, cholesterol, hypertension, diabetes, smoking, alcohol use, physical activity, kidney function, hormone replacement therapy and several medications. The models also incorporated weighting intended to account for differences in who remained in the study through follow-up and used multiple imputation for missing covariate information.
The clearest brain changes were regional
The researchers did not find evidence that lower vertebral bone density was associated with faster accumulation of white matter hyperintensities throughout the brain as a whole.
Among 408 participants with longitudinal white matter hyperintensity measurements, the association with total white matter hyperintensity accumulation was small and statistically nonsignificant. For every 0.1 g/cm³ lower vBMD, the estimated difference was 1.17% per year, with a 95% confidence interval ranging from −1.01% to 3.35% and a P value of .29.
The picture changed when the researchers looked at individual regions. In the corpus callosum, a major bundle of nerve fibers connecting the brain’s two hemispheres, each 0.1 g/cm³ lower vBMD was associated with a 12.8% faster annual increase in white matter hyperintensity volume. The 95% confidence interval was 4.2% to 21.4%, and the association remained statistically significant after false-discovery-rate correction, with an adjusted P value of .04.
The lowest bone-density quartile also showed faster corpus-callosum hyperintensity accumulation than the highest quartile, with an estimated difference of 14.6% and a P value of .01. Other regional white matter hyperintensity measures did not show statistically significant associations in the primary analysis.
The second imaging measure, fractional anisotropy, produced a similar pattern of some regional associations alongside a weaker overall result.
Among 405 participants with longitudinal FA measurements, every 0.1 g/cm³ decrease in baseline vBMD was associated with an additional 0.036 standard deviation per year decline in total white matter FA. The 95% confidence interval ranged from −0.065 to −0.006, with a nominal P value of .02. But that association did not survive the study’s false-discovery-rate correction. Its adjusted P value was .07.
The researchers estimated that this corresponded to about 0.2% of the baseline mean FA per year. The association was also smaller than the associations between FA decline and both age and diabetes in the same models.
One region produced a more robust result. In the anterior limb of the internal capsule, lower vBMD was associated with faster FA decline. Each 0.1 g/cm³ lower vBMD corresponded to a 0.048-standard-deviation-per-year faster decline, with a 95% confidence interval of −0.076 to −0.021. The nominal P value was less than .001, and the false-discovery-rate-adjusted P value was .006.
There were also nominal associations between lower vBMD and faster FA decline in the corpus callosum and temporal white matter, but these did not remain statistically significant after false-discovery-rate correction. Sensitivity analyses supported the association involving the anterior limb of the internal capsule, while the corpus-callosum association was less consistent when the follow-up weighting was removed.
The lowest vBMD quartile likewise had faster FA decline than the highest quartile in the temporal white matter and anterior limb of the internal capsule.
Lower bone density was also linked to cognitive decline
The cognitive results were more consistent at the global level.
Among 639 participants with valid global cognitive composite scores at both examinations, lower baseline vBMD was associated with faster decline. For every 0.1 g/cm³ lower vBMD, the global cognitive composite declined an additional 0.025 standard deviation per year. The 95% confidence interval was −0.042 to −0.009, with P = .002.
Over approximately five years, that association would correspond to about 0.15 standard deviation of additional decline for a 0.1 g/cm³ lower baseline vBMD, based on the study’s model. The authors noted that this was roughly one-third the magnitude of the overall annual time-related decline in the composite measure.
A similar association appeared with the Cognitive Abilities Screening Instrument. Among 675 participants, each 0.1 g/cm³ lower vBMD was associated with a β of −0.320 for the annual change, with a 95% confidence interval from −0.480 to −0.160 and P < .001. Because the CASI was analyzed on a logarithmic scale, the researchers interpreted this as about a 0.319% faster annual decline.
The relationship was not evident across the individual cognitive domains. There was no statistically significant association between vBMD and decline in digit symbol coding, digit span forward or digit span backward. The findings therefore centered on the broader measures of global cognitive performance rather than on a particular cognitive ability.
The quartile analysis was consistent with this pattern. Participants in the lowest vBMD quartile had faster declines in the global cognitive composite than those in each of the higher quartiles. Similar differences were seen for CASI when the lowest quartile was compared with the upper quartiles.
Diabetes changed one part of the relationship
The researchers also tested whether metabolic and other characteristics altered the associations between vBMD and brain aging.
The clearest interaction involved diabetes and white matter hyperintensity progression. The interaction between vBMD, time and diabetes status had a P value of .04. In analyses separated by diabetes status, lower vBMD was associated with faster total white matter hyperintensity accumulation among participants with diabetes, with an estimated 4.99% additional progression per year for each 0.1 g/cm³ lower vBMD. The 95% confidence interval was 0.40% to 9.57%, with P = .03. No corresponding association was found among participants without diabetes.
The authors treated these subgroup findings cautiously because the effect-modification analyses were exploratory, although they had been prespecified with a P < .10 threshold.
A possible sex difference also emerged for total white matter FA. The interaction between sex, vBMD and time had P = .08. In sex-specific analyses, the association between lower vBMD and faster FA decline was present among women but not men. The authors described this as suggestive rather than definitive. Other tested characteristics, including APOE-ε4 status, hypertension, diabetes, HDL, waist circumference and triglycerides, did not show evidence of modifying the vBMD-FA relationship.
For global cognition, the associations with vBMD were stronger among people with lower triglyceride levels. The CASI association was statistically significant in the lower-triglyceride group but not the higher-triglyceride group, and the global cognitive composite showed a similar pattern. These subgroup observations were also treated as exploratory.
The study does not show that low bone density causes brain decline
The researchers interpret the regional pattern as potentially consistent with a relationship between skeletal health and vulnerability of particular white matter structures. They point to the corpus callosum, anterior limb of the internal capsule and other regions as areas that can be susceptible to age-related white matter changes. They also discuss the possibility that microstructural changes detected through FA could precede the more visible lesions represented by white matter hyperintensities.
But those possible sequences are interpretations rather than something this study directly demonstrated.
The authors propose several possible explanations for the association. One is biological communication between bone and brain, including signaling by bone-derived factors such as osteocalcin. Another is that vBMD may act as an integrated marker of overall systemic health rather than directly affecting the brain. In that interpretation, lower bone density could reflect accumulated inflammation, metabolic dysfunction, physical inactivity or frailty that also contributes to brain aging. The study did not measure the bone-turnover markers, inflammatory mediators or bone-marrow adiposity needed to distinguish among these possibilities.
The observational design is another important qualification. Although the statistical models adjusted for many demographic, metabolic, behavioral and medical factors, the researchers state that causality cannot be inferred and that residual confounding remains possible, including from metabolic factors such as insulin resistance.
The study also cannot establish whether these associations eventually lead to dementia. The follow-up was long enough to detect changes in MRI measures and cognitive performance, but the authors note that it may not have been long enough to fully characterize late-life neurodegeneration or incident dementia.
Participation and survival could also have affected the results. People with the most severe disease may have been less likely to remain in the study, potentially weakening the observed associations. The inverse-probability weighting used by the researchers addressed differences in follow-up, but the authors note that residual healthy-survivor bias may remain.
The imaging approach has its own qualification. The CT-derived vBMD measurement was developed using the study’s research protocols and, according to the authors, still needs validation across scanners, CT protocols and clinical settings before routine use.
Most importantly, the study’s broadest white matter FA finding did not survive false-discovery-rate correction. The strongest corrected imaging associations were regional: faster FA decline in the anterior limb of the internal capsule and faster white matter hyperintensity accumulation in the corpus callosum. The authors describe these regional findings as hypothesis-generating and call for independent replication.
Taken together, the longitudinal data linked lower vertebral bone mineral density measured from noncontrast chest CT with modestly faster changes in specific measures of white matter integrity and with faster decline in global cognitive performance. The authors suggest that the association could reflect shared biological processes affecting bone and brain, rather than a direct effect of bone density on the brain, but the study does not distinguish between those possibilities.
The study was published in Radiology.
