Memory Variability Points to a Different Hippocampal Signal Than Average Recall in Older Latino Adults
Memory performance is often treated as a single number, but this preprint separates two features that can move in opposite directions: average score and memory intraindividual variability (IIV). In a cohort of 67 older Latino adults without dementia, the two measures did not behave the same way in relation to brain structure. Mean memory performance tracked hippocampal volume, while memory IIV aligned with microstructural integrity in the hippocampus-cingulum tract. The separation is clinically interesting because it suggests that inconsistency across memory tasks may capture a different neural vulnerability than how well someone performs on average.
The study drew participants from the Boston Latino Aging Study and paired cognitive testing with MRI. Investigators used linear regression to examine whether memory IIV related to hippocampal volume and to fractional anisotropy in the hippocampus-cingulum tract. They also ran parallel analyses with mean memory performance as the outcome, allowing a direct comparison between stable performance and fluctuating performance. In a field that often relies on broad screening scores, this design matters because variability can reveal early network dysfunction even when average performance looks acceptable.
One result stood out immediately. Higher mean memory performance was associated with greater hippocampal volume, with β = 0.24 and p = .03. Memory IIV did not show a meaningful relationship with hippocampal volume, with β = 0.028 and p = .84. In other words, better average recall tracked with larger hippocampi, but inconsistency did not. For clinicians and researchers, this hints that volume may reflect preserved gray matter reserve, while variability may arise from something less visible on routine structural imaging.
White Matter Integrity and Fluctuating Memory Performance
The opposite pattern emerged for the white matter measure. Higher memory IIV was associated with lower fractional anisotropy in the hippocampus-cingulum tract, with β = -0.34 and p < .01. Mean memory performance did not relate to that tract measure, with β = -0.07 and p = .52. A single sentence captures the core signal: fluctuating memory was tied to weaker hippocampal white matter integrity, not to gross hippocampal size.
This pattern fits neurobiological expectations. The hippocampus-cingulum pathway helps connect medial temporal memory circuitry with broader associative networks. When that white matter tract loses integrity, information transfer becomes less efficient, and performance can become erratic even before average ability drops sharply. That kind of instability may reflect disrupted coordination within memory circuits rather than a simple reduction in storage capacity. In practical terms, a person can still test within a normal range overall and yet show uneven, unreliable memory performance that points to underlying network disruption.
The finding also gives memory IIV a distinct anatomical signature. Rather than mirroring the same substrate as mean performance, variability appears to register the health of the connecting tract. This distinction is useful because hippocampal volume and white matter integrity are not interchangeable biological markers. Gray matter loss and tract degradation may move on different timelines, especially in aging brains where compensatory strategies can preserve average scores for a while.
Why Latino Aging Cohorts Matter for Memory Biomarkers
The focus on older Latino adults gives the study particular relevance. Latino populations remain underrepresented in many aging and dementia biomarker studies, even though inequities in risk exposure, access to care, and diagnostic delay can shape cognitive trajectories. A memory marker that behaves differently in this group may help refine how clinicians think about early detection and risk stratification. If average test scores miss the earliest network changes, memory variability could offer a more sensitive signal in people who are still performing well enough to avoid obvious clinical flags.
The sample size was modest, and the work remains a preprint, so the findings should be treated as preliminary rather than practice-changing. Still, the pattern is internally coherent. Mean memory performance was linked to hippocampal volume, and memory IIV was linked to hippocampus-cingulum tract fractional anisotropy. The divergence suggests that a standard memory score and a variability metric are not redundant; they may be readouts of separate parts of the same aging process.
For bedside translation, the most obvious question is whether neuropsychological testing should begin to include within-domain variability as a routine analytic feature, not just a byproduct of repeated measures. If a patient’s scores swing widely from one task or time point to another, that may signal white matter vulnerability even when the average score remains reassuring. Such a signal would not diagnose dementia on its own, but it could sharpen suspicion for early neurodegenerative change or other age-related network disruption.
Hippocampal Volume and Memory Mean Scores Tell Only Part of the Story
The study’s discussion points toward a useful conceptual split: memory IIV may be more closely tied to hippocampal white matter, while mean memory performance may better reflect hippocampal volume. That framing is biologically plausible and clinically tidy. A larger hippocampus can support stronger average encoding and recall, whereas preserved tract integrity may be what keeps performance stable from one trial to the next. When the tract falters, performance may wobble even before the overall average sinks.
This also raises a measurement issue for future work. Many cognitive studies collapse repeated observations into a mean, which can erase information about instability. Variability is not just statistical noise; in this context, it may carry neurobiological meaning. The present analysis suggests that ignoring that variability could miss a marker of hippocampal connectivity that is especially relevant in older adults who still appear cognitively intact.
Competing interests were limited; one author reported consultancy for Biogen, while the remaining authors reported no competing interests. The study was supported by the National Institute on Aging and Massachusetts General Hospital funding. Participants provided written informed consent, and the work received institutional review board approval at Massachusetts General Hospital.
The most important limitation is also the most basic: this is a cross-sectional preprint with 67 participants. Without longitudinal follow-up, it remains impossible to know whether memory IIV predicts later cognitive decline or simply reflects a stable individual trait linked to white matter architecture. The next step is clear enough from the data already in hand: follow these participants over time, determine whether variability precedes measurable decline, and test whether the hippocampus-cingulum tract can help explain who changes first.