BOSTON — How older adults approach cognitive test problems, such as the small mistakes they make and the strategies they use, can help predict who will later develop dementia, a finding that is helpful in detecting earlier any risk of cognitive decline, according to a recent study.
The study published in the Journal of the International Neuropsychological Society examined patterns of errors and response strategies used by older adults from the Framingham Heart Study during cognitive tasks. The researchers investigated whether these behaviors reflect underlying cognitive abilities and whether these abilities identify people who will develop dementia.1
Study authors are affiliated with VA Boston (MA) Healthcare System and the Boston University Chobanian & Avedisian School of Medicine in Boston.
Scores from neuropsychological assessments are widely used to identify underlying cognitive abilities. However, relying on a single score representing a specific cognitive domain can obscure the detection of subtle cognitive changes, which are critical for early detection of cognitive disorders, and increase risk of inaccurate assessment, according to a Boston University press release on the study.
There has been a gap between how clinicians actually interpret cognitive testing and how researchers analyze it, Brandon Frank, PhD, a clinical neuropsychologist at VA Boston Healthcare System and corresponding author of the study, told U.S. Medicine. While he said he always considers observations outside of test scores to help interpret performance, including strategies and patterns of errors, these features are rarely incorporated into research models. Many existing analytic approaches also require simplifying assumptions, which don’t reflect real-world cognitive performance. In this study, the researchers aimed to bridge the gap between clinical practice and research methodology, Frank explained.
“Many research studies focus on total test scores (for example, how many words someone remembers on a memory test),” Frank said in a press release. “We showed that looking closely at patterns of errors and response styles provides deeper information. In our study, subtle cognitive features predicted dementia more than five years before diagnosis.”
In the study, the researchers analyzed data from 2,363 dementia-free participants aged 60 and older who were part of the Framingham Heart Study Brain Aging Program. The program, which has collected cognitive data since 2005 and records final test scores, extensively documents how people perform on cognitive tests. This includes specific types of errors, problem-solving strategies and subtle response patterns. Participants were primarily female and non-Hispanic White.
The investigators used advanced statistical modeling and machine learning to organize hundreds of small test details into meaningful brain-related patterns. They also assessed the impact of demographics on several factors and the ability of these factors to predict future conversion to all-cause dementia, study authors reported.
The researchers found that how people complete cognitive tests, including the small mistakes they make and the strategies they use, can predict who will later develop dementia, even years before diagnosis. They demonstrated that these patterns improve early detection of dementia risk, the press release pointed out.
“Subtle cognitive changes are detectable earlier than we typically measure,” Frank told U.S. Medicine. “Process-level data (errors, strategies, response patterns) provide meaningful signals before clear impairment on summary scores. These findings reinforce what many clinicians already observe: that “how” a patient performs often matters as much as “how much” they recall. For systems of care managing high-risk populations (e.g., veterans with complex medical and psychiatric comorbidity), this approach may be particularly valuable for early identification and risk stratification.”
The authors discovered that errors and response strategies reflected a general ability, but also seven underlying cognitive processes were related to each of the cognitive tasks. This pattern was consistent across key demographics such as age, sex and education. In addition, older, less educated participants produced fewer accurate and strategic responses, while differences between males and females were varied. Overall, the study’s approach was helpful in identifying participants who would develop dementia, the researchers suggested.
Based on the results, process-level features may offer additional information beyond summary scores. This includes observing error types, problem-solving strategies and patterns of responses across tasks. The findings suggest that these process-level features capture subtle cognitive changes that may precede measurable decline on standard scores and are strongly associated with future dementia risk, Frank recommended.
By considering process-level features, clinicians could improve patient care by identifying patients who may benefit from earlier monitoring, intervention or referral, even when screening scores appear relatively normal. Ultimately, the approach supports more personalized and earlier detection-focused care. In research, it’s also beneficial to understand patterns, such as types of errors, problem‑solving strategies or response inefficiencies that might reflect subtle cognitive changes before they appear on summary scores, Frank pointed out.
The study’s results also could change practice for physicians. While brief screening tools, such as the Montreal Cognitive Assessment (MoCA) or Mini Mental State Examination (MMSE) are useful and important, these instruments can’t replicate the information provided by more comprehensive testing.
The analysis highlighted why comprehensive neuropsychological evaluation is often used in research settings to capture richer data than brief screens, and it contributes to ongoing discussions about how multidimensional data can complement existing clinical tools. The study demonstrates that incorporating these richer data improves sensitivity for detecting individuals who will later develop dementia, even when traditional measures already perform well, Frank suggested.
“This study reinforces that neuropsychological testing is far richer than a set of scores—it is a window into how the brain solves problems,” Frank said. “While advanced statistical and machine learning methods helped us demonstrate this at scale, the core message is simple and practical: clinicians already observe these patterns, and they matter.”
In the future, emerging digital tools might make it easier to capture these process-level features in routine care. Also, integrating these approaches could support earlier diagnosis, more targeted interventions and better patient outcomes. Even modest improvements in early detection can have meaningful impacts for patients and families, particularly as disease-modifying treatments continue to evolve, he explained.
- Frank B, Gurnani A, Hurley L, Guan C, Andersen SL, et. Al. Psychometric modeling of Boston process approach data for dementia prediction in the Framingham Heart Study. J Int Neuropsychol Soc. 2026 Apr 14:1-12. doi: 10.1017/S1355617726101921. Epub ahead of print. PMID: 41975557; PMCID: PMC13082455.