Cognitive decline in later life may not follow a single path, and a new study of Chinese older adults is drawing attention to the possibility that dementia prevention could become more precise by combining statistical profiling with artificial intelligence. Published in BMC Geriatrics, the research by Lai, Dong and Fu investigates how older adults can be grouped into distinct cognitive subtypes and how different risk factors interact within those groups. The work brings together two increasingly influential methods—latent profile analysis and explainable machine learning—to explore why some individuals experience greater cognitive vulnerability than others.
Rather than treating cognitive health as a simple scale ranging from “normal” to “impaired,” the researchers use latent profile analysis, or LPA, to search for hidden patterns in data. LPA is a statistical technique that sorts individuals into groups based on similarities across several measured characteristics. In the context of ageing, those characteristics may include memory, attention, executive function, daily functioning and other health-related indicators. The method does not require researchers to decide in advance exactly which categories exist; instead, mathematical models estimate the most plausible underlying profiles.
That approach is important because older adults with similar overall cognitive scores may have very different combinations of strengths and weaknesses. One person may show difficulties primarily in memory, while another may retain memory but struggle with attention or problem-solving. These distinctions can be missed when cognitive decline is summarized with a single score. By identifying patterns across multiple domains, latent profile analysis may help researchers describe cognitive ageing as a diverse collection of trajectories rather than one uniform process.
The second major component of the study is explainable machine learning. Machine-learning systems can detect complex relationships among large numbers of variables, but conventional models often operate as “black boxes,” producing predictions without making their reasoning clear. Explainable machine learning addresses this problem by estimating how strongly individual factors contribute to a model’s output. This can help researchers identify which characteristics are most closely associated with membership in a particular cognitive subtype and communicate the results in a way that clinicians and public-health specialists can evaluate.
The study’s focus on risk networks adds another layer to the analysis. Cognitive health is shaped by interacting biological, psychological, social and behavioural influences, rather than by one isolated cause. Education, physical activity, chronic disease, sleep, mood, social engagement and living conditions may combine in ways that are difficult to capture with traditional one-factor-at-a-time models. A risk network represents these relationships as connected elements, allowing researchers to examine whether certain factors appear central within a pattern of vulnerability or whether several weaker influences reinforce one another.
For China’s rapidly ageing population, this perspective could have particular significance. Ageing-related cognitive impairment creates challenges not only for individuals and families but also for healthcare systems and communities. Cultural expectations, regional differences, access to medical care, educational opportunities and patterns of family support can all influence cognitive outcomes. Research grounded in Chinese older-adult populations may therefore reveal relationships that are overlooked when evidence is drawn mainly from European or North American cohorts.
The potential value of the framework lies in its ability to move prevention toward a more individualized model. If distinct cognitive profiles can be reliably recognized, interventions might eventually be matched to the needs of each group. Someone whose risk pattern is dominated by vascular health may require a different strategy from someone whose profile is more closely associated with depression, inactivity or social isolation. Such applications remain dependent on clinical validation, longitudinal evidence and careful testing, but the analytical approach offers a route toward more targeted screening and support.
The study also illustrates a broader transformation in ageing research: the shift from asking whether a person is cognitively impaired to asking how cognitive difficulties are organized and what combination of factors may be driving them. By pairing a model that discovers hidden subgroups with one that explains predictive relationships, the researchers aim to make complex population data more interpretable. The findings could help generate hypotheses for future studies, although machine-learning associations should not automatically be treated as proof that one factor directly causes cognitive decline.
As artificial intelligence becomes more common in health research, transparency will remain essential. Explainable models can support that goal, but their conclusions still depend on the quality of the underlying data, the variables included, the representativeness of the sample and the statistical assumptions used. The work by Lai, Dong and Fu highlights the promise of combining advanced analytics with geriatric research while underscoring the need for independent replication. If validated, this type of risk mapping could help turn the enormous complexity of cognitive ageing into clearer, more actionable signals for prevention.
Subject of Research: Cognitive subtypes and associated risk networks among Chinese older adults
Article Title: Identification of cognitive subtypes and risk networks in Chinese older adults: based on latent profile analysis and explainable machine learning
Article References: Lai, L., Dong, B. & Fu, C. “Identification of cognitive subtypes and risk networks in Chinese older adults: based on latent profile analysis and explainable machine learning.” BMC Geriatrics (2026). https://doi.org/10.1186/s12877-026-08079-1
Image Credits: AI Generated
DOI: 10.1186/s12877-026-08079-1
Keywords: cognitive ageing, older adults, China, cognitive subtypes, latent profile analysis, explainable machine learning, risk networks, dementia prevention
