Dementia risk is often presented as a checklist: blood pressure, hearing, smoking, exercise, education. Useful, yes. Complete, no. A checklist centred on individual behaviour can miss the conditions that shape those behaviours—and the services available to change them.
A 2025 eBioMedicine Personal View pushes that point into the model itself. Its authors argue that poverty, wealth shocks, income inequality and HIV should sit alongside the 14 modifiable factors in the 2024 Lancet Commission framework. They also want sex, gender and regional context to influence how the framework is interpreted. [1,2][1,2]
What the paper actually does
This is not a trial, a prospective cohort or a new systematic review. The authors assemble observational studies and global indicators, compare them with the Commission’s framework and propose a broader model. Their argument is strongest as a critique of scope: evidence built largely in high-income settings should not automatically be treated as a universal map of dementia risk. [1][1]
The proposed additions are not interchangeable. HIV has biological and treatment-related pathways that matter for brain health. Poverty and income inequality influence education, nutrition, chronic stress, healthcare access and exposure to other risks. A wealth shock is an event, not a stable trait. Putting them in one framework is useful only if the model keeps those differences visible. [1][1]
Why the broader frame matters
For clinicians and service providers, social context is not background decoration. It can determine whether a person can afford transport, return for follow-up, use a hearing aid or manage vascular risk. For researchers, it can change who enters a study and whether a model travels beyond the population in which it was built. For entrepreneurs, it is a reminder that a prediction tool can look precise while measuring only the people and variables easiest to reach.
That does not mean poverty should become another box in an individual risk score by default. Structural exposures operate through several pathways and overlap with existing factors. A useful model must show whether a new variable adds information, changes an intervention or simply gives a new name to risk already counted elsewhere.
The 65% figure is a proposal, not a result
The paper suggests that adding the four factors could raise the potentially preventable share of dementia from about 45% to about 65%. That is the headline number. It is also the part that needs the most restraint. The paper does not report an intervention that prevented 65% of cases, nor does it present a newly validated global model that cleanly separates overlapping risks. [1][1]
The authors acknowledge the problem: these exposures are multifactorial, comorbid and difficult to assign unique contributions to. They call for causal inference, longitudinal modelling and regional calibration. An earlier Mendelian-randomisation analysis also found limited genetic support for causal effects across ten Commission risk factors, while carrying important limits of its own, including European-ancestry data and survivor bias. The sensible response is triangulation, not choosing one method as the final referee. [1,3][1,3]
Gender needs more than a count
The paper notes that eight of the Commission’s 14 factors are more prevalent in men, despite women carrying more dementia burden in many settings. After adding the proposed factors, it counts ten of 18 as more prevalent in women. This raises a valuable question. But counting which risks are more common is not the same as estimating how much dementia each one causes, how strongly it acts or how risks interact across the life course. [1][1]
A better gender analysis would connect exposure, effect size, access to prevention, competing risks and diagnostic patterns. It should also separate biological sex from gendered social conditions instead of asking one label to carry both.
What should happen next
The next model should earn its complexity. It needs diverse longitudinal datasets, explicit treatment of overlapping pathways, external validation by region and clear tests of whether the extra variables improve decisions. Services can already ask better questions about financial strain, care access and HIV status where relevant. They should not convert those answers into a precise dementia forecast until the model has been tested for that use.
The paper’s best contribution is not a bigger percentage. It is the insistence that dementia risk does not stop at the clinic door. The research task now is to turn that insight into models people can trust—and interventions that change outcomes rather than merely describe disadvantage.
