The need for a more holistic approach to dementia prevention

Public health action is needed to address the projected 2·7 times increase in global dementia prevalence in the coming decades.¹ This action should include targeting modifiable risk factors for dementia prevention.² An important measure of the preventative potential of risk factor reduction is the population attributable fraction (PAF), which estimates the proportion of dementia cases that could be prevented if a given exposure was eliminated in a population. The PAF incorporates the risk associated with the factor and its prevalence, which vary across populations and contexts.

In The Lancet Health Longevity, Blossom C M Stephan and colleagues³ report a systematic review and meta-analysis, in which they estimated global dementia PAFs for modifiable risk factors. Overall, they estimated that 32·0% to 55·0% of prevalent dementia cases were collectively attributable to seven modifiable factors (ie, low education, hypertension, obesity, smoking, physical inactivity, depression, and diabetes). Importantly, the authors included non-English language studies, which reduces potential reporting bias. Consistent with, and extending the 2020 Lancet Commission findings,² the study by Stephan and colleagues reaffirms the importance of addressing modifiable risk factors for reducing dementia risk. It highlights the vital role that clinicians, public health officials, and policy makers have in promoting and prioritising primordial prevention; for example, through lifestyle modification and social determinants.

The authors provide one of the most comprehensive summaries of the dementia PAF literature to date. However, we believe that important limitations exist with regard to common approaches to calculating and combining PAFs, which should be addressed to advance knowledge in this area. We argue for a more holistic approach for PAF calculations that better incorporates key population characteristics of person, place, and time. In terms of person, we need a different approach to address how dementia risk factors cluster in individuals, as many risk factors frequently co-occur. For example, up to 60% of people with diabetes have comorbid hypertension,⁴ and vascular risk factor clustering increases with age.⁵ Importantly, dementia risk increases with increasing vascular risk burden.^2,5 Current widely used methods⁶ to account for risk factor co-occurrence in the PAF, including those reported by Stephan and colleagues,³ systematically underestimate the combined effects of risk factors.⁷ This calculation is problematic as most older adults often have multiple risk factors. Therefore, the proportion of preventable dementia cases is likely to often be underestimated.^2,3 Future research is needed to identify the best methods to combine PAF estimates in longitudinal settings. We also need tutorials with sample statistical code, preferably from individuals in ageing research with expertise in epidemiological methods, to help researchers easily implement these methods.

With respect to place, because risk factor prevalence is a component of the PAF calculation, and prevalence will vary by location, the PAF is most relevant for public health planning when it is localised to the region in which the intervention will occur. We therefore urge researchers to report regional PAFs (eg, at city, state, division, and province levels) in addition to aggregate (eg, country and global) estimates when possible. Finally, regarding time, Stephan and colleagues rightly indicate that dementia prevention efforts require a life course approach. Both components of the PAF—risk factor prevalence and effect on dementia risk—vary over the life course. For example, hypertension is more common in later life (age ≥65 years) than midlife (age 45–64 years), but dementia is more strongly associated with hypertension in midlife than later life.⁸ We urge researchers to clearly define their study population for PAF calculations; if the goal is to estimate the number of dementia cases that could be prevented by targeting hypertension in a population of older adults, then risk factor prevalence estimates, and their associations with dementia (eg, relative risk), should be restricted to the appropriate age group. We note that the use of aggregate data from multiple large data sources to calculate PAFs is becoming increasingly common; researchers should be aware that inaccurate use of prevalence estimates and pooled relative risks that do not directly map to the age of the target population (eg, applying a risk factor prevalence estimate for adults aged ≥18 years and pooled relative risk from studies in adults aged 45–80 years to a PAF calculation) could conservatively or non-conservatively bias estimates, depending on the risk factor. Furthermore, nuance in interpretating and translating these age-dependent effects into real-world settings is key, particularly if successful risk factor prevention efforts in midlife (45–64 years) could paradoxically increase dementia incidence in later life.⁹

The findings reported by Stephan and colleagues emphasise the considerable value that targeting modifiable risk factors has for reducing the risk of dementia in older adults. However, to advance this field, we urge researchers to consider person (effect of co-occurring risk factors), place (reporting PAF estimates specific to the regions in which interventions are likely to be deployed), and time (a life course approach) when calculating and reporting PAF. This holistic approach will ultimately improve the alignment of PAF analyses and messaging of outcomes with the theory and intent¹⁰ of population-level approaches for the primary prevention of dementia.