Statement of principal findings
In England, although the PAF of modifiable risk factors for dementia showed a downward trend across 2004–2019, this temporal trend was slight and insignificant, with a substantial proportion of dementia cases (38%) still attributable, assuming causal relationships, to MRFs in 2018/2019. Hypertension has been the primary MRF driving the dementia burden, driving average 8.21% of cases of dementia during 2004–2019, followed by obesity (6.16%), social isolation (5.61%), hearing loss (4.81%), depression (4.72%), low education (4.63%), physical inactivity (3.26%), diabetes mellitus (2.49%), smoking (2.0%), excessive alcohol consumption (1.16%), air pollution (0.42%), and TBI (0.26%). However, only IW-PAFs of social isolation, low education, and smoking showed significant decreasing trends, while IW-PAFs of TBI, diabetes mellitus, and air pollution exhibited significant increasing trends. In contrast, the IW-PAFs of the remaining factors did not display any statistically significant changes over the study period. When disaggregated by sex, women have continuously higher PAF of MRFs than men, mainly attributed to later-life factors, notably social isolation, depression, and physical inactivity. Additionally, hearing loss, categorized as an early- to mid-life factor, also partially contributed to the observed sex disparity. Similar divergence was seen when examining PAF by SES, with people from low-income groups having continued higher PAF than those who are richer, also primarily linked to later-life factors, including social isolation, physical inactivity, depression, and smoking. Early- and mid-life factors, specifically low education and obesity, were also found to contribute moderately to the SES-based divergence in dementia risk. A critical examination of the temporal PAF or IW-PAF trends stratified by sex and SES suggest that the identified gaps across sex or SES have remained unchanged or increased over time.
Interpretation
The finding that ~ 40% of dementia cases can be attributed to 12 MRFs as of 2018/2019 is in line with prior studies, such as the extensive study of the Lancet Commission [10]. Our study adds to previous work by describing the temporal trend of the PAFs, and suggesting that previous attempts to address MRFs have been at best only marginally effective and also sheds light on the direction of future preventative strategies. The persistence of MRFs in influencing incident cases of dementia may stem from various factors. First, some MRFs such as obesity are closely tied to behavioral environments, which are often entrenched and challenging to change at the population level [10]. Additionally, societal changes such as an ageing population and urbanization might indirectly exacerbate some risk factors such as air pollution or social isolation [43]. Nevertheless, the small downward trend observed in our study provides some optimism that changing risk factors is possible. This is supported by previous studies, which showed reductions in risk factors like blood pressure and smoking [27, 28]. It suggests that dementia prevention initiatives, such as those aimed at enhancing awareness and early detection of risk factors of dementia may have a positive but yet insufficient, unevenly distributed, or inadequately sustained effect. In addition, when our results are viewed in a global context, the UK’s overall PAF for dementia MRFs (around 40%) falls in the middle-upper range of the spectrum—from Mozambique’s 24% to Australia’s 48% [6,7,8,9, 11,12,13,14,15,16,17,18,19,20]. This variation primarily reflect the disparate populations and the prevalence of risk factors across these regions. Nevertheless, the UK’s middle-upper position further supports the possibility of improvement.
Our findings on IW-PAFs of 12 MRFs further provide crucial insights to guide prevention efforts. Notably, like other countries [8, 17], metabolic risk factors, like hypertension and obesity, were the top factors in England. Indeed, hypertension alone accounted for a striking average 8.67% of cases of dementia throughout our study period in England. These findings underscore the urgent necessity for healthcare professionals and public health policymakers to maintain, if not amplify, their focus on preventative strategies—from diet and lifestyle modifications to appropriate antihypertensive therapies [44, 45]. Evidence from earlier work, which registered a decrease in dementia incidence over the past three decades partly due to efficient cardiovascular risk management [46], lends weight to the potential efficacy of this approach. Its cost-effectiveness was also validated by a UK-based modelling study [44] and addressing risk factors such as hypertension would have positive health benefits beyond dementia, for example in decreasing the incidence of heart disease or stroke.
However, it is also important to consider the potential impact of decreasing mortality in old age as a result of advancements in the prevention of dementia, particularly concerning hypertension. A recent simulation modelling study by Chen et al. (2023) explored the effects of changes in future hypertension prevalence on mortality, dementia, and disability simultaneously in England and Wales [47]. The study found that if the downward hypertension prevalence trend accelerates, with prevalence falling by 50% between 2017 and 2060, there would be a modest reduction in deaths and a small increase in dementia burden. This suggests that the beneficial effect of lower population blood pressure distribution on the incidence of dementia might not offset the expansion of the susceptible population due to reduced mortality. Therefore, while targeting hypertension remains crucial for dementia prevention, policymakers should also consider the potential trade-offs and plan accordingly to ensure adequate resources and support for an ageing population with potentially increased dementia prevalence.
The decrease in IW-PAFs of social isolation, low education, and smoking likely reflects the successes of public health interventions or national policy (for example in education) in these areas, yet their persistent contribution to the dementia burden emphasizes the need for sustained efforts. However, the upward trend or stagnation of other IW-PAFs is worrisome and diverges from the desire to decrease dementia risk via better management of modifiable factors [10, 24], especially given that cost-effective interventions have been identified for some factors, such as hearing aids for hearing loss [48]. This may suggest possible challenges, such as late detection, inadequate management of these conditions, or broader societal and environmental changes impacting these risk factors. Studies specifically examining each risk factor are needed.
Our findings on IW-PAFs of 12 MRFs, including their temporal trends and corresponding differences between early- and mid-life factors and later-life factors, underscore the importance of collective-level interventions using a life course approach. Such interventions that reduce population-level exposure to risks across the lifespan could more effectively mitigate an individual’s likelihood of developing dementia and related conditions, compared to interventions targeting sole risk factors [49]. For instance, age-related hearing loss demonstrates cumulative risk patterns beginning in early life. Similarly, hypertension risk correlates with behaviors like excessive salt intake and physical inactivity starting in youth that tracks into older age. In addition to the pharmaceutical treatments aforementioned, implementing upstream social and structural interventions at the population level that address modifiable dementia risk factors would likely have more impact than downstream individual-level interventions focused narrowly on diet and lifestyle changes alone [50].
The disaggregation of our findings by sex highlighted that there is more potential to prevent dementia cases in women than men at the moment. Similar sex disparity was also identified in Chile [17]. In our study, later-life factors were the primary drivers of sex disparity. Specifically, females demonstrated a higher overall MRF PAF, predominantly driven by social isolation, depression, and physical inactivity – later-life factors that are often intricately intertwined with gender roles, societal expectations, and ageing, but which are all modifiable. Females, particularly in their later years, maybe more susceptible to social isolation due to factors such as widowhood or caring for family members, which aligns with the findings of a recent study by Santini et al. (2020) that pointed to a higher prevalence of loneliness in older women [51]. Additionally, a systematic review by Guthold et al. (2018) showed a greater risk of physical inactivity among women [52], which is concerning considering our results. Similarly, a contribution from hearing loss, an early- to mid-life factor, to the gender disparity observed, supports the findings of the Lancet Commission [10], which highlighted the impact of hearing loss on increased dementia risk among women. Therefore, efforts to improve social support networks and mental health services, to encourage physical activity, and to identify hearing loss early and treat it, may need to be tailored differently for men and women, emphasizing more on women in these campaigns.
Our analyses by the SES quintile revealed that the burden of modifiable dementia risk is shouldered disproportionately by low-income groups, suggesting a higher potential for preventative measures within this demographic. This socio-economic divergence in dementia risk can also be attributed predominantly to later-life factors such as social isolation, physical inactivity, depression, and smoking. Early- and mid-life factors like low education and obesity also contributed, albeit moderately, to this SES disparity. This finding aligns with existing literature indicating a strong link between socio-economic status and health outcomes, including dementia risk [53]. What sets our study apart is that we present a detailed picture of the individual contributions of each risk factor to SES disparity, thereby informing priorities for intervention. Specifically, collective-level interventions that address education, social isolation, depression, and lifestyle-related factors like obesity, physical inactivity, and smoking, may require greater focus on lower-income groups.
Our study further extends the findings on sex and SES disparities by providing a temporal perspective. Crucially, the persistence or even widening of these disparities across sex and SES over time, as indicated by our temporal PAF or IW-PAF trends, warrants urgent attention as suggested above. The disparities we observed are consistent with those reported in studies on disability-free life expectancy in the UK [54, 55]. While the disparity itself is not a new discovery, our study uniquely contributes evidence specific to dementia. This stagnation or widening of the gaps contrasts with the fundamental principles of public health interventions, which envision an equitable and fair healthcare system. Notably, our findings revealed a diverging pattern between the lowest and 4th wealth quintiles, with a notable decline observed for the 4th quintile but not for the highest quintile. Our further by risk analysis indicted that compared to the highest quintile, the 4th quintile had a relatively higher decline in IW-PAF for low education, obesity, hearing loss, social isolation, depression, and smoking. This finding suggests that individuals in the 4th quintile may have benefited more from public health interventions targeting these risk factors than those in the highest quintile. It is possible that the 4th quintile had a higher initial prevalence of these risk factors, allowing for a greater margin of improvement. Additionally, public health interventions may have been more effective in reaching and influencing individuals in the 4th quintile, possibly due to factors such as health literacy and access to community-based programs, which can vary across socioeconomic groups [56]. We suggest future research examines the effectiveness of existing interventions and seeks innovative solutions to address these persistent disparities, with a particular focus on the gaps identified by our IW-PAF metrics. Whilst intervention for some risk factors may be difficult or may be delayed in impact for others (e.g. education), for some (e.g. hypertension) there are established mechanisms for identification and treatment that should be achievable within current healthcare settings. Our work presented here underscores the need for determined, consistent, up-to-date and targeted strategies to tackle these risk factors. Our work also stresses the importance of enhancing prevention efforts directed at women and individuals in low-income groups. These will be an essential part of ameliorating the dramatically increasing personal, economic and social costs arising from dementia.
Strengths and limitations
In addition to being the most up-to-date examination of this issue in the UK, our study brings sub-group and longitudinal views on modifiable risk factors for dementia, delivering several significant advances. Firstly, this study, to the best of our knowledge, is the first to map out the temporal trend of the MRF PAF for dementia in England from 2004–2019, hence providing a nuanced understanding of the evolution of dementia risk factors over a 15-year period. This longitudinal perspective not only offers a comprehensive view of the changing patterns of dementia risk factors but also delivers an evaluation of the effectiveness of past public health interventions. Secondly, our research extends the analysis by examining the role of sex and SES in dementia risk. We examine not just the overall disparity but also the differential contributions of each risk factor to this disparity. This sex- and SES-specific analysis allows us to explore underlying socioeconomic and gender issues tied to dementia risk, providing a more comprehensive understanding of the inequities in dementia incidence. Thirdly, the data and methodologies employed are robust. The RRs for each risk factor, sourced from recent meta-analyses, gather up the most compelling evidence currently available [10, 24]. Although these RRs do not adjust mutually for all other risk factors, we have ensured accounting for the non-independence of these factors via communality weights, a method known for its conservative accuracy in estimating the combined PAF. And unlike other studies, we have sourced all analysis components, including prevalence, communalities, and overall PAF, from a single information source, thereby maintaining high internal consistency, a factor not usually seen in other studies.
Several limitations need to be considered when interpreting our findings. First, the use of PAF posits a theoretical scenario where dementia risk can be wholly eliminated by removing risk factors. While entirely eradicating these factors is unfeasible, any reduction should theoretically forestall or prevent the onset of dementia, thereby decreasing prevalence. However, the PAF model does not account for potential increases in dementia prevalence due to longevity resulting from risk factor reduction [8]. Thus, our PAF-derived estimates of prevalence reduction might be overestimated. Nevertheless, it is also possible that those who live longer due to mitigating the twelve risk factors are less likely to develop dementia, so the age-related prevalence is decreased. A trend of diminishing dementia incidence has been observed in numerous countries over recent decades [21, 22], notwithstanding the potential augmentation in risk due to the simultaneous ageing of the population within this timeframe. In addition, one universal limitation in PAF studies pertains to the dichotomous presentation of relative risk data, as opposed to a continuous association between the magnitude of the risk factor and dementia risk [23]. Another universal limitation in PAF studies is the lack of consideration of the time lags between the measurement of risks and actual outcome in analysis. Moreover, our study could not distinguish the extent to which changes in PAF stem from shifting prevalence versus alterations in the shared variance among risk factors, given the assumption of constant relative risks over time. Addressing these three limitations will require the development of advanced statistical methods in future efforts.
Second, our study also faced limitations due to the lack of precise data on the prevalence of factors of air pollution, social isolation, TBI, and hearing loss. Direct measures of air pollution exposure and social isolation were unavailable, prompting us to employ proxies. The use of household fuel types as a proxy for air pollution, and cohabitation status for social isolation, has inherent drawbacks. For example, using home fuel types only addresses indoor air pollution, neglecting the outdoor aspect, and using cohabitation as a proxy for social isolation assumes that those who live alone have less social contact, although the increased risk of dementia in lifelong singles compared to married people [57] suggests this is reasonable. Another concern is that the ELSA did not include data on TBI, necessitating us to rely on average communality measures from other variables, as per the standard practice in dementia PAF research [10]. Moreover, TBI prevalence was estimated from the incidence rate of hospital admissions for head injuries with ICD10 codes S02, S04, S06, S07, S09, T04.0, and T06.0, which might underestimate its true prevalence, and thus potentially distort its PAF. Fortuitously, paralleling findings from other studies [6, 18], the contribution of TBI to the dementia population risk appears to be minimal; thus, any potential bias introduced due to our method of estimating TBI prevalence is unlikely to affect the practical implications of our results substantially. Moreover, the self-reported or observation-based measure of hearing loss used in our study may not align with the more stringent 25 dB threshold criterion [24], potentially leading to an underestimation of its contribution.
Third, the dataset we used is constrained by the presence of missing data. However, the robustness of our primary findings was affirmed by sensitivity analyses.
Fourth, the indicators used in our study do not encapsulate all potential hazards, such as accessibility of healthcare services [58]. The 12 MRFs used, pinpointed by the Lancet Commission for the general population, may not entirely elucidate the heightened dementia risk within socioeconomically disadvantaged demographics, as suggested by prior studies [9, 59].
Fifth, reverse causation could represent a potential confounding factor. For instance, depression could either precede or result from dementia, rather than being a causative agent [12, 14]. However, our sensitivity analyses, excluding participants with possible dementia during follow-up, supported our primary analyses.
Sixth, participants had the opportunity to engage in multiple iterations of the ELSA, it is important to note that, despite ELSA providing sampling weights to ensure representativeness in each wave, the representational efficacy of this approach may not be as robust as that achieved through a newly conducted cross-sectional survey. As indicated in Sup Table 2, it appears that the sampling weights over-adjusted the population, resulting in a younger adjusted demographic. This could potentially lead to an underestimation of the PAF. Nevertheless, this utilization is acceptable given that a recent study successfully projected the number of people with dementia in England up to 2040 based on ELSA data [33]. However, it is crucial to acknowledge the potential for survival bias in our study, which may lead to an underestimation of the PAF, particularly among individuals with lower SES. Considering the typically higher mortality rates observed among those with lower SES [55], as well as the fact that some of the MRFs we focused on (such as physical inactivity and smoking) were more concentrated in this subgroup [60], there is a possibility that participants with more risk factors were more likely to die during the study period. This could result in an underestimation of the corresponding prevalence and PAF disparities.
Seventh, the identification of probable dementia was based on a 25-point cognitive scale and a threshold of 1.5 standard deviations (SDs). Although this scale and cut-off have been extensively employed in studies within the United Kingdom [33, 61, 62], their validation is primarily documented in the United States [63, 64]. Notwithstanding this geographic specificity in validation, it is noteworthy that both our principal analysis and sensitivity analysis converged on the same conclusion.
Finally, our study was based on data collected prior to the COVID-19 pandemic, a condition with increased mortality and poor outcomes in older adults and those with cognitive impairment. How this might impact risk factors and incidence of dementia is unknown and not covered in the data reported here.