Incidence of Dementia and Associated Factors in the EPIC-Spain Dementia Cohort

INTRODUCTION

Dementia is a broad term that encompasses several mental disorders caused by brain disease or injury affecting memory, cognitive function, and behavioral and personality traits [1]. The most frequent type of dementia is Alzheimer’s disease (AD), which accounts for 60–70% of the overall cases [1]. Dementia is an age-related syndrome but it is not part of the normal aging process, and since there is no effective treatment to slow down its progression or ameliorate its symptoms [2], prevention remains as the most efficient strategy to reduce the burden of the disease. Several modifiable factors have been identified, such as education and cardiovascular risk factors [3], which seem to be related to the etiopathogenesis of AD and other dementia sub-types and are amenable to intervention.

The World Health Organization has recognized dementia as a public health priority, raising awareness of the dimension of the disease as a leading cause of death and disability and calling for larger research and innovation efforts. Near 50 million patients were estimated to have dementia in 2015 (overall prevalence in Europe is 7.1%) [4], but as the world population is steadily rising and aging, this figure is foreseen to more than double by 2050 [1]. Still, data on the incidence of dementia may not be accurate worldwide, as local (national or regional) estimates are lacking in many parts of the world. In Spain, there are no nationally-representative estimates on dementia incidence and the studies available show variability in their results, with overall incidence rates ranging from 7.6 to 13.1 per 1000 person-years (py) at 75–80 years, and from 12.6 to 25.4 per 1000 py in the 80–85 years-old group [5–7]. Literature on temporal trends is also heterogeneous, although some studies suggest a decline or stabilization of dementia incidence [8, 9].

In the absence of population-based dementia registries, cohort studies can be used to estimate the incidence of dementia. Although most cohorts have not been designed to be representative of the general population (which limits their external validity), when large enough they can be conveniently used to provide study-specific incidence estimates by population sub-groups, providing insightful data on incidence patterns and filling gaps in epidemiological knowledge of the disease. In addition, cohort studies are best suited for quantifying the magnitude of the association of potential risk factors, making possible to assess the dimension of the disease and to characterize their socio-demographic, lifestyle, and clinical correlates in a common study setting [10]. EPIC (European Prospective Investigation into Cancer and Nutrition)-Spain is a multi-center cohort study with a large sample size, long follow-up time, and geographical variability, which provides a convenient framework for obtaining incidence estimates of dementia and related factors. The purpose of this study was to assess the risk of dementia in the EPIC-Spain Dementia Cohort by sub-type, age group, and sex, and to assess how this risk varies according to selected variables.

METHODS

RESULTS

After a mean of 21.5 years, a total of 537,198 py were available for analysis. Within this period, 774 dementia cases were ascertained, 518 of which were validated as AD, and 256 as non-AD dementia cases. The median follow-up time of cases was 18.3 years.

Baseline characteristics of the sample are shown in Table 1. Dementia cases were more likely to be women, obese, to have a primary educational level or lower, and to suffer chronic diseases such as diabetes, hypertension, or hyperlipidemia. On the contrary, they were less likely to be alcohol consumers. Table 2 shows the incidence rates of dementia by sex and age groups. As shown, incidence rates increased exponentially with age for overall dementia and subtypes, and AD was the first cause of dementia representing 66.9% of all cases. Global incidence rates were higher in women for dementia and AD, while lower for non-AD (versus men). Overall rates of dementia and AD in people aged 65 and above were 4.2 (95% CI: 3.9–4.5) and 2.8 (7.0–9.4) per 1000 py respectively, while adjusted rates (ASRE) were 8.2 (7.0–9.4) and 5.5 (4.5–6.4) per 1000 py. As illustrated in Fig. 2, the models showed their steepest increase from the age of 75 onwards. Age-adjusted incidence rates of dementia were lower in Gipuzkoa (ASRE 65+6.9 (5.6–8.2) per 1000 py) and higher in Navarra (10.6 (6.2–14.9) per 1000 py) (Table 3), but there was not a clear geographical trend in dementia incidence, and differences by center were not statistically significant in multivariate models. Table 4 presents the cumulative incidence of dementia and AD in women and men for quinquennial age groups in the 50-65 age band. Noteworthy, 10.1% of the participants aged 60–65-years developed dementia in the following 20 years, whereas 7.5% developed AD.

OPEN IN VIEWER

Fig.1

Flow-chart of the dementia case ascertainment process in the EPIC-Spain Dementia Cohort.

OPEN IN VIEWER

Fig.2

Royston-Parmar model-based incidence rates and cumulative incidence (incidence proportion) of dementia, Alzheimer’s disease, and non-Alzheimer’s dementia, by sex, in the EPIC-Spain Dementia cohort (N = 25,015). Estimates based on a restricted cubic spline parametrization of the baseline hazard function with 3 degrees of freedom.

The results of the multivariate analysis showed that educational level and diabetes were the strongest factors associated to dementia and its sub-types. Risk of overall dementia or dementia sub-types was 29–37% lower among participants with higher education as compared to those with primary studies, whereas HR among participants with diabetes mellitus (versus non-diabetics) was 1.38 (95% CI: 1.10–1.72) for dementia, and 1.86 (1.32–2.62) for the non-AD subtype. Furthermore, hyperlipidemia was also significantly associated to AD (HR = 1.11, 1.01–1.21). On the other hand, alcohol consumers showed a 19% lower risk of overall dementia risk (CI: 4% to 32%) (Table 5), whereas no significant associations were found for other lifestyle variables, cardiovascular risk factors, or chronic diseases (myocardial infarction, stroke, or cancer) in multivariate models.

DISCUSSION

The main contribution of this work is the estimation of incidence rates of dementia according to sub-type, age, and sex in a large Spanish cohort. Our study is the first to contribute data on dementia incidence in areas from the north (Gipuzkoa and Navarra) and the south of Spain (Murcia), and one of the few studies available in the Mediterranean region. The estimated overall age-adjusted incidence rate of dementia in the EPIC-Spain dementia cohort was 6.7 to 8.2 cases per 1000 py (depending on the population standard). By sex, rates were higher in women, especially in younger age groups, and became comparable above 80 years old. Together with previous data, these results suggest that incidence of dementia in Spain is moderate as compared to other studies [18].

It is of concern that despite the importance of dementia as a public health problem worldwide, the epidemiology of the disease is not fully characterized, even in developed countries. While recent research has focused on the evolution of dementia trends, point estimates of its incidence are scarce, with most of the available data dating back one or two decades ago [19]. Assessing the incidence of dementia in different geographical areas and population groups is important for establishing the magnitude of the disease burden and detecting temporal trends to inform effective prevention policies.

Several studies have enriched the knowledge on dementia incidence in the last decades, most of them carried out in North America (such as the Framingham study, the MoVIES Project, the Mayo Clinic Study of Aging, or the Rochester Epidemiology Project) [9, 20–22], or Europe (MRC-CFAS, PAQUID, Rotterdam, or Lundby Studies) [23–26]. Unfortunately, only a few cover Southern European countries such as Spain [5–7, 27–29], and not in a geographically comprehensive manner. Furthermore, comparisons of overall rates across studies are difficult since dementia is a condition heavily determined by age and differences in the age structure of different studies could have a large impact in epidemiological estimates such as incidence or prevalence. The only one prior study in Spain which reported age-adjusted incidence rates of dementia, estimated an ASR of 10.6 cases per 1000 py, which is slightly higher than our results [6]. The fact that they had a significant proportion of participants in the oldest age groups (85–89 and ≥90 years) could partially account for their higher incidence. We had a small number of person-years in the oldest age band (≥85 years), and we admit this may limit the accuracy of our estimates in this group. We cannot discard that the lower incidence observed in our cohort could also be partially attributable to differences in the prevalence of risk factors or to methodological heterogeneity across studies. Of note, the cohort size and length of follow-up in our cohort study were much larger than those reported for other studies in our geographical area. Besides, our multi-center study includes data from three geographically dispersed populations within the country, which strengthens their external validity. While we did not perform repeated cognitive assessments of the participants (impractical in large cohort settings), case ascertainment of dementia was based on the revision of all medical records available for each potential case, also reducing the outcome misclassification bias inherent to single cognitive assessment tools. Some studies, such as the Rochester Project [22], used a similar case ascertainment approach, yet the number of participants and length of follow-up were much lower. It its worth considering the possibility that rates of dementia incidence could have changed throughout the two decades of follow-up. Some authors have suggested that incidence of dementia would be decreasing in developed countries [9], and this could partially account for a lower incidence rate in our cohort, since incidence of dementia would be decreasing in parallel with the ageing of the cohort. While such a possibility cannot be discarded, a recent meta-analysis suggested that the potential reduction in dementia incidence would be, if any, of small magnitude and not statistically significant, mainly due to the high variability across studies [8].

The cumulative incidence of dementia at different ages and time spans is an important estimator for evaluating the burden of the disease and for health planning and resource allocation. However, few data have been published on this matter so far. Estimates on lifetime risk of dementia around 15–25% for people aged 65 years have been published in the Framingham [30] or Rotterdam [31] studies, and by Lobo et al. in Spain [5]. Together, results highlight the exponential burden of dementia up to advanced ages, even when estimates are usually based on model projections. The long follow-up in our study allowed us to use a different approach, i.e., to estimate the cumulative proportion of dementia and AD cases, by sex, observed at 10, 15, and 20 years. Observed results showed that 12.5% of women and 9.1% of men 60–65 years-old developed dementia within a 20-year time frame, which again points out to a lower burden of dementia in our country. Of note, while a larger proportion of women suffered dementia overall, men were more likely to be diagnosed with non-AD at older ages (Fig. 2).

Education and diabetes were the major modifiable factors associated with dementia risk in our study. A lower educational level was associated to an approximately 30–35% higher risk of dementia, AD or non-AD (p < 0.01). A higher educational attainment contributes to the cognitive reserve of the brain (i.e., its ability to cope with the age-related decline in neurological functions), which has been widely recognized to prevent the onset and severity of dementia [32]. Our results thus agree with previous evidence showing the major effect of education on dementia risk. Of note, this association was independent of cardiovascular risk factors, such as diabetes mellitus and hyperlipidemia, which are known determinants of dementia and likely play a major role in the etiopathogenesis of the disease [3]. While our results confirmed the independent associations for these factors, other well-known determinants of the disease, such as obesity or hypertension [33] were not found to be independently associated with dementia or its sub-types in our cohort. While there was a larger proportion of hypertensive or obese participants among the cases, these associations were largely attenuated (and no longer significant) in multivariate models. Literature results on the association of BMI and dementia are conflicting [34, 35]. Previous evidence suggests that obesity in mid-life may increase the risk of dementia through increased production of inflammatory cytokines and hormones by the adipose tissue, whereas a higher BMI seems to be inversely associated to dementia risk when measured in late-life [35]. Furthermore, obesity is an upstream risk factor predisposing to diabetes, dyslipidemia, hypertension, and cardiovascular disease [36]. The weakening of statistical significance in the association of obesity in multivariate models accounting for other mediating cardiovascular risk factors would thus be an expected finding. On the contrary, the lack of a significant association for hypertension contrasts with most previous literature and needs to be considered with caution. Elevated blood pressure is recognized as a leading cause of neurovascular dysfunction, affecting cerebrovascular structure and ultimately impairing cognition [37]. The fact that the condition was self-reported (which implies some extent of misclassification) and the inability to account for the degree of pharmacological control or severity of the disease might partially explain our null results. Finally, we found alcohol consumption to be protective against dementia, as consumers had an estimated 18% lower risk than non-consumers in adjusted models. This result persisted after defining ‘never consumers’ as the reference group to avoid the sick-quitter bias. Although such an effect has been pointed out in previous works [6], we suggest caution in the interpretation of this finding, which warrants further analysis.

A number of strengths and limitations needs to be considered. Our results come from the largest cohort in Spain, and cover geographical areas in which only prevalence data on dementia had been available to date [38], thus helping filling the gaps in the epidemiology of the disease in Southern Europe. Major strengths of the study are the large sample size and number of cases, a long follow-up, geographical variability, and the ability to study socio-demographic, lifestyle, and clinical factors.

The main limitation of our study was the lack of a baseline cognitive assessment of participants. As the study EPIC was not initially designed for the study of dementia as a primary outcome, participants did not undergo a cognitive evaluation at baseline or during follow-up, so the diagnosis was based solely on clinical records and we could not fully discard baseline conditions related to cognitive decline, such as mild cognitive impairment (MCI). However, the mean age of the cohort was 49.3 (±7.8) years at baseline, and only 142 participants (0.57%) were between 65–69 years, in whom prevalence of MCI in our area was estimated below 9.7%, whereas prevalence of dementia was below 0.3% [38]. Furthermore, most of the cases were diagnosed after a long follow-up time: the median follow-up time of cases was 18.3 years, whereas less than 6.6% of the cases were diagnosed within the first 10 years. Finally, our results do not suggest an over-estimation of dementia incidence, as it have been the case if a significant proportion of MCI or dementia cases had been included at baseline. For the above-mentioned reasons, we consider that the lack of a baseline cognitive assessment, despite a limitation, did not have a significant impact in our estimates. As another limitation, a selection bias was implicit since we could only ascertain those cases that accessed the health system, and this might have resulted in an underestimation of dementia rates in our population, particularly in older participants with cognitive decline who may have not sought for a medical diagnosis. Nevertheless, the fact that healthcare in Spain is universal and free would have reduced the extent of such a bias (less than 1% of the Spanish population have private health coverage exclusively). Also, the scarcity or participants in the oldest age band limits the accuracy of incidence estimates in this group. Another limitation is that data on lifestyles and co-morbid conditions were only evaluated at baseline, and so we were unable to assess changes in these variables throughout study time. Also, a potential left-truncation bias due to death or emigration before enrollment could not be discarded. Furthermore, as in any observational study, residual or unmeasured confounding cannot be ruled out. Finally, the lack of representativeness of the cohort would limit the generalizability of our results.