Prevalence estimates of dementia in older adults in rural Kilimanjaro 2009–2010 and 2018–2019: is there evidence of changing prevalence?

Although limited, existing epidemiological data on dementia in sub‐Saharan Africa indicate that prevalence may be increasing; contrasting with recent decreases observed in high‐income countries. We have previously reported the age‐adjusted prevalence of dementia in rural Tanzania in 2009–2010 as 6.4% (95% confidence interval [CI] 4.9–7.9) in individuals aged ≥70 years. We aimed to repeat a community‐based dementia prevalence study in the same setting to assess whether prevalence has changed.


| INTRODUCTION
Dementia is a global health priority with an estimated 47 million people affected, 58% of whom live in low and middle income countries (LMICs). 1 The prevalence of dementia varies across world regions, 2,3 and a degree of this variation is attributed to potentially modifiable risk factors. 4 Evidence from some, but not all world regions, especially highincome countries, suggests that the age-adjusted prevalence of dementia may be falling, 2,5-7 a change attributed to better management of dementia risk factors, such as hypertension.
In sub-Saharan Africa (SSA), epidemiological data on dementia have been limited and only recently have sufficient data been available to allow meta-analysis of prevalence with acceptable precision. 8 Reviews based on these data suggest that 2.1 million people are affected and that dementia prevalence may be increasing. 7,9,10 However, it is not clear to what extent this reflects refinements of research methodology and culture-specific dementia assessment.
Demographic transition and population ageing estimates indicate that the SSA population aged ≥60 years will increase from 43 million in 2010 to 67 million by 2025. 11 Alongside this, there have been rapid increases in the prevalence of vascular risk factors for dementia in much of SSA and a gap in healthcare provision, particularly for non-communicable diseases (NCDs). [12][13][14][15] In a study conducted in 2009-2010 we have reported the community prevalence of dementia in rural Tanzania to be 7.5% (6.4% age-adjusted) in individuals aged ≥70 years using a two stage door-to-door study design. 16 We aimed to repeat a community-based dementia prevalence estimate in the same setting using a similar methodology to assess whether prevalence has increased in this setting during the intervening 9 years.

| Setting
The study was conducted in the Hai demographic surveillance site (DSS), located in the Kilimanjaro region of Tanzania. Tanzania is a low-income country (Gross National Income: $1020 in 2018 17 ) where life expectancy at birth is 64 years and 3% of the population are aged 65 years and over. The Hai DSS is a well-demarcated area with a highly organised village structure facilitating epidemiological research and follow-up. There is substantial local experience in NCD research and regular census completion since 1994. 18 The main languages spoken are Kiswahili (official language of Tanzania) and Kichagga (a local tribal language) and the economy is based around subsistence farming, with cash crops, such as coffee and tomatoes, grown by those with more fertile land. Levels of illiteracy are high in older adults 16 reflecting a previous lack of available schooling, and levels of population mobility are low, with most of the population being lifelong residents of Hai. For each participant, verbal information was provided about the aims of the study and the implications of taking part. An information sheet was read aloud, and written information made available for those preferring written materials. Participants were also given the opportunity to ask questions. Consent was then obtained by signature or thumbprint, depending on literacy status. Where capacity to consent was in doubt due to cognitive impairment, assent was sought from a close relative and assessments completed if the participant appeared willing to do so. Onward referral, and access to treatment (cognitive stimulation therapy and/or carer education) for those identified as having dementia was a key element of the study protocol.

| Study design
We conducted a census-based two-stage door-to-door community prevalence study using a similar (but not identical) design to that used for our 2009-2010 study. 16 Eligible participants aged ≥60 years underwent screening for dementia in phase I and a subsequent sample, stratified for screening performance, underwent detailed clinical examination for dementia by DSM-5 criteria in phase II. In contrast to the 2010 study (where participants were aged ≥ 70 years), a cut-off of ≥60 years was selected to ensure that YOSEPH ET AL.
-951 younger-onset cases were not missed and to allow comparability with other studies.

| Recruitment and timing
Twelve villages were randomly selected from the 80 within Hai DSS, with stratification for high zone, middle zone and low zone (depending on their position on the slopes of Mount Kilimanjaro). Soil is generally more fertile in high zone villages and this may result in a slightly higher economic status due to availability of cash crops. This stratification helped ensure that the selected villages were representative of the wider DSS.

| Data collection and management
Census and Phase I cognitive screening data were collected during a single visit using a mobile digital application (app) developed in Open Data Kit (ODK) software on a hand-held tablet device. People identified as aged ≥60 years, and who consented to screening, were assigned an identification number and underwent cognitive screening in their homes or at a local health facility, according to their mobility status. As part of the census, demographic data (age, sex, occupation and education level) were collected. Determination of age is recognised to be challenging in rural SSA elders who may lack formal identification documents. Age was calculated from year of birth and triangulated using a table of historical events, alongside ages at marriage and of first child (see online materials for table used). This method is well-validated in SSA. [19][20][21] Data from the tablets were uploaded weekly to a secure, Screening was overseen and organised by senior research clinicians experienced in previous Hai DSS surveys and resident in the local area (John Kissima, Aloyce Kisoli) with one-to-one support offered for initial visits where necessary (usually for those lacking confidence in computer tablet use).

| Cognitive screening
Those who consented, were screened using the IDEA cognitive screen, previously validated in low-literacy populations in Tanzania and elsewhere in LMIC countries. [22][23][24][25] The screen was developed from the community screening instrument for dementia (CSI-D) 26 alongside a version of the CERAD 10-word list, extensively validated in LMIC settings 27 and a matchstick construction task previously validated in Nigeria. 28 A brief version of the IDEA instrumental activities of daily living questionnaire (IDEA-IADL) for Tanzania was also administered. 29,30 The brief questionnaire has three questions that are asked to an informant who knows the participant well. Each question is scored as 0 (cannot do), 1 (can do with assistance) and 2 (can do easily), giving a total possible score of six in those with no IADL problems.

| Stratification for dementia assessment
Screening performance was stratified according to the combined performance in the IDEA cognitive screen and the IDEA-IADL questionnaire in those with an informant (score 0-10: classified as impaired 'probable dementia', score 11-13:classified borderline 'possible dementia', score 14-21: classified normal 'no dementia'). 29 In those without an informant, stratification was based only on performance in the IDEA cognitive screen (score 0-7: 'probable dementia', score 8-9: 'possible dementia', score 10-15: 'no dementia') (22,24). A list of all those screening as impaired or 'probable dementia', a random sample of 50% screening as borderline or 'possible dementia' and a random sample of 10% screening as normal or 'no dementia' was produced using a random number generator. We aimed to see as many people on the list as possible. The phase II assessment team were given the list in numerical order by village, with no indication which screening category participants fell into and blind to the screening score.
The prevalence of dementia generally increased with age and was higher in females and those with lower educational level (Table 2). However, for those seen in Phase II, only age remained significantly associated with dementia in multivariable logistic regression modelling (Table 3)

| DISCUSSION
We report a prevalence of dementia of 6.1% (4.6% age-adjusted) in those aged ≥60 years and 10.2% (8.6% age-adjusted) in those aged ≥70 years in rural Kilimanjaro. This is an increase, although not a statistically significant one, from the previously reported crude prevalence estimate of 7.5% (95% CI 6.0-9.0) and 6.4% age-adjusted (95% CI 4.9-7.9) in those aged ≥70 years in 2009-2010. As in 2009-2010, age is the most significant predictor of dementia and, although dementia is associated with female sex and low/no education, these are no longer significant after controlling for age.
Our prevalence estimate is substantially higher than other rural and semi-urban studies in Nigeria 21,38 and Benin, 33 but similar to YOSEPH ET AL.
-953 those conducted in urban francophone west Africa 34 ; see Table 5. In contrast, a single-stage study conducted in South Africa, and another in rural Uganda (published but non-peer reviewed) report substantially higher prevalence (up to 20%), but used only the brief CSI-D to diagnose dementia, without confirmatory diagnosis. 39 The brief CSI-D was developed through modelling on LMIC populations but with little data from SSA, and has not been specifically validated in this setting. 40 This is therefore likely to be an over-estimate, since other conditions that could affect performance in cognitive function tests (e.g., depression) cannot easily be distinguished from dementia during a brief screen. Indeed, our own screening, using a tool validated for use in SSA identified 13.6% of those screened as probable dementia.
Similarly, other SSA dementia prevalence estimates using screening without confirmatory diagnosis report high prevalence; 10.1% in Nigeria, 41   It would have been ideal to collect data on vascular risk factors on the entire screened population, but this was felt to be of limited utility given the well-known limitations of self-report and awareness of NCDs in this setting. Similarly, on clinical assessment determination of vascular and other risk factors through blood tests and other investigations (e.g., electrocardiograph) would have been ideal, but was not feasible given logistics and available resources. We were able to measure blood pressure in the majority (79%) but not all participants, again due to resources. Similarly, other potentially emergent risk factors such as air pollution were not measured.
Available data are scarce but suggest Tanzania has lower levels of air pollution than other areas of SSA. 49 Nevertheless, this is likely to be a potentially emergent public health factor to be examined in future studies. 50 Finally, we emphasise that our findings should be extrapolated beyond the narrow setting of Hai district with caution. Hai is known to have high rates of cardiovascular risk factors such as stroke and hypertension as discussed. 12,47,51 Nevertheless, the district is similar to much of rural SSA, with traditional lifestyles predominating, food insecurity common and limited coverage of healthcare services.

| CONCLUSION
In contrast to some high-income settings, there is no evidence of a reduction in dementia prevalence in Kilimanjaro, Tanzania, since 2010 and prevalence may well be increasing. The population is generally older, with lower levels of illiteracy than seen in 2009-2010. Dementia is likely to continue to be a significant and growing issue in Tanzania and further studies on potentially-modifiable risk factors in this setting are needed.