Dementia Prevalence in a Rural Region of South Africa: A Cross-Sectional Community Study

INTRODUCTION

Alzheimer’s disease (AD) and other dementias are a growing public health concern globally. The World Alzheimer Report 2016 estimated that there are approximately 46.8 million people living with dementia worldwide [1]. A greater percentage (58%) of people with dementia are currently living in low- and middle- income countries (LAMIC) and this is expected to rise to 71% by 2050 [2]. This increase is driven by population growth and demographic aging. By contrast, there are new reports of a decline in the incidence of dementia in Europe, possibly due to better management of modifiable lifestyle risk factors [3]. In the US, the prevalence has dropped from 11.6% in 2000 to 8.8% in 2012, the drop being possibly related to increased levels of education [4].

The 10/66 Group reported a four-fold variation in the prevalence of dementia from 2.1% to 8.5% for those aged 60 years or older in 21 Global Burden of Disease regions across the world [5]. A recent meta-analysis confirmed these estimates, with most between 5% and 7% [6]. Rates were highest in Latin America (8.5%) and lowest in four Sub-Saharan African (SSA) regions (2–4%). The reasons for the low prevalences are unclear, but they may be due to the different methods and screening tools used in the studies, or to different biological or environmental risk factors. It may be that milder dementia is under-detected in LAMIC because of low awareness, high levels of support routinely provided to older people, and reluctance to report social and occupational failings to outsiders [7]. However, a more recent SSA study reported a prevalence of 6.4% [8]. A systematic analysis of 11 studies in SSA revealed a broad range of prevalence estimates from 0% to 10.1% [9]. Prevalence was generally higher in urban compared with rural areas.

In South Africa, there has been very limited research with prevalence data published from a few small study samples. A survey of 200 participants conducted in an older urban black community in the Bloemfontein area reported a 6% prevalence of dementia [10]. A small patient-based study of mixed race persons over the age of 65 years [11] reported 8.6% of patients with possible dementia. The recent World Alzheimer’s Report [1] estimated that there were 4.4 million people over the age of 60 years living in South Africa, with approximately 187,000 living with dementia. This number is predicted to rise to 250,000 by 2030, with a concurrent increase in the number of older people (60 years plus) to 7 million.

With increasing numbers of older people, there is a corresponding need to accurately determine not only the burden of dementia in the country but also the resources that are needed to support and care for the affected persons, their families, and carers. This study not only looked at prevalence in this area –the focus of this paper –but also piloted a program of community care and support for affected people and their families. The latter will be the subject of another paper.

Measurement instruments used in epidemiological dementia prevalence studies in SSA have commonly included the Community Screening Instrument for Dementia (CSID) [5], with and without additional measures such as the Five Words Test, animal fluency, 10 word delayed recall test [12], Stick Design Test, and Blessed Dementia Scale with clinical criteria. Other screening protocols included the 10 item semi-structured home interview (CHIF) [13], the MMSE, and the Tygerberg Cognitive Battery [14]. Screening with subsequent diagnosis has been performed as either a one- or two-phase process. The one-phase process requires clinicians to apply diagnostic criteria to all people screened, while the two-phase process only applies diagnostic criteria to those who screen positive in phase one. This 2-phase process is seen to be more affordable and efficient in countries where staff resources and finances are low [8]. The DSM-IV criteria have commonly been used to confirm dementia diagnoses in prevalence studies [15].

The 10/66 Dementia Research Group has reported and validated algorithms [16] for diagnosis using a screening protocol containing the CSID, validated for use in LAMIC worldwide. The CSID, with both a cognitive screen and informant interview to assess change in functional status from normal, is reported to be reliable for dementia detection and suitable for use by non-clinicians. However, the 10/66 screening process generally produces higher prevalence estimates than the clinically applied DSM-IV criteria, possibly due to the detection of early dementia and mild cognitive impairment [17]. A depression questionnaire, EURO-D was shown to be useful in discriminating depression from dementia [18].

We aimed to conduct a large screening study to estimate the prevalence of dementia in a low income rural population in South Africa.

METHODS

A cross-sectional, population survey on dementia prevalence for isiXhosa speaking people 60 years and over was conducted using a brief version of the CSID [19] with participants and their informants. The study was approved by the Faculty of Health Sciences Human Research Ethics Committee, University of Cape Town and the Department of Health, Eastern Cape. Informed consent was obtained from each participant and informant.

This study was conducted through collaboration between three organizations: 1) The University of Cape Town research group with research skills provided the survey tools, data collection training, and statistical expertise; 2) Dementia SA, a non-governmental organization (NGO) with expertise in dementia provided dementia awareness training and resources for the study; 3) The Donald Woods Foundation (DWF), a local NGO which runs a community healthcare program called Health In Every Hut, which is working to promote health in the catchment areas. DWF have expert knowledge about the area, with Community Health Outreach Workers (CHOWs) who assisted with conducting the study.

RESULTS

Demographics

From the study population screened (n = 1,394), there were 1,382 useable results. The participants included 948 (68.6%) females. The mean age±standard deviation of the total sample was 71.3±8.3 y. Little or no education was reported by 69.8% (level 1), 15.3% had some primary school, 9.4% completed primary school (7 years), 5.1% completed high school (12 years) and 0.4% (n = 6) had post-school education (level 5). Of the men, 37.4% had some formal education, while 26.8% of women had some education.

Screening was terminated before the full study population of 2,000 persons could be screened in the funded timeframe due to strikes in the region, potential participants not being home for visits and other unforeseen difficulties that affected working capacity. Of the 606/2,000 people not screened, 197 were male. There were 7 refusals or ineligible participants. There was no intentional selection bias for screening, as shown by topographical maps (Supplementary Figure 1).

Fig.1

Dementia-positive percentage by age group in the study population.

The distribution of cogscores, informant and total scores was tabulated by education level (1–5) (Table 1). This revealed a significant effect of education on cogscore (p < 0.0001) and on total score (p = 0.0001), but not on the informant score (p = 0.08). There was no difference in cogscore between men and women (p = 0.55).

Table 1

Score distribution by education level; p-values obtained from Kruskal-Wallis tests

Variable	Levels	n	Min	Q1	Median	Mean	Q3	Max	sd	IQR
Cogscore	1	965	0	7.0	8.0	7.7	9.0	9	1.5	2.0
	2	211	3	8.0	8.0	8.1	9.0	9	1.1	1.0
	3	130	1	8.0	9.0	8.1	9.0	9	1.4	1.0
	4	70	4	8.0	9.0	8.3	9.0	9	1.0	1.0
	5	6	6	8.2	9.0	8.3	9.0	9	1.2	0.8
p < 0.0001	All	1,382	0	7.0	8.0	7.8	9.0	9	1.4	2.0
Inforscore	1	964	0	0	1.0	1.3	2.0	6	1.5	2.0
	2	211	0	0	1.0	1.0	1.0	6	1.3	1.0
	3	130	0	0	0.0	1.0	2.0	6	1.4	2.0
	4	69	0	0	0.0	0.9	1.0	5	1.3	1.0
	5	6	0	0	0.5	1.5	1.8	6	2.3	1.8
p < 0.08	all	1,380	0	0	1.0	1.2	2.0	6	1.5	2.0
Totalscore	1	965	–6	5.0	7.0	6.5	8.0	9	2.6	3.0
	2	211	–2	6.0	8.0	7.1	9.0	9	2.0	3.0
	3	130	–2	6.0	8.0	7.0	9.0	9	2.4	3.0
	4	70	0	6.0	8.0	7.4	9.0	9	1.9	3.0
	5	6	0	7.2	8.0	6.8	8.8	9	3.4	1.5
p < 0.0001	all	1,382	–6	6.0	7.0	6.7	8.0	9	2.5	2.0

Q, quartile; sd, standard deviation; IQR, χ-adjusted mean. Level 1: Little or no education; 2: some primary school; 3: completed primary school; 4: completed high school; 5: post-school education.

DISCUSSION

This is the first large dementia prevalence study reported in South Africa. It was conducted in a low income, rural Xhosa-speaking population where knowledge and awareness of dementia is poor. However, the collaboration between the three organizations was a great advantage to achieving the screening.

The prevalence estimates, determined with a simple brief screening tool, are considerably higher than the estimated 4% in the World Alzheimer Report, 2016 [1]. However, they are not too different from those of small studies in South Africa for persons 60 years and over, including rates of 8.3% in a small mixed ancestry patient sample in Cape Town [11], 6% of 250 Black African community participants in the Free State, [10] and 7.9% in a heterogeneous sample of 140 residents in five care homes in KwaZulu Natal [22]. The two prevalence estimates we report of 8% and 12% for those ≥60 years, and 11% for those ≥65 years need consideration. Although these rates may not be nationally representative, based on our lowest rate, the number of people 60-plus estimated to be living with dementia in South Africa would be 352,000, rather than the 187,000 reported by Alzheimer’s Disease International [1].

The discrepancy between the two different prevalence estimates is difficult to explain, especially as it falls away for those 65 years plus. It may relate to the higher number of participants in the 60–64 age range in our study compared with the data used for the Prince study [19]. However, our age distribution was similar to that of the WHO’s worldwide standard distribution [23]. While very good sensitivity and specificity values for cases and non-cases were achieved, the PPVs were lower. This means that in an unknown population, a proportion of dementia cases could go undetected with the brief CSID. However, with 100% NPV, the likelihood of cognitively normal people screening positive is very low. This indicates that a screening tool using a set cut-off, even including an informant score, is typically not sufficient for diagnosis.

In the first method, where PPV was lowest, it may be that the cut-off point needs adjusting to the local population according to age, education level, or another factor, such as depression score. In the second method, excluding those participants with cogscores above 6 from being classified as dementia-positive regardless of their informant score, left only 68 participants with a true negative classification compared with 121 true negatives derived from the sensitivity/specificity formula (first method).

The full CSID has been shown to be more sensitive to mild dementia than DSM-IV diagnosis [17]. This sensitivity could also occur with the brief version we used, thus increasing the prevalence with the second method. Although both prevalence calculations were based on data analysis from the same study with confirmed clinical diagnoses [19], we consider that the published sensitivity/specificity values appear more likely to give an accurate estimate than the two cut-off scores (second method). However, as the prevalences converged at 11% for those 65 years and over, there is a certain amount of confidence in the second method as well.

It could be argued that the informant report on a person’s functional state is a better estimate of dementia status than a cognitive test score, which is likely to be influenced by various factors at the time of assessment. This may be especially true when the cognitive test is too brief to assess all the cognitive domains affected by dementia. The study informants reported most problems for screen-positives with ‘ability to think and reason’ and ‘remembering what they did the day before’ and least for ‘difficulties with dressing’. Thus, the results are typical of what one expects with dementia. Possibly, many of our participants had only early dementia or mild cognitive impairment. Difficulties with dressing are more likely with moderate to severe dementia. Informant reports may also have been influenced by cultural expectations of older person’s functional abilities [22] or by the age of the informant. Perceptions of difficulty with activities of daily living may be reported as worse by younger informants, but as normal by older informants who may see these difficulties purely as age-related change. This would need further analysis.

The biggest risk factor for dementia is older age. However, lower life expectancy in South Africa might be expected to reduce the prevalence. The risk of dementia due to increased age is complex. It is subject to various contributing factors across the life-course that may differ between those living in high and low-income regions [22].

Another important risk factor may be the exceedingly low number of older people having received formal education. The cognitive reserve and more informed lifestyle choices resulting from education that includes literacy and numeracy are thought to be protective against cognitive decline. We found an initial trend for increasing education level to be negatively associated with dementia. However, when age group and education level were entered in a logistic regression, the association of dementia with education level was no longer evident. Thus, age was the over-riding factor. In general, older people in the study received less formal education than the younger participants. The brief CSID retains the culture-fair properties of the full CSID. It is not skewed with high educationally rated items. No reading, writing, or drawing is required. Items in our study with the lowest scores were typically the 3-word delayed recall and naming the day of the week.

Without the full clinical picture, some dementia positive screens may have been due to reasons other than cognitive impairment. We included an abbreviated depression scale, which should help to discriminate dementia from depression. However, depression may be either a risk factor for, or a consequence of, memory loss [24]. We found that the screen positives had higher self-reports of depressive symptoms than the screen negatives. The prevalence estimate may have been increased by the high numbers of screen positives reporting depression, thus some participants could have depressive pseudo-dementia It is also possible that depressive symptoms were confused with memory problems by some informants [18, 25]. This is an aspect we will explore further when we have more data.

This study was primarily aimed at determining the prevalence of dementia in a low-income rural isiXhosa-speaking community. Together with the additional component of the study aimed at piloting care and support for individuals with dementia and their families, our findings will hopefully raise enough concern to promote public health policies for improving access to services and interventions.

The question of the primary causes or types of dementia has not been addressed here but is the subject of further investigations. Other studies of causes of dementia in LMICs have reported a greater prevalence of vascular dementia than Alzheimer’s disease. This is thought to relate to risk factors such as untreated hypertension, diabetes, hypercholesterolemia, and stroke [26]. These factors may contribute to the dementia prevalence in our sample, but this needs further study Other risk factors for dementia that may be pertinent to investigate in this population would include the ApoE4 genotype, known to increase the risk of Alzheimer’s disease in North America and Europe, but with reported protective effects in Nigeria [27]. Poor nutrition in our study region and reliance on staple, unfortified food, such as maize, may contribute to low levels of vitamin B12, and consequently, high blood levels of homocysteine, which is associated with increased cognitive impairment and Alzheimer’s disease. HIV-positive status in older people also confers a risk for dementia [28]. The prevalence of HIV-associated dementia among those presenting at HIV clinics has been reported to range from 15–30% [29, 30]. It is not known what the contribution of HIV is to the overall prevalence of dementia, especially in older people [28]. Once the primary causes of the different types of dementia have been identified for this population, risk-focused initiatives should be implemented to lower the dementia risk in the future.

A limitation of our study was the lack of clinician resources to provide a DSM-IV diagnosis of dementia. However, the use of recommended sensitivity and specificity values and cut-off scores on the brief CSID helped to estimate the likely number of true positives. Another limitation of our study is that we used the brief CSID rather than the full validated CSID. The 6 informant questions in the brief version were mostly related to cognitive impairment rather than basic functional impairment. Thus, informant reports may have had limited utility for corroborating dementia classification. Screening tests need to be brief for large studies, and this was very advantageous in allowing us to screen a substantial population sample. Although we didn’t achieve the targeted sample size due to funding constraints, the prevalence calculations are still robust. Although the results are not generalizable to the whole South African population, they are probably representative of this large homogeneous, low income, rural community.