The economic burden of dementia in low- and middle-income countries (LMICs): a systematic review

1Previous review and studies on the economic burden of dementia are heavily based on high-income countries, while a larger proportion of people with dementia lives in low- and middle-income countries (LMICs).

2No reviews on the economic burden of dementia or determinants of the cost have been published before.

3This is the first systematic review focused on the economic burden of dementia in LMICs.

4Our review established that in LMICs, indirect cost is the major contributor to the high economic burden of dementia. The costs increase as the severity and number of comorbidities increases.

5Our study findings emphasise that the studies from LMICs faced methodological challenges, especially in the recruitment of participants, standardisation of measured costs items, and indirect cost components.

6Based on our study findings, we recommend that future research studies on the economic burden of dementia in LMICs use standardised methods to measure the direct and indirect costs.

7Inclusion of reduced working time data from patient and caregiver and informal caregiving time outside the reduced work time without double counting in the indirect costing calculation are encouraged.

8Nationally representative samples of dementia patients, including community-dwelling and institutionalised people, are required to allow for comparisons between countries.

9The health systems in LMICs should focus on the slowdown of dementia progression and control of comorbid conditions to reduce the direct cost of dementia in the long run.

10Assistance for caregivers of people living with dementia in these regions should be enhanced to reduce the burden of caregiving, thus, the indirect cost.

11Dementia is a syndrome causing deterioration in memory, thinking, behaviour and the physical functioning. It is estimated that 50 million people are living with dementia globally. More than two-thirds of them were in low- and middle-income countries (LMICs). In 2015, the estimated global economic burden of dementia was US$818 billion or 1.09% of aggregated global gross domestic product (GDP). This figure is projected to reach US$2 trillion by 2030. However, these estimates are heavily based on studies from high-income countries (HICs). Indeed, only 10% of the studies in this estimation were from LMICs. As such, current cost of global dementia may not be representative of the LMICs.

12Levels of mental functioning and dependency deteriorate as dementia progresses to a severe stage. Therefore, the cost will vary with disease severity. Existing reviews have shown differences in the direct medical, direct non-medical and indirect costs associated with dementia. Studies from HICs have reported that the indirect costs are higher than direct costs. The main drivers of dementia costs in HICs are home based care and nursing costs. There is lack of such studies from LMICs, but the costs are expected to be different from HIC due to differences in services, infrastructure and cultural perception of ageing and dementia (eg, disease vs not a disease). A thorough review of the dementia costs and its contributors in LMICs is also necessary to inform stakeholders to plan the healthcare and social care delivery for people affected by dementia in these countries.

13Given the urgency for increased efforts to improve outcomes of dementia in LMICs, the importance of good-quality economic data is unquestionable. Also, identifying the challenges in estimating the dementia costs specific to LMICs will be crucial to develop methodological recommendations to improve the quality of economic data from these settings. This review aimed to systematically review existing evidence of dementia costs in LMICs and to conduct methodological assessment of the included studies.

14The systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). A comprehensive search in six electronic databases (EconLit, EMBASE, PubMed, Cochrane Review (DARE), ERIC and PsycINFO) was performed. The search strategy was based on three broad search strings (‘cost of illness’) AND (‘Dementia’ OR ‘AD*’) AND (LMICs). The complete search strategy can be found in online supplemental material pages 1-6. Potentially relevant publications were retrieved from the databases from inception to September 2020, without language restriction. Additional literature was identified by snowballing the reference list of the eligible studies and via a grey literature search of government and international agencies reports.

15We included studies based on three criteria: original research article, reported the economic burden for all-cause dementia and/or its types, and the study setting was in LMICs. The search included 138 LMICs according to the World Bank in 2020. Quantitative studies which reported any cost of dementia, including direct medical, direct non-medical and indirect costs were included. We did not assess the intangible cost as it was challenging to quantify their monetary value. Studies on economic evaluation of specific dementia interventions, editorials, animal studies, reviews, case studies and case series were excluded. Selection criteria were not limited to any specific language, thus minimising language bias.

16Two reviewers (SMM and AMM) independently screened the title and abstract, followed by full-text screening. Disagreements were resolved by consensus or by consulting a third reviewer (DM). The PRISMA flow chart illustrating the screening process is shown in figure 1. For studies in which the price year was not reported, the publication year was used. One reviewer (SMM) collected data from each report, and the research team confirmed data validity.

17Data on author(s), publication year, type of dementia, study design, aims, participants (sample size, mean age, gender), economic components, data sources, cost unit(s), estimates of total costs, currency, comorbidities reported, price year and key findings were extracted. The risk of study bias was assessed using the Larg and Moss method as shown in online supplemental table 1.

18Data were synthesised under the following domains- study characteristics, key findings, studies’ methodological quality and estimation of dementia costs. All costs were inflated from the reported year to 2019 values using country-specific Consumer Price Index (CPI) data from the World Bank and subsequently converted to US dollars (US$) according to the recommendations of Turner 2019. The most updated data were used in the final cost summary measures for studies estimating cost from the same data source at different time points.

19To further facilitate comparison of dementia costs at the national level across countries, we estimated the cost of dementia, as a percentage of the country’s GDP in 2019. Unweighted means and means weighted by population (to account for the differences in population size) were calculated. Data on population size, CPI, exchange rate and GDP were obtained from the World Bank website and, if unavailable, were taken from the International Monetary Fund and other appropriate available sources listed in the online supplemental table 2.

20Patients were not involved in the design or conduct of this systematic review.

21We identified 14 095 studies after removing duplicates. After title and abstract screening, 196 references were selected for full-paper screening. Of these, 172 articles were excluded (as shown in figure 1) based on the inclusion and exclusion criteria, including two near-missed articles. One of the near misses was an editorial with a brief estimation of Pakistan’s dementia cost using data from India in 10/66 Dementia study. The second one was a letter to the editor with an estimation of dementia cost in Nigeria; both of the articles were not published elsewhere. In total, 22 study articles and 2 published theses were selected for inclusion.

22Dementia costs from 122 LMICs (out of 138) were reported within the included studies as shown in figure 2. Table 1 summarises the key characteristics of the 24 included studies. The studies were categorised based on the cost estimation sources. The first group includes 16 studies using collected original data, the second group consists of 6 studies using prevalence estimates of dementia, and the third group includes two studies using a combination of original data and prevalence estimates. The first group reported patient-level cost of dementia, except in two studies which showed both patient-level and total national expenditure estimate, and one study showed dementia cost as per episode of care. In contrast, all the second and third groups of studies, showed patient level and total national expenditure estimate, except one study that measured change in net present value per person and predicted China’s dementia burden from 2011 to 2050.

23Out of all studies, 7 studies reported combination of LMICs (five studies were worldwide economic burden from World AD Report (WAR) released by the AD’s Association and 2 from 10/66 Dementia Group study), 16 studies were from upper-middle-income countries (UMICs) and 1 study from lower-middle-income countries (LRMICs). The studies from WAR are the most comprehensive cost estimation report of the worldwide cost of dementia, but contribution from LMICs countries’ original data within these reports is lacking; thus costs imputations were done for the countries with no data. Fourteen studies assessed the cost of all-cause dementia, while eight studies assessed the cost of AD and two studies assessed the cost of cause-specific dementia.

24Most studies (n=18) costed dementia from a societal perspective, while four studies used a healthcare payer perspective, and two studies used a healthcare provider perspective. The most common study design was cross-sectional (n=16), followed by five studies with worldwide cost estimation report from WAR, two longitudinal, and one simulation modelling (not shown in table 1). Study data were frequently sourced from the patient’s medical records and/or patient/carer interviews (n=16), followed by national prevalence based estimates (n=6), insurance claim data (n=1), and a combination of medical records and cost assumption based on dementia governing body (n=1). Although comorbidities are major issues among for older dementia patients, only eight studies assessed comorbidities either in descriptively or as a determinant of dementia costs.

25Of the 18 studies that collected patient-level primary data, 14 studies recruited both patients and informal caregivers; however, 2 studies did not specify the caregiver information, and two studies did not include caregivers (due to data from hospital medical records or insurer database). The study participants in all the 18 studies were aged 60-year-old and above and with any severity of dementia except for one study that excluded severe dementia. Most studies (n=11) included patients from rural and urban locations. Four studies with patient-level data had recruited control samples.

26The full assessment of the risk of bias is shown in the online supplemental table 1. In general, most studies exhibited a low risk of bias in the analytical framework and methodology sections, while moderate risk was observed in the analysis and reporting section of the Larg and Moss framework.

27In the analytical framework component, all studies exhibited low risk except for two studies in which identification of the non-trivial cost components were limited. Zencir et al did not report hospitalisation and formal care costs but were acknowledged in the study limitation section. Meanwhile, Mould-Quevedo et al did not report the direct non-medical cost and indirect cost in monetary value even though it is a societal perspective study and misclassified paid caregiver under the indirect cost category.

28Low risk of bias was observed in the methodology and data section except in three areas. First, most studies (n=22) did not measure additional or excess costs due to dementia. Second, the data sampling by nine studies was not representative of the study population stated in the aim. Out of 18 studies conducted using primary data, only four studies obtained their data through random sampling, while the remaining studies used convenient sampling. However, all 18 studies include all dementia severity levels, except one study in which severe dementia patients were not represented. Third, two studies included cost components unrelated or not specific to dementia cost (ie, ancillary study).

29A moderate risk of bias was identified in the analysis and reporting section. Twelve out of 24 studies did not provide the range of cost estimates. Ten studies did not report any sensitivity analysis. Ten studies reported sensitivity analysis on the important parameter estimates and key assumptions, while four studies performed sensitivity analysis only on the important parameter estimates. None of the studies performed a sensitivity analysis on the point estimates.

30Two methodological criteria were assessed in detail: quality of healthcare resources and indirect cost estimation. The quality of healthcare resources evaluation and quantification were done appropriately in most studies. However, variations in the individual cost components were observed as shown in online supplemental tables 3-5.

31All studies reported direct medical cost, but only 19 listed the individual cost components. The studies from WAR did not separately report the direct medical and direct non-medical cost components. Important direct cost components that were not reported in some studies were medications (n=7), hospital stays (n=3), and outpatients visits cost (n=2)

32One of the major challenges regarding healthcare quantification of direct cost is whether the data were collected from private or public health facilities. Ten studies collected data from private (n=2) or public (n=2) or both (n=4), including two studies which quantified private and public direct cost separately. These two studies showed patients who went to private healthcare had a higher cost than those who went to public healthcare providers.

33Direct non-medical cost was included in 21 studies. Sixteen studies reported individual components including transportation (n=11), paid caregivers (n=9), nursing home or institutionalisation (n=6), meals or special diet (n=5), special equipment (n=4), and nappy (n=2)

34Indirect costs components included productivity loss (income loss) of the caregiver and/or patient and informal caregiver time and were assessed in 20 studies. Most studies utilised an opportunity costing method (n=12) compared with a replacement cost method (n=5) to estimate caregiving burden. One study used the human capital approach to value productivity loss, and two studies did not report their indirect costing method.

35Out of 20 studies assessing indirect cost, caregiving time was assessed in 15 studies, productivity work loss of informal caregiver was included in 10 studies, and productivity work loss of patient was included in 3 studies. Rarely, health services for caregiver treatment were included (n=3).

36The economic burden of dementia was presented in the included studies in various ways. While all studies reported the cost of dementia as absolute cost, two studies included the attributable cost of dementia. The cost of dementia was also presented using the catastrophic health expenditure in one study. The dementia costs estimation were reported as per person/per capita expenditure (n= 13), both total national expenditure and per person costs (n=9), per episode of care, or change in net present value per person and predicted dementia cost value year 2011–2015.

37The cost of dementia reported is summarised below, separately as patient level and national level costs, and derivative of absolute cost.

38Of the 22 studies which reported cost of dementia per patient, 5 studies provided per capita cost of all-cause dementia, 9 studies reported absolute cost of dementia by severity level, 1 study presented both absolute and attributable costs of dementia by severity, and 1 study provided the attributable cost of dementia by severity. Two studies estimated per patient cause-specific dementia costs. The WAR studies estimated per capita costs at the regional level.

39The most recent predicted annual total costs of dementia per patient in 2015, in UMICs is US$10 467, in LRMICs is US$3865 and in low-income countries (LICs) is US$939. Table 2 shows the annual absolute and attributable cost of dementia per patient by severity (n=11 studies), inflated to the year 2019. The total annual absolute per capita cost ranges from US$590.78 (mild) to US$25 510.66 (severe). On average, the indirect cost is the biggest contributor to dementia burden compared with direct cost; however, both cost categories show wide variation. The absolute annual direct cost ranges from US$439.23 (mild) to US$6193.22 (severe). Meanwhile, the indirect cost ranges from US$0 (mild) to US$7428.87 (severe). The total attributable cost ranges from US$0.03 (mild) to US$15.704.82 (severe).

40All studies except one showed the costs increase with disease severity, as shown in table 2. We did not extrapolate the cost of patient-level data to national estimates due to a high risk of overestimation or underestimation due to non-representativeness and the small size of samples.

41Eleven studies with costs of dementia per patient were not included in table 2. Two studies that reported the cost per patient by three common types of dementia showed conflicting results. Frontotemporal dementia (FTD) patients had the highest cost compared with AD and vascular dementia (VaD) in Peru; however, in Argentina, VaDpatients had the highest annual direct cost compared with AD and FTD. There was significant variation in total costs of dementia of the same country as per the results of five studies from China. The total annual costs of dementia in China as inflated to the year 2019 ranged from US$1332.95 to US$52 163.37 per patient, with median costs of US$7458.36.

42Nine out of 24 studies reported the total national expenditure or world regional classifications of dementia costs. We calculated the annual cost of dementia as the proportion of the country’s GDP for each country inflated to 2019 as shown in online supplemental table 6. We excluded four studies due to the lack of country-specific estimates and the availability of the most recent study from China. The estimated annual total national costs of dementia ranged from US$1.04 million in Vanuatu to US$195 billion in China. The highest total cost in percentage of GDP was 4.114% in Liberia, and the lowest was 0.001% in Venezuela and Western Sahara.

43Table 3 summarises the dementia cost as percentage of GDP for all LMIC groups. The mean total cost of dementia as a proportion of GDP in LMICs was 0.45%. On average, the indirect cost was about 58% of the total costs of dementia. The total costs of dementia as percentage of country’s GDP ranged from 0.35% in LRMICs to 0.46% in LICs. LICs had the lowest direct (0.15%) and the highest indirect cost (0.44%) of dementia. LICs also had the highest total costs percentage of GDP (0.46%), followed by UMICs (0.43%) and LRMICs (0.35%). The most updated total cost of dementia reported in LMICs was US$148.2 billion in 2017, with the highest cost coming from UMICs. The total LMIC cost inflated to 2019 was US$264.8 billion.

44Two studies have reported the extent of catastrophic health expenditure caused by dementia. A study in Peru outlined that the monthly cost of a dementia patient is 2.5 times greater than the current legal minimum wage. Meanwhile, it was found only 0.5% of dementia patients in Malaysia faced an out-of-pocket cost exceeding the 40% threshold for health expenses due to almost total subsidisation of government healthcare.

45Only one study reported the mean direct medical cost of dementia per episode of care, which was MYR10 034 (SD: MYR7604) (approximately US$2414 (SD: US$1830) in year 2017. Severe cases had the highest direct medical cost per episode of care compared with moderate and mild dementia. Meanwhile, one study showed the national level cost of dementia as the average change in the net present value of GDP of China (US$253.40 per patient in 2015), and the annual predicted value of dementia costs exceeds US$1 trillion by 2050.

46Twenty studies measured both direct and indirect costs of dementia, and 13 studies showed the driver of total costs is indirect cost ranging from 42% (direct non-medical cost had the same share) to 94% of the total costs. Nationally represented data in table 3 also shows that indirect cost is commonly greater than the direct cost in LMICs.

47Various factors were identified as the determinants of dementia costs in LMICs. Determinants of greater costs across studies that collected patient-level data are higher dementia severity; n=6, greater number of comorbidities (n=4), prescription medication (n=3), longer treatment and hospitalisation (n=2), older age (n=1), having a paid caregiver (n=3), an older informal caregiver (n=1), and lower education level (n=1). Studies using national dementia prevalence or nationally representative data also highlighted greater prevalence of dementia (n=3), increase in population ageing (n=1) and hospital location outside the capital state (n=1) as the determining factors of higher dementia costs at the national level.

48This is the first study to systematically review studies of dementia costs in LMICs, providing a comprehensive understanding of its economic burden. We estimated that the total costs of dementia in LMICs as percentage of GDP is 0.45% in 2019, as compared with 1.4% in HIC in 2017. The patient-level data showed the annual total costs could be as little as US$590.78 (for mild dementia) to about US$25 510.66 (for severe dementia) per patient. This wide variation of economic burden may be due to the varied number of samples and cost components assessed. Indeed, many of the included studies included patients with greater access to healthcare, with moderately measured indirect costs and assumptions of data sources.

49The estimated costs of dementia in LMICs are lower than HIC. The lower cost of dementia in LMICs compared with HICs could be partly due to the differences in the social determinants of health, like the poorer access to clinical care and dementia medications, and lack of knowledge of these services. Most LMICs are under- resourced to provide specialist and hospitalisation services to dementia patients. Family caregivers offer most of the support even for advanced dementia patients. These factors could contribute to the lower total cost and a higher proportion of indirect cost in LMICs. However, we cannot prove this as the available data from the articles in the review do not provide any direct evidence to draw this conclusion.

50The annual cost of dementia among different countries at the patient level varied widely. The wide range of estimates may be due to the differences in the cost prices, number of samples and specific items being measured in the cost calculation. Although China has multiple economic burden of dementia studies, the costs reported widely varied across studies. This may be due to the inclusion and exclusion of cost components and the method of estimating indirect and direct costs.

51Only two studies have assessed the attributable cost of dementia, due to the absence of control sample cost. Even so, one study managed to include additional expense of dementia associated with the progression of disease using multivariate regression analyses with mild patients as the reference category (no controls in this study). It is difficult to differentiate the real attributable cost of dementia from other health condition. Comparing cost between cohorts with and without dementia is the best alternative to measure attributable cost.

52We re-estimated the burden of dementia as a percentage of GDP in 2019 using previous studies reports and found our estimates are similar to the average total costs of dementia as a percentage of GDP in LMICs reported in the most recent WAR 2017 (ie, 0.36% of GDP in LMICs). However, this is expected because only a handful of studies in LMICs provide nationally representative cost data, and our reported figure is heavily based on the previous year WAR estimates. LMICs spend less on dementia compared with the major non-communicable diseases, namely cancer, diabetes, respiratory and cardiovascular diseases which are estimated to be about 4% of the GDP.

53The main determinants of dementia costs in LMICs were slightly different from the HICs. Patient care setting, dementia severity and included costs categories and components are some factors outlined by studies from HICs. We also found that the most important factor is dementia severity. Stratification of costs based on dementia severity is important and has been shown as the driver of dementia cost in previous studies. Besides, informal care cost increase 18% per year as dementia progress. Our finding is also consistent with the previous studies, which employed various diagnosing tools and slightly different cut-off points. Comparison between studies could be better if homogeneous cut-off points and dementia severity diagnosis tool were used. The second most mentioned factor is the number of comorbidities. Cost increases as the number of comorbidities increases. Studies in HICs show that dementia patients with comorbidities are more likely to be hospitalised, have longer hospital stays, and spend more on comorbidities than people without dementia.

54The cost of dementia studies in this review largely represents the UMICs group. Lack of nationally representative data from LMICs is a concern, despite most studies having minimal risk of bias and detailed analyses. Most of these studies use convenient sampling or non-random participant recruitment from a single healthcare centre or dementia society. These patients tend to have advanced dementia, their caregivers typically report higher strain levels, and the families have greater access and use more healthcare and community support services. Limited financial and human capacity is the most common barrier that lead to challenges in conducting clinical studies in developing countries. In LMICs, convenient sampling may cause significant bias in the cost estimation of dementia, as patients have inequitable access to healthcare. Therefore, the results shown may have low generalisability to broader dementia patients within the same country.

55Indirect costs are often challenging to be estimated. One of the reasons is dementia is prevalent in older adults and mostly retirees; therefore, productivity loss (loss of income due to unpaid leave/absenteeism, early retirement, or death) of the patient is less often described in the cost of illness studies on dementia. Our review found that only a few studies measured productivity loss of a caregiver but with no justification for the exclusion of patient’s productivity loss. Indirect costing was based only on informal caregiving time, which may have underestimated the magnitude of caregivers’ and patients' loss of resources. It is important that future studies justify their decision in the methodology if productivity loss of either patient or caregiver is excluded.

56The indirect cost due to the poorer health of caregivers and lost wages associated with caregiving are often under-reported in these studies. Provision for the training of professional and family carers is a key strategy to reduce the negative outcomes associated with caregiving. However, the biggest challenge would be to mobilise resources to cater to a large number of carers with the limited resources available. In this context, digital health interventions to provide distant training, many of which have been proven effective in the COVID-19 pandemic, can be adapted to increase the reach of carer training in dementia.

57The impact of caregiving on informal caregivers’ health and its associated cost is often overlooked. Thus, data of indirect cost from health deterioration due to caregiving even from HICs are unavailable for cost modelling. A hidden cost component associated with dementia caregiving is the long-term cost of deteriorating health and the resulting impact on the health and productivity of the carers. There should be standardised ways of capturing this cost in dementia costing studies. Caregiving time often been highlighted as the major contributor to the cost of dementia, regardless of the economic capacity of the country. In HICs such as the USA, a combination of informal care and paid formal care had the highest contribution to total dementia costs. In LMICs, indirect cost from informal caregiving is the major contributor to the total cost burden of dementia compared with direct cost. Better estimation and inclusion of indirect cost parameters may increase the actual dementia costs estimate in LMICs.

58The strength of this study is that this is the first systematic review of dementia costs from LMICs regardless of the publication year and language. The methodological aspect of the included studies was also evaluated, providing recommendations for future research studies. There are some limitations to our study. Although we wanted to include as many studies as possible from LMICs, only 15 countries had original published costs data that were included in the review. We inflated the cost of dementia at the national level to the year 2019 and showed them as a percentage of GDP without accounting for current dementia prevalence, as there is limited data on the current prevalence of dementia in each country. This may have resulted in an underestimation of cost. All included studies had participants who were at least 60 years old, implying a lack of data on costs associated with early-onset dementias in LMICs. Lastly, most studies used data from UMICs. This may limit the generalisability of the summarised data.

59Dementia is associated with a significant cost in LMICs. However, is difficult to draw accurate conclusions on the actual economic burden due to the lack of available data and standardised cost assessment methods. Future studies on dementia costs needed to be standardised and inclusive of all important cost items. The stakeholder in LMICs should focus on the delivery of holistic primary healthcare to slowdown dementia progression and prevention of comorbidities, that may reduce the direct cost of dementia in the long run. Besides, LMICs should enhance the assistance offered to caregivers of people living with dementia to reduce the burden of caregiving and thereby the indirect cost.