Research evaluating the effectiveness of dementia interventions in low‐ and middle‐income countries: A systematic mapping of 340 randomised controlled trials

More people with dementia live in low‐ and middle‐income countries (LMICs) than in high‐income countries, but best‐practice care recommendations are often based on studies from high‐income countries. We aimed to map the available evidence on dementia interventions in LMICs.


| INTRODUCTION
Dementia will be one of the biggest global health challenges of the 21 st Century. Tremendous successes in public health and healthcare over recent decades have made it possible for more people to live into old age. In consequence, there are rapidly more people at risk of age-related conditions, of which dementia is one of the most challenging through its effects on disability, quality of life, and costs. 1 Moreover, dementia can also affect younger people. 2 Causing lifechanging disability and dependency for millions of people, as well as impacting on the lives of family carers, dementia is seen as a public health priority by the World Health Organization (WHO). 3 A silver bullet to target this global issue is not in sight; even if effective disease-modifying treatments are developed, their availability and affordability in low-resource settings must be in serious doubt.
Most people with dementia reside in LMICs. 4 With high prevalence of some risk factors, dementia prevalence in LMICs is expected to increase more rapidly than in high-income countries. 5,6 Need for support in everyday activities means that dementia affects not only the person with the condition but also family members and other close contacts who provide care and support. Consequently, informal care is the largest economic cost of dementia in LMICs. 4 These impacts provide a clear rationale for creating care and support systems in LMICs for people with dementia that improve the quality of life for affected individuals and their carers.
In the absence of an effective disease-modifying treatment, other interventions (pharmacological and non-pharmacological) can improve cognition and quality of life, and are recommended by WHO and in clinical practice guidelines. [7][8][9][10] However, evidence on what interventions work seems to mostly come from high-income countries. Indeed, previously proposed packages of dementia care for LMICs were largely based on evidence from high-income countries. 11,12 Findings from studies in high-income settings may not be applicable in low-resource settings with different cultures, social environments, diagnosis rates, and less well-resourced health and care systems. 3 It is unclear what dementia interventions have been studied in LMICs, and how extensively.

A comprehensive map of dementia intervention research in
LMICs is missing. We therefore systematically reviewed studies of interventions aimed at improving the lives of people with dementia or MCI and their families in LMICs.

| MATERIALS and METHODS
We conducted a systematic mapping study to describe which interventions for people with dementia or MCI and for their carers have been studied using randomised controlled trials (RCTs) in LMICs between 2008 and 2018. Studies involving people with MCI were included due to their high propensity to develop dementia. 13 This mapping study forms part of a systematic review and meta-analysis on the effectiveness of dementia interventions in LMICs, registered on PROSPERO (CRD42018106206). 14

| Search strategy and selection criteria
We searched 11 electronic peer-reviewed and grey literature data- The search syntax for each database was developed in collaboration with a library information specialist (AF) and tested for sensitivity against a set of 10 previously identified records. Searches were structured around four search blocks ("dementia"; "intervention"; "study design"; "LMIC"), combining, as available, free text with variants, controlled vocabularies, and filters. The full search strategies for four databases are available in the supplementary material (Tables S1-S4).
Database searches were conducted throughout October 2018 and updated through 9 January 2019. To complement database searches, we manually reviewed lists of studies included in 75 previous systematic reviews (Table S5) and all dementia intervention reviews indexed in the Cochrane Database of Systematic Reviews as of September 2018.
We included RCTs, including cluster-RCTs, published 2008-2018 of any intervention aiming to improve the lives of people with dementia or MCI, or their carers, in any LMIC (as defined by the Organisation for Economic Co-operation and Development at the time the RCT was conducted). 15 We followed study authors' definitions of dementia and MCI to identify eligible participants in each study. We imposed no restrictions on interventions to capture all previously studied interventions. Similarly, we did not specify any outcomes or imposed sample size restrictions. Inclusion criteria in the PICOS (Participants, Intervention, Comparator, Outcomes, Study design) format are in Table 1. Detailed exclusion criteria are in our published protocol. 14 Briefly, we excluded studies that were either not conducted in LMICs, not focussing on people with dementia or MCI or their carers (i.e., non-dementia-specific interventions or primary prevention interventions), not assessing an intervention, as well as non-original reports (reviews), or studies not published in a language spoken by a member of our global team (51 researchers covering 15 languages).
Records identified from database searches and previous reviews were collated and de-duplicated. We applied a combination of semiautomated and manual screening of abstracts and titles. 16 We first uploaded all records to the Rayyan platform for systematic reviews to obtain prediction scores for probability of a record matching inclusion criteria. 17 The prediction algorithm was trained using a set of 4000 inclusion/exclusion decisions made by hand. After assessing performance of the algorithm for 1000 records around a conservative cut-off prediction score (1.5 out of 5 stars, where 1 indicated minimal and 5 maximal relevance), all records below that score were screened by one reviewer while remaining records were screened independently by two reviewers. Full text review was performed independently by two reviewers. For full texts in languages other than English, at least one reviewer was a native speaker of that language. Conflicting decisions were arbitrated by a member of the core review team (DM, MS-K, GW, CS).
We deviated from the published review protocol by focussing on RCTs only and excluding non-randomised studies. This decision was taken after inclusion and exclusion decisions had been made for the full set of studies meeting our initial eligibility criteria, revealing an unexpectedly large number of eligible studies (approximately twothirds randomised and one-third non-randomised). Extracting information from and appraising all studies was not feasible and we therefore focussed on eligible RCTs as they would be expected to provide the most internally valid evidence.
Data extraction was performed independently by two reviewers using a standardised data extraction form. At least one of the two reviewers was a native speaker of the language of the main study paper/report. We extracted information on study characteristics (location, funding, care setting), design (including outcomes measured), participants (baseline characteristics), interventions, results, and risk of bias. Conflicting data extractions were resolved by a senior researcher (DM, MS-K, CS).
Risk of bias assessment was conducted independently by two reviewers using the Cochrane risk of bias 2.0 tool. 18 For each study, risk of bias was judged as low, moderate ("some concerns") or high in five domains according to a series of signalling questions. As recommended by Cochrane, high risk of bias for any of the domains resulted in an overall study-level "high risk" judgement. Conflicting judgements were resolved by a member of the core review team (DM, MS-K, CS). For all included studies, arbitration by a senior reviewer also served as an additional quality assurance mechanism.
Study inclusion decisions, data extraction, and risk of bias assessment were managed using Covidence systematic review software (Veritas Health Innovation, Melbourne, VIC, Australia). Researchers based in 13 countries contributed to the review. We report total number and proportion of included studies by country and category of intervention. Inclusion of multi-country trials and studies investigating more than one experimental intervention means that reported proportions may add up to >100%.

| Role of funding source
The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

| Patient and public involvement
People with dementia/MCI, their carers, or the public were not directly involved in this research. Representatives from associations advocating for people with dementia and their carers were actively involved in the research and are named as members of the STRiDE Evidence Review Group.

| RESULTS
A total of 63,585 records were identified through database searches

F I G U R E 4
Summary plot of risk of bias assessments. Plot shows the proportion of all included randomised controlled trials with "low risk of bias", "some concerns", or "high risk of bias" judgements for five risk of bias domains and the overall judgement.
research, yet we found that dementia intervention RCTs were also conducted in low-income countries. When accounting for population size, there was a trend of more trials conducted in countries with higher GDP per capita and higher share of over 65-year-olds in total population. While this may indicate some alignment between research priorities and ageing populations among countries already active in dementia research, it is notable that we did not identify a single published RCT for the vast majority of LMICs. This absence of evidence may suggest that dementia is not considered a priority in many countries.
Given the trajectory of ageing populations, it is essential to develop and test interventions that can affordably be implemented at scale. Efforts to develop effective disease-modifying treatments for dementia are apace, but even if available, there must be questions about access and affordability in many countries. For example, aducanumab, approved in the US, was initially priced at US$56,000 per patient annually. 36 Identifying effective, less costly nonpharmaceutical interventions that improve the lives of people with dementia and their carers is therefore essential.
There is a need for targeted research investment to generate robust evidence to inform future care pathways. Methodologically, RCTs are usually the most powerful tools for assessing intervention effectiveness, 37  shown in other disease areas. 38,39 Western pharmaceuticals, along with traditional medicines and supplements, also accounted for the majority of RCTs in our study. Yet, the robust evidence generation standards of RCTs need not be restricted to the approval of drugs.
Our study demonstrates that some RCT evidence already exists for non-pharmaceutical dementia interventions. This is potentially good news for the increasing number of LMICs planning to implement or already implementing national dementia plans. 40

| CONCLUSION
In conclusion, in this mapping study of dementia intervention RCTs in LMICs, we found that evidence-generation is concentrated in few countries, the body of evidence is skewed towards selected interventions, and overall it is subject to high risk of bias. Given expected demographic trajectories in LMICs, more and better studies are needed, and a more coordinated approach to evidence-generation-both within and across countries-would improve the likelihood of robust, relevant and impactful findings.