Effectiveness of interventions for improving livelihood outcomes for people with disabilities in low‐ and middle‐income countries: A systematic review

Abstract Background People with disabilities—more than a billion people worldwide—are frequently excluded from livelihood opportunities, including employment, social protection, and access to finance. Interventions are therefore needed to improve livelihood outcomes for people with disabilities, such as improving access to financial capital (e.g., social protection), human capital (e.g., health and education/training), social capital (e.g., support) or physical capital (e.g., accessible buildings). However, evidence is lacking as to which approaches should be promoted. Objectives This review examines whether interventions for people with disabilities result in improved livelihood outcomes in low‐ and middle‐income countries (LMIC): acquisition of skills for the workplace, access to the job market, employment in formal and informal sectors, income and earnings from work, access to financial services such as grants and loans, and/or access to social protection programmes. Search Methods The search, up to date as of February 2020, comprised of: (1) an electronic search of databases (MEDLINE, Embase, PsychINFO, CAB Global Health, ERIC, PubMED and CINAHL),(2) screening of all included studies in the instances where reviews were identified,(3) screening reference lists and citations of identified recent papers and reviews, and(4) An electronic search of a range of organisational websites and databases (including ILO, R4D, UNESCO and WHO) using the keyword search for unpublished grey to ensure maximum coverage of unpublished literature, and reduce the potential for publication bias Selection Criteria We included all studies which reported on impact evaluations of interventions to improve livelihood outcomes for people with disabilities in LMIC. Data Collection and Analysis We used review management software EPPI Reviewer to screen the search results. A total of 10 studies were identified as meeting the inclusion criteria. We searched for errata for our included publications and found none. Two review authors independently extracted the data from each study report, including for the confidence in study findings appraisal. Data and information were extracted regarding available characteristics of participants, intervention characteristics and control conditions, research design, sample size, risk of bias and outcomes, and results. We found that it was not possible to conduct a meta‐analysis, and generate pooled results or compare effect sizes, given the diversity of designs, methodologies, measures, and rigour across studies in this area. As such, we presented out findings narratively. Main Results Only one of the nine interventions targeted children with disabilities alone, and only two included a mix of age groups (children and adults with disabilities. Most of the interventions targeted adults with disabilities only. Most single impairment group interventions targeted people with physical impairments alone. The research designs of the studies included one randomised controlled trial, one quasi‐randomised controlled trial (a randomised, posttest only study using propensity score matching (PSM), one case‐control study with PSM, four uncontrolled before and after studies, and three posttest only studies. Our confidence in the overall findings is low to medium on the basis of our appraisal of the studies. Two studies scored medium using our assessment tool, with the remaining eight scoring low on one or more item. All the included studies reported positive impacts on livelihoods outcomes. However, outcomes varied substantially by study, as did the methods used to establish intervention impact, and the quality and reporting of findings. Authors' Conclusions The findings of this review suggest that it may be possible for a variety of programming approaches to improve livelihood outcomes of people with disabilities in LMIC. However, given low confidence in study findings related to methodological limitations in all the included studies, positive findings must be interpreted with caution. Additional rigorous evaluations of livelihoods interventions for people with disabilities in LMIC are needed.


| What studies are included?
This review includes studies that evaluate the effects of interventions on livelihood outcomes for people with disabilities in LMICs. The authors found nine interventions which used eligible study designs.
Countries represented are Bangladesh, India, Nigeria, Ethiopia, Brazil, China and Vietnam. All included studies have some important methodological weaknesses.

| What are the main findings of this review?
All included studies reported positive impacts on livelihoods outcomes. However, due to variation between studies, we did not conduct as analysis of effects across studies. As such, it is hard to draw firm conclusions about what works, for whom and how.
Most studies focused on improving access to the workplace.
For example, people without disabilities were involved in programmes to improve their social attitudes to working with people with disabilities. People with certain disabilities were provided with wheelchairs. And some people with disabilities were placed in supported employment.
Studies examined the effects of vocational training programmes, a 'motivation to work' programme, community-based rehabilitation and social skills training. All of these approaches showed positive impacts on livelihood outcomes, including finding employment and gaining social skills for work.
The included studies all reported that their programmes improved outcomes related to the livelihoods of people with disabilities, including acquisition of skills for the workplace, access to the job market, employment in formal and informal sectors, and access to the formal and informal social protection measures.
Future research should evaluate these approaches with more rigorous study designs. This would develop a firmer evidence base, which would also inform the delivery of interventions at scale.

| What do the findings of this review mean?
In general, there is not a great deal of evidence on interventions to improve livelihood outcomes for people with disabilities in LMICs, so more studies are needed. Researchers should work with organisations of persons with disabilities and other non-governmental organisations to identify priority interventions to evaluate. For instance, online and community-based delivery of livelihood interventions could be explored, to bridge gaps in coverage of programming and reach rural populations (who were underrepresented in this review).
There are other specific research gaps that need to be filled. The geographic scope of studies should be expanded. There were no studies from Europe, Central Asia, the Middle East or North Africa.
Programmes should integrate impact evaluations to improve the evidence base. Research evaluating programmes for people with disabilities other than those with physical impairments are needed.
Overall, there is a need for more and better data to inform policy and practice, including data on a broader range of impairment types.

| How up-to-date is this review?
The review authors searched for studies up to February 2020.
2 | BACKGROUND 2.1 | The problem, condition or issue The United Nations Convention on the Rights of Persons with Disability defines disability as 'long-term physical, mental, intellectual or sensory impairments which, in interaction with various barriers, may hinder [a person's] full and effective participation in society on an equal basis with others' (UN, 2006). More than one billion persons in the world have some form of disability (World Health Organization, 2011).
This figure corresponds to about 15% of the world's population.
Disability and poverty are strongly linked. On a global level, 80% of people with disabilities live in LMIC (World Health Organization, 2011). Within countries, disability disproportionately affects the most disadvantaged sector of the population . Disability is significantly associated not only with poverty, but also lower educational attainment, lower employment rates, and worse healthcare access (Mitra et al., 2013).
Consequently, scholars identify the risk of experiencing 'multidimensional poverty' (poverty across multiple domains) as extremely high in this population (Mitra et al., 2013). This relationshipbetween disability and poverty-is bidirectional, and driven by a number of factors and proposed mechanisms; for instance there are high costs associated with many of types of impairments (e.g., costs of rehabilitation), and people with disabilities are often excluded from opportunities to learn and earn, so that people with disabilities may 'fall into' poverty (Braithwaite & Mont, 2009;Mitra, 2018;Mitra et al., 2011Mitra et al., , 2013Palmer, 2011). Conversely, people who are living in poverty may be more vulnerable to injury and illness, and have worse healthcare access, and thus at increased risk of acquiring an impairment and experiencing disability (Groce et al., 2011;Palmer, 2011;Trani & Loeb, 2012).
Of relevance to our review is the first of these pathways, from disability to poverty. The widespread exclusion of people with disabilities from livelihood opportunities is one of the drivers of the relationship of disability to poverty and is the focus of a substantial literature (Banks & Polack, 2014;World Health Organization, 2011).
The 2018 UN Flagship Report on Disability and Development reported that across 8 geographical regions, the employment to population ratio for people with disabilities aged ≥15 years was 36% compared to 60% for people without disabilities. Indeed, an employment gap between people with and without disabilities is observed in the vast majority of countries (Mitra & Yap, 2021). The exclusion of people with disabilities from employment is also HUNT ET AL. | 3 of 34 repeatedly shown in the broader literature, as illustrated in Figure 1, although these international comparisons must be made with caution due to differences in how disability and employment (especially informal employment) are measured.
There are complexities to the relationship between employment and disability. Disability is not a homogenous category and the experience of exclusion from employment will vary by gender, impairment type and context. Women already frequently face discrimination in terms of livelihood inclusion, and this may be compounded for women with disabilities (World Health Organization, 2010). For instance, the World Health Surveys used consistent methods to measure these constructs across 51 countries, and showed that employment levels were lower in men with disabilities (53%) compared to men without disabilities (65%), but that the overall level of employment was lower the gap higher when comparing women with disabilities (20%) to women without disabilities (30%) (World Health Organization, 2010). Exclusion may also vary by impairment type, as people with mental health conditions or intellectual impairments or particularly stigmatising conditions may be at higher risk of exclusion from employment (Van Beukering et al., 2021; World Health Organization, 2010), or face resistance when requesting necessary employment accommodations (Prince, 2017). Although data are lacking, people with disabilities may be particularly left behind within humanitarian settings in terms of livelihood inclusion. Another consideration is that employment level alone is not the only pertinent measure of exclusion. Multiple studies have shown that when people with disabilities do work it is more likely to be in the informal sector, part-time and for lower wages (Banks & Polack, 2014;World Health Organization, 2011). This pattern is illustrated by Figure 1, again with the caveat that differences in measurement of disability and employment (especially informal employment) make international comparisons difficult. The inequity in employment associated with disability occurs even though almost all jobs can be done by people with disabilities if the right supports are in place. However, it is unclear which interventions are most effective at improving employment inclusion and outcomes among people with disabilities in LMIC, and this question has not been previously explored through a systematic review.
It is important to focus beyond waged employment alone, to livelihood more broadly. Livelihood encompasses the means through which individuals or households can meet their basic needs. It F I G U R E 1 Employment-to-population ratio for persons with and without disabilities: Most recent data close to year 2010. Source: ILO (2018). encompasses people's capabilities (Sen, 1993), assets, income and activities required to secure the necessities of life (Hebinck & Bourdillon, 2001). A livelihood is sustainable when it can cope with, and recover from, stress and shocks, and when it can maintain or enhance its capabilities and assets both now and in the future, while not undermining the natural resource base (Chambers & Conway, 1991). Livelihood, therefore, also includes social protection and financial support, as well as individual's skills to be included in employment.
Social protection includes programmes and policies designed to reduce poverty and vulnerability, for instance, by providing social assistance or by promoting efficient labour markets. Social protection can therefore assure that low-income and vulnerable populations are able to maintain a basic livelihood, including people with disabilities. Indeed, many countries offer a disability allowance or similar scheme (Walsham et al., 2019). In Korea, for instance, there is a means-tested and noncontributory public assistance grant, called the National Basic Livelihood Security System (NBLSS) (emphasis added) (Jeon et al., 2017). The aim of this grant is to support livelihoods-to mitigate poverty and improve the quality of life and capacity to maintain a minimal standard of living, for the low-income families and vulnerable groups (including people with disabilities) (Jeon et al., 2017). Social protection interventions need to address the inequalities and the processes of social exclusion that people with disabilities face in attaining a livelihood to have a meaningful impact on their livelihood (de Haan, 2017;Stienstra & Lee, 2019). Yet, evidence is lacking on whether social protection or other similar interventions are effective at improving livelihoods for people with disabilities, as most studies have focused on interventions to improve waged employment alone (Banks, Mearkle, et al., 2017;Cramm & Finkenflugel, 2008).
The financial benefits for people with disabilities of inclusion in livelihood opportunities are obvious ( Figure 2) (Banks & Polack, 2014). Improving livelihood outcomes will help people to meet their basic needs. People who are employed will earn income, whether financial or in kind, which will reduce their poverty levels.
These benefits will extend beyond the individual to his/her household, as they contribute to the household economy. Financial benefits are also reaped by employers, as they are able to select employees from the full range of skills and abilities, and as evidence suggests that people with disabilities may be particularly loyal and committed employees (UNenable, 2007). Society will also see financial benefits through tax generated from the salary of people with disabilities (Deloitte, 2011). For instance, a report commissioned in 2011 by the Australian Network on Disability showed that closing the gap between labour market participation rates and unemployment rates for people with and without disabilities by one-third would increase Australia's GDP by $43 billion over the following 10 years (Deloitte, 2011).
The nonfinancial benefits of improving livelihood opportunities for people with disabilities must also be emphasised ( Figure 2).
Employment is a cornerstone of social inclusion and facilitates friendship and engagement in society. It also promotes human dignity and social cohesion. Fulfilling the right to livelihood inclusion may also help other rights to be met-for instance, the workplace is a key provider of healthcare, and receipt of social protection may help health care and educational costs to be met (as evidence from F I G U R E 2 How livelihood can reap gains for people with disabilities. Source: Banks and Polack (2014).

| The intervention
The intervention considered in this review are those that improve livelihood outcomes for people with disabilities. We used the WHO's Community Based Rehabilitation (CBR) Guidelines (World Health Organization, 2010) as our starting point for conceptualising the kinds of interventions which may be considered as livelihood interventions. CBR, which is promoted by the WHO to improve the lives of people with disabilities, has 'livelihood' as one of its five pillars (World Health Organization, 2010). Within the 'livelihood' pillar, there are five specific components which we used to initially elaborate a list of the types of interventions which might be included in this review: wage employment, skills development, self-employment, access to financial services (e.g., micro-credit schemes, access to bank accounts), and inclusion in social protection programmes. Each of these categories has specific interventions which are named in Table 1 (e.g., vocational training, job placements, and birth registration). However, given that our review is not only concerned with CBR programmes, the CBR served only as a guiding framework for the intervention categories, these were piloted and refined against a set of studies before use. During this process, we added two categories to the livelihood pillar, namely Health and Rehabilitation, and Policies, as potential approaches to improve livelihood inclusion.
We considered interventions that specifically target people with disabilities, as well as mainstream programmes that are inclusive of people with disabilities and present disaggregated outcomes for people with disabilities.

| How the intervention might work
It is important to consider the barriers to livelihood opportunities experienced by people with disabilities, to identify how these may be overcome. People with disabilities are not a homogenous group, and the reasons for exclusion will vary for women and men, in different settings, and for people with different impairment types. Nevertheless, barriers can be broadly categorised as being experienced at the level of the System, the Programme (Workplace), and the Individual (the Family or the Person) (Wapling, 2016).
System-level barriers include the lack of legislation or policies to support the inclusion of people with disabilities in livelihood opportunities. Even where there are good policies, these may not be implemented due to failure to monitor inclusion or to implement incentives or penalties to promote inclusion. Another important concern is inadequate resource allocation to support inclusion (e.g., lack of funds for access to work schemes). Policies may also be inappropriately formulated so that they penalise people with disabilities who work (e.g., create a benefits trap) or establish over-protective labour laws that discourage firms from employing people with disabilities.
Programme-level barriers include lack of reasonable accommodation (including assistive technology), physical accessibility of the workplace, transport or toilets, or the existence of negative attitudes from employers and co-workers towards people with disabilities.
Programmes, such as micro-credit schemes, may also explicitly exclude people with disabilities (e.g., making people with long-term health conditions ineligible).
Individual-level barriers include the frequently lower level of training or skills of people with disabilities, following their higher risk of exclusion from education, which may make livelihood opportunities more difficult to obtain. People with disabilities may also experience poor health, and require treatment and rehabilitation, which can make full-time employment more challenging. Depending on the impairment type, people with disabilities may have difficulties with different skills needed in many work environments, such as concentrating and controlled behaviour, and this may reinforce negative attitudes that people with disabilities are not capable of learning or worth investing in. People with disabilities may experience Approaches to improve livelihood inclusion and outcomes for people with disabilities must act by targeting the barriers that they experience. In other words, they must operate at the level of the system (e.g., improving policy and legislation), the programme (e.g., making reasonable accommodations) and/or individual (e.g., providing training in new skills). These interventions should address inclusion in livelihood opportunities in the broadest sense, and not focus on employment alone. The World Report on Disability describes different approaches to addressing barriers and thereby enhancing livelihood opportunities (World Health Organization, 2011).
At the systems-level, most countries have laws and regulations in place protecting people with disabilities from discrimination in employment, 1 but they should be implemented where they are lacking or improved if they are inadequate. Systems-level interventions may also include instituting requirement for reasonable accommodation in the workplace, implementation of quotas for employment of people with disabilities, establishment of tax incentives to employers, mainstreaming disability into public employment services, or promotion of affirmative action. A concern is that regulations can act as disincentive to the employment of people with disabilities (e.g., due to expense of providing specialist resources, of strong protection of workers' rights), and this must be avoided.
Examples of programme-level interventions include supported employment (e.g., specialist job training, social firms), sheltered employment (e.g., employment in segregated facilities), social protection (e.g., disability grants), and micro-finance (e.g., group loans or small business loans). 14 eligible studies, which generally found positive impacts of the interventions, despite concerns about the quality of the data. While this latter review is relevant to the current proposed review, it did not include interventions aimed at people with psychosocial disabilities, nor did it address broader livelihood outcomes (e.g., social protection, access to financial services). There are also likely to be relevant papers published since these reviews were undertaken.
There is a broader existing pool of reviews which focus on specific aspects of the central question of which interventions are effective at improving livelihood outcomes for people with disabilities. These reviews are restricted in terms of: -Impairment type/condition included: Several reviews have been undertaken, or are planned, which focus on livelihood outcomes for people with specific impairments or conditions, including people with musculoskeletal conditions (Alexander et al., 2017;Seeberg et al., 2019), Autism, (Westbrook et al., 2013) acquired brain injury (Batavia et al., 2017), Stroke (Chan et al., 2013) or mental health conditions (Suijkerbuijk et al., 2017). However, reviews are lacking addressing disability holistically. what is effective to improve inclusion in social protection programmes for people with disabilities . Here too, data are lacking despite the fact social protection programmes and financial schemes are widely promoted globally in efforts to alleviate poverty.
-Other socio-demographic restrictions: Several reviews exist focused only on interventions for young adults (Jetha et al., 2019).
There is consequently a need for a review assessing the overall literature on effectiveness of interventions to improve livelihood for people with disabilities, including broad livelihood outcomes.
This review should be focused on LMIC, as this is where 80% of people with disabilities live and the particular challenges and opportunities with respect to livelihood may differ from highincome settings.

| OBJECTIVES
The question posed by this review was 'What works to improve livelihood outcomes for people with disabilities in LMIC?'. The objectives of this review were to answer the following research questions: Eligible designs included those in which one of the following was true: a) participants were randomly assigned (using a process of random allocation, such as a random number generation), b) a quasi-random method of assignment was used, c) participants were non-randomly assigned but matched on pretests and/or relevant demographic characteristics (using observables, or propensity scores) and/or according to a cut-off on an ordinal or continuous variable (regression discontinuity design), d) participants were non-randomly assigned, but statistical methods have been used to control for differences between groups (e.g., using multiple regression analysis or instrumental variables regression), e) the design attempted to detect whether the intervention has had an effect significantly greater than any underlying trend over time, using observations at multiple time points before and after the intervention (interrupted time-series design), f) participants receiving an intervention were compared with a similar group from the past who did not (i.e., a historically controlled study), or g) observations were made on a group of individuals before and after an intervention, but with no control group (single-group before-and-after study).

| Types of participants
The target population were people with disabilities living in LMIC, including people with physical, sensory, intellectual, cognitive, and psychosocial (i.e., arising from a mental health condition) impairments.
We also included studies which were concerned with family members or carers of people with disabilities, and service providers working with people with disabilities, in LMIC (although these studies were only included where a relevant livelihood outcome among people with disabilities was included). Population sub-groups of interest included women, children (particularly vulnerable children, e.g., those in care), different impairment groups, conflict and post-conflict settings, migrants/refugees/internally displaced people, and ethnic minority groups. The LMIC context, and opportunities for people with disabilities, are considerably different from those in high-income countries, hence the need for a separate review.

| Types of interventions
The WHO CBR matrix served as a guiding framework for the intervention and outcome categories, as described above. There were no restrictions on comparators/comparison groups, however, a study must have both an eligible intervention and an eligible outcome to be included. Eligible intervention types related to livelihoods, targeted at the system-, programme-and/or individual-level, and are presented in Table 1.

| Types of outcome measures
Eligible outcomes were also developed from those included in the livelihood pillar of the CBR matrix. All outcomes were relevant regardless of whether they were primary outcomes, or secondary outcomes of the study. It is important to note that if a study did not have both an eligible intervention and an eligible outcome then it was excluded. The outcomes of interest included those experienced at the system-, programme-and/or individual-level, and are presented in Table 2.

| Duration of follow-up
Any duration of follow-up was included.

| Types of settings
All settings were eligible, provided that the study is situated within a low-and-middle-income country, as defined by the World Bank (https://datahelpdesk.worldbank.org/knowledgebase/articles/ 906519-world-bank-country-and-lending-groups).  (Saran et al., 2020). The EGM present studies on the effectiveness of interventions for people with disabilities in LMIC. We updated the database search and screened the references to identify additional studies. This review was based on the updated searches performed for the map February 2020. As the inclusion/exclusion criteria for this review were narrower in scope than the scope of the EGM, the review team independently screened all studies included in the map to meet the predetermined eligibility criteria outlined previously. In April 2022 we also updated our searches using Open Alex in EPPI-reviewer to ensure that nothing had been missed, adding livelihoods-specific search terms to our search strategy using the Campbell Collaboration's machine learning system for identifying studies. This process did not yield any additional studies within the time frame of this review (i.e., before

| Search methods for identification of studies
February 2020).
The search comprised (1) an electronic search of databases and sector-specific websites, (2) screening of all included studies in the instances where reviews are identified, (3) and citation searching of included reviews (including both forward and backward searching).

| Electronic searches
A search of the following electronic databases was conducted by the author: • Web of Science (Social Sciences Citation Index) • WHO Global Health Index MEDLINE, Embase, PsychINFO, and CAB Global Health were searched through OVID and ERIC and CINAHL through Ebsco.
PubMED through NCBI.
Search strategies were tailored for each of the databases (see Supporting Information: Annex 2). No restrictions were placed. The main search strategy was as follows, using English as the search language: POPULATION: (disable* or disabilit* or handicapped) OR (physical* or intellectual* or learning or psychiatric* or sensory or motor or neuromotor or cognitive or mental* or developmental or communication or learning) OR (cognitive* or learning or mobility or sensory or visual* or vision or sight or hearing or physical* or mental* or intellectual*) adj2 (impair* or disabilit* or disabl* or handicap*) OR (communication or language or speech or learning) adj5 (disorder*) OR (depression or depressive or anxiety or psychiat* or well-being or quality of life or self-esteem or self perception) adj2 (impair* or disabilit* or disabl* or handicap*) OR mental health OR (schizophreni* or psychos* or psychotic or schizoaffective or schizophreniform or dementia* or alzheimer*) adj2 (impair* or disabilit* or disabl* or handicap*) OR (mental* or emotional* or psychiatric or neurologic*) adj2 (disorder* or ill or illness*) OR (autis* or dyslexi* or Down* syndrome or mongolism or trisomy 21) OR (intellectual* or educational* or mental* or psychological* or developmental) adj5 (impair* or retard* or deficien* or disable* or disabili* or handicap* or ill*) OR (hearing or acoustic or ear*) adj5 (loss* or impair* or deficien* or disable* or disabili* or handicap* or deaf*) OR (visual* or vision or eye* or ocular) adj5 (loss* or impair* or deficien* or disable* or disabili* or handicap* or blind*) OR (cerebral pals* or spina bifida or muscular dystroph* or arthriti* or osteogenesis imperfecta or musculoskeletal abnormalit* or musculo-skeletal abnormalit* or muscular abnormalit* or skeletal abnormalit* or limb abnormalit* or brain injur* or amput* or clubfoot or polio* or paraplegi* or paralys* or paralyz* or hemiplegi* or stroke* or cerebrovascular accident*) The screening process is reported using a PRISMA flow chart 3. Study design: Is the study one in which participants are randomly assigned or quasi-randomly assigned, or where nonrandom assignment has been done, but participants have been matched on pre-tests and/or relevant demographic characteristics or statistical methods have been used to control for differences between groups; or where the design attempts to detect whether the intervention has had an effect significantly greater than any underlying trend over time, using observations at multiple time points before and after the intervention (interrupted time-series design); or where participants receiving an intervention are compared with a similar group from the past who did not (i.e., a historically controlled study); or where observations are made on a group of individuals before and after an intervention, but with no control group (single-group before-and-after study).  Confidence in study findings was rated high, medium, or low, for each of the criteria, applying the standards as shown in Table 3.

| Assessment of risk of bias in included studies
Overall confidence in study findings was determined to be the lowest rating achieved across the criteria-the weakest link in the chain principle.

| Measures of treatment effect
We found that it was not possible to conduct a meta-analysis, and generate pooled results or compare effect sizes, given the diversity of designs, methodologies, and outcome measures across studies in this area, as well as poor reporting of parameters required to calculate standardised measures of effect. However, when effect sizes cannot be pooled, study-level effects were reported in as much detail as possible.

| Unit of analysis issues
The unit of analysis of interest to the present review was individual people with disabilities, their caregivers, carers, or those working with them. If a study was included with more than two intervention arms, we included only intervention and control groups that met the eligibility criteria. Where multi-arm studies were included, we ensured not to double-count participants, and separately report eligible interventions and their respective outcomes.

| Criteria for determination of independent findings
Multiple publications of the same study were examined as a single study.

| Subgroup analysis and investigation of heterogeneity
We did not conduct subgroup analyses as part of a meta-analysis given the high level of heterogeneity in reporting and effect sizes.

| Sensitivity analysis
No sensitivity analyses were conducted.

| Treatment of qualitative research
We did not include qualitative research. impact evaluations were found to be eligible for inclusion in livelihood review.
As noted in the methods section, to identify any relevant articles that may have been missed during the EGM processes, we also ran the searches with search terms specific to livelihood review using Open Alex in EPPI reviewer. We identified an additional 252 studies, the results were deduplicate and we identified 247 studies that were screened for title and abstract. Only 13 studies were included for full text review. Only one study was included for data extraction, however, it was a paper from a study that was already included and hence the papers were linked (discussed below).

| Included studies
Included studies are summarised in Tables 4 and 5, and discussed in detail below.

| Impairment groups
Seven of the included interventions targeted individuals with a single type of impairment, and only one intervention catered to people with a range of impairments . There was also one programme for people without disabilities who were service providers    and visual impairments , respectively.

| Region
Ten countries were represented across the studies, some of which were multi-country. Two studies were conducted in the East Asia and   and none at the level of the system.

t-Test
Pereira-Guizzo et al. The research designs of the studies included one randomised controlled trial , one quasi-randomised controlled trial (a randomised, posttest only study using propensity score matching (PSM) , one case-control study with PSM , four uncontrolled before and after studies , and three posttest only studies , one where PSM was also used, and two with implied baselines as being unemployed was a prerequisite for being enroled into the programmes.

| Subject assignment
In two studies, subject assignment was individual random , and in one study whole group random . In one study, there was matched, nonrandom  subject allocation. However, in most cases, nonmatched and nonrandom subject assignment was used in determining intervention participation and/or participation in the associated impact evaluation . Allocation was not reported in two studies , and no allocation was used in the case-control with PSM .

| Intervention characteristics
We used the above-detailed table when extracting data on the included studies, loosely based on the interventions associated with the livelihoods pilar of the CBR matrix. Many of the interventions were multicomponent, and so fell into several categories. In terms of skills development, three interventions aimed to improve training opportunities for employment such as vocational training , and three provided social, life and communications skills training . In the area of improving access to waged employment, two programmes aimed to facilitate physical access to the workplace , and one included job placement as an intervention component . Two vocational training programmes were included, and both targeted formal employment as well as equipping people with disabilities to sell goods and services . One programme, conducted with individuals without disabilities, aimed to remove social and attitudinal barriers to access for people with disabilities . Finally, four interventions aimed to improve livelihoods by improving access to rehabilitation Mauro et al., 2014), or assistive technology .
Several categories of possible intervention, including financial services, social protection and policy change were absent from the included studies. Overall, eight of the interventions focussed on the individual-level, one at the programme-level  and none at the level of the system. Additional detail on intervention setting, delivery, and implementation, as well as dosage, are presented in Table 5.

| Outcome characteristics
The outcomes of the included interventions were mapped in a similar manner to the interventions (i.e., against a table loosely based on the CBR matrix). The main outcome of two programmes were 'acquisition of skills for the workplace', specifically social and communication skills needed for work .
In the domain 'access to the job market', two studies examined outcomes to do with the capacity of people with disabilities to engage in job searching , and three physical and social barriers to employment . Most outcomes fell into the category of 'employment in formal and informal sector', with six studies examining entrepreneurship and informal sector participation as well as waged employment and formal sector participation Mauro et al., 2014;. Four interventions used outcomes related to 'income and earnings from work' Mauro et al., 2014). Finally, one study used the outcome of access to formal and informal social protection . No studies reported on outcomes related to the development of inclusive policies, or access to financial services such as grants and loans. Table 6 presents a summary of the findings of this review, by outcome of interest.

| Excluded studies
Exclusions are recorded in the PRISMA diagram above. Excluded studies with the associated reason for exclusion are presented in Annex A. Common reasons for exclusion included that the study was not an impact evaluation, presented a protocol for which there were no associated results, focused on an ineligible population, had a livelihoods intervention but no livelihoods outcomes, presented only qualitative data, and-in one case-otherwise relevant findings were not disaggregated for people with disabilities.

| Risk of bias in included studies
Our confidence in the overall findings is low to medium on the basis of our appraisal of the studies. Two studies  scored medium using our assessment tool, with the remaining eight scoring low on one or more item Mauro et al., 2014;. There is diversity within low ratings as we employed the weakest link in the chain principle to T A B L E 6 Summary of findings by outcome HUNT ET AL.
| 23 of 34 assess confidence in study findings (Table 7). However, the findings of a study receiving a low rating on a single item (e.g., Zhang et al., 2017 for reporting of attrition) should not be treated in the same manner as those derived from a study rating low on multiple items. The latter approach allows for valuable learnings not to be overlooked due to an overall 'low' confidence in study findings score, in studies which had many areas of strength.

| Study design
Most studies were rated 'low' on study design , as many used before and after designs, often without a control group.
T A B L E 7 Confidence in study findings appraisal Furthermore, two of the included before and after studies, by Nuri et al.  and , only had implied baselines and data was only collected after the intervention was administered. However, entry into the interventions in these cases required meeting of certain baseline criteria (e.g., unemployment), and so both were treated as before and after studies. Three studies were rated medium  in our assessment of confidence in study findings based on design, as they employed PSM (in a quasi-randomised controlled trial and in a case-control study) or double difference techniques to mimic the conditions of a more rigorous design. There was only a single randomised controlled trial , and so only one 'high' rated study on design.

| Masking
For most studies, masking was not an applicable measure of confidence in study findings, as few were RCTs. However, for the single randomised controlled trial, masking was implemented and reported . For the quasi-randomised controlled trial, masking was not reported, and so a rating of 'low' was recorded . Nuri et al. , loss to follow up was not applicable as the data were collected at one time point only.

| Disability/impairment measure definition and reliability
One of the areas which received relatively good ratings across studies was the use of disability measures or definitions which were consistently clear and reliable. No studies received a rating of 'low'.
Most studies received ratings of 'medium' . In these papers, impairment type was mentioned, and associated diagnoses listed (but how these were arrived at was not reported), or single items were used to determine disability status (i.e., 'do you have a spinal cord injury'), but more rigorous disability assessments or clearer operational definitions were omitted. In four studies, rigorous and replicable criteria were used, and high ratings were given Mauro et al., 2014;. In Zhang et al.'s (Zhang et al., 2017) study of individuals with psychosocial disabilities (schizophrenia), The Brief Psychiatric Rating Scale was used to assess the participants' psychiatric status. In the study of , every individual in the sample had been seen by a physician and had been deemed physically in need of a wheelchair.
Finally,  defined disability following the instructions of the World Health Organization's Community-based Rehabilitation Manual.  targeted participants who were people without disability working with individuals with disabilities, rather than people with disabilities themselves, and as such this publication was not assessed for this criterion. It is worth noting that reliability and validity of specific outcome measures used was scarcely discussed in the publications, and so it was not possible to systematically extract information on these indices.

| Outcome measure definition and reliability
Outcome measures were largely well-defined, perhaps reflective of the tendency of the studies to be outcome-driven interventions, and so primarily concerned with operationalizing and then acting upon, a particular dimension of livelihoods. All but two studies  received high ratings on this item.

| Baseline balance
Baseline balance was only relevant for four of the studies; the randomised controlled trial , the two studies using PSM Mauro et al., 2014), and a two group before and after study . The randomised controlled trial and the studies using PSM reported acceptable baseline balance and were coded as high on this item. However, the two group before and after study by  was scored as medium on baseline balance as there was evidence of baseline balance but the sample size per group was very small (n = 8) for each group. ) received a 'low' rating on study design, as they only collected data at a single time point, postintervention.

| Appraisal by study
However, admission to the intervention was predicated on lack of employment, and the outcomes measured at the assessment time point included employment gained, and so the study was implicitly a before and after design, and therefore included in this review. A 'medium' rating was given to the definition of disability used, as the authors did list conditions (impairments resulting from a variety of physiological conditions, cosmetic disfigurements, spinal cord dysfunctions, musculoskeletal losses, sensory impairments, and various types of chronic diseases) which merited inclusion but did not report how eligibility was determined. Outcome measures were not clearly defined, but a rating of medium was given as percentages of certain binary measures (secured employment/did not secure employment) were reported. Masking, attrition, and baseline balance were not relevant criteria, given the design, and so not scored. An overall score of 'low' was assigned to confidence in these study findings.
HUNT ET AL.
| 25 of 34 The study of  received a design rating of low, as the study employed a before and after experimental group design. The study received a rating of high for its reporting of losses to follow up, as before and after scores were reported for the total sample, indicating no attrition. The disability/impairment measure rating for this study was medium as the authors note that they enroled visually impaired students from the School for Handicapped Children in Ibadan and Osogbo, Nigeria, but did not record what degree of vision impairment was included, or whether there was variation in the group. The study received a high rating for its definition and reliability of outcome measures, as the 'Work Value Inventory' was an existing tool developed by Salami (2000). Masking and baseline testing were not relevant given the study design, and so not scored. The overall confidence in study findings score for this study was low.
The study of  received a study design score of medium, as the authors reporting using covariate matching, seeming unrelated regressions (SUR), and a series of robustness checks for endogeneity within their two group (intervention and control) design. Losses to follow up was rated as medium, given that attrition was not explicitly reported, and for some outcomes observations were noted to be for the full sample (n = 261), while for other outcomes one observation was missing (n = 260). This could be read to imply the loss of one individual to follow up for some items, which-had it been reported-would have rendered a rating of high on this item. However, given the failure to report attrition, the score was downgraded to a medium. The definition of disability/ impairment was rated as high, as the authors noted that the sample comprised individuals with a range of physical impairments, and described the range of aetiologies and impairment types included in the broader group (polio, infections, work accidents, war victims, muscular dystrophy and leprosy identified by wheelchair recipient lists and waitlists). The outcomes of interest were also score high, as they were clearly defined: more hours per day for work, fewer hours per day for street begging, and percentage increase in income.
Masking and baseline testing were not relevant given the study design, and so not scored. The overall finding regarding confidence in these study findings was medium, driven by item ratings for robustness of the study design, and reporting of attrition.
Hansen et al.  received a rating of 'low' on study design as the study employed an uncontrolled before and after design. A rating of 'medium' was given for the authors' definition of disability/impairment, as they did not report using a standardised measure, but did note that all participants had spinal cord injuries.
Definition and reliability of outcome measures was rated 'low' in this study, as definitions of key outcomes were lacking. Masking, loss to follow up, and baseline testing were not relevant given the study design, and so not scored. Overall confidence in study findings was rated 'low' for this study.
Biggeri et al.  and Mauro et al. (Mauro, 2014) reported on the same intervention, but using slightly different designs.
As such, both similar ratings for many criteria. A rating of 'medium' was given to both on study design as Biggeri et al.  employed a case-control design with PSM, and Mauro et al. conducted a quasi-randomised controlled trial design with PSM, to evaluate the intervention. The anomaly between the two studies has to do with reporting of masking. Because Biggeri et al.  framed their study as a case control, they were not assessed on this criterion.
However, Mauro et al. (Mauro, 2014) called their evaluation a quasirandomised trial, and so they were expected to have reported on masking, but did not and so a 'low' rating was given to the latter for the masking criterion. The disability/impairment measure used in both was clearly defined and reliable, and so a rating of 'high' was recorded for this criterion. The same applied to the outcome measures used, which were clearly defined and reliable. PSM was used in both studies to build treatment/case and control groups with balanced pretreatment covariates, and so a rating of high was given to this study on the balance criterion. An overall of 'medium' was given to Biggeri et al. . However, a 'low' rating was recorded for the Mauro et al. (Mauro, 2014) study, given deficits in reporting of masking and attrition.
Vilela and Leite ) received a rating of 'low' on study design for conducting an uncontrolled before and after study.
However, losses to follow up were presented and acceptable (participant scores were recorded for all items at both time points).
Outcome measures were clearly defined and reliable (the Conceptions of Disability Inventory). The target participants in this intervention were people without disability, and so the definition of disability criterion was not applicable, and neither were masking, nor baseline balance given the design. The overall confidence in study findings was rated 'low'.
In the study of , a rating of low was given for study design, as although they employed a two group before and after design, they did not specify how allocation to the groups had occurred. Losses to follow up were not reported and so a rating of 'low' was given. A 'medium' rating was given for the disability/impairment measure used, as physical impairment was stated and a list of conditions was given, but it was not clear how eligibility was established. Outcome measures were clearly defined and reliable, as the authors reported using the Professional Social Skills Observation System and the Social Skills Inventory, and as such the study was rated high on this criterion. Finally, baseline balance was evidenced, but the sample sizes were very small for each group (n = 8) and so a rating of medium was given, rather than one of high, as would have been the case had the sample sizes been larger.
Masking was not relevant for this study and so not rated. Overall, a rating of 'low' was recorded.
Shore and Juillerat (2012) employed an uncontrolled before and after design, thus receiving a 'low' rating for study design.
Losses to follow up were not reported, nor was a clear and reliable measure of disability reported, leading to two additional 'low' scores. However, the outcome measures were clearly defined and reliable, and so a 'high' rating was recorded on this criterion.
Neither masking nor baseline balance were relevant given study design. Overall, a low rating for confidence in study findings was allocated to this study.
Finally,  was the only study to receive a high rating on study design, being the only randomised controlled trial included in the review.  received high ratings across all indices of our tool, bar one.
Unfortunately, losses to follow up were not reported, and so a 'low' rating was given on this criterion. This study is exemplary of the one weakness of the weakest-link-in-the-chain principle in administering a confidence in study findings appraisal tool, as an otherwise well-designed and well-reported study is assigned a low rating overall, based on a single failure to report.

| Effects of interventions
All the included studies reported positive impacts on livelihoods outcomes. However, outcomes varied substantially by study, as did the methods used to establish intervention impact, and the quality and reporting of findings.
Vilela and Leite  reported that a criticalreflexive intervention for people without disabilities resulted in more positive social attitudes of employees and administrative staff towards the participation of people with disabilities in the workplace (as measured using the Conceptions of Disability Inventory).
However, statistically significant improvements were only reported for one construct (normality) from the pre-and postintervention scores on the Inventory (p = 0.01).
Nuri et al.  reported that a vocational training programme in Bangladesh led to a 60% increase in employment rate among participants and a 74% improvement in self-reported capacity of participants to provide for their families. Statistical significance of these findings was not evaluated by the researchers.
Shore et al.  reported that, following 12 months of using a wheelchair, the percentage of respondents in their study who reported having some employment had increased (p < 0.001), as had the percentage who reported adequate income (p < 0.001). In another evaluation of the impact of wheelchair allocation, Grider and Wydick et al.  reported that people with disabilities given access to a wheelchair allocated 1.75 more hours per day to work (p < 0.001), 1.40 fewer hours per day to street begging (p = 0.0004), and realised a 77.5% increase in income (p = 0.0001), all of which were statistically significant.
Students in Eniola and Adebiyi's  intervention for youth with visual impairments showed a significant increase in the level of motivation post-intervention compared preintervention across the whole sample (p < 0.05).

| Synthesis of results
A quantitative synthesis was not undertaken.

| Summary of main results
We identified, coded, evaluated, analysed, and narratively summarised the findings from 10 studies that evaluated 9 interventions to improve the livelihoods of people with disabilities in LMIC. These studies served as the data for this review and are reported according to the interventions and outcomes identified across all studies. Due to the heterogeneity of outcomes and the low level of confidence in study findings, a meta-analysis was not deemed appropriate to this review, and so findings were presented narratively.
These findings are discussed, broadly, according to participants and programmes.
Children, older people, and service providers were underrepresented in the studies included in this review. While it is perhaps understandable that children were not often targeted for employment- HUNT ET AL.
| 27 of 34 (Cullinan et al., 2013;Kwan & Walsh, 2018), and so may benefit from inclusion in social protection programmes. It also appears that programmes targeting very low-income participants were lacking, although this was possibly a function of poor reporting. Given the relationship between disability and poverty (Braithwaite & Mont, 2009;Groce et al., 2011;Mitra, 2018;Mitra et al., 2011Mitra et al., , 2013Trani & Loeb, 2012), it is surprising to see that few of the included studies reported the socioeconomic status of the participants.
The studies overrepresented people with physical impairments (five out of nine interventions only included people with physical impairments). This focus is possibly due to the perception that they are a relatively easy group for delivery of programmes and conduct of research, as there is no need to overcome communication or cognitive difficulties. However, there are many opportunities for meaningful intervention with people with other impairment types.
Moreover, people with intellectual impairments may experience the greatest barriers to employment and livelihoods and the greatest socioeconomic disadvantage of all impairment groups (Gouvier et al., 2003;Kavanagh et al., 2015). Indeed, a recent report by Mitra and Yap (Mitra & Yap, 2021) found that people with mobility, cognitive and self-care difficulties had the lowest employment rates.
As such, more programming targeting individuals with cognitive and self-care difficulties may be warranted. pensions. There were also no programmes aimed at improving policy involvement and provision for people with disabilities. Social protection interventions were most glaringly absent, given the importance of this intervention for people with disabilities and the high profile which these types of interventions hold in disability discourse (Banks, Mearkle, et al., 2017;Palmer, 2013). The interventions were almost all aimed at the individual level, only one at programme level and none at system level. There was therefore implicitly a focus on 'fixing' people with disabilities rather than addressing broader barriers and facilitators.
Further, while several of the included studies reported on 'hard' outcomes, such as increases in rates of employment or higher income, there were no studies which examined people with disabilities' access to financial services such as grants and loans.
Given the lack of interventions in this area, there were also no outcomes related to the development of inclusive policies. to work, and time spent working. This pattern seems to suggest that it may be possible for a variety of programming approaches to improve outcomes related to the livelihoods of people with disabilities. It is imperative that future research evaluate these approaches with more rigorous study designs, to develop a firmer evidence base to inform intervention at scale.

| Overall completeness and applicability of evidence
The evidence presented here highlights examples of potentially effective interventions to improve the livelihoods of people with disabilities in LMIC. However, these interventions need to be evaluated using rigorous study designs before firm conclusions about their effectiveness can be drawn.

| Quality of the evidence
The quality of the included studies is generally low, as assessed by the confidence in study findings tool. Most study designs employed were unable to consider many potential confounders. Losses to follow up and other important dimensions of study rigour were either poorly recorded or poorly reported. Although some studies did attempt to undertake robust analyses on quantitative outcomes, the ability to make definitive judgements about programme effect was undermined by the absence of controlled trials. Given our assessment of confidence in study findings, it is difficult to draw definitive conclusions from the papers included in this review.
6.4 | Potential biases in the review process

| Agreements and disagreements with other studies or reviews
Our review contributes to the body of literature concerning the livelihood component of the CBR matrix, as a previous systematic review (Iemmi et al., 2015) focussed on CBR-specific interventions for people with disabilities did not find any studies which specifically addressed livelihoods (although one included study had livelihoods elements as minor components). However, our review had in common with that of this earlier one by Iemmi et al. (Iemmi et al., 2015), a finding of heterogeneity of interventions and scarcity of good-quality evidence. A similar dearth of evidence, variety of studies, and low quality of evidence was found by Tripney et al. (Tripney et al., 2015) in their review of interventions to improve the labour market situation of adults with physical and/or sensory disabilities in LMIC. These authors also found that the majority of studies were conducted in a limited range of LMIC (in Asia, Africa and Latin America), and that most programmes were focused on persons with physical impairments (Tripney et al., 2015). This review also found a preponderance of single-group pre-test/posttest designs.

| Implications for research
• There is not a great deal of evidence on livelihoods interventions for people with disabilities in LMIC and so more studies are needed.
• Researchers should work with OPDs and NGOs to identify priority interventions to evaluate.
• Social protection programmes, particularly, need to be rigorously evaluated and outcomes reported for subgroups of beneficiaries including people with disabilities.
• Generally, methodological details are reported poorly, making it difficult to judge inclusion, and assess risk of bias. Consistent use of outcome measures and clear reporting (e.g., standard deviations and sample sizes for treatment and control groups) would help support their inclusion in systematic reviews • Future research could usefully evaluate those programmatic elements associated with livelihoods outcomes to distil the core components of programming which are responsible for change in key livelihoods outcomes.
• Future research could also focus on livelihoods outcomes not examined in the literature yet (interventions-financial services, social protection and policy change were absent from the included interventions; outcomes-the development of inclusive policies, or access to financial services such as grants and loans), in particular those targeted at the programme or system level, rather than the individual level.
• Overall, there is a need for more and better-quality data to inform policy and programming, including data from people with a broader range of impairment types and from different parts of the world.
Central to this project will be the collection, by countries, of disability-disaggregated data through national information management systems. Such data could form the platform from which livelihoods programming could then be developed, saving research-

DECLARATIONS OF INTEREST
The authors have no interests to declare.

DIFFERENCES BETWEEN PROTOCOL AND REVIEW
There are no differences between the protocol and the review, aside from the addition of one intervention category (equipping people with disabilities to sell goods and services) which was deemed important for the coding sheet to capture relevant data from all studies.

Internal sources
• No sources of support provided

External sources
This systematic review is supported by the Commonwealth and Development Office (FCDO) under its support for the Centre for Excellence for Development Impact and Learning (CEDIL) and the Programme for Evidence to iNform Disability Action (PENDA).