A review of screening tools for the identification of autism spectrum disorders and developmental delay in infants and young children: recommendations for use in low‐ and middle‐income countries

Without intervention, developmental delay (DD) and autism spectrum disorders (ASDs) severely restrict children from reaching their developmental potential. Monitoring child development through the use of screening tools can help identify children who need further assessment or intervention. Screening has been widely encouraged to identify children with ASD or DD, and a large variety of screening instruments are suggested in the literature. There is a lack of consensus around which screening tools are most effective, especially where tools are used in cultures other than those in which they were created. We conducted a review of the literature for screening tools for DD and autism to make recommendations for tool selection and use in low‐ and middle‐income countries (LMIC). We included 99 screening tools in the review and created profiles for each tool to evaluate their properties and determine which tools could be effectively used in various LMIC. Our review identified a substantial number (35 for DD and 6 for ASD) of screening tools from LMIC. We identified 10 tools which show promise for use across settings; these tools are brief, low‐cost and can be implemented by paraprofessionals or lay community health workers. Routine screening is an important first step toward addressing the need for services in LMIC, but high‐quality tools take time to be conceptualized, developed, piloted, and validated, before implementation can happen. A focus on improving the scientific rigor of early detection approaches and on enhancing the reach to underserved populations should be prioritized. Autism Res 2019, 12: 176–199 © 2019 International Society for Autism Research, Wiley Periodicals, Inc.


Introduction
Children who experience developmental disabilities are among the most vulnerable members of a society. Without intervention, these difficulties severely restrict children, both academically and socially, from reaching their developmental potential. Developmental delay (DD) and neurodevelopmental disorders such as autism spectrum disorders (ASDs) encompass a range of difficulties that infants and young children may experience in the areas of cognitive, language, social-emotional, behavioral, and neuromotor development (Bellman, Byrne, & Sege, 2013). The prevalence of global DD in children is reported as 1-3% (Bellman et al., 2013), while the global prevalence of ASDs is estimated to be 1 in 132 (Baxter et al., 2015). Children living in circumstances characterized by adversities such as poverty and malnutrition are also at significantly higher risk of experiencing disability (UNICEF, 2013;WHO, 2011). There are a paucity of community-based data on developmental status and disability from low-and middle-income countries (LMIC), despite the fact that most children with disability live in these countries (Durkin et al., 2015;WHO, 2013). Little is known about the epidemiology and clinical presentation of ASD in South-East Asia, South America, and Africa (Baxter et al., 2015;de Vries, 2016;Elsabbagh et al., 2012). For children who are developmentally delayed, prevalence rates are likely even higher than reported, since children with milder and more subtle signs are likely to go unnoticed (Sajedi, Vameghi, & Kraskian Mujembari, 2014). Given the increasing developmental burden in LMIC (Lawn et al., 2014), it is essential to identify at-risk and affected children as early as possible.
The under-identification of children with disabilities is of concern, as early identification and initiation of treatment have been shown to improve child outcomes for DD (Berlin, Brooks-Gunn, McCarton, & McCormick, 1998;Hwang, Chao, & Liu, 2013) and for autism (Filipek et al., 2000;Stahmer & Mandell, 2007). The World Health Organization (WHO, 2012) promotes developmental monitoring (also referred to as developmental surveillance by the American Academy of Pediatrics) as a process for the early detection of developmental difficulties, specifically for LMIC. One of the suggested ways of monitoring children's development is through formal screening for DD or neurodevelopmental disorders, as part of a step-wise approach to diagnosis and provision of care (American Academy of Pediatrics, 2006). Evidence from high-income countries (HIC) suggest that incorporating screening tools into routine health care visits can result in earlier and more accurate identification of children who need help, compared to relying on clinical impressions only (Hamilton, 2006;Sheldrick, Merchant, & Perrin, 2011). This may be particularly relevant for LMIC, where care providers are often less experienced in identifying DD or disorders (Desai & Mohite, 2011).
Regular screening during health care visits for autism or DD offers an easily administered means of early detection, while enabling referral for further evaluation and intervention where needed. However, despite its promise, early detection remains a challenge in both HIC and LMIC (Barton, Dumont-Mathieu, & Fein, 2012;Durkin et al., 2015;King et al., 2010;Macy et al., 2014). Identification is difficult in early life, when changes in development are rapid, domains overlap, and early signs are often subtle (Mukherjee, Aneja, Krishnamurthy, & Srinivasan, 2014). Both primary health care staff and caregivers in LMIC settings may have limited knowledge of more subtle delays or specific disorders such as autism. Autism is a prevalent and well-known neurodevelopmental disorder in HIC, but many communities in LMIC have little awareness of the disorder (Abubakar, Ssewanyana, & Newton, 2016), and affected children are less likely to be identified by primary care providers (Wallace et al., 2012). Also, establishing a relevant set of screening criteria to identify autism across different cultures and socio-economic backgrounds is difficult (Wallice & Pinto-Martin, 2008).
Screening requires adequate financial and human resources for implementation. Factors that may impede screening in LMIC include costs, lack of resources, staff limitations, and insufficient training (Morelli et al., 2014;Pinto-Martin, Dunkle, Earls, Fliedner, & Landes, 2005;Rydz et al., 2006;Sand et al., 2005). Importantly, screening needs to be linked to psychoeducation and counseling, follow-up services and treatment (Grossman et al., 2010;King et al., 2010). False-positive screen results can lead to unnecessary stigma, anxiety, and excess costs for the family and the health care system, whereas falsenegative results can lead to delays in treatment and worse outcomes. Ideally, surveillance and screening would be the starting point of a comprehensive developmental monitoring process, whereby the screening results guide decisions about intervention services that may help mitigate or minimize the severity of a child's delay or disability (Ali, Mustafa, Balaji, & Poornima, 2013;Pinto-Martin et al., 2005;Zwaigenbaum et al., 2015).
Another barrier to early identification through screening revolves around the selection process of the screening instruments themselves (Drotar, Stancin, Dworkin, Sices, & Wood, 2008;Warren et al., 2016). Screening tools may be general, encompassing multiple domains (e.g., the Ages and Stages Questionnaire (ASQ-3), Abo El Elella, Tawfik, Abo El Fotoh, & Barseem, 2017) or specific to a disorder such as autism (e.g., the Modified Checklist for Autism in Toddlers Revised with Follow-up (M-CHAT-R/F); Robins et al., 2014). As awareness of concerns about child development and specifically autism has increased, screening has been widely encouraged to identify children with ASD or DD, accompanied by a large variety of instruments suggested in the literature (Moodie et al., 2014;Ringwalt, 2008;Rydz et al., 2006;Semrud-Clikeman et al., 2017). A lack of consensus exists around which screening tools will be most effective to detect developmental disability in different settings. While significant improvements have been made in the development, validation, and implementation of screening tools for use in LMIC, most tools have been developed in North America or Europe and are increasingly being used in cultures other than those in which they were created (Soto et al., 2015). There is a scarcity of validated tools available to identify children with autism in LMIC (Durkin et al., 2015) and Africa in particular (Abubakar et al., 2016).
An important challenge in early identification of developmental disability is having tools that respond to local differences, including cultural perceptions in meaning of disability (Fischer, Morris, & Martines, 2014). Crossculturally appropriate and affordable tools with good psychometric properties remain limited (Goldfield & Yousafzai, 2018), and using tools developed in HIC for LMIC settings may not always be appropriate. Applying Western-based norms to other cultural contexts may be problematic, since there is a tendency to over-identify children as delayed. In addition, many of these tools are copyrighted and require permissions and payment for translation into other languages (Durkin et al., 2015), thus further limiting their use in LMIC. An ideal screening tool for LMIC would be a brief, inexpensive tool with developmentally appropriate items and good psychometric properties (Goldfield & Yousafzai, 2018), available in local languages where it is used, validated on representative healthy children of the particular population, and requires minimal training (Lansdown et al., 1996). These criteria apply to tools used to detect autism, as well as more general DD. It is not clear which existing tools are best suited for this, or where further tool development and research is most needed. We conducted a review of the literature for screening tools for DD and ASD. This review had the following objectives: 1. Identify current screening instruments for DD and ASD. 2. Create screening tool profiles in order to consolidate the available information on characteristics and use. 3. Make recommendations for screening for DD and ASD in LMIC.

Search Strategy
We conducted online searches, using various databases (PubMed, Web of Science, EBSCO, and Google Scholar) to identify publications related to the identification of children with DD or ASD. The search was conducted in two phases, with each phase consisting of two parts. We conducted Phase 1 of the review in 2014; searching for tools published up to October 2014 (we did not specify a start date). Search terms included "screening," "screening tools," "autism spectrum disorders," "autism," "developmental delay," "developmental disability," and "lowand middle-income countries." In August 2017, we applied the same search terms to update the review, in order to identify and include new tools that have been developed or published since 2014. Given that in most of the peerreviewed literature the name of the screening tool is not mentioned in the title or even as a key word, we also conducted individual searches to identify tools. Therefore, during each phase, the search for screening tools (Part 1) was followed by an individual search (Part 2), using the name of each tool identified in the general search results. The initial search results generated a large volume of studies and reviews related to developmental screening processes and instruments. Search results yielded guidelines and recommendations for the use of screening tools to identify children with DD or ASD (e.g., American Academy of Pediatrics, 2006;Charman & Gotham, 2013;King et al., 2010) and reports of screening tools used in different populations (e.g., Barton et al., 2012;Bello, Quartey, & Appiah, 2013;Grossman et al., 2010;Perera, Wijewardena, & Aluthwelage, 2009). The search results included a large number of studies that described tool development and validation (e.g., Allen, Silove, Williams, & Hutchins, 2007;Bhave, Bhargava, & Kumar, 2010;Durkin et al., 1994Durkin et al., , 1995 or adaptation of screening tools (e.g., Gladstone et al., 2008;Kakooza-Mwesige et al., 2014;Soto et al., 2015), as well as comparisons between screening tools (e.g., Mayes et al., 2009;Murray, Mayes, & Smith, 2011;Snow & Lecavalier, 2008). Using existing publications, as well as our focused literature search, we compiled an alphabetical list of all the tools used to identify children with DD and ASD. We used this list to conduct an individual search on each tool for more detailed information on the tool's properties. If any other tools were mentioned during the individual searches, they were added to the list and an individual search for the newly identified tool was also conducted. The inclusion criteria for screening tools were as follows: 1. Suitable for use with children between 0 and 7 years of age. 2. Studies on the tool's use published in English. 3. Intended use is screening or rapid assessment, not formal diagnosis. 4. Targets at least one of the following developmental domains: motor, language, cognitive, socio-emotional, or behavioral domains. 5. Information on the tool's performance available for a minimum of four characteristics (e.g., screening domain, age range, format, and items/length).
Because our focus was on developmental monitoring, we excluded tests used for diagnostic purposes such as the Autism Diagnostic Observation Schedule, the Mullen Scales of Early Learning, or the Bayley Scales of Infant Development. However, search parameters were relaxed for tools developed for LMIC because of the limited evidence-base from many of these countries. Tools that were designed to screen for children with specific disabilities (e.g., hearing or vision impairment) and tools designed for specialist settings such as inpatient rehabilitation centers were also excluded, as the purpose of the review was to identify screening tools that could be used effectively in general or at-risk populations. Information on screening was not always optimally available; therefore, the decision to include a particular tool was based on current best knowledge. Following the individual searches, some tools were removed because they had been replaced by a newer, improved version. An example of this was the Kilifi Developmental Checklist, used in Kenya to screen for DD, which had been replaced by the Kilifi Developmental Inventory (Abubakar, Holding, Van Baar, Newton, & van de Vijver, 2008).
Profiles for each tool were then created in order to determine the tool's feasibility for use in LMIC. We gathered the information on screening instruments from several sources. We consulted test reviews and articles that describe the psychometric properties published in peer-reviewed journals, practice guidelines developed by professional societies, administration manuals, technical documents, and information from the test publishers or distributors. Profiles were populated with information, age ranges, whether the tool used a rater report (e.g., completed by parent or care provider) or direct assessment (e.g., observing the child's behavior), the instrument properties (number of items, type of response, reliability, and validity data) and information on cost, administration, and scoring. We also included information on the training involved in administration and the level of qualification required, if any. Where information was available on the tool's strengths and limitations, this was incorporated into the tool's profile as well. There was a considerable amount of contradictory information regarding some of the tools and their properties (e.g., time of administration, number of items, or the various training and administration requirements). In these cases, MM, MT, and CS came to a consensus about how to populate the profile.
Following this process, screening tools were divided according to those used to screen for ASD, more general DD and screening tools specifically developed for LMIC/ non-Western settings. The final set of tools were organized into four categories (DD screening tools developed for LMIC and non-Western settings; general DD screening tools; ASD screening tools for LMIC and non-Western settings and ASD screening tools). The tools were collated into a table, and each tool was assessed according to areas screened for, age range, tool format (rater report or observation), length of test or the number of items, and the training required in order to administer and score the test. Checkmarks (√) in the columns were used to represent the presence of the following criteria:

Specificity and Sensitivity data:
Tools that have both specificity and sensitivity data above 70% receive double checkmarks (√√). Tools with only one score above 70% received a single checkmark (√).

Sample size:
If a tool was studied in a sample of 300 participants or more, it received a checkmark (√). According to Bujang and Adnan (2016), a sample of 300 participants is a sufficient rule-of-thumb to determine the specificity and sensitivity of most screening tests.

Free:
If a tool is freely available for use, it received a checkmark (√). Tools that appear to be free (i.e., no purchase cost involved or tool described as low-cost), received a checkmark with an asterisk (√*) to indicate that it could potentially be implemented at no or low-cost outside of the research setting. 4. Used in LMIC/non-Western settings: If a tool has been adapted, validated, or developed for use in a low-or middle-income country, based on the World Bank classification of countries, it received a checkmark (√). Tools received a checkmark with an asterisk (√*) if the tool was designed for a non-Western setting or aboriginal populations within in a HIC.

CHWs:
If there was evidence in the literature that the tool has been used for screening by a lay community health worker (CHW), it received a checkmark (√).

Results
A total of 99 screening tools were included in the review ( Fig. 1). We identified 59 tools used to screen for more general DD, and 40 tools intended to screen for ASD. Thirty-five screening tools used to identify DD were developed specifically for LMIC/non-Western settings (Table 1), and 24 tools used for more general DD originated from HIC (Table 2). Only six ASD screening tools were developed specifically for LMIC/non-Western settings (Table 3), while the majority of ASD screening tools were developed in and for HIC (Table 4). Most tools have been developed in HIC (out of 58 screening tools from HIC-41 are from the USA and 3 from Canada). There are a number of screening tools used for DD from LMIC (35 tools), but ASD tools for LMIC remain limited (only six identified in our review). Tools used to screen for ASD in LMIC are often derived from existing tools: for example, the HIVA screening tool used in Iran (Samadi & McConkey, 2014 includes items from the GARS-2 and the M-CHAT screening tools, while the Three-Item Direct Observation Screen (TIDOS; Oner et al., 2013) used in Turkey to screen for ASD in young children, combines the parent-report items from the Social Communication Questionnaire (SCQ; Allen et al., 2007;Chandler et al., 2007;Oosterling et al., 2010;Snow & Lecavalier, 2008) with three observational items. The 23-item screener used in Uganda (Kakooza-Mwesige et al., 2014) is an adaptation of the Ten Questions Screening Instrument (TQSI; Durkin et al., 1995), including an additional 13 items to identify children with ASD and to increase screening capability for visual, hearing, and seizure impairments. The Pictorial Autism Assessment Schedule (PAAS; Perera et al., 2017) used in Sri Lanka was an attempt to overcome cultural barriers to identifying symptoms of ASD by adding a visual aid to facilitate the recognition of autism.

Psychometric Data
Tools varied significantly in their psychometric performance and feasibility. Most studies sought to assess whether the screening instrument could differentiate the ASD (or DD) group from other groups. Sensitivity and specificity analysis were also widely used (primarily using the ROC curve), although small sample sizes often prevented comprehensive reliability or validity testing of screening tools. Over 80% of screening tools for DD were studied in a sample of 300 or more, while 70% of ASD tools were studied in a sample of 300 or more. Only 45% of tools for DD had both specificity and sensitivity data above 0.7, while over 70% of tools for ASD had specificity and sensitivity data above 0.7.

Cost and Access to the Instrument
Most of the tests developed and licensed in HIC are strictly protected by copyright. Examples of such tools are the Battelle Developmental Inventory Screening Tool (BDI-ST; Elbaum et al., 2010;Glascoe & Byrne, 1993), the Baby and Infant Screen for Children with aUtism Traits (Matson et al., 2010;Matson, Fodstad, et al., 2009;Matson, Wilkins, et al., 2009), or the Checklist for Autism Spectrum Disorders (Mayes et al., 2013). The majority of screening tools developed in the USA require payment for use (e.g., the ADEC, ASRS-SF, CARS-2, GARS-3, GADS, KADI, PDDST, SCQ, SRS-2, STAT, ASQ-3, BDI-ST, BINS, BITSEA, BRIGANCE-II, DDST, ECI-4, ESI-R, Greenspan, or PEDS). In many cases, a licensed psychologist is the only person that is permitted to purchase the tests from the publishing companies. Copyright laws prohibit any use of the tests (including photocopying) without explicit permission or purchase, which prevents many researchers working in LMIC from using these standardized tools. Furthermore, translation is not allowed without additional approval. Costs are often prohibitive for use in low-resource settings and screening at population level. A few exceptions that are freely available for download include the AQ, ASAS, A-TAQ, Childhood Asperger's Syndrome Test (CAST), M-CHAT R/F, ITC, POSI, SSI, BPSC, EDI, ESSENCE-Q, PPSC, PSC, and SWYC.

Adaptation and Translation for Use in LMIC
Methods used to translate or revalidate screening tools for different settings varied widely. Some tools developed in HIC have been adapted for use in LMIC, such as the ASQ, PEDS, and M-CHAT screening tools: The ASQ has been used in India (Chaudhari & Kadam, 2012;Juneja, Mohanty, Jain, & Ramji, 2012), Taiwan (Tsai, McClelland, Pratt, & Squires, 2006), Brazil (Filgueiras, Pires, Maisonette, & Landeira-Fernandez, 2013), Turkey (Kapci, Kucuker, & Uslu, 2010), Thailand (Saihong, 2010), and Iran (Vameghi et al., 2013). The PEDS has also been used to detect DD in LMIC (Woolfenden et al., 2014), and has been translated for use in Tanzania (Kosht-Fedyshin, 2006), India (Malhi & Singhi, 2002), Thailand (Theeranate & Chuengchitraks, 2005), and Indonesia (Gustawan & Machfudz, 2010). The M-CHAT remains one of the most widely used screening tools for the detection of autism and has been translated for use in Mexico (Albores-Gallo et al., 2012), Albania (Brennan, Fein, Como, Rathwell, & Chen, 2016), nine Arabic speaking countries (Seif Eldin et al., 2008), and Sri Lanka (Perera et al., 2009). However, in Sri Lanka, effort was made to examine the tool rather than just use it, and the M-CHAT demonstrated unacceptably low specificity (Perera et al., 2009). For an extensive review on the modification and adaption of tests for use in lower-income settings than those of the population the tests were standardized on,

Training and Use by CHWs
We also included information on the training involved with administering screening tests. For screening tools that use a parent report format, this may seem arbitrary.
For example, studies that used tools such as the First Year Inventory (Reznick, Baranek, Reavis, Watson, & Crais, 2007;Watson et al., 2007) or the Childhood Asperger's Syndrome Test (CAST; Scott et al., 2002;Williams et al., 2005) mailed the questionnaires to parents, and therefore no training was conducted for administration of the tool. Nonetheless, information on training procedures or stipulations about who can administer and score screening tests was an important consideration for this review, since we were looking specifically for tools that can be used by lay health workers in LMIC.
Of the 99 tools, only 26 had been used by CHWs and most of these were developed for LMIC. In HIC, administrators of screening tests are usually required to complete training on how to administer and score the test and are often professionals who regularly interact with children in some capacity (e.g., pediatricians, psychologists, or teachers). However, other personnel with relevant backgrounds (community health workers, social workers, etc.) can also be trained to conduct these tests (Fernald et al., 2009), even though there is limited literature available on tools from HIC used by lay health workers.

Selected Tools for Use in LMIC
From the tools included in the review and indexed as per the above indices, we selected 10 tools that adhered most closely to our feasibility criteria to screen children for ASD or DD in LMIC (Table 5). We selected tools that: • Take 30 min or less to administer; • Cover multiple domains of development; • Are free to access and can be implemented at low cost; • Can be implemented by paraprofessionals or lay community health workers; • Have successfully been used/easily adapted for use in more than one LMIC.
For the screening and detection of ASD specifically, we identified three tools, namely the Modified Checklist for Autism in Toddlers, Revised with Follow-up (Robins et al., 2014), the PAAS, (Perera et al., 2009(Perera et al., , 2017 and the TIDOS (Oner et al., 2013). To identify children with, or at risk of DD, we selected seven tools for use in LMIC Tools that appear to be free (i.e., no purchase cost involved or tool described as low-cost), received a checkmark with an asterisk (√*). Tools received a checkmark with an asterisk (√*) if the tool was designed for a non-Western setting or aboriginal populations within in a HIC. Tools that appear to be free (i.e., no purchase cost involved or tool described as low-cost), received a checkmark with an asterisk (√*). Tools received a checkmark with an asterisk (√*) if the tool was designed for a non-Western setting or aboriginal populations within in a HIC. et al., 2010, 2013, 2014); TQSI (Durkin et al., 1994(Durkin et al., , 1995(Durkin et al., , 1998Thorburn et al., 1992); Caregiver-Reported Early Development Index (CREDI;McCoy et al., 2017); INTERGROWTH-21st Neurodevelopment Assessment (Fernandes et al., 2014), and the 12-month screener (Biasini et al., 2015). The Engle Scale and Survey (Verdisco et al., 2015) and the East-Asia Pacific Early Child Development Scales (EAP-ECDS; Rao et al., 2014) have been identified as promising tools, although limited information in the peer-reviewed literature is currently available.

Discussion
Monitoring child development through screening in LMIC can provide valuable data on rates of developmental difficulties in order to ensure interventions can be appropriately targeted, their effect monitored and the need for further interventions determined (Engle et al., 2007;Mung'ala-Odera & Newton, 2007). Identifying atrisk and affected children should be a key priority, especially for countries where children with DD or disability frequently remain undetected and untreated. The World Health Organization (WHO, 2012(WHO, , 2013 has stated that developmental monitoring needs to be integrated in routine maternal and child health care, in the context of growth monitoring, early childhood development and provision of comprehensive care for children with specific needs and their families. In most LMIC, developmental surveillance is currently not a common feature of health service delivery, and there is a lack of standardized practice in screening of DD and ASD. A focus on improving the scientific rigor of early detection approaches and on enhancing the reach of such approaches to underserved populations should be prioritized (Daniels, Halladay, Shih, Elder, & Dawson, 2014). The purpose of this review was to identify available tools from the literature used to screen children for ASD or more general DD, in order to make recommendations for tool selection and use in LMIC. The information on available tools provided here could inform decisionmaking related to developmental monitoring in LMIC, while considering heterogeneous realities, available resources and local health systems' capacities within different LMIC. We included over 90 different screening tools in our final review, and consolidated information on their properties to determine which tools could be effectively used for screening of either ASD or DD in various LMIC. An important challenge in early identification of developmental disability is having tools that respond to local differences, including cultural perceptions in meaning of disability and that can be used across countries (Fischer et al., 2014). As a result of the many challenges in determining cross-cultural validity of tests  Tools that appear to be free (i.e., no purchase cost involved or tool described as low-cost), received a checkmark with an asterisk (√*). Tools received a checkmark with an asterisk (√*) if the tool was designed for a non-Western setting or aboriginal populations within in a HIC.    developed in HIC, screening tools developed in local areas of study have accelerated, focusing on questions and testing methods that are culturally appropriate for children in LMIC (Semrud-Clikeman et al., 2017). Our review identified a substantial number (35 for DD and 6 for ASD) of screening tools from LMIC. We identified 10 tools which show promise for use across settings in LMIC. Three tools specifically for ASD (M-CHAT-R/F; PAAS; TIDOS) and seven for more general DD (CREDI; GMCD; INTERGROWTH-21st Neurodevelopmental Assessment; MDAT; RNDA; TQSI; 12-month screener) were selected. These tools most adequately adhered to our feasibility criteria to screen children for ASD or DD in these settings. Furthermore, the newly developed Engle Scale and Survey (Verdisco et al., 2015) and the EAP-ECDS (Rao et al., 2014) also show promise, although to the best of our knowledge no peer-review publications are currently available.
Despite its potential benefits, screening presents numerous challenges. In LMIC, many children do not regularly see medical or mental health professionals in the early years, making regular screening or surveillance difficult (Biasini et al., 2015). Community health workers have limited knowledge about age-appropriate developmental milestones and early warning signs, which means that problems are often only picked up when children come in contact with the primary health care system. Also, primary care staffs often have limited training and experience in recognition of early neurodevelopmental delays (Lian, Ho, Yeo, & Ho, 2003). The use of formal screening tools as part of developmental surveillance can assist health workers in this regard, but training and supervision need to accompany screening for it to be effectively implemented. Screening tools, including parent-report tools, should involve training and supervision for staff, particularly in terms of providing feedback of the screening results to caregivers. Given the human resource shortages in most LMIC, training community health workers to conduct screening and developmental surveillance is essential.
When selecting an existing screening tool, policy makers, researchers, and interventionists must consider its affordability, feasibility, and cultural appropriateness for the intended setting. The selection and validation of an appropriate screening tool requires considerable time and effort, research personnel, and financial resources , and the adaptation process is more complex than simple translation. Determining the psychometric properties of a tool in a new context is expensive and requires research expertise and capacity. Tools comprised of a large number of items and that take more than 30 min to administer may further limit its feasibility for low resource settings. A large number of tools included in this review had over 100 items, challenging their usefulness for brief screening. In terms of  √√ *Level-2 screening tool; Tools that appear to be free (i.e., no purchase cost involved or tool described as low-cost), received a checkmark with an asterisk (√*). Tools received a checkmark with an asterisk (√*) if the tool was designed for a non-Western setting or aboriginal populations within in a HIC. can be used to screen for ASD TQSI (Durkin et al., 1994(Durkin et al., , 1995 Multiple LMIC administration, combining rater report with observation items in a screening tool may be beneficial for LMIC settings, given that both rater report and direct administration methods have drawbacks. Caregiver or parent reports may not be as reliable in LMIC due to poor literacy levels, lack of knowledge about milestones and the possibility of parents providing socially acceptable responses for fear of social stigma (Fernald et al., 2009;Robertson, Hatton, & Emerson, 2009;WHO, 2012).
Checklists about milestones and caregiver concerns may not be sufficient to identify developmental disabilities in LMIC (De Lourdes et al., 2005). Although several observational or direct assessment screening tools have been developed, they may be too costly in time and effort for wide-scale use (Barton et al., 2012). Routine screening will not be a panacea to the problem of non-detection. Not all children who screen positive for a DD or disability will be diagnosed, and not all children who screen negative are certainly clear of a diagnosis (Sheldrick & Garfinkel, 2017;Veldhuizen, 2017). If a child is screened and it is decided that they need to undergo formal assessment, there are very few specialists available who can make these assessments and reach a diagnostic decision. For example, in South Africa, families will typically wait 18 months for a basic diagnostic assessment for ASD in a specialist clinic (de Vries, 2016). Finally, linking screening and diagnosis with appropriate treatment services does not exist in many settings. If treatment and intervention is not available, screening may seem futile, especially to families and care providers (Collins et al., 2017). However, screening may provide crucial data as a means to understand the disease burden in order to plan and then monitor services. Routine screening is an important first step toward addressing the need for services in LMIC.

Limitations
Only publications in English were considered for inclusion, which may limit the generalizability of the findings. Given the large proportion of LMIC that do not have English as a primary language, it is possible that some promising tools may have been missed in this review. Second, tools were included in the review regardless of the size and quality of studies on screening tools. However, to account for this limitation, we included information in the tables on the sample size and specificity and sensitivity data reported in the studies. The search terms used in this review was broad, which means that tools designed for more specific delays or other neurodevelopmental disabilities may have been excluded. Finally, we included screening tools designed for population-level assessment, as well as for individual screening.
It should be noted that even the recommended tools have limitations. Previous studies using the MCHAT in Mexico (Albores-Gallo et al., 2012) and Egypt (Mohamed et al., 2016) have noted that there are cultural differences in responses, which may limit its acceptability for use in LMIC. However, we are recommending the MCHAT-R/F, which includes a simplified scoring procedure, paired with a flow chart with open-ended follow up questions that facilitate a second-stage screening process. The TQSI is only for children over the age of 2 years and has limited sensitivity for less severe disabilities. More research is needed on its use in more subtle DD. The RNDA has mixed sensitivity and specificity in the younger age group, and more research is needed from other countries. Although the MDAT has shown good sensitivity and specificity, it takes between 30 and 40 min to apply.

Conclusion
We suggest that great care needs to be taken when considering tools designed for research settings or diagnostic purposes as part of developmental monitoring efforts. This review was positioned broadly, in order to present findings of use to policy makers and interventionists considering screening as part of developmental monitoring in LMIC. Screening should ideally be conducted at two levels-routine general screening followed by a structured interview for those whose scores exceed a locally validated cut-off point. The adoption of strengths-based assessment and bio-psychosocial approaches whereby assets and risks in the family and broader environment are considered, and families are empowered with appropriate knowledge, skills and support, are recommended. An approach such as this will require substantial health system changes in most LMIC in order to deal with the scarcity of financial resources, low numbers of health workers skilled and trained in ASD and DD, cultural barriers to identification and the increasing costs of training. It will be important to remain mindful that high-quality tools take time to be conceptualized, developed, piloted, and validated, before implementation can happen. To do this, we will need expert centers across the globe that can compare novel instruments against "gold-standard" instruments. We should not risk introducing inferior quality tools into low-resource environments (de Vries, 2016). We believe that these profiles may assist researchers and practitioners to evaluate whether a developmental screening tool is appropriate, affordable, and feasible, while highlighting where further research or reporting is needed.