Introduction
Autism is a global phenomenon. An estimated 1–2% of children worldwide lie on the autism spectrum, with approximately 52 million autistic
1 individuals across the globe (Baxter et al.
2015). These estimates are largely driven, however, by prevalence estimates from high-income countries (HIC). Virtually no data exist on the population prevalence of autism in low-income countries (LIC) and only one in a LIC rural setting (Uganda: Kakooza-Mwesige et al.
2014). In fact, there is a paucity of research examining autism generally in these regions at all (Abubakar et al.
2016). Studies that have been conducted in lower-middle and upper-middle income countries have produced varied results reporting prevalence estimates ranging from 0.32 (China: Tao
1987) to 250 per 10,000 (China: Ren et al.
2003) and more recently 90 per 10,000 (India: Raina et al.
2017). These discrepancies are possibly due to a variety of reasons, including the fact that autism is a spectrum condition (American Psychiatric Association
2013), the variety of traits; changing definitions of autism; varying levels of awareness in different countries; cultural variation in expectations and understandings of children’s behavior; different methodological approaches used to assess prevalence; and the lack of availability of culturally-sensitive diagnostic tools and year of assessment (Elsabbagh et al.
2012).
Variation in methodological approaches to assessing autism prevalence includes differences in case-finding techniques, from population-based sampling (Baird et al.
2006) to sampling from clinics and healthcare registers (Croen et al.
2002). Variation is also found in the method of diagnosing or screening cases for autism, including a mix of relying on healthcare or educational reports (Idring et al.
2012) and/or researchers assessing for autism first-hand using comprehensive diagnostic tools including the Autism Diagnostic Observation Schedule (ADOS) and the Autism Diagnostic Interview (ADI) (Kim et al.
2011), to less sensitive screening tools or questionnaires, based on different standards and diagnostic criteria (Stewart and Lee
2017). Furthermore, while many studies have investigated the prevalence of autism in Western societies, the general consensus is that there is an urgent need for more studies investigating the prevalence of autism in LICs (Abubakar et al.
2016; Elsabbagh et al.
2012; Mpaka et al.
2016; Ruparelia et al.
2016), in particular LIC rural settings using low-cost, community-based screening tools that can be administered by lower cadres of health workers. Without these tools and studies, it is impossible to draw an accurate prediction of the global prevalence of autism and develop appropriate services that can cater for the needs of autistic children and adults and their families.
Sixty-one percent of the world’s population of children and young people live in LICs or lower- to middle-income countries (LMIC). Nepal is one of the poorest LICs (World Bank
2016) and one of only four LICs outside of sub-Saharan Africa. The majority (estimated 82%) of its population lives rurally (World Bank
2014). Nepal has a growing population of about 26.7 million people due to high birth rate and declining death rate. The Autism Care Nepal Society website states that there is “no reliable estimate for the prevalence of autism in Nepal as autism is not known to many people” (Autism Care Society Nepal). Indeed, recent qualitative evidence from this population shows limited understanding of many aspects of atypical child development, in particular autism—but a strong desire to advocate for and increase support for these children and their families (Heys et al.
2016).
The overarching aim of the current study was to establish the prevalence of autism in school-aged children living in Nepal. In a preliminary screening-adaptation and acceptability study, we first sought to identify a low cost, short, population-based screening tool with good sensitivity and specificity that could potentially be delivered by all cadres of health care workers to detect possible autism in children and young people in a rural Nepali setting. We then sought to develop a Nepali-language adaptation of the identified screening tool, the AQ-10 (Allison et al.
2012), which would be acceptable for Nepali parents.
In the prevalence study, we used the resulting tool to estimate the prevalence of autism in 9–13 year-old Nepali children living in Makwanpur District, a rural hill area in the central region in which most households are dependent on subsistence agriculture (population > 500,000 in 2014). We also assessed the validity of the adapted AQ-10 within the same population by comparing results of a screening tool for childhood disability, which included questions on social and communication skills, and was delivered 6 months previously to the same cohort of children.
Our research aims were twofold: (1) to examine whether the AQ-10 is acceptable for use with Nepali parents and (2) to determine the estimated prevalence of autism in a rural population of Nepali children using the adapted AQ-10.
Discussion
This is one of only two published studies to date to have estimated the prevalence of autism in a rural LIC setting. The study showed that the adapted AQ-10 was acceptable to groups of parents of children both with and without a known diagnosis of autism. Of the 4098 children sampled, 14 scored positive for autistic symptomatology. The demographic, anthropometric, clinical and educational characteristics and gender ratio of these 14 children indicate that the AQ-10 administered with a threshold score of 6 or above in this population is likely to identify children with complex needs and more likely more severe autism. 66 out of 422 children screened positive for the three questions most intuitively associated with autism symptomatology from the MCFD disability screening tool. A comparison of scores from the AQ-10 to those from the MCFD administered 6 months earlier provided some evidence of the clinical validity of the AQ-10. Likewise, the comparison of scores from these two tools provided evidence of the potential of MCFD, at least the section pertaining to child behavior, to identify successfully children within the population with atypical child development and behaviors.
There is no population-based study with which to compare these AQ-10 scores. The original paper describing the AQ-10 as a short screening tool was tested on UK cases (autistic children and adults) and controls (children and adults without a known diagnosis of autism) (Allison et al.
2012). The autistic adolescents scored a mean of 8.40 (SD 1.69) and adolescent controls a mean of 1.78 (SD 1.80). The AQ-10 scores for UK cases were therefore not dissimilar to those from this cohort who screened positive with AQ-10 scores (M = 7.9; SD 1.5). However, the Nepali population mean scores were much lower than those of the control group tested in the UK study.
If the AQ-10 screening tool is as sensitive and specific in the Nepali population as it is in the UK, the current results would give an estimated true prevalence of 3 in 1000 (95% confidence interval: 2–5 in 1000) (Brown et al.
2001). The current population of children under 18 years in Nepal is a little over 11.6 million. If confirmed, these prevalence estimates would equate to 34,803 children and young people currently living in Nepal (range 23,203–58,007) with a potential diagnosis of autism. The number of children with a current diagnosis of autism (n = 107); Kathmandu Valley, 2012 estimates (Autism Care Society Nepal) is substantially lower than this figure. This estimated prevalence of 3 per 1000 is lower than that in HICs, which is reported as 10–20 per 1000 (Elsabbagh et al.
2012), and lower than the only other LIC study showing Ugandan population estimates in 1–10-year-olds to be 12–13 per 1000 (Kakooza-Mwesige et al.
2014). Nevertheless, this estimate is higher than global median estimates (1.7/1000) and estimates from a recent study in rural India 0.9 per 1000 for all children in the full cohort of 11,000 children aged between 1 and 10 years and 1.1 per 1000 in the rural cohort (Raina et al.
2017). In the latter study, urban rural abode and higher socio-economic class were associated with reduced prevalence. The closest areas geographically, other than India, in which there is some (but still not a significant amount of) research on autism prevalence available are Sri Lanka and Indonesia. One, relatively old Indonesian-based study estimated 1.7 cases of autism per 1000 of the population (Wignyosumarto et al.
1992). In comparison, a study based in Sri Lanka placed the estimate as high as 10 per 1000 (Perera et al.
2009).
Of those children who screened positive for autism symptomatology, almost all also screened positive for physical, learning and behavioral disabilities. Given these two screening tools were delivered approximately 6 months apart, this finding provides preliminary evidence that our modified AQ-10 is a valid measure of atypical child behavior. It also suggests that those children who screened positive for autism symptomatology on the AQ-10 have evidence of multiple impairments and so most likely represent the severe end of the spectrum of potential clinical presentations of autism. This is also reflected in the gender ratio (1.4:1, M:F), showing a much higher number of females than is typically reported in the full spectrum of autism, around 4:1 (Fombonne
2009); though see (Loomes et al.
2017). In autistic children who are cognitively less able, the gender ratio falls from 10:1 to be closer to 1:1 (Volkmar et al.
1993). It is therefore likely that our estimated prevalence of 3 per 1000 is an
underestimation of the true prevalence of autism within this population. This is also reflected in the number of children (66 of 4222) screening positive for difficulty in at least one of the areas of coping with changes, social relationships and playing with others. These questions have not yet been validated or indeed tested in any way as a screen for autism. Yet if it were assumed that a positive screen was indicative of likelihood of autism, this figure would equate to an estimated prevalence of 16 per 1000, much more similar to HIC estimates and those from Uganda. Indeed, the Ugandan study used an adaptation of the Ten questions questionnaire that was the basis for the MCFD used in this study (Kakooza-Mwesige et al.
2014).
In addition to being more likely to screen positive for physical, learning and behavioral disabilities, children who screened positive for autism symptomatology were more likely to be stunted and to have cognitive difficulties (as measured by marked difficulties completing the forward and backward digit recall, a measure of working memory). The majority of these children were not attending education and were unlikely to be receiving any financial support despite having significant difficulties. An 83% stunting rate is substantial. Nutritional deficits in children with disabilities and learning difficulties are common and can not only be a cause of cognitive deficits, but also contribute to the failure to reach full developmental potential in the presence of a developmental condition. These limited, but important descriptive data support the general opinion that children with all kinds of atypical child development and disability are a highly vulnerable, disadvantaged group, especially in resource-limited settings such as Nepal (UNICEF
2013).
The strengths of our study include (1) its novelty—to our knowledge there is only one other published study of prevalence estimates in a LIC; (2) the interview of families within households, thus including children who were not present in school; (3) our restricted age range, thus rendering age range less of a confounding variable; and (4) our cultural adaptation of the screening tool using qualitative methods in collaboration with local stakeholders (Stewart and Lee
2017). Also, to our knowledge, no other study has incorporated cultural constructions of mind/emotions and local notions of childhood/children in the translation and piloting of the AQ-10 or similar tools. We also situated our prevalence estimate within the context of other important data, including disability screen, school attendance and growth with which to explore characteristics of those screening positive for AQ-10.
There are, however, limitations to our study. First, the cohort is derived from mother-infant dyads who were enrolled in a perinatal trial. Nevertheless, children who screened positive for the AQ-10 were equally as likely to have been born into the villages enrolled in the intervention as those villages enrolled in the control. Second, the MCFD is under development and indeed since the commencement of this study minor edits have been made to the questions. In addition, early testing in India and Cameroon, coupled with the prevalence found here, suggest cultural interpretation of degree of difficulty may greatly influence reporting (Mactaggart et al.
2016). For instance, prevalence of reporting
at least some difficulty in at least one domain was 35% and 64% in India and Cameroon, respectively, whereas prevalence of reporting
a lot of difficulty or
cannot do in at least one domain was 4 and 9% again in India and Cameroon, respectively (Mactaggart et al.
2016). Thus, our comparison of MCFD-extended questions with AQ-10 here should be considered exploratory. Notably, however, our prevalence estimate of 7.4% is in keeping with a systematic review of the global estimates of childhood disability in LMICs which suggested that despite a wide range in estimates (0.5–18%), the majority clustered around 5–10% (Maulik and Darmstadt
2007).
Future research is required to validate the AQ-10 and the MCFD-core and -extended modules through in-depth comprehensive assessments of high-scoring children and a representative sample of low-scoring children. Given the wide range of perinatal data available for this cohort, we also have the unique opportunity to conduct an exploratory study around the association of perinatal factors with likelihood of autism symptomatology in a population of children with poor nutrition (40% stunting at mean age of 11.5 years). Such prevalence research, however, also raises serious ethical concerns, including the possibility of disclosure of likelihood of and/or even a diagnosis of autism in a population for whom there is no term for autism (see Heys et al.
2016) and there is very little health or education provision to support children and families with a diagnosis of autism. These issues surrounding the ethics and logistics of diagnostic disclosure were precisely the ones that our focus group participants perceived to be of concern. There are also no validated diagnostic tools available for use in the Nepali population, rendering a validation study even more challenging. Thorough examination of these issues with Nepali parents and practitioners is essential prior to pursuing further the validation study of the AQ-10, including examining the potential impact of disclosure of a diagnosis of, or likelihood of a diagnosis of autism about which very little is known in most resource-poor settings like Nepal. Finally, future research in this area should explore the potential impact of caste/ethnicity and rural/urban divide on understanding of autism and its implications.
In our qualitative study of Nepali parents’ and professionals’ understanding of typical and atypical child development (with an emphasis on autistic symptomatology), we found that parents of children without a diagnosis of autism and professionals in general had little explicit awareness of autism (Heys et al.
2016). Only parents of autistic children, pediatricians and the disability sector worker identified behaviors typically associated with autism as ‘autistic’. Other participants, including parents of children without an autism diagnosis, primary and early child development teachers, community health workers and faith healers, used distinctive terms, such as “stubborn” and “insisting” to distinguish vignettes of autistic children from vignettes of children with other developmental conditions. Most participants felt that environmental factors, including in-utero stressors and birth complications, parenting style, and home or school environment, were key causes of atypical child development and further called for greater efforts to raise awareness and build community capacity to address autism. Thus, the preliminary prevalence estimate reported herein, combined with complementary research showing the lack of awareness of autism by Nepali professionals and parents, demonstrates a substantial unmet need and stresses the importance of developing services to support families and children with atypical development in LIC settings.
Acknowledgments
We are grateful to all of the participants who so generously gave up their time to take part in the study, to Autism Care Society Nepal for hosting training days for the researchers and to the Mother and Infant Research Activities, Nepal, senior research team: Mangalmaya Manandhar, Rita Shrestha, Dhurba Adhikari, Jyoti Bamjan, Bharat Budhathoki, Dej Krishna Shrestha and Sagar Khadka. We also thank Dr. Melissa Neuman for her data cleaning of the original trial dataset and Claudia Cappa, Statistics and Monitoring Specialist Statistics and Monitoring Section, Division of Policy and Strategy, UNICEF, USA and Mitchell Loeb from the Office of Analysis and Epidemiology, National Center for Health Statistics, USA, and the UN/Washington Module on Child Health Functioning group for sharing the Module on Child Functioning and Disability (MCFD) with us and Prof. J. Katz for sharing their group’s Nepali translation of the Ten questions questionnaire and Amy Alexander for her help with analysis of the qualitative data. The facilitator training and focus group work for this project was funded by a grant awarded to M.H., E.P., and A.C. from the Institute of Education (IOE)/UCL Strategic Partnership Incubator Fund. M.H. was funded by the National Institute for Health Research, UK, during this project. E.P. was funded by a Philip Leverhulme Prize from the Leverhulme Trust, UK. Research at the Centre for Research in Autism and Education (CRAE) is supported by The Clothworkers’ Foundation and Pears Foundation. The study was funded by a Wellcome Trust strategic award. Ref: 085417/Z/08/Z, Project Title: “Population Science of Maternal and Child Survival.”