Language Development

We approached the same panel members who had formed part of the CATALISE consortium for our previous Delphi on criteria. As detailed by Bishop et al. (2016), we restricted consideration to English‐speaking countries, and there was a predominance of speech‐language therapists/pathologists (SLT/Ps). Of the original panel, two declined to take part in CATALISE‐2 for personal reasons, leaving a panel of 57 individuals, whose characteristics are shown in Table 1. Nine panel members had a close relative with impaired language development.
The first two authors (DVMB and MJS), both psychologists with considerable experience in the area of children's language problems, acted as moderators: they did not contribute rankings, but agreed on modifications to statements on the basis of feedback from the panel. The third author (PT) set up the online Delphi, controlled the anonymisation and analysed responses to produce reports for panel members. The fourth (TG), an expert in primary health care who was familiar with the Delphi method acted, as methodological advisor.

Two groups of parents took part in the study: parents of children with a diagnosis on the autism spectrum and parents of families who only had children who were developing typically. The parents of children with an ASC were recruited from families who registered on the Cambridge University Autism Research Centre volunteer database between 2002 and 2009 (online at http://www.autismresearchcentre.com). Ethics approval for the database and questionnaire collection was provided by the Cambridge Psychology Research Ethics Committee. Parents gave informed consent to take part in the study electronically.
Parents registering on the database are asked to state the diagnosis of their child, who made the diagnosis, and where and when. In addition, parents are asked to provide information about the IQ and language development of their child, although this is not obligatory. After registering, the parents are invited to complete an AQ, and are asked to encourage the other biological parent to also register and complete an AQ. The advantage of using an online website to collect these data is that large samples can be collected. The disadvantage is that diagnosis cannot be validated in every case. All parents included in this study reported that their children had received their diagnosis from experienced clinicians in recognized clinics and according to the criteria from the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV) or the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10).
In total, 1582 families took part, with 571 fathers and 1429 mothers completing the AQ. There were questionnaires from 418 couples. The mean age of the fathers for whom these data were available (n = 551) was 44.0 ± 6.9 years and for mothers (n = 1228) was 41.2 ± 7.5 years. The numbers of diagnosed and developing typically children in the 1582 families are shown in Tables 1 and 2. There were 1752 children with a diagnosis of ASC:1472 male and 280 female, giving a male:female ratio of 5.3:1. of he 1752 children, 727 were reported to have autism, 725 to have Asperger's syndrome (AS), 185 to have high functioning autism (HFA) and 115 to have pervasive development disorder. As a further check on diagnostic subtype, the registration process asks about comorbid language delays or learning difficulties. A diagnosis of AS is only registered if the parent indicates that the child did not have any language or general developmental delay. A diagnosis of HFA is only registered if the parent indicates that there was no general developmental delay even if language was delayed.
For the control group, the AQ was sent to the parents of 1255 children (633 girls, 622 boys) who were participating in a large epidemiological study of social and communication skills in primary school-age children [30 (link)-32 (link)]. This part of the project was ethically approved by the National Health Service Suffolk Research Ethics Committee, and written informed consent was obtained from participating parents. Two copies of the questionnaire and two consent forms were posted to each family: one set for the mother and one set for the father. A post-paid envelope was included for participants to return the questionnaires and consent forms. This sample was originally ascertained by inviting 136 mainstream primary schools in Cambridge City, east and south Cambridgeshire, and Fenland in the UK to participate in research (n = 92 (68%), agreed). In total, 1012 questionnaires were received from 669 families (661 from mothers, 351 from fathers) with 343 pairs, a 40% response rate. Two families were excluded because they reported a child with a diagnosis or suspected diagnosis on the autism spectrum. One family was excluded because they reported a child with Down's syndrome. Questionnaires from 666 families (558 mothers and 349 fathers) with 341 pairs were analysed. The mean age (at the time of completing the AQ) of the fathers for whom these data were available (n = 344) was 44.6 ± 5.0 years and for mothers (n = 644) was 42.4 ± 4.7. The children in these families were, according to parental report, all developing typically, and the numbers are shown in Table 3; in total, there were 1532 children in these families.

In November 2002, UNICEF convened a panel of 25 international experts (Expert Panel I) with expertise in human development, anthropology, nutrition, and measurement to (a) develop a framework of domains of family care practices and resources important for young children's development, (b) evaluate possible items to use in pilot testing, and (c) define priorities for testing. The selected family care domains and items were culled from the multidisciplinary literature on the practices and resources identified as important for the motor, social, emotional, cognitive, and language development of young children, and caregiver resources.
The majority of candidate items were selected from instruments that have shown good psychometric properties across a variety of samples in the USA (e.g. HOME; Early Childhood Longitudinal Study measures, see www.nces.ed.gov/ecls/; National Household Educational Surveys, see www.nces.ed.gov/nhes/) or in developing counties (46 -48 ). Where no suitable candidate items could be found in the literature, these were suggested by panel members with expertise in that domain.

Although the field is at an early stage of understanding sex differences in autism, there is considerable evidence that they exist in a variety of biological and cognitive domains (Ferri et al., 2018 (link)). Moreover, there are known sex differences in language development, even if they are not as great as previously believed (Etchell et al., 2018 (link)).

Parental education is known to affect both language development (Roberts et al., 1999 (link)) and social development (Hediger et al., 2002 (link)), and therefore may confound hypothesized relationships between the two.

Data were extracted from an existing electronic healthcare record (EHR) database housed at a research university’s school of nursing and developed in partnership with a large academic healthcare system in the Southern United States.^{[39 ,40 (link)]} The EHR database contains 811,771,561 unique electronic records from 1,089,586 unique patients, which is a 17% simple random sample of all patients seen in the healthcare system between 2012 and 2017 with no restrictions on demographics, diagnoses, or services received. The records were taken from the Clinical Data Warehouse and deidentified to remove all protected health information through shifting dates, truncating zip codes, anonymizing patients identification numbers, and removing patients names and contact information in compliance with safe harbor methods by the data solutions group.^{[41 ]} The Emory University Institutional Review Board provided a waiver for this study of deidentified data.
The database was accessed via the relational database management system PostgreSQL. EHR data regarding patients demographic information, healthcare services, laboratory results, medication histories were extracted from the database into comma-separated values files and analyzed using RStudio, and an integrated development environment for the R programming language.