Identifying language disorder in bilingual children aged 2.5 years requires screening in both languages

2.1 Participants

Children were recruited from three child health centres in the city of Gävle, Sweden. The child health centres were located in areas where about 40%-50% of the families were foreign-born. Gävle has 75 000 inhabitants, including 7000 children aged 0-5. A total of 129 bilingual children were recruited consecutively from November 2015 to June 2017, at the time of the regular 2.5-year health visit. Only children born after gestational week 37 and without any known disability were included. Children whose parents had Swedish as mother tongue were excluded. Before the clinical validation, 18 children dropped out (Figure 1). The remaining 111 (51% girls) were between 29 and 33 months old at the time of screening. Parental languages were Somali (n = 42), Arabic (n = 40), Kurmanji (n = 11), Farsi (n = 7), Sorani (n = 6) or Turkish (n = 5). All children except one attended preschool, but 54% of them did not attend regularly. Sixty per cent attended preschools, where many different languages were represented (Table 1). The mother's educational level was compared with that of all foreign-born women in Sweden with a permanent residence permit. The maternal educational level in the sample was significantly lower than that of foreign-born women aged 24-44 in the general population (P < .01).¹⁸

2.2 Material

The screening consisted of an assessment of language comprehension and an observation of language production. The screening was conducted in the child's mother tongue and in Swedish. The results were analysed for this study according to four models: the child's mother tongue (model one), Swedish (model two) and a combination of both languages (model three). Lastly, screening in Swedish was combined with parental reports regarding the child's language development in mother tongue (model four). In order to minimise the learning effects between tests, the comprehension items were replaced by new items of equal difficulty translated into the six languages included in the study (Table 2). The child indicated understanding with a gesture or verbally. There were no formal grammatical requirements for two-word utterances because grammatical parts are often given by the context. All type of constituents, which the child put together in order to express itself while interacting with others, were approved as two-word utterances. For example, ‘daddy car’ may stand for ‘daddy is driving the car’, ‘Up mummy’ stand for ‘mummy lift me up’ and ‘no eat’ may stand for ‘I don't want to eat’ depending on the context. However, echoing and frozen phrases, meaning a fixed combination of two or more constituents, which did not occur separately, like ‘good morning’ or ‘bye-bye’ did not count as two-word utterances. Combinations of words or constituents in two languages were approved. In accordance with existing clinical routine, parents completed a questionnaire with focus on production and comprehension in the child's mother tongue (Table 2). Intelligibility of the child's speech in both languages and parental concerns about the child's language development were also recorded. The test administrators also recorded the child's ability to cooperate during the comprehension part of the screening.

The examination by the speech and language pathologist (SLP) included a structured observation during a play session of the child's ability to communicate and talk in multiword utterances. To examine language comprehension, the Swedish version of the receptive part of Reynell Development Language Scales III^{19, 20} was applied, here, referred to as Reynell. In order to minimise variations in interpretation, the questions in Reynell were carefully translated into the six languages included in the study. Preschool teachers provided information on the child's language development, ability to play and participation in everyday social activities. The parents reported on the child's medical background, family history, and early language and communication development.

2.3 Procedure

The screening in Swedish was performed by 10 nurses who had been working in child health care for at least 1 year. Furthermore, 12 bilingual preschool staff were trained to be able to perform the screening in the child's mother tongue without the nurse's participation. They also helped to interpret the test items during the clinical testing of the child in his/her mother tongue, which was performed by the SLP. The nurses and bilingual staff took part in a 2-day workshop on DLD and screening, including practical exercises.

The screening took place during two sessions within 2 weeks at the child health centre. At the first visit, the nurse performed the screening in Swedish. The parents answered questions about the child's language development in both languages. At the second visit, one of the bilingual staff performed the screening in the child's mother tongue. The parents answered additional questions about their education, the child's preschool setting and the child's exposure to both languages. A research assistant received all screening protocols and thereafter blinded the identity of the child. The blinded protocols were forwarded to the SLP (first author, LN), who scored the protocols according to the screening criteria (Table 2). In order to pass the screening, the child was required to pass the comprehension and the production criteria (Table 2). If the child did not cooperate in either of the screenings, he/she was classified as positive.

All children were clinically examined by the SLP (LN) within 2 months from the screening. The clinical examination took place at the child's preschool during two sessions: one in Swedish and the other in the child's mother tongue, with an interval of 2-4 weeks. Thus, all children were assessed four times in total, twice in their first language and twice in their second language, Swedish.

2.4 Statistics

Sensitivity, specificity, positive and negative predictive values and positive and negative likelihood ratio were calculated with 95% confidence intervals using MedCalc in order to determine screening accuracy (MedCalc Software Ltd.). Associations between parents' education and clinical outcomes were calculated by logistic regressions in SPSS, version 24 (IBM Corp.).

2.5 Ethical considerations

Information about the study and consent forms, both in Swedish and translated into the family's language, were sent to parents prior to their child's routine health visit at 2.5 years. During the health visit, the nurse also provided oral information. The study was approved by the Regional Ethical Review Board in Uppsala (dnr 2015/199).

4.1 Main findings

Screening in one of the child's languages only, as demonstrated in models one and two, led to many false positives. However, the number of children identified as being positive decreased significantly when both the mother tongue and Swedish, presented as model three, were used for screening. Moreover, specificity and positive predictive value increased radically with model three, which is also supported by previous studies.^{4, 21} Screening in Swedish, combined with information from parents about the child's mother tongue, presented as model four, yielded a high specificity but a very low sensitivity. In other words, numerous children with DLD passed the screening, while children with typical language development could be correctly identified. Many nurses conduct the screening in this way today. The shortcommings of model 4, is unfortunate because screening in the child's mother tongue using the parental reports could have been faster than a full bilingual screening.

Information provided by parents about the child's language development in their mother tongue indicated that they had poor ability to identify DLD in their child. However, they were better at identifying a child that did not have language difficulties. Hence, it is important to follow-up on parental concerns; however, it would lead to massive under-referral standing alone.

4.2 A complex linguistic environment

We found that DLD was three times as frequent among bilingual children in this study than in monolingual peers from the same district.¹⁷ Bilingual children also had a higher risk of severe language difficulties compared to monolingual peers.^{22, 23} Similar to monolingual children aged 2.5 years, DLD was overrepresented among non-cooperating bilingual children.¹⁷

Bilingualism per se does not lead to higher prevalence of DLD.² However, a background with multiple risk factors has negative impact on language development. Bilingual children living in segregated areas may develop their second language at a slower rate, while the development of their mother tongue tends to be less affected.²⁴ Bilingual school-aged children from socio-economically disadvantaged environments, who attended schools in these areas, displayed lower test results in their second language than bilingual children from socio-economically advantaged environments.²⁵

The receptive part of the Reynell test, used by the SLP in this study, has been validated for Swedish speaking bilingual children aged 2.5-3.5 years.^{19, 26} The results showed that the monolingual norms could be useful for bilingual children, although they generally score lower than monolingual peers. However, this was taken into account by the SLP during the assessment. Moreover, because children were assessed in both their languages, the risk for overidentification was low. Instead, the high prevalence of severe DLD in this study may have been related to complex mechanisms associated with the migration process,²⁷ and opportunities for migrants to integrate into society.

First, all children in this study had a second language acquisition with a varied and often insufficient exposure depending on their preschool. Half of them went to preschool sporadically, and 60% of all children (n = 66) attended preschools where more than 80% of the children had other mother tongues than Swedish (Table 1). Consequently, a majority of them were exposed to a complex linguistic environment, surrounded by many languages and a lower quality and quantity of language input.²⁸ It is well known that such factors aggravate any existing DLD in bilingual children.^{2, 8}

Second, the impact of factors associated with socio-demographic conditions and migration on the child's language development should be considered. A Swedish epidemiological study highlighted specific risk factors for bilingual children.²² One example is parental need to have an interpreter despite several years of residence in Sweden, which may serve as proxy for difficulties with literacy or language. About 66% of parents who lived in Sweden for more than 5 years, reported need for an interpreter when they visited health care (Table 1).

Finally, low socio-economic status has been considered as a factor affecting children's linguistic and cognitive ability.^{6, 29} The mothers in this study displayed considerably lower education compared to the immigrant female population of the same age in Sweden; about 10% of mothers had never attended school. However, in this study, parents' educational level showed no significant association with the clinical outcome on the individual level. Here, we studied a group of less socio-economically advantaged bilingual children. Perhaps, the prevalence of DLD would have been lower for bilingual children in affluent areas with parents having higher education. However, the findings indicate clinically relevant problems in multicultural areas of low socio-economic status.

4.3 Clinical implications

Bilingual children are a very heterogeneous group, with different family backgrounds, and social and linguistic circumstances. All these aspects have significant influence on language development and on severity of DLD.⁷ Almost every third bilingual child, growing up in a low socio-economic environment, displayed DLD in our study. Consequently, the screening should always be conducted in both languages used by the child, particularly if the child does not pass the screening in one language. The current procedure to screen the child in his/her second language, Swedish, combined with information received from parents is not a reliable enough approach. However, now, there is a validated test for bilingual children at 2.5 years of age.

Given that a high percentage of children with suspected DLD in early age have shown neuropsychiatric disorders at school start,³⁰ early detection and support are vital. Educational efforts are needed for child health services related to language assessment, language development, risk factors and signs of DLD. The linguistic environment provided by the preschools constitutes a modifiable risk factor amenable to political decisions on improved quality.

To determine whether different living conditions and social class affiliation influence language development in bilingual children, there is a need for future studies. In addition, diagnostic accuracy is more challenging with younger children compared to older.¹⁷ Future studies should therefore determine later outcomes of bilingual children diagnosed with DLD at age 2.5.

4.4 Strengths and limitations

This is the first study of bilingual children in which all screened children were assessed in both their languages, followed by a blinded clinical examination as a gold standard for DLD diagnosis. A further strength was that the children were not selected but consecutively recruited during 20 months. The study followed the same procedure and recruited children from the same child health centres as a previous study of monolingual children,¹⁷ allowing for some comparisons between the studies.

Furthermore, the Reynell test used by the SLP in the present study had previously been evaluated for bilingual children in a number of studies. The translations of Reynell and items in the screening regarding comprehension were reviewed by the assessing SLP together with bilingual staff and others with diverse languages. As a rule, translation of tests is not recommended and is usually considered a limitation. The children in this study, however, were given the test items by trained bilingual staff using carefully controlled translations of the items. Hence, this procedure is likely to have mitigated this limitation. The fact that the evaluation of the screening outcomes in either language was performed by an SLP, not by a nurse at the child health centre, has both advantages and disadvantages. It ensured conformity and that the evaluation was transparent and unaffected by the wait-and-see strategy found in earlier studies.¹³ At the same time, it constituted a deviation from the ordinary procedure of how the screening is evaluated.

The relatively small sample size, the restriction to six languages and the fact that all three child health centres recruiting children served low socio-economic areas limit the generalisability of the study. Some restriction on the number of languages is clearly needed, and the six languages included are presently the most frequently spoken by immigrant families in the region.