Specific Language Impairment and Child Care by a Domestic Helper: Title and subTitle BreakA Case-Control Study in Chinese Children FREE

Specific language impairment (SLI) is the most common developmental disorder in childhood characterized by impaired language acquisition expected for chronological age with significant discrepancy between language development and cognitive function, not due to hearing impairment, mental retardation, emotional disturbance, or neurological or psychiatric disorders.

The prevalence of an SLI has been estimated to lie between 2.6% and 16% in preschool children, depending on geographical area and the criteria for defining SLI.^{1 - 2} In some children, the language and communication difficulties can persist into adulthood.² In addition, children with an SLI have elevated risks of reading and learning difficulties,³ behavioral problems,⁴ and psychiatric disorders.⁵ Although genetic factors are thought to be more important,^{6 - 8} environmental and sociodemographic factors have been found to be associated with the development of an SLI.^{9 - 11}

Previous studies have suggested that the quality of language input in early childhood is critical to optimal language development.¹² Therefore, the quality of early childhood day care is likely very important. In developed countries it is a common practice to send a young child to a day care center when both parents need to work. Whether such practice may increase the risk of language impairment has not been directly investigated in a controlled manner. In Hong Kong, instead of sending a child to a day care center, many nuclear families employ full-time domestic helpers imported from Southeast Asia to perform household chores and look after their children if both parents have to work. The children in these families are separated from their mothers most of the time and are cared for by somebody who usually does not speak the local dialect (Cantonese for Southern Chinese) well and who is busy with other household duties. Hence, an optimal environment conducive to language development may be unavailable.

In this study, we aimed to investigate whether there is any association between SLI and childcare by a domestic helper, and whether the severity of SLI is related to this factor, after adjustment for sociodemographic confounders. To our knowledge, there has been no study directly addressing this question in the literature in any ethnic group.

We conducted a retrospective case-control study during a 4-year period (1999-2003) by reviewing consecutive medical records of all Chinese children younger than 5 years assessed in a university-affiliated Child Assessment Center in the Duchess of Kent Children’s Hospital, Hong Kong. These children were referred from both primary and secondary health care settings all over Hong Kong for various neurodevelopmental or behavioral problems.

All children were assessed by developmental pediatricians using the Griffiths Mental Developmental Scale (GMDS), a standardized test for assessment of a child’s development in 6 domains, namely, locomotor, eye-hand coordination, language, social, performance, and practical reasoning. A domain quotient is calculated by dividing the mental age in that domain by the chronological age multiplied by 100. Therefore, a domain quotient of 100 indicates equivalence of the domain mental age and the chronological age. The general developmental quotient (GQ) is the average of the 6 domain quotients. In a large field trial of GMDS in normal children, the GQ had a mean (SD) of 100 (12) and each domain quotient had a mean (SD) of 100 (16).^{13 - 14} The GMDS results of our clinic attendees were used for the definition of cases and controls.

The GMDS was developed originally in the United Kingdom and has been validated in different ethnic groups for the assessment of a child’s development from 0 to 8 years old,^{15 - 16} indicating that it is a robustly valid instrument for developmental assessment. Although it has not been formally validated for use in Chinese children, it has been tested in a Chinese population^{17 - 18} and has been widely used for more than 20 years in Hong Kong. All developmental pediatricians administering the test were trained to perform it in a proper standardized way and the developmental assessment was assisted with a detailed written test administration guide.

For children exposed to multilingual environments, the language domain of the GMDS was tested with all languages to which the child was exposed. We first established the dominant language of the child and started the evaluation process in that language. We also scored all important milestone achievements for the other languages and actively elicited the child’s ability in all languages in vocabularies and sentence construction. Instead of just assessing the “best” language of the child or taking an average of the child’s ability in all his or her languages, we considered and examined the child’s abilities in all languages. In other words, if the child achieved a milestone of an item in the language testing scale in one language but not another, it would still be counted as that the child had achieved that particular language milestone; that is, we adopted the highest level achieved in any language for multilingual children. Similarly, vocabularies in all languages were summarily considered while avoiding “double counting,” that is, same vocabulary in different languages. In this way, we were able to assess the overall language achievement of the child that represented his or her global language developmental stage, irrespective of the dominant language and the total number of languages. In case the assessing pediatrician was unfamiliar with any of the child’s language, an interpreter would be engaged, who was most commonly the accompanying parent or domestic helper, or one of the staff in our unit.

There is no universally accepted definition of SLI.¹⁹ It is difficult to define what constitutes a significant discrepancy between language and cognitive domains that are not easily distinguishable based solely on psychological testing.¹⁹ Previous researchers defined SLI as language ability below −2 (lowest 2.5 percentile),²⁰ −1.25 (lowest 10 percentile),²¹ or −1 (lowest 15.9 percentile) SD²² from the population mean and normal nonverbal intelligence.

In this study, we defined SLI cases as children with normal GQ and language quotient (LQ) more than 1 SD below the mean and more than 1 SD below the GQ. All consecutive children satisfying these criteria were included, except those with a pure phonological disorder. These children were then separated in 2 subgroups based on the severity of language impairment. Mild SLI was defined as an LQ between 1 to 2 SDs below the mean, severe SLI as an LQ more than 2 SDs below the mean.

Controls were all consecutive children referred for other problems such as behavioral problems and assessed by a developmental pediatrician to have normal GQ and normal domain quotients, and that no significant discrepancies exist between the GQ and LQ, or among any of the 6 domain quotients. After clinical assessments, children considered to have significant disturbing behaviors that impair their social functioning were referred to a clinical psychologist or psychiatrist for further evaluation and were excluded.

Children with all types and levels of hearing impairment, mental retardation, history of prenatal or postnatal brain insults, autistic spectrum disorders, psychiatric disorders, neurological diseases, or syndromal disorders were excluded. Hearing impairment was ruled out by standard hearing tests performed by an audiologist. We also excluded children who were exposed to more than one language at home apart from those spoken by domestic helper, to exclude multilingual exposure not purely related to the presence of a domestic helper.

Childcare by a domestic helper was defined as childcare by an employed full-time domestic helper who is the main caretaker of the child during daytime (±nighttime) for at least 1 year before the initial evaluation of the child. Other independent variables included age and gender of the subjects, number of siblings, family history of developmental language disorder in a first- or second-degree relative, and parental age, educational level, and occupational status. The educational level of each parent was divided into the following 3 grades: primary school level or below, high school level, and tertiary level or above. The occupational status of each parent was classified into 8 grades according to the modified International Classification of Occupational Status: administrator, professional, associate professional, office worker, service worker, skilled manual worker, unskilled manual worker, and unemployed.²³ These data and the child’s medical and developmental history were collected by a standardized assessment interview checklist at the same session when the developmental assessment was performed. These data were mainly based on parental self-report, which were verified as much as possible by documentary evidence, such as medical records of siblings and relatives for verification of data on family history.

χ² Tests for trend were used to test for an association between severity of an SLI and various independent variables. Analysis of variance followed by post hoc tests were used to test for difference in GQ and LQ between different groups. Crude and adjusted odds ratios (ORs) for a diagnosis of an SLI for each independent variable were estimated by univariate and multivariate binary logistic regressions, respectively. For ordinal independent variables, a test for linear trend in the ORs of different levels of the variable was performed. If there was a statistically significant linear trend, the variable would be entered as a continuous covariate into the multivariate binary logistic regression model so as to minimize the number of parameter estimates and reduce the standard errors. After stratifying cases into mild vs severe SLI, ordinal logistic regression was used to investigate whether childcare by domestic helper was associated with higher severity of SLI after adjustment for confounders.

Version 10.0 of the SPSS (SPSS Inc, Chicago, Ill) was used for all analyses. All statistical tests were 2-tailed with .05 as the threshold level of significance.

This study was approved by the institutional review board of the University of Hong Kong which complied with the Declaration of Helsinki. All values are given as mean (SD).

A total of 496 children were recruited composed of 237 cases of SLI (hereafter referred to “SLI cases”) and 259 controls. One hundred fifty-six cases had mild SLI; 81 cases had severe SLI. Missing values constituted 0.5% of all data, all of which were because of incomplete data capture in clinical records.

The mean ages of SLI cases and controls were 2.51 (0.58) years (age range, 1.42-4.67 years) and 2.89 (1.26) years (age range, 1.00-5.83 years), respectively. Boys constituted 73.4% of SLI cases and 60.2% of controls. The mean ages of father and mother for SLI cases were 36.02 ( 5.80) years and 31.60 ( 4.83) years, respectively. For the controls, the mean ages of father and mother were 34.81 (5.51) years and 31.79 (4.90) years, respectively.

The distributions of independent variables are given in Table 1. The means of developmental LQ, developmental language age, and LQ are listed in Table 2. Domestic helpers mainly came from the Philippines (78.9%); Indonesian (13.7%) and Thai (7.5%) helpers were less common.

Concerning the type of language impairment, we had information in 48.9% of cases. Expressive delay and mixed receptive and expressive delay constituted 25% and 75%, respectively.

The crude and adjusted ORs are tabulated in Table 3. Children cared for by a domestic helper had a 20% excess risk of an SLI before adjustment for other confounding variables. This excess risk increased to 71% after adjustment for all other covariates. Ethnic origins of domestic helper did not affect its association with SLI. Male gender, positive family history of language disorder, fewer siblings, and advanced paternal age were independently associated with an SLI in the logistic regression model. There was a significant inverse trend between paternal occupational status in terms of socioeconomic gradient and risk of an SLI. The Hosmer-Lemeshow goodness-of-fit test showed satisfactory model fitness (P = .96). Residual analysis did not suggest any pattern of unexplained variance. Interaction between domestic helper and other variables were tested for and no significant interaction was identified.

We have also stratified the case group according to the type of delay (mixed or expressive) and compared them with the control group separately. The crude ORs of childcare by a domestic helper were similar in both strata. The crude OR was 1.26 for the mixed delay stratum and 1.21 for the expressive delay stratum, both numerically close to the combined crude OR of 1.20 (Table 3), indicating that the association between childcare by domestic helper and SLI was similar in magnitude and direction whether the child got the expressive or the mixed type of language disorder.

Table 3 gives the adjusted ORs for increasing severity (ie, normal vs mild vs severe) of an SLI. For children cared for by domestic helper, the excess risk of a higher severity of an SLI was 56%. Estimates for other covariates were similar to those obtained from binary logistic regression. The model did not significantly deviate from the proportional odds assumption (P = .92) and demonstrated satisfactory model fitness (P = .71).

To our knowledge, this is the first study demonstrating that children cared for by a full-time domestic helper had significantly higher risk and severity of an SLI after adjustment for sociodemographic confounders. Interestingly, multivariate analyses revealed a stronger significant association but univariate analysis did not. This discrepancy was probably due to confounding by other variables. Parents with higher educational levels or higher occupational status tended to have employed a full-time domestic helper (P<.001 for all associations). Since these socioeconomic factors and domestic helper generally have opposite effects on the risk of an SLI, the effect of domestic helper was less apparent in univariate analysis where socioeconomic status was not adjusted.

There are 2 possible mechanisms by which the domestic helper may increase the risk of an SLI. First, research has shown that greater attention to early language development by parents and carers enhances the development of literacy and communication skills.²⁴ Mothers naturally provide predominant influences in children's early years and mothers’ response to their children's sounds can reinforce early language development.²⁵ However, whether the domestic helper can replace a mother’s facilitation on the child’s language development is questionable. In addition, domestic helpers usually have to perform most household duties rather than interact with the children. If facilitating the child’s development is not a high priority for the domestic helper, the quality and quantity of language input from the helper may be suboptimal and this may increase the risk of an SLI. Second, the domestic helpers in Hong Kong usually speak a foreign language which may adversely affect the child’s language development. Although young children have inborn cognitive mechanisms to acquire a first language rapidly without teaching,²⁶ learning 2 languages simultaneously may pose a higher cognitive demand on the learner. Some research demonstrated that bilingual children were worse than monolingual speakers in verbal tests in both languages,^{27 - 29} and there was association between SLI and bilingual environment.³⁰ However, there were also studies showing that bilingual speakers might perform better than monolingual speakers in verbal ability tests if fair testing in both languages was used.^{31 - 32} Whether the higher risk of an SLI in children raised by domestic helper is related to bilingual exposure requires further studies.

It is also possible that the observed higher rate of childcare by a domestic helper in children with an SLI was because these children were more difficult and demanding and had higher care needs, representing a kind of reverse causality. However, this is considered unlikely since employing a full-time domestic helper in Hong Kong is mainly because both parents need to work.

We found that boys have a higher risk of an SLI. This might have represented a gender distribution imbalance in our sample. However, the inferior language aptitude of boys during childhood is a common observation,² which may be related to structural and functional differences in the language areas of the brain.³³ The presence of family history moderately increased the risk of an SLI. Apart from genetic predisposition of an SLI,^{6 - 8} parents having residual language impairment may adversely influence their children’s language development.

We found that the father’s but not the mother’s occupational status was associated with an SLI.³⁴ This is possibly because the father’s job tends to be the principal determinant of social class.³⁵ A mother who is a homemaker did not seem to have a protective effect for an SLI, suggesting that it might not be the absence of the mother, but the presence of a full-time domestic helper that makes the difference. Alternatively, we might have insufficient statistical power to show the protective effect of having a mother who is a homemaker.

Our results suggested that having more siblings may protect a child from an SLI. This finding is opposite to most observations reported in the literature.^{11 ,36} Yet, there were studies reporting that family size was not significantly associated with an SLI.³⁷ Whether this is related to cultural difference between different populations requires further studies.

Our study had certain limitations. First, we have only demonstrated a significant association between childcare by a domestic helper and an SLI and a risk gradient across the severity levels of an SLI, but have not ascertained a causal relationship, which requires additional criteria according to Bradford-Hill.³⁸ Second, the GMDS is only a simple assessment tool for language development and it cannot give a detailed analysis of the children’s language abilities. Moreover, in applying the GMDS to children exposed to more than one language, we pooled the children’s abilities in all languages to assess the global language development. Such pooling rendered it impossible to determine the children’s ability in each language. However, this method was found to be more reliable in assessment of multilingual children³¹ ; and it was particularly applicable since we were concerned with the overall language development rather than the ability in each language. Third, because of the retrospective nature of the present study, there is limitation on the details of the language disorder in each patient. Information on specific deficits such as pragmatic or semantic defects was unavailable for most patients. However, we have at least demonstrated that child care by a domestic helper increased the risk of expressive and mixed delay to a similar extent. The retrospective nature of the present study also limited the number of confounding variables that could be adjusted, including behavioral problems that were possibly present in some children. Hence, the extent to which behavioral problems influence the observed association between childcare by a domestic helper and an SLI was unknown. Although it is likely that behavioral problems are associated with an SLI in preschoolers, it is yet to be determined whether behavioral problems are associated with childcare by domestic helpers. Since we did not have much detailed information on behavioral problems that have large variations in their nature and severity, we were unable to control for this complex variable. Further studies are needed in these respects. Fourth, recall bias is possible in any case-control study. However, the chance of recall bias was probably low in our study as we had a standardized assessment interview checklist and always asked a simple robust question of any childcare provided mainly by a domestic helper for more than 1 year in the past, which was unlikely to be influenced by biased memory or subjective interpretation. Fifth, there might have been referral bias since we analyzed subjects from one center only. However, we received referrals from primary and secondary care settings all over Hong Kong, thus having a wide representation of different groups of children. We also included all eligible consecutive subjects instead of choosing a sample haphazardly to minimize selection bias. Nevertheless, selection bias was possible if the decision to refer a subject was influenced by whether the child was looked after by a domestic helper. Sixth, the control group came from an at-risk population and might not represent an entirely normal control group. Nevertheless, they were assessed by a valid developmental test to have normal development in various domains. Seventh, the information about the caregivers is sparse. The quality and quantity of language and social interaction between the domestic helper and the child, and the relative contribution of the mother and other family members and schools were not assessed, although these were important determinants of the child’s language development. Other information about the domestic helper such as the age, educational level, and the experience with childcare of their own or other people’s children are also potentially important determinants of the language development of the children they look after. Future prospective studies should aim at further evaluation of the above factors on the child’s language development. Since cultural factors are likely to be relevant in the pathogenesis of an SLI, whether our findings in Chinese children can be generalized to other ethnic groups is questionable.

We cannot undermine the importance of foreign domestic helpers in assisting with childcare. In fact, most children cared for by domestic helpers do not suffer from an SLI. This is also consistent with the observation that children in day care centers generally develop normally if they are raised in high-quality settings. However, for those with a genetic or socioenvironmental predisposition to an SLI, exposure to an additional risk factor of being raised by a domestic helper might result in significant language impairment. Further studies are needed to better quantify and confirm the potential adverse effect of childcare by domestic helper. Long-term follow-up studies are essential to determine whether early childcare by domestic helper has long-lasting effect on language development. Intervention studies might also be required before recommendation can be made against childcare by domestic helper. Although the day care setting in other countries represents a somewhat different environment, the potential contribution of care in such setting to higher risk and severity of an SLI warrants further research.

Correspondence: Daniel K. L. Cheuk, MRCPCH, Department of Pediatrics and Adolescent Medicine, Queen Mary Hospital, University of Hong Kong, Pokfulam Road, Hong Kong SAR (cheukkl@hkusua.hku.hk).

Accepted for Publication: February 15, 2005.