Autism-spectrum disorders, characterised by the presence of a triad of impairment of social cognition, communication and imagination, are considered to be the most common of the severe developmental disabilities. The rate of diagnosis has increased substantially with a recent estimated prevalence over 1%. ^{Reference Baird, Siminoff, Pickles, Chandler, Loucas and Meldrum1} However, it is unclear whether this higher prevalence represents a true increase or could be attributed to changes in diagnostic practices and complex issues relating to service provision. ^{Reference Baird, Siminoff, Pickles, Chandler, Loucas and Meldrum1–Reference Keen and Ward3}

The aetiology of autism-spectrum disorders is now largely considered to be multifactorial. Although genetic factors clearly play an important role, genetic study reveals an increasingly complex picture ^{Reference Gillberg and Coleman4} and no unifying processes have been established ^{Reference Harrison, O'Hare, Campbell, Adamson and McNeillage5} or alleged cause empirically supported. ^{Reference Baird, Siminoff, Pickles, Chandler, Loucas and Meldrum1} Of possible risk factors, some specific environmental factors operating at the intrauterine level such as proximity to pesticides ^{Reference Roberts, English, Grether, Windham, Somberg and Wolff6} and adverse perinatal events such as growth restriction and newborn hypoxia would appear to be of significance. ^{Reference Hultman, Sparen and Cnattingius7–Reference Reichenberg, Gross, Weiser, Bresnahan, Silverman and Harlap10}

There is an increasing literature supporting the role of advancing paternal and maternal age, although data for the latter are possibly inconsistent, ^{Reference Kolevzon, Gross and Reichenberg9–Reference Tsuchiya, Matsumoto, Miyachi, Tsujii, Nakamura and Takagai13} with a greater than twofold increase in risk with each 10-year increase in paternal age. ^{Reference Kolevzon, Gross and Reichenberg9} It has been suggested that first-born offspring of two older parents have a threefold risk of autism-spectrum disorders compared with younger siblings. ^{Reference Durkin, Maenner, Newschaffer, Lee, Cunniff and Daniels12}

Other social and environmental factors may also have a possible role in fuelling the increased prevalence and merit serious consideration. For example, the risk of autism appears to be associated with urbanisation. ^{Reference Lauritsen, Pedersen and Mortensen11,Reference Gillberg, Steffenberg, Borjesson and Anderson14} Political and economic influences on patterns of population immigration and ethnic diversity have an increasingly important impact on society, with resultant implications for patterns of health and disease as well as for service access and provision. One recognised association with immigration is the substantially increased risk of schizophrenia in immigrants, which is not solely accounted for by psychosocial explanations and appears to be independent of genetic vulnerability, ethnic group or country of origin. ^{Reference Cantor-Graae and Selten15}

Over the past 30 years or so there have been sporadic reports, initially of small populations or clinic samples, of an increased frequency of autism in children of immigrant parents, leading to the suggestion that immigration may be a factor contributing to the real rise in the prevalence of autism. ^{Reference Gillberg, Steffenberg and Schaumann16} One of the first published reports describes a disproportionate number (21–29%) of children with infantile autism born to Greek and German parents emigrating to Australia ^{Reference Haper and Williams17} and this remains the only study relating to European immigrants. Subsequently, European studies reported similar observations relating to immigrants from distant geographical areas or ‘exotic’ countries. ^{Reference Gillberg, Steffenberg, Borjesson and Anderson14,Reference Wing18} A UK clinic study found four to six times the rate of autism in second-generation African–Caribbean children. ^{Reference Goodman and Richards19} An Australian study ^{Reference Williams, Helmer, Duncan, Peat and Mellis20} found an increased risk associated with having a mother born outside Australia, with the greatest risk in those born in North East and South East Asia. Maternal place of birth outside Europe or North America has also been reported to be associated with a relative risk of autism of 1.4–3.0 for children born in Denmark and Sweden respectively, ^{Reference Kolevzon, Gross and Reichenberg9} and a three to four times greater prevalence in children of mothers born in Somalia emigrating to Sweden. ^{Reference Barnevik-Olsson, Gillberg and Fernell21}

Complex paternal sociobiological factors may act as confounders in relation to both parental age and immigration. Three processes operating in men with genetic vulnerability to autism have been suggested: seeking an immigrant partner who may be less aware of autism-related social problems than someone raised in his own cultural milieu; having greater difficulty finding a partner (with resultant older paternal age); or selective immigration of those at higher risk of autism-spectrum disorders. ^{Reference Gillberg, Schaumann and Gillberg22,Reference Gillberg and Gillberg23} One study examining the relationship between autism-related personality traits and delayed paternity has not found an association. ^{Reference Puleo, Reichenberg, Smith, Kryzak and Silverman24} There is some support for a trend to seeking an immigrant partner but not for selective immigration, in the finding that maternal country of origin has a greater effect than paternal country of origin. ^{Reference Lauritsen, Pedersen and Mortensen11} These key studies are specific to the Nordic countries, and replication in other geographical regions may be necessary before the results can be generalised. ^{Reference Kolevzon, Gross and Reichenberg9}

In contrast to these reports, suggested relationships linking autism-spectrum disorders with demographic and ethnicity factors have been dismissed as being based on small numbers and ‘not convincing’. ^{Reference Volkmar, Lord, Bailey, Schultz and Klin25} Some large population prevalence studies from the USA have also not supported a relationship between either ethnicity or immigrant status and rates of autism, and in particular the rate of autism-spectrum disorders in Black children was found to be comparable with other ethnic groups, ^{Reference Yeargin-Allsop, Rice, Karapurkar, Doernberg, Boyle and Murphy26} and neither maternal ethnicity nor immigration were an influence. ^{Reference Croen, Grether, Hoogstrate and Selvin27,Reference Croen, Grether and Selvin28}

The inconsistency between reports raises the question as to whether the effect of ethnicity and/or parental immigration on the risk of autism-spectrum disorders differs between countries or regions, such as between the USA and Europe, as has also been suggested in the case of schizophrenia. ^{Reference Cantor-Graae and Selten15}

There are a limited number of studies looking at the relationship between parental immigration and other types of developmental disability. However, where this has been explored, it appears that children of immigrant parents are at risk from a greater severity of neurodisability in general, whether physical, intellectual, social or sensory functioning, ^{Reference Goodman and Richards19,Reference Morton, Sharma, Nicholson, Broderick and Poyser29,Reference Akinsola and Fryers30} and this has also been suggested for autism-spectrum disorders. ^{Reference Barnevik-Olsson, Gillberg and Fernell21} High levels of recessive genetic conditions have been observed in some specific populations, ^{Reference Morton, Sharma, Nicholson, Broderick and Poyser29} and there has been repeated suggestion of a possible role for intrauterine infection in relation to autism and disability in general, ^{Reference Kolevzon, Gross and Reichenberg9,Reference Goodman and Richards19} but these and other aetiological factors have not been investigated systematically.

Despite the emerging research findings, immigration as a possible aetiological factor in the increasing prevalence of autism-spectrum disorders has been underinvestigated. This study aims to examine the hypotheses that maternal ethnicity and/or immigration are linked to childhood autism-spectrum disorders in the second generation, in the context of conflicting evidence relating to ethnicity and parental immigration as risk factors for autism.

Method

The study population was drawn from children presenting to the paediatric child development services of two adjacent south London boroughs, Wandsworth and Lambeth. These boroughs were selected because of differences in the demographic and economic profile, the former being relatively affluent with a Black and minority ethnic population of 22%, and the latter with higher levels of social adversity and a Black and minority ethnic population of 38%. ³¹

Retrospective analysis of case notes was undertaken. Participants were all children diagnosed with autism-spectrum disorders during a 6-year period (1 September 1999 to 31 August 2005). The inclusion criteria were that children were born in the UK and resident in the borough at the time of diagnosis.

The child development teams were organised differently, but were both established multidisciplinary services providing autism-specific assessments through child observation, individual assessment and clinical interview (using tools such as the Autism Diagnostic Interview (ADI–R), ^{Reference Lord, Rutter and Le Couteur32} Diagnostic Interview for Social and Communication Disorders (DISCO) ^{Reference Wing, Leekam, Libby, Gould and Larcome33} and Autism Diagnostic Observation Schedule (ADOS)). ^{Reference Lord, Risi, Lambrecht, Cook, Leventhal and DiLavore34} In most cases a diagnosis was made following a multidisciplinary team assessment, and in both services by reference to ICD–10 criteria. ³⁵ A very small number of older children were diagnosed by other services in the boroughs during the same period (e.g. psychology or mental health services) prior to being known to the paediatric service. In two cases the exact date of diagnosis was unclear and the age at diagnosis was given in whole years. The diagnosis was recorded as autism or a pragmatic category of other autism-spectrum disorders for those with atypical autism, high-functioning autism or Asperger syndrome, as these descriptive terms were used inconsistently.

Details of parental ethnic group and country of birth were routinely recorded at assessment in both centres. Parental country of birth was assigned to region according to the 2001 Census classification of countries by continent for Europe, Africa and Asia. We created a separate subgroup born in the Caribbean because of the substantial numbers from that region. Other countries in the Americas and Oceania have been grouped as ‘elsewhere’ because of extremely small numbers. Because of the lower rates of recording for fathers, a pragmatic decision was made to use the mother's ethnic group and region of birth as the primary markers.

The rates of autism-spectrum disorders by maternal region of birth and ethnic group were calculated with reference to the 2001 Census population data, made available from the London Health Observatory and the Office for National Statistics. Rates were calculated by dividing the number of mothers in each ethnic or regional subgroup by the 2001 female population aged 16–39 years in that borough, representing women of approximate childbearing age. Mothers of mixed Black/White or Asian/White ethnicity were added to the Black and Asian ethnic groups respectively, due to small numbers, and the same groups were combined in the population data.

Correction was made for possible variation in rate of autism-spectrum disorders resulting from differences in family size in different ethnic subgroups. Average family size, estimated from the total period fertility rate (TPFR), was obtained for ethnic groups from a report by the Data Management and Analysis Group of the Greater London Authority, 2002. ^{Reference Klodawski36} The TPFRs for detailed ethnic subgroups were weighted according to the distribution of these subgroups in each borough's population, to give an overall TPFR for each of the Black and Asian populations in the two boroughs. Information on TPFR was available only by ethnic group and not for country of birth.

Relative risks for having a child with an autism-spectrum disorder were calculated relative to White ethnic group and UK country of birth. The 95% confidence intervals for relative risks were obtained using the Confidence Intervals Analysis computing package, version 1.2 in Windows XP. ^{Reference Gardner, Gardner and Winter37} Relative risks were considered statistically significant if the confidence interval did not include one. In the text, the effect size is generally quoted as the lower end of the confidence interval, giving a conservative estimate of the increased risk.

This study was approved by the Research and Development Committee of South West London and St George's Mental Health NHS Trust and permission to access clinical data was given by the Lambeth Primary Care Trust Caldicott Guardian.

Results

Data were collected on all 428 children diagnosed with autism or autism-spectrum disorders over the 6-year period, of which 267 were in Wandsworth and 161 in Lambeth.

Demographic characteristics

About four-fifths of children with a diagnosis of autism-spectrum disorder were male in both boroughs (Table 1). The average age for diagnosis was higher in Lambeth than in Wandsworth (6.2 v. 4.6 years respectively). Mean age at diagnosis did not differ significantly between males and females in either borough, but showed a significant variation across ethnic groups in Wandsworth, with the Black ethnic group having the youngest age at diagnosis.

Table 1 Age at diagnosis of autism–spectrum disorder, by gender and maternal ethnicity, within borough of residence^a

	Lambeth (n = 161)				Wandsworth (n = 267)
		Age at diagnosis, years				Age at diagnosis, years
	n (%)	Mean (s.d.)	Range	P	n (%)	Mean (s.d.)	Range	P
Overall		6.2 (2.4)	2.2-15.6			4.6 (2.4)	1.4-18.0
Gender				0.49				0.96
Male	124 (77.5)	6.3 (2.5)	2.2-15.6		225 (84.6)	4.6 (2.3)	1.4-18.0
Female	36 (22.5)	6.0 (1.8)	3.5-12.2		41 (15.4)	4.6 (2.9)	2.0-16.0
Mother's ethnicity				0.68				0.002
White	27 (16.8)	6.1 (2.4)	3.0-13.0		163 (61.0)	4.9 (2.5)	1.4-16.0
Blackb	97 (60.2)	6.3 (2.4)	2.8-15.6		78 (29.2)	3.7 (1.4)	1.8-10.0
Asianc	13 (8.1)	6.1 (2.4)	4.0-12.5		16 (6.0)	5.3 (4.2)	2.1-18.0
Other/not knownd	24 (14.9)	6.0 (2.7)	2.2-12.4		10 (3.7)	4.9 (2.1)	2.5-8.5

For those children in whom a diagnostic subtype was identified, around two-thirds were classified as having autism and a third as having autism-spectrum disorder (Table 2). There was little difference in these proportions between males and females, but in Wandsworth the proportion with autism was significantly higher for the children of Black and Asian ethnic groups compared with the White group.

Table 2 Diagnostic subgroups by gender and maternal ethnicity, within borough of residence

	Lambeth (n = 161)		Wandsworth (n = 267)
	n (%)	P	n (%)	P
Overall
Autism	104 (70.3)		159 (62.6)
Other autism-spectrum disorders	44 (29.7)		95 (37.4)
Unconfirmed subtype	13		13
Gender a		0.98		0.34
Male
Autism	80 (70.8)		131 (61.2)
Other autism-spectrum disorders	33 (29.2)		83 (38.8)
Unconfirmed subtype	11		11
Female
Autism	24 (70.6)		27 (69.2)
Other autism-spectrum disorders	10 (29.4)		12 (30.8)
Unconfirmed subtype	2		2
Mother's ethnicity b		0.77		0.047
White
Autism	17 (65.4)		86 (56.2)
Other autism-spectrum disorders	9 (34.6)		67 (43.8)
Unconfirmed subtype	1		10
Black
Autism	66 (72.5)		55 (71.4)
Other autism-spectrum disorders	25 (27.5)		22 (28.6)
Unconfirmed subtype	6		1
Asian
Autism	8 (72.7)		12 (75.0)
Other autism-spectrum disorders	3 (27.3)		4 (25.0)
Unconfirmed subtype	2		0

Incidence of autism-spectrum disorders by region of birth of mother

The estimated annual incidence of having a child with an autism-spectrum disorder was calculated according to the region of birth of the mother (Table 3). Compared with mothers born in the UK, there was a significantly increased risk of having a child with an autism-spectrum disorder for immigrant mothers born in Africa, the Caribbean and Asia; this result was seen consistently in both Wandsworth and Lambeth. The highest relative risk was observed for those born in the Caribbean, where the risk was estimated to be at least five times that of UK-born mothers, in both boroughs. There was no significant increase in risk for those born in other European countries.

Table 3 Incidence and relative risk of autism-spectrum disorders, 1999–2005, by region of birth of mother (immigration effect)

Region of birth of mothera	Mothers of children with autism-spectrum disorders, n	Female borough population aged 16-39, n	Annual incidence of having a child with an autism-spectrum disorder (per 1000)	Relative risk v. UKb	95% CI for relative riskb
Lambeth (n = 137)
UK	48	44015	0.18	1.00	—
Other Europe	8	5581	0.24	1.31	0.62-2.78
Africa	56	6482	1.44	7.92	5.39-11.6
Caribbean	14	1283	1.82	10.01	5.53-18.1
Asia	10	2310	0.72	3.97	2.01-7.84
Elsewhere	1
Wandsworth (n = 258)
UK	152	48 858	0.52	1.00	—
Other Europe	20	5276	0.63	1.22	0.76-1.94
Africa	47	4621	1.70	3.27	2.36-4.53
Caribbean	13	470	4.61	8.89	5.08-15.5
Asia	22	3402	1.08	2.08	1.33-3.25
Elsewhere	4

Incidence of autism-spectrum disorders by ethnic group of mother

The estimated annual incidence of having a child with an autism-spectrum disorder was calculated for different maternal ethnic groups (Table 4). Compared with White mothers, there was a significantly higher risk for mothers of Black ethnic origin, with at least a fivefold increase in risk observed in Lambeth and a threefold increase in Wandsworth. The relative risk for mothers of Asian ethnic origin compared with White mothers was only statistically significant in Lambeth, where at least a threefold increased risk was indicated.

Table 4 Incidence and relative risk of autism-spectrum disorders, 1999–2005, by ethnicity of mother

Ethnicity of mother	Mothers of children with autism-spectrum disorders, n	Female borough population aged 16-39, n	Annual incidence of having a child with an autism-spectrum disorder (per 1000)	Relative risk v. Whitea	95% CI for relative riska
Lambeth (n = 137)
White	27	41 332	0.11	1.00	—
Blackb	97	17 932	0.90	8.28	5.41-12.7
Asianc	13	3219	0.67	6.18	3.19-12.0
Wandsworth (n = 259)
White	163	55 404	0.49	1.00	—
Blackb	78	6908	1.88	3.84	2.93-5.02
Asianc	16	4416	0.60	1.23	0.74-2.06

Incidence of autism-spectrum disorders by region of birth and ethnic group of mother

In order to differentiate the effect of ethnicity from the effect of region of birth, each ethnic group was further divided into mothers who were born in the UK and those not born in the UK. The relative risk of having a child with an autism-spectrum disorder was calculated for each subgroup compared with a baseline of White women born in the UK (Table 5). There was a significantly increased risk of having a child with an autism-spectrum disorder for mothers of Black ethnic origin born outside the UK, of at least ninefold in Lambeth and fivefold in Wandsworth. Similarly, the risk was significantly increased for mothers of Asian ethnic origin born outside the UK by a factor of 5 in Lambeth and 1.4 in Wandsworth. Conversely, most of the subgroups of mothers born in the UK failed to show a significant increase in the risk of autism-spectrum disorders regardless of ethnic group, the exception being mothers of Black ethnic origin born in the UK in Lambeth. There was also no significant difference in risk for White immigrants to the UK.

Table 5 Incidence and relative risk of autism-spectrum disorders, 1999–2005, by ethnicity and immigration status of mother

Ethnicity of mother and whether born in the UK or not	Mothers of children with autism-spectrum disorders, n	Female borough population aged 16-39, n	Annual incidence of having a child with an autism-spectrum disorder (per 1000)	Relative risk v. White UK-borna	95% CI for relative riska	Estimated average family size, n	Relative risk v. White UK-born, adjusted for family sizea	95% CI for adjusted relative riska
Lambeth (n = 137)
White						1.24
UK-born	19	30 973	0.10	1.00	—		1.00	—
Immigrant	8	10 359	0.13	1.26	0.55-2.87		1.26	0.55-2.87
Missing	0
Blackb						2.31
UK-born	25	10 581	0.39	3.85	2.12-6.99		2.07	1.14-3.75
Immigrant	69	7351	1.56	15.30	9.21-25.40		8.21	4.94-13.63
Missing	3
Asianc						2.38
UK-born	2	1680	0.20	1.94	0.45-8.32		1.01	0.23-4.33
Immigrant	10	1539	1.08	10.59	4.93-22.70		5.52	2.57-11.83
Missing	1
Wandsworth (n = 259)
White						1.24
UK-born	132	41 888	0.53	1.00	—		1.00	—
Immigrant	31	13 516	0.38	0.73	0.49-1.08		0.73	0.49-1.08
Missing	0
Blackb						2.30
UK-born	19	4384	0.72	1.38	0.85-2.22		0.74	0.46-1.20
Immigrant	57	2524	3.76	7.17	5.27-9.75		3.86	2.84-5.26
Missing	2
Asianc						2.30
UK-born	0	1994	—	—	—		—	—
Immigrant	16	2422	1.10	2.10	1.25-3.52		1.13	0.67-1.90
Missing	0

Since the probability of having a child with an autism-spectrum disorder is likely to increase with family size, we adjusted these relative risks for the larger average family size among Black and Asian ethnic groups. The same adjustment was applied to both UK and non-UK subgroups, as a differential adjustment factor was not available by region of birth. The relative risk for mothers of Black ethnic origin born outside the UK remained statistically significant in both boroughs, with an increased risk of approximately three- to fivefold relative to White UK-born mothers. A significant effect was also still observed in Lambeth for mothers of Asian ethnic origin born outside the UK, and for mothers of Black ethnic origin born within the UK.

Discussion

This study found a significant association between maternal immigration and the risk of having a child with an autism-spectrum disorder. The immigration effect varies by maternal region of origin, with the highest risk observed for immigrants from the Caribbean, followed by those from Africa and then Asia. There is no apparent effect for immigrants from Europe.

There was little evidence of an independent ethnic effect on the risk of autism-spectrum disorders in the absence of maternal immigration, since the association with ethnicity did not remain significant when considered simultaneously with region of birth, except for UK-born Black individuals in Lambeth where the result just reached significance. Therefore it appears that it is immigration, not ethnicity, which is the main candidate for a risk factor.

The relationship between maternal immigration and risk of autism-spectrum disorders did however vary by ethnicity, with a stronger association observed for Black immigrants than for Asian immigrants, and no apparent effect for White immigrants. Whereas the increased risk appears unequivocal for Black immigrant mothers, the message is less clear for Asian immigrant mothers, where a significant effect was only seen in one borough, Lambeth. This might be explained by differences in composition of the Asian groups in each borough: in Lambeth, all Asian immigrant mothers originated in the Indian subcontinent, whereas in Wandsworth this proportion was only 56% (9/16), the remainder originating in the Far East, Middle East or Africa.

The variation in risk of autism-spectrum disorders by ethnicity among immigrant mothers may simply be a confounding effect with country of origin. For example, the high rates of autism-spectrum disorders observed for the children of Black immigrant mothers reflect the high rates found for mothers immigrating from the Caribbean and Africa. Alternatively, there may be a more complex explanation, with a combined effect of ethnicity and region of origin among immigrants from these countries.

These findings appear to be different from those reported by US studies. This may partly be related to differing population characteristics. In contrast to the Black immigration patterns in the USA, the substantial proportion of Black immigration into Europe has occurred mainly over the past 50 or so years. UK-born Black mothers are therefore likely to be second- or third-generation immigrants. The increased risk of autism-spectrum disorders in Black mothers born in the UK was lower than that observed for their Black immigrant counterparts. This raises the question of whether we might be observing evidence of a progressive attenuation of the effect of immigration generation on generation. This effect might account for the differences in the findings relating to settled ethnic populations between European studies and those from the USA, but would not explain differences relating to recent immigration.

It is unclear as to how immigration may exert an increased risk and clues from research into immigration and other aetiological factors in related neuropsychiatric or neurodevelopmental disorders are sparse. Autism-spectrum disorders are one element of an interwoven tapestry of overlapping and co-occurring neurodevelopmental disorders such as attention-deficit hyperactivity disorder, developmental coordination disorder and Tourette syndrome. There does not appear to be literature suggesting a link between parental immigration and the risk of any of these disorders.

However, there are interesting parallels with the literature on schizophrenia and psychosis. An increased rate of psychosis is reported to be present in all immigrants irrespective of ethnicity, suggesting an explanation that is environmental rather than genetic. However, similar to the findings of this study, there is a particularly high risk seen in the second-generation African–Caribbean population. ^{Reference Singh and Burns38} This raises the question as to what aetiological factors might be common to both psychosis and autism-spectrum disorders that are not shared by those co-occurring neurodevelopmental disorders. Increased paternal age and urban upbringing are also factors associated with increased risk in both psychosis and autism-spectrum disorders. Other diverse hypotheses put forward to explain increased risk in psychosis include selective immigration of parents with higher risk, obstetric complications resulting from cephalopelvic disproportion due to better nutrition, exposure to novel neurotropic virus infection and prenatal exposure to vitamin D deficiency caused by dark skinned immigrants' reduced exposure to sun in Northern countries. ^{Reference Cantor-Graae and Selten15} Most of these factors have also been suggested as proposed aetiological triggers for risk of autism-spectrum disorders although few have been systematically examined ^{Reference Gillberg and Coleman39} and therefore remain speculative. For schizophrenia, none have withstood systematic examination. ^{Reference Cantor-Graae and Selten15} There is no clear emerging picture as to possible aetiological factors in either schizophrenia or autism-spectrum disorders.

It is important to recognise that the term ‘ethnicity’ as used in population and health research is a multifaceted and largely socially determined concept. Ethnicity may not map directly to ethnic and biological make-up or genetic, allelic or epigenetic differences likely to be important in understanding the cause of increased risk. This is clearly an area that requires more research. ^{Reference Singh and Burns38}