Prevalence of autism spectrum disorders among children in China: a systematic review.

BACKGROUND: There are no reliable estimates of the prevalences of autism and autism spectrum disorders (ASD) in China.
OBJECTIVE: Combine results across studies to estimate the prevalences of autism and ASD among Chinese children under the age of 18, and assess variations in the prevalences with respect to gender, ethnicity, and urban versus rural residence.
METHODS: Based on pre-defined inclusion and exclusion criteria, studies were identified by searching the following databases: Chinese National Knowledge Infrastructure, Chongqing VIP database for Chinese Technical Periodicals, WANFANG DATA, Chinese Biological Medical Literature Database, Pubmed, and Web of Science. Statistical analysis was conducted using R-2.15.2 software.
RESULTS: The 24 studies meeting inclusion criteria included 5 registry studies from Taiwan and Hong Kong (covering a total of 14570 369 children) and 19 community-based screening and diagnostic studies from mainland China (with a combined sample of 771 413 children). The annually reported prevalence of autism in the registry studies ranged from 1.8 to 424.6 per 10 000. A meta-analysis of 18 of the studies from mainland China (excluding a large nationwide study with the lowest prevalence of autism) with a range in rates from 2.8 to 30.4 per 10 000 generated an estimated pooled prevalence of autism of 12.8 per 10 000 (95%CI, 9.4 to 17.5). The pooled prevalence of ASD estimated from 5 of these studies (which had a range in rates from 7.3 to 75.3 per 10 000) was 24.5 per 10000 (95%CI, 10.4 to 57.4). The reported prevalence of autism varied substantially by gender, location of residence, date of publication, and source of the sample.
CONCLUSION: The huge difference between the rates for autism reported from registry systems in Hong Kong and Taiwan (a 200-fold difference) and the large differences in rates reported from community-based screening studies in mainland China (a 10- to 15-fold difference) highlight the urgent need for establishing standardized methods for estimating the prevalences of autism and ASD. Until these methodological improvements have been made, it will not be possible to develop evidence-based prevention and treatment strategies for the management of these uncommon but seriously disabling conditions.

Introduction

Autism is a developmental disorder with an onset during early childhood that results in social deficits, communication deficits, stereotyped interests, and repetitive behaviors.[1] Recently the term ‘autism spectrum disorders’ (ASD) has been used to describe a group of disorders that include autism and similar types of disorders.[2] The conditions included under the ASD label vary slightly between the three diagnostic systems commonly used in China – the 3;[3] the 10;[4] and the 4 [5] In all three diagnostic systems, ASD includes autism, atypical autism, Rett syndrome, childhood disintegrative disorder, Asperger syndrome, and pervasive developmental disorder not otherwise specified (PDD-NOS). The ASD diagnosis in ICD-10 also includes ‘overactive disorder associated with mental retardation and stereotyped movements’. Rett syndrome will be excluded from the ASD diagnosis in the forthcoming DSM-5.[6]

Prevalence studies conducted by the United States Centers for Disease Control and Prevention[7]–[9] and others centers[10]–[12] have documented a rapid increase in the prevalence of ASD. A meta-analysis conducted by Williams and colleagues in 2006[13] reported a pooled estimate of prevalence of 7.1 per 10 000 for autism, and 20.0 per 10 000 for ASD among individuals under the age of 18. Another meta-analysis conducted by Fombonne and colleagues in 2009[14] summarized data from 57 studies and yielded a pooled prevalence of 22 per 10 000 for autism and 60 to 70 per 10 000 for ASD. Elsabbagh and colleagues[15] reviewed studies conducted after the year 2000 and found a median prevalence of 17 per 10000 for autism and 62 per 10 000 for ASD.

Information from China has not figured prominently in these prevalence estimates for autism and ASD: the 2006 analysis by Williams[13] included no data from China; the 2009 analysis by Fombonne[14] included one study from Hong Kong; and the 2012 report from Elsabbagh[15] combined data from China and Japan together as the ‘Western Pacific region’. Most studies on the prevalence of ASD in mainland China have been provincial studies with relatively small sample sizes that report wide variations in prevalence.[16]–[18] The only national study, a study conducted by Li and colleagues in 2011,[19] reported a very low prevalance of autism (2.4 per 10 000). The current report is a systematic review and meta-analysis of prevalance studies on autism and ASD from mainland China, Hong Kong and Taiwan.

Methods

Identification of studies for inclusion in the meta-analysis

The process of identifying studies for inclusion in the meta-analysis is shown in Figure 1. Two authors (YW and QH) first conducted electronic searches in the following databases: the Chinese National Knowledge Infrastructure database (CNKI, 1979-2013), the Chongqing VIP database for Chinese Technical Periodicals (1989-2013), the WANFANG DATA database (1990-2013), the Chinese Biological Medical Literature Database (1978-2013), Pubmed (1966-2013), and Web of Science (1950-2013). All reports published by 28 February 2013 were included in the search. All articles in which any word denoting autism (including ‘autism spectrum disorders’, ‘pervasive developmental disorders’, ‘autism’, ‘autism disorder’, ‘zi bi zheng’ [an older term for autism in Chinese], ‘Asperger’, ‘Asperger syndrome’) in either Chinese or English occurred with any word denoting prevalence (including ‘prevalence’, ‘detectable rate’, ‘incidence rate’, and ‘epidemiology’) in either Chinese or English were identified. Reference lists of identified studies were hand-searched.

Figure 1.

Identification of studies included in the meta-analysis

Included studies were epidemiological studies about the prevalence of autism or ASD (as defined by CCMD[3], ICD[4] or DSM[5] diagnostic criteria) among Chinese individuals (including those from mainland China, Hong Kong, Macau, and Taiwan) under the age of 18 who were identified from the general population, from clinical populations, or from student populations. Non-human studies, unpublished reports, reviews, and case reports were excluded. The report of the study had to provide the sample size and estimates of the prevalence of autism or ASD based on the use of a valid diagnostic tool or data from an authoritative health monitoring system.

First, all studies were imported to the literature management software Endnote X5 to eliminate duplicated records. Two authors (WY and HQ) independently conducted a preliminary screening of reports by reading titles and abstracts and then the full texts of potentially relevant articles were downloaded for the second round of screening. The above inclusion criteria were used to select studies for the analysis. There were six articles in which the initial screening result about whether or not to include the article differed between the two raters; after discussion, a third reviewer (TL) made the final decision about inclusion of these articles.

Data extraction

Two authors (WY and HQ) independently extracted and entered relevant data about the included studies. For each study the basic characteristics of the study (i.e., name of first author; year of publication; location of study; source of sample, sampling method, sample size and number of potential subjects not screened; breakdown of sample by gender, ethnicity, and urban versus rural residence; method of screening and diagnosis) and the reported prevalences of autism and ASD were recorded. Reports of ‘current prevalence’ and ‘point prevalence’ were both included.[20] If prevalence by gender, age group, urban versus rural residence, or ethnicity (Han versus other) was provided, this was also recorded.

The quality of the reports of the included studies was assessed using the guidelines recommended for Strengthening the Reporting of Observational Studies in Epidemiology (STROBE).[21] This guideline lists 22 criteria, covering specific elements of the methods, presentation of the results, and interpretation of the results that are considered important for assessing the integrity of a study. One point was given for each of these elements that was present in the report, so the total score for quality ranged from 0 to 22 points. The inter-rater reliability of the two raters for this quality score was excellent (intraclass correlation coefficient [ICC], 0.96).

Statistical analysis

After checking for consistency, the Metaprop module in the R-2.15.2 statistical software package was used for the meta-analysis; the prevalence reported in each study was logit transformed prior to computing the pooled prevalence.[13] Tests of heterogeneity were conducted to decide which method would be used to pool the results. Results were considered homogenous when the I2 statistic (the percentage of variance due to heterogeneity) was less than 50% and the p-value for the test of heterogeneity was ≥0.10, in these cases a fixed-effect model was used to compute the pooled estimate of prevalence. In all other cases (i.e., I2 >50% or p<0.10) the studies were considered heterogeneous and a random-effect model was used to compute the pooled prevalence.[22] When heterogeneity was present, a sensitivity analysis was conducted to inspect possible reasons for heterogeneity. Subgroup analyses were conducted by gender, urban versus rural residence, source of the study population, and year of publication. Begg's rank method was used to assess potential publication biases.[23]

Results

Characteristics of identified studies

As shown in Figure 1, 30 studies were identified that met the inclusion criteria; 23 of these publications were in Chinese and 7 in English.[16],[17], [19],[24]–[50] Several sets of reports were from the same study: Wang (2002)[47] and Wang (2003)[26]; Huang (2010)[48] and Chen (2010)[35]; Li (2010a)[44] and Li (2010b)[49]; Wu [33] and Wang[50]; and Zhang (2005)[39], Zhang (2004)[28] and Guo (2004).[27] Thus, a total of 24 separate studies were identified[16],[17],[19],[24]–[26],[29]–[46] including 19 population-based screening and diagnostic studies in mainland China[16],[17],[19], [24]–[26],[29]–[39],[44],[45] (with a combined sample of 771 413 children) and 5 prevalence reports based on health registry data in Hong Kong and Taiwan[40]–[43],[46] (covering a total of 14 570 369 children). Details of the included studies are shown in Table 1.

Table 1.

Characteristics of the 24 included studies

Study	region	Sampling methoda	Population sourceb	Diagnostic criteriac	Sample size	Age range	Autism events	ASD events
Jiang 2000[24]	Henan/Luoyang	R	E	DSM-III	10 140	3-14	3	—
Luo 2000[25]	Fujian	CL,R	G	CCMD-2-R, DSM-III-R	10 802	0-14	3	—
Wang 2003[26]	Changzhou, Yizheng	S,CL,R	G	CCMD-2-R	7344	2-6	9	—
Guo 2004[29]	Gansu/Dingxi	R	G	DSM-IV	3776	2-6	3	—
Zhang 2005[39]	Tianjin	R	G	DSM-IV	7345	2-6	8
Liu 2007[16]	Beijing	S,CL	G	DSM- IV	21 866	2-6	14	16
Yang 2007[30]	Guizhou/Zunyi	R	E	DSM-IV	10 412	3-12	6	—
Wong 2008[40]	Hong Kong	—	REG	DSM-III-R, DSM-IV	4 247 206	0-14	—	682
Zhang 2008[31]	Jiangsu/Wuxi	S,R,CL	C,E	DSM-IV	25 521	1-6	25	—
Lin 2009[41]	Taiwan	—	REG	—	2000: 5 850 535	0-17	1846	—
					2001: 5 711 309		2276
					2002: 5 583 696		2170
					2003: 5 442 398		3354
					2004: 5 376 458		4040
					2005: 5 267 440		4710
					2006: 5 140 471		5407
					2007: 5 054 196		6179
Zhang 2009[32]	Guizhou/Yunyan	S,CL	G	CCMD-2-R	4999	0-6	5	—
Chen 2010[35]	Daqing	S,R,CL	C,G	DSM-IV	7034	2-6	10	17
Li 2010a[44]	Tianjin	R,CL	C	DSM-IV	8274	1.5-3	22	—
Wu 2010[33]	Lianyungang	R,S,CL	C	DSM-IV	8532	0-3	9	—
Yu 2010[34]	Haerbin	S,CL	E,G	DSM-IV	7059	2-6	15	16
Chien 2011[42]	Taiwan	R	REG	ICD-9	1996: 268 753	0-17	—	48
					1997: 264 191			68
					1998: 259 255			102
					1999: 253 671			177
					2000: 249 336			259
					2001: 245 666			358
					2002: 241 252			429
					2003: 237 361			486
					2004:233 365			565
					2005: 229 454			659
Hsu 2011[43]	Taiwan	R	REG	ICD-9	162 171	0-15	6886	—
Liu 2011[45]	Shanghai	—	G	DSM-IV	770	1.5-2	1	—
Li 2011[19]	Nationwide	S,M,CL	G	ICD-10	616 940	0-17	77301d	—
Su 2011[36]	Tianjin	S,CL	C	DSM-IV	7904	1.5-3	22	—
Wang 2011[17]	Guangzhou	S,R,CL	E	DSM-IV	6111	2-6	18	46
Lai 2012[46]	Taiwan	—	REG	DSM-IV-TR DSM-IV, DSM-III-R	2004: 4 664 310	3-17	—	3995
					2005: 4 601 833			4684
					2006: 4 387 827			5345
					2007: 4 395 283			6119
					2008: 4 268 630			6771
					2009: 4 157 940			7429
					2010: 4 044 433			8072
Wei 2012[37]	Shenzhen	—	C	DSM-IV	3624	1.5-2	7	10
Zhou 2012[38]	Shenzhen	—	G	—	2960	1-2	9	—

a R, random; CL, cluster; S, stratified; M, multiphase

b E, educational services; G, general population; REG, registration data; C, clinical services;

c DSM, Diagnostic and Statistical Manual; CCMD, Chinese Classification of Mental Disorders; ICD, International Classification of Diseases

d weighted number of cases to account for complex survey sample design; the prevalence (95% CI) estimated using the Taylor series linearization method was 2.38 (1.92-2.84)

Reported prevalence of autism in the registry studies from Taiwan and Hong Kong

Wong and colleagues[40] reported an interval prevalence of 16.1 per 10000 in Hong Kongduring the period of 1986 to 2005. In Taiwan Hsu and colleagues[43] reported a 12-month prevalence of 424.6 per 10 000 individuals under the age of 15 in 2007; Lin and colleagues[41] reported that the 12-month prevalence of autism in individuals under the age of 18 increased from 3.2 to 12.3 per 10 000 from 2000 to 2007; Chien and colleagues[42] reported that the 12-month prevalence of ASD increased at a rate of 1.8 per 10 000 annually from 1996 to 2005, reaching 28.7 per 10000 in 2005; and Lai and colleagues[46] reported that from 2004 to 2010 the 12-month prevalence of ASD increased steadily from 8.6 to 20.0 per 10 000.

Pooled prevalence estimates of autism and ASD from studies in mainland China

The registry-based data from Taiwan and Hong Kong were not suitable for inclusion in the meta-analysis because, unlike the studies from mainland China, there was no screening process used to identify cases. Thus, only the 19 population-based screening and diagnostic studies from mainland China[16],[17],[19], [24]–[26],[29]–[39],[44],[45] were considered for the meta-analysis. However, the heterogeneity of these studies was great (I=94.9%, p<0.001) so we first used sensitivity analysis to identify the causes of heterogeneity prior to pooling the results. Based on these analyses, the 2011 study by Li and colleagues[19] – the largest and only nationwide study (the other studies were conducted in provinces and had much smaller samples) – was excluded because it was the cause of substantial heterogeneity in the estimated prevalence. After excluding this study, the I2 for the remaining 18 studies decreased from 94.9% to 76.3%. Using a random-effect model to pool results from the 18 remaining studies (after excluding the study by Li [19]), the pooled sample was 154473 individuals, and the current prevalence of autism was 12.8 per 10000 (95%CI, 9.4 to 17.5 per 10 000). These results are shown in the Forest plot in Figure 2.

Figure 2.

Forest plot of prevalence estimates of autism and 95% confidence intervals from 18 studies in mainland China

The five studies from mainland China that reported the prevalence of ASD were also quite heterogeneous (I=94.4%, p<0.001). However, sensitivity analysis did not identify any factors that substantially influenced the heterogeneity of the results (i.e., removal of the identified study with the factor did not result in a substantially reduced I2), so all 5 studies were included in the meta-analysis. Using a random-effect model to pool results from the 5 studies, the pooled sample was 45 694 individuals and the current prevalence of ASD was 24.5 per 10 000 (CI, 10.4 to 57.4 per 10 000). These results are shown in the Forest plot in Figure 3.

Figure 3.

Forest plot of prevalence estimates and 95% confidence intervals from 5 studies of autism spectrum disorders in mainland China

Among studies in mainland China, one study[17] reported a prevalence of 40.9 per 10 000 for Asperger syndrome; two studies[16],[35] reported a prevalence of atypical autism of 0.46 and 8.53 per 10 000, and a prevalence of Rett Syndrome of 0.46 and 1.42 per 10 000; and three studies[17],[34],[37] reported a prevalence of PDD-NOS of between 1.4 and 8.3 per 10000. Due to the small number of studies for these subtypes of ASD, separate, diagnosis-specific meta-analyses could not be conducted.

Study quality and publication bias

Only 8 of the 19 (42%) studies from mainland China[17],[26],[32],[34]–[36],[39],[44] considered the influence of non-response in the estimation of prevalence and 3 of the 19 (16%) studies[37],[38],[45] did not provide a description of the sampling methods. Based on criteria listed in the STROBE, the quality score of the 19 community-based prevalence studies from mainland China (with a theoretical range of 0 to 22) ranged from 9 to 19 with a mean (sd) of 13.1 (3.2). Four studies[24],[30],[37],[45] with a score of <11 (i.e., less than 50% of the theoretical maximum score) were classified as ‘poor quality’. Of the 22 items from the STROBE assessment, the most common problems were a failure to estimate the required sample size (which was done in only 2 of the 19 reports), and the poor generalizability of the results (which was considered in only 4 of the 19 reports).

A minimum of 10 studies are needed to assess potential publication bias so it was only possible to conduct this analysis for the 18 studies used to estimate the prevalence of autism in mainland China and for the subgroup of 13 studies that assessed the prevalence of autism using DSM-IV as the diagnostic criteria. The Begg's funnel plots are shown in Figure 4. Based on the plots, the studies with smaller sample sizes tended to report a higher prevalence of autism. The Z-test for the plot of all 18 studies of the prevalence of autism was 2.95 (p=0.003) and that for the 13 studies that used DSM-IV criteria to make the diagnosis of autism was 2.95 (p<0.001); this indicates that publication bias was present in both analyses.

Figure 4.

Begg's funnel plots of publication bias for 18 studies of the prevalence of autism in mainland China (left) and the subgroup of 13 studies that used DSM-IV diagnostic criteria (right)

Prevalences of autism and ASD in mainland China by subgroup

The pooled estimates of the prevalences of autism and ASD in different subgroups of individuals are shown in Table 2. There are significant differences in the estimated prevalence of autism by gender (male prevalence is more than 2-fold female prevalence), residence (urban prevalence is 3-fold rural prevalence), age (children under 2 havea lower prevalence than children 2to6years of age), diagnostic criteria (the prevalence reported in studies using DSM-IV or CCMD criteria is much higher than in the single study using DSM-III criteria), year of publication (reported prevalence is higher in more recent publications), and source of population (estimated prevalence is highest in clinical populations and lowest in samples identified from schools). The reported prevalence of ASD was also much high in males than in females and higher in more recent publications than in older publications. However, the difference in the urban versus rural prevalence of ASD and the difference in the prevalence of ASD in Han children versus that in children from other ethnic groups were not statistically significant.

Table 2.

Prevalences of autism and Autism Spectrum Disorder (ASD) in different subgroups

	number of studies	n	events	I2	p	prevalence (per 10 000)	95% CI	U-value	p
AUTISM
current prevalence	18	154 473	189	76.3%	<0.001	(R)12.80	9.38-17.47	—	—
males	10	51 017	99	75.2%	<0.001	(R)19.51	12.82-29.68	5.09	<0.001
females	10	45 837	28	15.5%	0.300	(F)7.46	5.15-10.80
urban	5	29 471	39	25.0%	0.255	(F)15.10	11.03-20.66	2.44	0.014
rural	5	29 815	28	82.2%	<0.001	(R)8.54	3.17-22.96
sample 0 to 2 years old	7	23 404	21	60.2%	0.020	(R)10.23	4.57-22.92	-1.98	0.047
sample 2 to 6 years old	11	91 811	134	82.5%	<0.001	(R)15.78	10.35-24.06
use CCMD criteria(A)	2	12 343	14	0%	0.716	(F)11.40	6.75-19.23	A>B 2.28	0.022
use DSM-III criteria(B)	1	10 140	3	—	—	2.96	0.61-8.64	A v C -0.83	0.407
use DSM-IV criteria(C)	13	118 228	160	75.5%	<0.001	(R)14.32	10.26-19.97	B<C -2.99	0.003
study published 2000-2005 (A)	5	39 407	26	54.2%	0.068	(R)6.94	3.81-12.62	A<B-2.63	0.009
study published 2006-2010 (B)	8	93 697	106	75.6%	<0.001	(R)11.91	7.97-17.79	A<C-6.39	<0.001
study published 2011-2012 (C)	5	21 369	57	0%	0.807	(F)27.11	20.92-35.13	B<C-5.13	<0.001
source: educational system (A)	3	26 663	27	90.7%	<0.001	(R)8.47	1.99-36.03	A<B-3.79	<0.001
source: clinical services (B)	4	28 334	60	56.8%	0.074	(R)21.00	14.00-31.48	A v C -0.61	0.542
source: general population (C)	8	59 862	52	64.0%	0.007	(R)10.02	6.16-16.29	A<D -2.02	0.043
source: mixed populations (D)	3	39 614	50	64.8%	0.058	(R)14.06	8.65-22.85	B>C 4.20	<0.001
								B>D 2.19	0.029
								C v D -1.86	0.063
AUTISM SPECTRUM DISORDER (ASD)
Overall prevalence	5	45 694	105	94.4%	<0.001	(R)24.45	10.40-57.41	—	—
Males	5	24 005	85	93.7%	<0.001	(R)36.72	14.85-90.48	5.52	<0.001
Females	5	21 689	20	38.3%	0.166	(F)11.05	7.17-17.03
Urban	3	21 516	38	66.8%	0.049	(R)18.46	10.59-32.16	1.77	0.077
Rural	3	14 443	11	78.2%	0.010	(R)11.20	2.72-45.95
Han ethnicity	2	27 500	31	91.9%	<0.001	(R)12.95	3.72-44.94	-0.83	0.407
other ethnicity	2	1400	2	35.9%	0.212	(F)21.88	6.33-75.34
study published 2006-2010	3	35 959	49	86.3%	0.001	(R)15.91	7.46-33.90	-5.80	<0.001
study published 2011-2012	2	9735	56	88%	0.004	(R)47.40	17.75-125.95

I, heterogeneity coefficient (proportion of variance in results due to heterogeneity of included studies); CI, confidence interval; (R), computed using random-effects model; (F), computed using fixed-effects model; DSM, Diagnostic and Statistical Manual

Discussion

Main findings

Based on available registry data from Hong Kong and Taiwan, the reported prevalence of autism over the period from 2000 to 2010 ranged from a low of 1.8 per 10 000 to a high of 424.6 per 10 000, a more than 200-fold difference. Most of the studies reported increasing rates over time, but there must be substantial methodological differences in these studies to result in such a huge range in the estimated prevalence, so it was not possible to pool the results of the registry studies or to integrate them with the community-based screening and diagnostic studies from mainland China.

The 19 identified studies from mainland China had an estimated prevalence of autism ranging for 2.38 per 10 000 to 30.41 per 10 000, a 13-fold difference. Sensitivity analysis found that by eliminating the study with the lowest prevalence – the large nationwide study – the extremely high heterogeneity of the studies improved substantially (I2 decreased from 94.9 to 76.3%), so the meta-analysis only included the remaining 18 studies in which the estimated prevalence of autism ranged from 2.78 per 10 000 to 30.41 per 10 000 (an 11-fold difference). The resulting pooled prevalence of autism from the 18 included studies from mainland China was 12.8 per 10 000 individuals.

Five of these 18 studies also reported the prevalence of ASD in mainland China. The reported prevalence ranged from 7.32 per 10 000 to 75.27 per 10 000, a 10-fold difference. The heterogeneity of these studies was also quite high (I=94.4%), but sensitivity analysis did not identify factors that could exclude specific studies, so all 5 studies were included in the meta-analysis. The pooled prevalence of ASD estimated from these studies was 24.5 per 10 000.

Our estimated prevalences of autism and ASD are substantially lower than the corresponding prevalences reported by most meta-analyses from other countries,[14],[15] but some international meta-analyses[13] have reported lower prevalences than those estimated in this study. There are many possible explanations for these cross-national differences in the reported prevalences of autism and ASD. Methodological differences – particularly in the source population, in the diagnostic criteria, and in the methods of identifying cases – probably explain the majority of the difference. But there may also be biological and cultural determinants that affect the ‘true’ prevalences of these conditions. Moreover, there appears to be an upward trend in the trajectory of the prevalences of these conditions over time that is more than a simple increase in clinical awareness or an increase in care-seeking. If this is the case, it is certainly possible that part of the reason for the lower prevalences in China is that China is at an earlier stage in this trajectory than other regions. However, one major caveat to the suggestion that rates in mainland China are increasing with time is that the large nationwide study conducted in 2011 (with a sample several times the combined sample of the remaining 18 studies) had the lowest reported prevalence of autism (2.38 per 1000).

Subgroup differences in the prevalences of autism and ASD found in this study were similar to those reported in other countries. Male prevalence was three-fold female prevalence, in line with findings from elsewhere.[9],[51]–[53] The prevalences of autism and ASD in urban communities was higher than that in rural communities (though the difference was not statistically significant for ASD), a finding that is also reported in high-income countries.[13] Our study found a higher prevalence of autism among children who were 2 to 6 years of age than in younger children; some authors suggest that this may be related to the different clinical manifestations of the condition at different developmental stages during childhood (and the difficulty of establishing the diagnosis at younger ages). [13],[54] The one study that used DSM-III criteria to diagnose autism (in 10 140 individuals) had a much lower reported prevalence than the 13 studies that used DSM-IV criteria or the 2 studies that used CCMD; these differences may have been related to different characteristics of the samples, but other authors have suggested that the use of different diagnostic criteria can result in substantial differenes in the estimated prevalences of these conditions.[10]

Limitations

Our analyses identified several weaknesses in the included studies, so the results need to be interpreted with caution. Many of the included studies were either based in school systems (where rates are much lower) or in clinical populations (where rates are much higher), so their representativeness of the population as a whole is uncertain. The substantial heterogeneity of the results suggests that there were important methodological differences across studies. There was a clear suggestion of publications bias, with smaller studies tending to report higher rates. And the formal assessment of quality of the reports of the study used in the meta-analysis based on the STROBE criteria identified several limitations. Few of the studies reported the prevalences of the subtypes included under the ASD rubric so it was not possible to make robust estimates of the prevalences of these conditions. The huge difference between the rates reported from registry systems (200-fold differences) and the large differences in rates reported from community-based screening studies (10 to 15-fold differences) indicate that substantial improvements and standardization of the methodology for estimating the prevalences of these uncommon but serious conditions are needed.

Significance

Studies from other countries have reported a clear upward trajectory in the prevalences of autism and ASD over time. The current systematic review and meta-analysis identified 24 studies published since 2000 that had estimated the prevalences of autism or ASD in Chinese children. There were widely varying rates reported in these studies, presumably due to methodological differences across studies, so it is not certain whether or not China is also experiencing the upward trend in the prevalences of autism and ASD reported from high-income countries. There is, however, an increased awareness of the importance of autism since the first reported cases in mainland China in 1982.[56] Improved study methodology is needed to provide valid estimates of the prevalences of these conditions over time. Until these methodological improvements have been made it will not be possible to develop evidence-based prevention and treatment strategies for the management of these seriously disabling conditions.