Prevalence of Serious Emotional Disturbance Among U.S. Children: A Meta-Analysis

Accurate estimates of the prevalence of psychiatric disorders among children and adolescents are essential for effective service planning, allocation of resources, and development of public policy (1,2). Of special concern to clinicians and policymakers are prevalence rates of youths who experience severely impairing psychiatric disorders that cause substantial disruption in daily functioning (3). Clinically, there is evidence that youths who experience substantial impairment due to a psychiatric disorder have a different course and prognosis than their less impaired peers, are at greater risk for negative long-term outcomes, and require specialized interventions and more intensive levels of care (4,5). Despite this, many of these youths do not receive necessary mental health services (3).

Increasing knowledge of the prevalence of these conditions is a first step toward improved identification and treatment. From a policy standpoint, prevalence estimates of psychiatric disorders characterized by severe impairment are necessary to determine levels of need for services and benefits, accurately assess service gaps, and evaluate the magnitude of unmet need (6).

In the United States, numerous state and federal agencies target mental health services to the population of youths who experience psychiatric disorders that cause substantial functional impairment—a condition referred to in federal regulations as serious emotional disturbance (SED) (PL 102–321). As defined by the Substance Abuse and Mental Health Services Administration (SAMHSA), criteria for SED include persons under the age of 18 who experience a psychiatric disorder that causes substantial impairment in one or more functional domains during the previous 12 months (7,8). States are required to submit SED prevalence estimates in their applications for mental health block grant funds, and federal agencies such as the Centers for Medicare and Medicaid Services often use SED criteria to determine eligibility for services (9). Some federal agencies, such as the Social Security Administration, employ criteria similar to SED, but with more stringent requirements for severity of impairment, to determine eligibility for disability benefits (10). Clinicians are often directed to apply SED criteria when assessing youths as one component of establishing eligibility for mental health services or determining an appropriate level of care (3).

Despite the importance of SED to mental health practice and policy in the United States, precise estimates of the prevalence of SED are lacking. Prior research estimated that 4% to 17% of youths in the United States meet criteria for SED; however, the breadth of these estimates significantly undermines their usefulness for clinical practice or policy (3,8,11). Furthermore, these reviews did not incorporate several recent nationally representative surveys of youth psychopathology sponsored by the National Institute of Mental Health and the Agency for Healthcare Research and Quality (1,12–14). During the past 20 years, several epidemiological studies have been conducted that provide important data points for estimating the true population prevalence of SED in the United States (9,15). However, each study relied on a unique sample and methods and, as a result, produced a unique prevalence estimate. An optimal population prevalence estimate can be derived by pooling these studies by using meta-analytic methods to model sampling variation and other forms of study heterogeneity (16). Nevertheless, none of the narrative reviews summarizing this literature has quantitatively synthesized the results (11,17,18).

The goals of this study were to determine pooled prevalence estimates of SED, based on a systematic review and meta-analysis of population-representative studies conducted in the United States, and to explore predictors of heterogeneity among these estimates.

Methods

Data Sources, Study Selection, and Inclusion Criteria

The search process identified epidemiological studies of psychiatric disorders among community samples of children and adolescents in the United States. We followed PRISMA and MOOSE recommendations for the transparent reporting of systematic reviews and meta-analyses (19,20). With the assistance of a reference librarian, we conducted electronic searches in the MEDLINE and PsycINFO databases, which provide optimal coverage for psychosocial problems (21), accessing articles from January 1, 1980, the publication year of DSM-III, which included explicit diagnostic criteria, through January 1, 2017. Search terms focused on prevalence (“epidemiology” OR “prevalence”), psychiatric disorder (“psychiatric disorder” OR “mental disorder” OR “psychopathology” OR “serious emotional disturbance” OR “severe emotional disturbance” OR “emotional disturbance” OR “severe emotional disorder”), and child and adolescent populations (“child*” or “adoles*”). In addition, studies were extracted from nine previous reviews of childhood psychiatric epidemiology (2,3,6,8,11,17,18,22,23). References from these reviews were extracted and screened, as were the references from a recent National Academy of Sciences report on SED (15). Because of the focus on prevalence of SED in the United States, searches were limited to peer-reviewed articles published in English.

After the initial screening of abstracts and titles, two authors (NW and LS) independently reviewed the full text of articles considered for inclusion and resolved disagreements through discussion with the third author. Studies were selected for inclusion if they collected original, prospective data on the point to 12-month prevalence of psychiatric disorders among persons ages 18 and younger; included a probabilistic community sample in the United States; derived DSM or ICD diagnoses on the basis of structured assessment procedures or assessed youth psychopathology via a standardized, nationally normed quantitative measure; assessed three or more disorders or symptom clusters representing both externalizing and internalizing psychopathology; and included measures of functional impairment. Studies with insufficient data to estimate standard errors were excluded, as were studies that relied on clinical records to make diagnoses, reported lifetime prevalence, were conducted with samples outside the United States, or incorporated response rates less than 50% in the final sampling stage (mean response rate of 75%, range 56%−96%). For longitudinal studies, the most recent wave of data was used whenever possible in order to obtain the most up-to-date estimates.

Data Extraction and Coding

Studies that met the selection criteria were coded for sample characteristics, methodological approaches, and case definitions. Coded variables were selected a priori on the basis of their importance for assessing the reliability and validity of prevalence estimates and their association with variation in prevalence estimates in prior studies. Consistent with prior reviews and conceptual research on SED, prevalence estimates of SED were extracted from the studies at two levels of functional impairment—domain-specific impairment and global impairment (8,11). Domain-specific impairment indicates substantial disruption in role functioning secondary to a psychiatric disorder in at least one functional domain of family, peers, educational settings, or the community (11). This definition meets the minimum criteria set by SAMHSA to identify a youth with SED (24). Global impairment is more severe and indicates substantial impairment of role functioning in multiple domains, operationalized here as two or more domains (11). Because of differences in the prevalence estimates of these two populations as well as likely differences in service needs, estimates of SED prevalence were extracted for both levels of impairment whenever possible. In addition, information was extracted on the measure and criteria (for example, cut score) used to operationalize each level of impairment in each study.

Consistent with prior research, SED estimates that relied on the Children’s Global Assessment Scale (CGAS) to assess functional impairment were classified as domain-specific if they used a cut score of ≤60, which indicates “disturbance would be apparent to those who encounter the child in a dysfunctional setting or time but not to those who see the child in other settings” (4,13). Estimates were classified as global if they used a CGAS cut score of ≤50, which indicates the youth experiences “interference in functioning in most social areas” due to the psychiatric disorder.

In addition to measuring functional impairment, studies were required to include diagnostic procedures for establishing the presence of psychopathology or psychiatric disorder. The literature on youth psychopathology is characterized by two approaches to assessment, one representing a taxonomic approach that relies on structured interviews to derive DSM or ICD diagnoses and the other representing a quantitative approach that relies on elevated symptom scores on standardized rating scales (9). Both approaches have demonstrated adequate sensitivity and specificity for population-based epidemiological studies, and there is considerable interest in examining the equivalence of these approaches; consequently, we included both types of studies in the meta-analysis.

For studies that used a taxonomic approach, we included SED estimates that relied on any decision rule to generate diagnoses (i.e., parent report only, youth report only, or parent or youth report [“or rule”]). For studies that used a quantitative approach, we required that the standardized measure have strong evidence of reliability and validity plus national norms for scoring and clinical cutoff criteria. Validity studies of the two standardized scales included in our meta-analysis (the Strengths and Difficulties Questionnaire and the Columbia Impairment Scale) confirm that they have adequate sensitivity and specificity for detecting psychiatric disorders among youths, and consequently the possibility of over- or underestimating prevalence based on the cut scores used was low (14,25–28).

Data Synthesis

Because of differences across studies in their samples, procedures, measures, and case definitions, the assumption of homogeneity that underlies fixed-effects meta-analytic models was untenable. Consequently, we relied on a random-effects meta-analytic model to estimate the population prevalence of SED at two levels of impairment—domain-specific impairment and global impairment. Each study contributed only one prevalence estimate per analysis. After estimating the prevalence of SED at each level of functional impairment, we subsequently fitted mixed-effects regression models (metaregression) to identify moderators that might explain variation in the prevalence estimates (16,29,30). All models were fitted by using the metafor package in R (31).

Results

Study Pool

The search process yielded 12 studies that estimated the prevalence of SED in the United States (1,4,5,11–14,32–46). [A PRISMA flow diagram of study identification, screening, and selection is available as an online supplement to this article.]

The studies included 10 estimates of SED with domain-specific impairment and eight estimates of SED with global impairment. Six studies provided estimates of SED at both levels of impairment. Most studies, which spanned 1989 to 2015, assessed eight- to 17-year-olds. Four studies incorporated U.S. national samples, and eight included regional samples. Only two studies derived SED prevalence estimates on the basis of quantitative measures. Table 1 presents the study characteristics and estimates of SED at each level of impairment for each study.

Prevalence of SED With Domain-Specific Impairment

The pooled prevalence of SED with domain-specific impairment was 10.06% (95% confidence interval [CI]=8.60%–11.51%). This estimate is based on 10 studies incorporating 32,015 youths. Figure 1 presents the forest plot of each study’s prevalence estimate and CI. Consistent with our random-effects analytic approach, there was significant variance between studies in their prevalence estimates (Q=56.96, df=9, p<.001), suggesting the importance of examining potential moderators.

Table 2 presents the moderator analyses examining variation in the prevalence rates of domain-specific SED by study characteristics. Results of the metaregression analyses indicated that domain-specific SED prevalence did not differ by assessment method (taxonomic versus quantitative), diagnostic time frame (12 months versus less than 12 months), representativeness of the study sample (regional versus national), year of data collection, age range of youths, measure of impairment, sample size analyzed, sampling frame (for example, households versus schools), type of informant, or diagnostic interview, although all of these analyses were considered exploratory because of the small number of studies.

We also calculated separate meta-analytic point estimates and CIs for domain-specific SED prevalence rates for targeted subsets of studies, for example, studies with national samples or studies with a 12-month time frame [see online supplement]. These data provide additional information for evaluating the relationship between study characteristics and domain-specific SED prevalence rates. Results of these analyses are consistent with the metaregression moderator tests but offer increased precision for readers interested in specific prevalence estimates based on particular types of studies.

Prevalence of SED With Global Impairment

The pooled prevalence of SED with global impairment was 6.36% (CI=5.78%–6.93%), based on eight studies incorporating 38,939 youths. Figure 2 shows the forest plot of studies estimating the prevalence of SED with global impairment. There was no significant variance in the prevalence estimates of these studies, suggesting they were homogeneous.

Table 3 presents the moderator analyses of the prevalence of SED with global impairment. Results indicated that prevalence rates of SED with global impairment did not differ by study or sample characteristic. [Separate meta-analytic point estimates and CIs for the prevalence of SED with global impairment in selected subsets of studies are available in the online supplement.]

Discussion

This systematic review and meta-analysis of population-representative studies conducted in the United States suggests that 10.06% of youths experience SED with substantial impairment in one or more functional domains and that 6.36% of youths experience SED with substantial impairment in two or more domains. The large numbers of youth and families affected by SED, the significant long-term consequences of these conditions, and the considerable costs and complexity of treating these disorders underscore the need for comprehensive and effective prevention and treatment services.

Results from this study have important implications for mental health services. First, whereas clinicians in primary care might expect a relatively small proportion of pediatric patients to present with substantial impairment related to a psychiatric disorder, these results suggest that one in 10 youths will likely require treatment or referral to appropriate mental health services. Collaborative care models may be promising in regard to engaging these patients (47,48). Second, the significant challenges in daily functioning experienced by these youths highlight the importance of research to develop and implement community-based services that provide care in the least restrictive setting. Wraparound service models, which feature a growing evidence base, are a promising alternative to restrictive out-of-home placements and have been increasingly embraced by states (49). Third, the category of SED belies the complexity of mental health problems faced by these youths due to a specific diagnosis or set of comorbid diagnoses. Youths with SED are more likely to experience comorbid psychiatric disorders and are at increased risk of treatment failure and dropout (13); consequently, it is important for clinicians to recognize the heterogeneity of diagnoses associated with SED and provide evidence-based care built around valid assessment and diagnosis (50). Examples of evidence-based approaches for youths with SED include multisystemic therapy, functional family therapy, a range of cognitive-behavioral therapies for specific diagnoses, and community-based wraparound services (51,52).

Another area of concern given these estimates is the shortage of child psychiatrists to provide care in community settings (53). Approximately 8,700 child psychiatrists deliver services in the United States, implying an untenable 620:1 ratio of youths with SED per child psychiatrist (54). This situation is exacerbated by the low number of child psychiatry residents and fellows in the United States, who numbered just 869 in 2015, and by an uneven distribution of child psychiatrists across the United States such that more rural and impoverished areas have particularly acute shortages (55–57). Currently, only about 50% of youths with SED receive any mental health treatment within a one-year period (12). Increasing workforce capacity is critical to improving access to mental health services for youths with SED.

Findings from this study also have implications for policy. First, these estimates provide a basis for states to determine rates of SED and thereby establish proxy classes or service targets. Second, the small differences between prevalence estimates based on standardized rating scales versus structured diagnostic interviews suggest that relatively cost-effective standardized measures may be useful for estimating SED prevalence in targeted population areas. Third, the large discrepancy between prevalence of SED with domain-specific impairment and SED with global impairment confirms the importance of impairment criteria in defining this target population. Fourth, these findings suggest that recent concerns regarding the overuse of benefits programs such as Supplemental Security Income (SSI) for youths with major mental disorders may be unwarranted (58). Combining the estimated prevalence of SED with global impairment from this study with data on SSI enrollment suggests that a maximum of 19% (N=654,370) of the 3.4 million American youths with SED with global impairment received SSI benefits in 2013 (10,59).

Limitations of this study included three issues common to systematic reviews and meta-analyses. First, the primary studies included in the analysis may have had flaws that influenced the results. For example, differential nonresponse of youths and caregivers of youths with SED may have attenuated the prevalence estimates of primary studies. If this occurred systematically across studies, the meta-analytic estimates would be downwardly biased.

Second, it is possible that some epidemiological studies of youth psychiatric disorders were not identified during the search process. Studies could have been missed during the search or because they were published outside of our inclusion dates. This threat was diminished, however, by several factors. First, the combination of MEDLINE and PsycINFO databases provides optimal coverage for systematic reviews of psychosocial disabilities (21). By combining searches of these databases with reference harvesting from nine prior reviews, it is unlikely that major prevalence studies were missed. Second, epidemiological studies of disease prevalence do not involve allegiance effects and are therefore less likely to result in unpublished studies (the “file drawer” problem) (16). Third, the pattern of results in both forest plots—which include high and low prevalence estimates distributed around a pooled mean—increases confidence that a range of estimates have been reported in the literature and were systematically identified by the search process.

Third, the heterogeneity in study methods and samples raised the “lumping and splitting” question that is common to meta-analysis—that is, were the included studies all measuring the same construct (16)? Several factors support the inclusion criteria used. First, the federal definition of SED supports the broad inclusion of studies with respect to diagnoses, age range (up to 18), and time frame for diagnosis (up to 12 months). Second, the moderator analyses, although relatively underpowered, offered little evidence to suggest that prevalence rates differed by sample or methodological characteristic. Third, findings from this meta-analysis are consistent with other studies that show that, with the exception of poverty, few youth characteristics are systematically related to rates of SED (2,9). Fourth, the application of uniform criteria developed a priori to classify SED estimates increases confidence that the studies were measuring the same construct.

Conclusions

This meta-analysis of 12 population-representative studies of youths varying in age from two to 18 found that 10% of youths met federal criteria for SED in at least one area of functioning, suggesting that one in 10 pediatric patients is likely to require treatment or referral to appropriate mental health services. The estimates of SED presented here provide clinicians and policy makers with necessary evidence to guide clinical decision making and mental health policy in the United States.