Introduction

Despite federal and local efforts to address unmet mental health needs of students (National Institute of Mental Health’s Advisory Council, 2001; President’s New Freedom Commission, 2003), many students still lack access to or underutilize services, leading to untreated and likely worsened mental health problems (Brenner et al., 2014). Estimates indicate that up to 30–46% of youth in the USA experience some mental health problems and 21% experience severe mental health problems (Merikangas et al., 2010; Kessler et al., 2009). However, more than 50% of youth in need do not receive adequate care (Ghandour, Kogan, Blumberg, Jones, & Perrin, 2012; Merikangas et al., 2011), likely leading to a host of negative outcomes such as health-related concerns, learning difficulties, social problems, and vocational challenges during transition into adulthood (Brooks, Harris, Thrall, & Woods, 2002; Davis, Young, Hardman & Winters, 2011). This unmet need also places a substantial burden on families and society through costly interventions and services delivered years after the initial needs were identified and escalating problems potentially could have been prevented (Soni, 2009).

Predictors of Service Use

Seminal research dating back to the 1990s has examined the issue of unmet mental health need to identify children and youth most at risk of not accessing needed care. Researchers identified differences in service receipt among sociodemographic groups of children and youth, between service settings, in relation to identification and referral practices of parents and educators, and between problem types. To begin, some minority youth are over-identified for special education services in the public school setting (i.e., African-American and Latino youth; Zhang, Katsiyannis, Ju, & Roberts, 2014); however, other sociodemographic groups are under-identified and continue to experience a large, unmet mental health need (i.e., Asian American youth; Guo, Kataoka, Bear, & Lau, 2014). Studies have also documented demographic differences in service receipt where youth from minority backgrounds, rural or urban areas, and/or low-income families are less likely to receive services compared with their white, suburban, and high-income family counterparts (Ghandour et al., 2012; Lewit, Terman, & Berhman, 1997).

Further, nearly all youths with behavioral and mental health needs are more likely to receive mental health care in the school setting than in the community or specialty sector (Burns et al., 1995; Hogue & Dauber, 2013). In fact, research has frequently documented limited access to and utilization of community-based mental health services for all youth (Hoagwood & Johnson, 2003; Lyon, Ludwig, Vander Stoep, Gudmundsen, & McCauley, 2013), especially for minority children and youth in relation to services that cost (e.g., specialty mental health services, general medical services; Costello, He, Sampson, Kessler, & Merikangas, 2014). Service receipt in outpatient mental health settings is associated with parent ratings of youth impairment where higher levels of parent-rated impairment related to accessing outpatient services (Langer et al., 2015). However, in the school setting, school engagement in early identification and referral strategies is associated with service use for adolescents in need (Green et al., 2013).

In relation to problem type, researchers have consistently demonstrated that youth with externalizing behavioral problems are typically referred for and receive services across treatment setting at higher rates than youth with emotional or internalizing concerns such as depression (Chang & Sue, 2003; Costello et al., 2014; Gresham & Kern, 2004; Rothi & Leavey, 2006). Educators also report being more concerned about and more likely to refer youth demonstrating acting out externalizing behavior problems than those experiencing internalizing concerns (Chang & Sue, 2003; Loades & Mastroyannopoulou, 2010). In the community setting, youth with attention, hyperactivity, and behavior-related disorders have higher odds of accessing speciality or community-based mental health services and higher parent-rated impairment than youth with depressive symptoms and other internalizing disorders (Merikangas et al., 2011; Wu et al., 1999).

These studies examining service use and referral practices in relation to the type or dimensions of behavior and mental health problems are critical for best directing the identification and referral efforts of parents and educators. Typically, parents and educators, as well as other informal (e.g., friends) and formal providers (e.g., mental health specialists, child welfare or juvenile justice caseworkers, primary health provider), facilitate children and youths access to mental health care (Stiffman, Pescosolido, & Cabassa, 2004). Stiffman et al. (2004) refer to those who facilitate access to care as “gateway providers” and prior research has found their knowledge, awareness, assessment, and concern for youth’s symptoms and impairment are essential leverages predicting whether or not they take steps to facilitate care for youth in need (Costello, Pescosolido, Angold, & Burns, 1998; Horwitz, Leaf, Leventhal, Borsyth, & Speechley, 1992; Stiffman, Chen, Elze, Dor, & Cheng 1997).

By better understanding the behavioral and mental health characteristics of children and youth who do and do not receive needed care, we can better work with parents and educators to use identification and referral strategies that increase the likelihood all youth in need access services. However, the current portfolio of such studies is limited by almost exclusively relying on variable and/or disorder centered analytic approaches that led to heterogeneous groupings of youth who may have the same, singular diagnosis but exhibit widely different problem behaviors (Brodbeck et al., 2014). For example, two children with social anxiety may manifest symptoms differently (e.g., refusal to come to school versus reluctance to answer questions in large group setting) and/or demonstrate comorbid depressive, inattentive, and/or conduct behavior problems such that teachers and parents may not recognize the youths’ problem as the same. The focus on singular disorders also fails to reflect a general trend of comorbid mental health concerns where more than 50% of people with a mental health disorder meet criteria for at least one other disorder (Demyttenaere et al., 2004; Kessler, Chiu, Demler, & Walters, 2005) and associations between disorders within and between the internalizing and externalizing domains are strong (Kessler et al., 2011). Importantly, comorbidity and higher impairment associated with the presence of multiple disorders are more predictive of higher rates of service use than any one disorder diagnosis (Zahner & Daskalakis, 1997). To better capture the scope and variability of complex mental health needs demonstrated by referred youth and specify groupings that exhibit homogenous behavior patterns across and within problem behavior domains, research using person-centered analyses is needed (Bonadio, Dynes, Lackey, Tompsett, & Amrhein, 2016).

Latent Profile Analysis and Service Use

Latent profile analysis is a person-centered approach that groups children and youth together who are exhibiting similar patterns of behavioral and mental health concerns. Thus, the groups or profiles that emerge each represent a group of children and youth with similar symptoms where the similarity within the group is greater than the similarity between groups. In other words, children and youth in a group exhibit behavioral and mental health problems more like one another than children and youth in another group (Bonadio et al., 2016). This type of analysis may be advantageous in informing the efforts of parents and educators concerned with ensuring all youth in need of mental health services access care (Bonadio et al., 2016; Walrath et al., 2004). By grouping children and youth by similar symptom presentations rather than diagnostic categories, parents and educators not trained to diagnose mental health problems may be better able to identify youth in need of mental health services and facilitate access.

Researchers using person-centered approaches have most often grouped similar behavioral and mental health profiles along the lines of problem type, including three to five groups of children and youth exhibiting some combination of externalizing and/or internalizing behavior problems. For example, Bonadio et al. (2016) identified five groups of symptom profiles among youth ages 9 to 18 and parent dyads receiving services at a community mental health center. The symptom profile groups included a normative group where neither parent nor youth reported behavior in the risk or clinical range, externalizing group where acting out symptoms were reported in the moderate range, internalizing group where internalizing symptoms were reported in the moderate range, high hisk where externalizing and/or internalizing symptoms were rated in the severe range, and delinquent where only externalizing behavior problems were rated in the severe range (Bonadio et al., 2016). Stephens, Petras, Fabian and Walrath (2010) identified three symptom profile groups when considering the clinician-rated functional impairment of children and youth ages 5 to 18. Similar to Bonadio et al. (2016), Stephens et al. (2010) also identified a no problem, or low impairment, group, as well as groups primarily exhibiting externalizing or internalizing concerns. Likewise, Bradshaw, Buckley, and Ialongo (2008) examined the behavioral and academic functioning of children in Grade 1 and also identified three symptom profile groups characterized by no problems or symptoms, primarily internalizing concerns but average educational performance, and primarily externalizing concerns but low educational performance.

Hogue and Dauber (2013) examined symptom profiles of adolescents aged 12–18 who were exhibiting behavioral and/or mental health problems but not receiving treatment. Within their sample of youth with unmet mental health need, Hogue and Dauber (2013) examined person-centered patterns of mental health diagnoses and identified five groups using both parent and adolescent self-report. The basic externalizers group included adolescents characterized by high rates of conduct problems and oppositionality. The comorbid externalizers group included adolescents with high rates of conduct problems, as well as moderate rates of impulsivity and depression. The adolescent distress group included adolescents who self-reported high rates of oppositionality, depression/dysthymia, and inattention, as well as moderate rates of anxiety and traumatic stress. The severely distressed group included adolescents with high rates of nearly every diagnosis. Nearly every member of this group self-reported both defiance and depression. Finally, the parental concern group included adolescents who did not self-report a high level of any diagnosis but caregivers reported moderate rates of oppositionality, defiance and inattention.

Although these studies indicate clear symptom profile patterns along the dimensions of internalizing and externalizing concerns, with more comorbidity and differentiated levels of distress in samples including older youth (Bonadio et al., 2016; Hogue & Dauber, 2013) than those including younger children (Bradshaw et al., 2008; Stephens et al., 2010), few have examined differences in service use history between profiles identified. Similar to existing variable-centric research (Chang & Sue, 2003; Costello et al., 2014), Bradshaw et al. (2008) found children in a person-centered symptom group of externalizing concerns and low educational performance were most likely to receive school-based mental health and special education services. However, those in an internalizing average educational performance group were no more likely to receive school-based services than those with no symptomatology (Bradshaw et al., 2008). Although this study represents one examination of service use differences between person-centered symptom profile groups that are both more homogenous and more identifiable by parents and educators, it is limited by only considering school-based mental health and special education services. Additional research is needed to consider a broader scope of treatment options available and accessed by children and youth in need of mental health services. Although researchers have historically found most children and youth who receive mental health services do so in the school setting (Burns et al., 1995; Lyon et al., 2013), Costello et al. (2014) found nearly equal utilization rates across school (23.6%) and specialty mental health services in the community (22.8%).

Current Study

The purpose of this study is to examine differences in services use between person-centered symptom profile groups when comprehensive list of mental health service settings and options are considered. This approach is advantageous because a person-centered approach will create more homogenous groups of behavioral and mental health concerns, while also pinpointing behavioral patterns identifiable to parents and educators who are most likely to facilitate children and youths access to mental health services. To this end, our research questions were: among adolescents referred for mental health and classroom interventions for their emotional and behavioral problems and school functioning impairment: (1) do different profiles of internalizing and externalizing symptoms exist? If so, how many profiles can be identified and what symptoms characterize each profile?, and (2) do these profiles differ as a function of mental health services received and/or adolescents’ sociodemographic characteristics?

We hypothesized that different internalizing and externalizing behavior symptom profiles would emerge, including at least three to five profiles characterized by externalizing behavior concerns, internalizing behavior concerns and comorbid symptomology, consistent with previous research (Bonadio et al., 2016; Bradshaw et al., 2008; Hogue & Dauber, 2013; Stephens et al., 2010). Prior research examining symptom profiles in a sample of adolescents most similar to participants in our study identified five groups inclusive of youth with comorbid symptom presentation and differentiated levels of symptom distress (i.e., moderate to severe impairment; Bonadio et al., 2016; Hogue & Dauber, 2013). Thus, we anticipate detecting three to five groups distinguished between externalizing and/or internalizing behavior patterns and impairment levels.

Further, we hypothesized differences in sociodemographic characteristics and service use history between these groups. Based on previous variable-centric research, we anticipated referred youth across all profile groups would access most services in the school setting (Lyon et al., 2013); however, we also hypothesized referred youth with externalizing behavior and comorbid problems would be more likely to have accessed community services and more intensive treatments (i.e., pharmaceutical interventions and inpatient care) than youth with internalizing behavior problems or elevated, but less severe concerns, given prior research indicating youth with externalizing behaviors and/or high levels of parent-rated impairment, are most likely to access services in the community sector (Garland et al., 2001; Langer et al., 2015; Wu et al., 1999; Zahner & Daskalakis, 1997).

Method

Participants

Data for the study came from baseline data collection of a randomized control trial funded by the US Department of Education’s Institute of Education Sciences (Center for Adolescent Research in Schools [CARS]; Kern, Evans, & Lewis, 2011). Participants included 647 adolescents who met eligibility criteria and provided consent. Most participants were male (66%, n = 430), in the 9th or 10th grade (82%, n = 531), and low income (annually at or below $20,000 = 35%, n = 227; $20,000–$40,000 annually = 31%, n = 200; and exceeded $40,000 annually = 29%, n = 190). Approximately half were white (52%, n = 337), 39% were black (n = 250), and 5% were Hispanic (n = 34) with equal distribution across suburban locations, rural, and urban areas. Nearly half of study participants were identified as having a special education label (48.5%, n = 314). For the 314 youth with a special education label, the most common disability category indicated on their Individualized Education Plans was learning disability (50%, n = 156) followed by emotional disturbance (25.5%. n = 80), other health impairment (19%, n = 60), and other (5%; n = 15).

Procedures

The purpose of CARS was to develop and evaluate a package of mental health and classroom interventions for high school students with comorbid emotional and behavioral problems, and school impairment. Following a 3-year development phase, the intervention package was implemented within a randomized controlled trial (RCT) for 2 years in 54 schools located across five states (Kansas, Missouri, Ohio, Pennsylvania, and South Carolina). Data for the current study were collected at baseline of the RCT prior to intervention implementation. School personnel (e.g., school counselor, administrator, or special education teacher) in participating schools were asked to nominate approximately 25 students who would be attending 9th through 11th grade in Year 1 of the study and were exhibiting the most severe behavioral, emotional and school problems. The nominating school personnel contacted parents to request permission for study staff to contact them. Study staff met with interested parents to obtain parental consent and youth assent for screening and study participation. Consent for screening was obtained from a total of 857 adolescents. To participate, adolescents had to meet the following inclusion criteria: (a) parent, teacher or self-report measures of emotional or behavioral problems were at least in the at-risk range and (b) at least two areas of school functioning (e.g., discipline, attendance, course grades) were impaired. Adolescents were ineligible if they had a documented intellectual disability or Autism Spectrum Disorder, or an IQ below 70 because of concern they would be unlikely to benefit from cognitive intervention strategies being implemented. Additionally, at least one parent/guardian had to be capable of speaking fluent English in order to complete assessments. Baseline data collection was completed by trained research assistants who typically administered study measures in person in the youth’s home, school, or neutral location (e.g., library or restaurant) but in some cases by mailing the assessment packet home.

Measures

Psychosocial Behaviors

Internalizing and externalizing behaviors were measured by adolescent self-report of depression (14 items) and anxiety (14 items), and parent report of hyperactivity (10 items), aggression (11 items), and conduct problems (9 items) on the Behavior Assessment System for Children, Second Edition (BASC-2; Reynolds & Kamphaus, 2004). The BASC-2 measures emotional and behavioral functioning in children and adolescents (ages 2–21) across clinical and adaptive broad (e.g., externalizing, internalizing) and narrow band (e.g., attention problems, depression) subscales. It is a commonly used measure with high internal consistency (∝ = .84–.96) and test–retest reliability (∝ = upper .70–low .80s) on all composite scales. The BASC-2 Self-Report of Personality—Adolescent (SRP-A) includes 176 items rated on a 4-point Likert scale (Never to Almost Always). Only the depression and anxiety scale scores from the BASC-2 SRP-A were used for the current study. The BASC-2 Parent Rating Scale-Adolescent (PRS-A) includes 150 items also rated on a 4-point Likert scale. The hyperactivity, aggression, and conduct problem scale scores from the BASC-2 PRS-A were used in analyses for the current study.

Prior Mental Health Services Received

Mental health services use was collected from two parent report measures: the Services Assessment for Children and Adolescents (SACA; Hoagwood et al., 2000) and Services for Children and Adolescents–Parent Interview (SCAPI; Hoagwood et al., 2004; Jensen et al., 2004), and extant school records data. Parents responded to three sections of the SACA to capture inpatient, outpatient, and school-based service receipt (e.g., has [child] ever stayed overnight in a residential treatment center/detention center or jail/emergency shelter; has [child] ever received outpatient help or treatment from a community mental health center/ psychiatrist/probation officer), date of service initiation, and closure or ending. Adequate validity, including high concordance (kappas .50–1.0) among reporters and medical records, and test–retest reliability for parent report (kappas .82–.94) have been established for the SACA (Hoagwood et al., 2000).

Parents also completed two categories of the SCAPI to capture pharmacological treatment and other services adolescents received during out of school time. Adequate test–retest reliability (kappas .49–1.0, overall kappa value for all services of 0.97) has been established for the SCAPI; seven of 10 service types had kappa values of 0.75 or higher, indicating excellent reliability (Hoagwood et al., 2004).

Extant school records data were collected at baseline and included identification of the student as special education or general education (0 = general education, 1 = special education).

Participants’ receipt of behavioral health services was examined based on scoring whether the adolescent had ever received a particular service or not (0 = no, 1 = yes). Services assessed by the SACA, SCAPI and school records data for each adolescent were grouped into the following types: community-based psychosocial, school-based psychosocial, pharmacological, and inpatient services.

Sociodemographic Characteristics

Adolescent age and gender at the start of the study was collected by parent report on a demographic information form developed for the study.

Analysis

Analyses were conducted in MPlus 7.31 statistical package (Muthén & Muthén,, 2011). Latent profile analysis (LPA) was used to group participants based on behavioral and emotional problems in combination with theoretical consideration. Five composite scores for behavioral and emotional problems were used: adolescent report of symptoms of (a) anxiety and (b) depression, and parent report of (c) hyperactivity, (d) aggression, and (e) conduct problems. LPA is a person-centered approach that, in combination with theoretical interpretation, identifies groups of youth or profiles differing in experience of symptoms across these five indicators. The result is a group or profile of adolescents similar to one another on their levels of each of the five indicators of internalizing and externalizing behaviors but different from the individuals in the other profiles (McCutcheon, 1987). Multiple model fit indices were examined to determine the appropriate number of profiles, although there is not a definitive test that determines the number of profiles as results must be interpreted in conjunction with substantive theory about the meaningfulness of the profiles (Nylund, Asparouhov, & Muthén, 2007). That is, the substantive meaning of the classes is important in decision making when selecting the solution, especially when the model indices do not clearly support a particular solution. It is important that the model selected results in meaningful characteristics of the profiles that differentiate between the classes and is consistent with theory and existing research. Model fit indices examined included comparative fit indices which reflect values that are relative or compare models rather than being compared to absolute values of fit. Specifically, the Akaike information criterion (AIC), Bayesian information criterion (BIC, Schwartz, 1978) and sample size adjusted Bayesian information criterion (ABIC, Sclove, 1987) were considered, allowing for models of varying numbers of profiles to be compared with lower values indicating better fitting models. Also, the Lo–Mendell–Rubin (LMR) likelihood ratio test (LRT) and the sample size adjusted LMR LRT were used to consider the difference in the likelihood statistic in relation to degrees of freedom between the model tested and the model with one less profile, with significant p values indicating a better solution than the model of comparison (Nylund et al., 2007; Tofighi & Enders, 2008). Although entropy is not used to determine the number of profiles to be selected, it is used as an indicator of how well the profiles adequately represent the data with values greater than .80 being acceptable and those approaching 1 reflecting clear separation of the latent classes or better classification (Celeux & Soromenho, 1996). Multinomial regression was used to compare the service use histories and sociodemographic characteristics (i.e., adolescent gender and age) as predictors contributing to the likelihood of profile membership in one versus another.

Results

Descriptive statistics for study variables including means, standard deviations, and correlations are presented to describe the sample. Adolescents in this sample reported on average normative levels of anxiety symptoms (M = 50.22; SD = 11.06) and depression symptoms (M = 53.43; SD = 12.66) relative to adolescent population norms. Parents reported their adolescents experienced at-risk levels of hyperactivity (M = 66.19, SD = 18.94), aggression (M = 61.81, SD = 13.29), and conduct problems (M = 65.42, SD = 15.23). Domains of functioning were significantly correlated in the expected direction. For example, all three indicators of externalizing problems were positively correlated such that on average higher levels of hyperactivity were associated with higher levels of aggression (r = .698, p < .001) and conduct problems (r = .637, p < .001), and higher levels of aggression were associated with conduct problems (r = .629, p < .001). Adolescent reports of anxiety were also significantly correlated with depression (r = .597, p < .001). There were no significant correlations among indicators of internalizing and externalizing behaviors for this sample.

Model Fit Indices for Internalizing and Externalizing Behavior Profiles

To examine the number of adolescent profiles for the sample of youth with emotional/behavioral problems referred to the RCT for SMH interventions, we examined the results of a single profile through six-profile solution. Although the selection of the best fitting model was not clearly supported by all model fit indices, the five-profile solution was selected based on comparing fit indices across models, ensuring adequate sample size in each group and the interpretation of the meaningfulness of the solution. As summarized in Table 1, although AIC, BIC, and ABIC values declined across an increasing number of profiles up to the four-profile solution, the four-profile solution did not fit significantly better than the three-profile solution. However, the five-profile solution fit better than the four-profile solution based on the LMRLRT and adjusted LMRLRT. Additionally, the five-profile solution yielded variation in profiles that were a better match to previous research and study hypotheses (Bonadio et al., 2016; Hogue & Dauber, 2013). Specifically, the five-profile solution included a profile characterized by internalizing problems that were masked by the three- and four-profile solution. The three-profile solution yielded three profiles reflecting elevated externalizing behaviors which limited their utility to differentiate between dimensions of internalizing and externalizing behaviors. Thus, in determining the most appropriate model and taking into account all model fit indices, previous research, and the theoretical meaningfulness of the profiles derived by each solution, the five-profile solution was selected.

Table 1 Model results for latent profile analyses of internalizing and externalizing behaviors
Full size table

Symptom Characteristics of Behavioral and Mental Health Problem Profiles

To describe the characteristics of the adolescent profiles of internalizing and externalizing behaviors, we present the means of each indicator for each profile (Table 1 and Fig. 1). Profile 1 was the smallest group of adolescents (7.18%) and their mean ratings of anxiety and depression fell in the normative range while their mean ratings of externalizing behaviors were the highest levels of the five identified profiles, including clinically significant levels of hyperactivity, aggression, and conduct problems. As the T-score averages are in the 80s and 90s (see Table 2), this profile is characterized by its Severe Externalizing Behaviors. Profile 2 had the largest number of adolescents in the sample (37.78%) and yielded means in the normative range for internalizing and externalizing behaviors (Normative Behaviors). Note, despite the average T-scores for adolescents in this profile being in the normative range on the broadband BASC-2, these adolescents met eligibility criteria on another narrow-band assessment measure also utilized for eligibility criteria, including either the Disruptive Behavior Disorder scale (Pelham, Evans, Gnagy, & Greenslade, 1992), Multidimensional Anxiety Scale for Children (MASC; March, Parker, Sullivan, Stallings, & Conners, 1997), and/or Reynolds Adolescent Depression Scale (Reynolds & Mazza, 1998). Adolescents in Profile 3, comprising 12.76% of the sample, demonstrated the highest level of internalizing behaviors across the profiles, with anxiety at the cutoff for the at-risk range and depression in the clinical range, and normative range levels of hyperactivity, aggression, and conduct problems (Internalizing Behaviors). Profile 4 was comprised of a third of the adolescents in the sample (33.01%) and had mean ratings of anxiety and depression in the normative range, but elevated mean ratings of externalizing behaviors including hyperactivity and conduct problems in the clinical range and aggression in the at-risk range. The T-score averages for this profile’s externalizing behaviors are in the clinical and at-risk range, however, not as high as the Severe Externalizing Behaviors profile. This profile is referred to as Externalizing Behaviors. Adolescents in Profile 5 (9.25% of the sample) were characterized by elevated internalizing and externalizing behaviors, with anxiety in the at-risk range, and depression, hyperactivity, aggression, and conduct problems all in the clinical range (Comorbid Behaviors).

Fig. 1
figure 1

Mean T-scores of emotional and behavioral problem profiles

Full size image
Table 2 Profile means and standard errors for the five-profile solution
Full size table

Service Use Among Behavioral and Mental Health Problem Profiles

To describe services received based on internalizing and externalizing behavior profiles, the percentage of adolescents who had received a particular type of service was examined as a function of profile membership and can be seen in Fig. 2. Notably, the Normative Behaviors and Internalizing Behaviors had the lowest percentage of adolescents who had ever received any service at about 2/3 (66.8 and 69%, respectively). About 4/5 of adolescents in the Externalizing Behaviors and Comorbid Behaviors profiles had received a service prior to being identified for the CARS study (82.2 and 82.7%, respectively). The Severe Externalizing Behaviors profile had the highest rate with 87.5% of adolescents having previously received a service.

Fig. 2
figure 2

Percentage of adolescents who reported having ever received each service type based on profile membership

Full size image

To test if there were significant differences in the likelihood of having received a particular type of service based on profile membership, a multinomial regression analysis was conducted. The regression model also included adolescent age and gender as covariates predicting profile membership. The overall multinomial regression model was statistically significant, indicating that the set of service variables and covariates reliably distinguished between profile membership (χ2 = 87.28, df = 28, p < .000, R2 = .15). The effect of community-based psychological services (χ2 = 10.65, df = 4, p < .05) and inpatient care (χ2 = 19.81, df = 4, p < .001) were both significant while pharmaceutical treatment, special education services, and school-based psychological services were not significant predictors. Adolescent age also significantly contributed to profile membership (χ2 = 12.62, df = 4, p < .05).

Special Education Services

Special education services received by adolescents in this sample did not significantly differ between the five behavior profiles.

Community-Based Psychosocial Services

Compared to the adolescents with Normative Behaviors (Profile 2), adolescents with Severe Externalizing Behaviors (Profile 1) were significantly more likely to have received community-based services (Wald statistic = 5.02, p < .05) with a relative risk ratio of 2.768 (CI 1.136–6.745) as were adolescents with Externalizing Behaviors (Profile 4) (Wald statistic = 5.25, p < .05) with a relative risk ratio of 1.765 (CI 1.086–2.870), as well as those with Comorbid Behaviors (Profile 5) (Wald statistic = 5.01, p < .05) with a relative risk ratio of 2.42 (CI 1.116–5.248). There were no differences between adolescents with Internalizing Behaviors (Profile 3) and other profiles in having received community-based psychosocial services.

School-Based Psychosocial Services

The five internalizing and externalizing behavior profiles did not significantly differ in the extent to which adolescents had ever received school-based psychosocial services.

Pharmaceutical Treatment

Compared to adolescents with Normative Behaviors, those with Externalizing Behaviors were significantly more likely to have received pharmaceutical treatment (Wald statistic = 5.83, p < .05) with a relative risk ratio of 1.869 (CI 1.125–3.106). Adolescents with Externalizing Behaviors were also significantly more likely to have received pharmaceutical treatment in comparison with those with Severe Externalizing Behaviors (Wald statistic = 3.86, p < .05) with a relative risk ratio of 2.387 (CI 1.002–5.688).

Inpatient Treatment

Compared to the adolescents with Normative Behaviors (Profile 2), the adolescents with Severe Externalizing Behaviors were significantly more likely to have received inpatient care (Wald statistic = 9.50, p < .001) with a relative risk ratio of 3.369 (CI 1.556–7.293). Relative to the adolescents with Severe Externalizing Behaviors, adolescents with Internalizing Behaviors were significantly less likely to have received inpatient care (Wald statistic = 14.35, p < .000) with a relative risk ratio of .133 (CI .047–.378). Also, relative to the Severe Externalizing Behaviors profile, adolescents with Comorbid Behaviors were significantly less likely to have received inpatient care (Wald statistic = 4.73, p < .05) with a relative risk ratio of .358 (CI .142–.904). Finally, in comparison with the adolescents with Internalizing Behaviors, adolescents with Externalizing Behaviors were significantly more likely to have received inpatient care (Wald statistic = 8.53, p < .01) with a relative risk ratio of 3.649 (CI 1.531–8.700).

Demographic Covariates

Adolescent age as a covariate was also a significant predictor of profile membership such that being younger was associated with being in the Externalizing Behaviors profile compared to the Normative Behaviors profile ((β = − .36, p < .001) and the Internalizing Behaviors profile ((β = − .35, p < .05). There were no significant differences between profiles based on adolescent gender.

Discussion

This study, the first of its kind to explore profiles of internalizing and externalizing symptomology among a group of referred adolescents and service use histories across a comprehensive range of treatment options, found five distinct profiles of adolescent functioning, which had significant associations with service use. These findings have the potential to provide a better understanding of service use patterns among youth in need of mental health services to inform the identification and referral practices of parents and educators who are most likely to facilitate children and youth’s access to care. To this end, it is important to consider study findings in light of both the symptom profiles identified and service use histories examined.

Profiles of Social, Emotional, and Behavioral Symptoms

Consistent with our hypotheses, we detected groups of adolescents displaying elevated internalizing behavior problems and comorbid behavior problems. We also detected two groups of youth with externalizing behavior problems differentiated on level of severity—one with clinical severity (T-scores ranging from 67 to 72) and one with severely elevated clinical externalizing behavior problems (T-scores ranging from 86 to 95). Combined, these groups displaying elevated to severely elevated externalizing behavior problems comprised approximately 40% of the referred sample. In comparison, only 12.8% of the sample displayed heightened internalizing problems and 9.25% displayed comorbid behavior problems. Although this distribution is consistent with prior research examining symptomology in relation to SMH services (Bradshaw et al., 2008; Hogue & Dauber, 2013), it is inconsistent with nationwide prevalence estimates indicating youth ages 6–17 are more likely to be diagnosed with an internalizing behavior problem (nearly 8%) than an externalizing behavior problem (5.4%) in their lifetime (Ghandour et al., 2012). For older youth (ages 12–17), the odds of an internalizing diagnosis are nearly double that of younger children, but externalizing problems remain relatively unchanged across age groups (Ghandour et al., 2012). It is important to note that the rates reported by Ghandour et al. (2012) are based on parent-reported diagnoses by a health care professional rather than health records or diagnoses by a mental health professional and thus may not be not representative of the rates of the problems in the population. It may be that our findings are more reflective of the prevalence of these problems in schools than studies that only rely on healthcare professionals’ diagnosis of a problem. Alternatively, due to limitations of our sampling strategy (i.e., referral from school personnel; see Limitations and Future Research Directions section below), it may also be that we over-identified youth with externalizing behavior problems and under-identified youth with internalizing behavior problems, a typical problem when relying on school-based referrals (Bradshaw et al., 2008).

In light of these findings, schools should consider expanding their identification strategies. Youth with internalizing behavior problems were not referred at expected rates and thus universal screening and mental health literacy training may be needed to adequately identify all youth in need of mental health services (Dowdy et al., 2015; Papandrea & Winefield, 2011). Universal screening would help to proactively examine emotional and behavioral patterns of all youth regularly such that any problems, both internalizing and externalizing, could be identified early (Dowdy et al., 2015). Further, training educators and even parents in identifying less observable internalizing behaviors (De Los Reyes & Kazdin, 2005) and being concerned for such problems, despite no perceived educational impact (Bradshaw et al., 2008), may be necessary to close the gap in unmet need between youth with internalizing and externalizing concerns (Green et al., 2017; Papandrea & Winefield, 2011).

There was also a large group of youths within the referred sample who displayed normative or average levels of internalizing and externalizing behavior problems as measured by the BASC-2. Although this finding is consistent with previous research where a no problem or no symptom group was also detected (Bonadio et al., 2016; Bradshaw et al., 2008; Stephens et al., 2010), it was not expected based on our sampling procedures where school professionals were asked to refer youth with the most severe behavioral, emotional and school problems and adolescents only qualified if they met inclusionary criteria requiring impairment in both (a) emotional or behavioral problems and (b) school functioning. These youths were eligible for the study based on elevated internalizing or externalizing behavior problems rated on other narrow-band measures. Their behavior was not rated in an elevated range across all study measures and, thus, may present with less severe symptomology or functional impairment. We discuss the service use history and intervention needs of adolescents in this class in the next section.

Patterns of Service Use Histories and Demographic Characteristics

Previous studies have indicated a large unmet mental health need, as well as differences in service receipt based on diagnoses of internalizing and/or externalizing disorders and demographic characteristics (Bradshaw et al., 2008; Burns et al., 1995; Lyon et al., 2013). However, no recent studies have examined the relationship between a profile of presenting problem(s) and service receipt across a full range of treatment or service setting options. Of concern, 11 (Severe Externalizing Behaviors) to 33% (Internalizing Behaviors and Normative Behaviors) of adolescents in each class never received a single service. However, all symptom profiles had higher receipt of any service than those reported more than 20 years ago in the Great Smoky Mountain Study (Burns et al., 1995). Whereas Burns and colleagues reported that only about 30% of children and youth with mental health diagnoses and/or impairment received services in the last 3 months, 66–87.5% of our sample reported some lifetime receipt of mental health services. The more than twofold increase is encouraging; however, important methodological differences should be noted. Burns and colleagues measured receipt of services within the last 3 months, whereas we queried lifetime mental health services. Further, Burns and colleagues used a representative sampling and behavior screening technique, whereas we used school referral and broadband assessment of internalizing and externalizing behavior problems to identify participants. Our procedures may be more likely to identify youth known to school personnel as being in need because of prior intervention services than the sampling technique used by Burns et all. (1995). Both these differences would likely lead to higher service receipt rates among our sample than the Great Smoky Mountain Study sample (Burns et al., 1995). Nonetheless, the increase in reported mental health service receipt is substantial and should be acknowledged.

Those profiles without elevated levels of externalizing behaviors (i.e., Profile 2 Internalizing Behaviors and Profile 3 Normative Behaviors) had the lowest rates of service use among groups (at 66.8% and 69% of the sample, respectively), while the Severe Externalizing Behaviors profile had the highest level of service use (87.5%). Thus, not only were youth with internalizing behavior problems referred at lower rates than those with externalizing concerns, they also reported receipt of fewer lifetime services. Again, identification strategies such as universal screening may be needed in order to identify this group of adolescents with unmet mental health need given the difficulty parents and educators have in observing such behaviors (De Los Reyes, & Kazdin 2005). However, it may also be that parents and educators are less concerned about these behaviors or are not aware of their functional impairment on the adolescents’ life (Loades & Mastroyannopoulou, 2010). Green et al. (2013) found parent ratings of impairment to be the strongest and most consistent predictor of adolescent mental health service use across mental health disorders and treatment options. Thus, raising the concern and awareness of parents and educators may be needed to move from universal screening for identification to actual service receipt (Papandrea & Winefield, 2011).

It is expected that youth with fewer behavioral and mental health concerns or less impairment (i.e., Profile 3 Normative Behaviors) would not have high rates of historical mental health service use. Nonetheless, a large number of youth were referred for services using the RCT’s sampling procedures yet demonstrated normative behavior patterns. From a prevention framework, these youths may represent an opportunity to provide early intervention services and mitigate the development of more severe and impairing concerns. In order to identify youth with similar behavior patterns (i.e., emerging or mild symptomatology), early identification practices such as regular universal screening may be needed. Green et al. (2013) found schools engaged in early identification strategies were associated with early intervention service use among adolescents with mild behavior problems. Further, school and community mental health systems may need to expand their capacity to provide early intervention services to children and youth of all ages. Although such services are common in early childhood settings, findings from the current study indicate the need to both continue monitoring behavioral and mental health needs into adolescence and have early intervention services available.

Although prior research has found youth are more likely to access school-based mental health services than clinic-based services, including special education and psychosocial intervention, participants in our sample reported accessing community-based psychosocial and pharmaceutical treatments at equal or higher rates than school services (Burns et al., 1995; Hoagwood & Johnson, 2003; Lyon et al., 2013). Receipt of special education and/or school-based psychosocial services was equal across symptom profiles; yet, those displaying some level of externalizing behavior concerns (Profile 1 Severe Externalizing Behaviors, Profile 4 Externalizing Behaviors, and Profile 5 Comorbid Behaviors) reported significantly higher rates of accessing community-based psychosocial and/or pharmaceutical services than youth in the Internalizing Behaviors (Profile 3) and Normative Behaviors (Profile 2) groups.

Although these differences in service receipt between symptomology groups are consistent with prior findings, our finding favoring more receipt of community psychosocial and/or pharmaceutical treatments than school-based services across symptom profiles is not. This could be related to the fact that most of the schools in the sample had very few mental health professionals working in the buildings prior to the RCT (Hoover-Stephan, Weist, Kataoka, Adelsheim, & Mills, 2007). In fact, Green et al. (2013) found fewer school-based mental health professionals associated with lower rates of school-based service use. Mental health systems for children and youth should consider strategies for increasing the availability of mental health professionals in school settings. School mental health services have demonstrated positive outcomes (Bradshaw, Mitchell, & Leaf, 2010; Walker, Kerns, Lyon, Bruns, & Cosgrove, 2010; Wells, Barlow, & Stewart-Brown, 2003) and enhanced access especially for those who are more likely to go undetected or untreated (Atkins et al., 2006; Bruns, Walrath, Glass-Siegel, & Weist, 2004; Catron, Harris, & Weiss, 1998; Merikangas et al., 2011). Further, for those youth who receive mental health services, school mental health services are most likely to be the first sector accessed and a gateway to services in the community or specialty mental health sector (Wood et al., 2005).

Nearly 57% of the sample reported pharmaceutical treatment for mental health concerns with highest rates reported by those in the comorbid (Profile 5), externalizing (Profile 3) and severe externalizing (Profile 1) groups. Across groups, all rates of accessing pharmaceutical treatment were higher than those reported for school-based psychosocial services. This is particularly troubling given school-based psychosocial services have less associated risk and, when implemented well, are more accessible (Comer, Olfson, & Mojtabai, 2010; Correll et al., 2009).

Limitations and Future Research Directions

Despite notable and informative findings, there are some limitations to our study. First, participants were from a larger RCT (see Kern et al., 2015), with entry criteria based on having serious emotional and behavioral problems as operationalized by scores on standardized measures, and school problems, such as discipline, attendance, and/or academic performance concerns indicated on school records data. Normative scores of 37.8% of participants on the BASC-2 alone suggest that many youths in the sample may not have had serious emotional and/or behavioral problems. It is also possible that broad band measures do not sufficiently detect particular types of emotional and behavioral problems that were better detected on the RCT’s narrow band measures. On the other hand, our sampling procedures may have resulted in a generally more severe sample, resulting in high utilization rates of specialty mental health services than school-based services. Therefore, there remains a need to replicate the study in samples of youth without such specific eligibility requirements. At the same time, considerations of both emotional and behavioral functioning and school performance are common in referral and intervention decision making (Bradshaw et al., 2008) and thus our findings may actually reflect referral patterns in samples independent of any ongoing research study.

In addition, assessment of referred youths’ externalizing and internalizing behavior problems was limited to one time point for purposes of this study; hence, stability of these symptom profiles cannot be assured and ratings from parents and adolescents could be reflective of recent concerns, lifetime or ongoing concern, or even re-occurrence of a problem previously treated and in remission. Conversely, participants’ receipt of mental health services was collected as a lifetime, retrospective construct. In most prior studies, the sample was only inclusive of youth currently receiving services (Bonadio et al., 2016) or only queried service use within the last three to 12 months. However, prior service use is a strong predictor of current and future use (Langer et al., 2015) and we aimed to capture all occurrences of mental health service receipt to best examine who has and has not accessed each service sector. Despite this study strength, our results should still be interpreted with caution given the mismatch of measuring service use as a lifetime construct and symptomatology as a current problem. Additionally, our measurement of mental health service receipt may be subject to some inaccuracies related to parents’ difficulty recalling their adolescents’ lifetime mental health service utilization and our use of one data source for each construct studied. That is, internalizing behavior problems were measured by youth self-report, externalizing behavior problems by parent report, and mental health service use by parent recall. Taken together, these measurement issues may have impacted our findings. For example, our finding of low utilization of school-based mental health services may be due to the fact that parents are typically less involved in and sometimes even unaware such services, leading to difficulties recalling them. Future research should seek to replicate this study with more robust measurement of behavioral and mental health concerns, as well as ongoing, rather than historical, reporting of services received (see Bradshaw et al., 2008).

With regard to our analytic strategy, we chose to examine demographic characteristics, including gender and age, as covariates of service type to predict profile membership. We considered also using these demographic variables to predict service type given existing research identifying gender and age differences in mental health services accessed and received (Cohen & Hesselbart, 1993; Zwaanswijk, Van der Ende, Verhaak, Bensing, & Verhulst 2003). However, we examined this research question in prior analysis with the CARS dataset and found gender was not a significant predictor of differences in the likelihood of adolescents having received community-based, school-based psychosocial, pharmacological or inpatient treatment (George, Zaheer, Kern & Evans, under review). Instead, white adolescents and those in special education were more likely to have received services, particularly community-based and pharmacological treatment. These limitations notwithstanding, the study included a large sample of high school youth presenting emotional and behavioral challenges, from 54 high schools across five states, providing support for the generalizability of findings.

Conclusion

Our findings point to important considerations for improving children’s mental health services such that the gap in unmet need is closed. Alarmingly, we found 11–33% of referred youth reported no receipt of mental health services and those who did mostly accessed services outside of the school building. More specifically, youth receipt of community-based psychosocial services and pharmaceutical treatments was equivalent to or higher than reported rates of accessing services in the school. Research indicates that services outside of the school are often times less accessible, costlier, and have higher associated health and behavior risks (Correll et al., 2009). At the same time, receipt of school-based mental health services is associated with higher numbers of mental health professionals working in the school (Green et al., 2013). Although the service receipt rates we identified are overall higher than those identified more than 20 years ago (Burns et al., 1995), large unmet mental health need and access disparities remain. To offer healthier, more accessible treatment alternatives, there remains a need for children’s mental health systems to expand their school-based capacity, making school mental health services the norm rather than the exception in schools nationwide.

We also found concerning disparities in referral and service receipt between youth with internalizing and externalizing behavior problems, as well as a group of youth referred for services but without elevated behavioral or emotional problems. These findings emphasize the need to improve identification, referral and early intervention practices. Universal mental health screening should be considered as an additional referral mechanism, rather than solely relying on teacher identification and referral (Bradshaw et al., 2008). Additionally, schools and mental health professionals should train teachers and parents to identify and refer youth with mental health concerns with specific attention to the presentation of internalizing concerns and early warning signs (Papandrea & Winefield, 2011).