Behavioral Health Diagnoses Among Children and Adolescents Hospitalized in the United States: Observations and Implications

Mental and substance use disorders (behavioral disorders) affect 13%−20% of children and adolescents in the United States (1). Studies indicate that 3%−10% of hospital discharges are for children with a primary diagnosis of a behavioral disorder and that these conditions were among the top five most costly conditions for children ages 17 or younger in 2012 (2–4). Such findings underestimate the full extent of behavioral disorders among hospitalized children because they omit disorders that are comorbid to other primary diagnoses. In contrast, this article provides nationally representative estimates of hospitalizations for children under age 21 in general and children’s hospitals with any behavioral health diagnosis, which we refer to herein as behavioral disorders or diagnoses. By considering secondary diagnoses of behavioral disorders, this study offered a more accurate picture of behavioral health services for hospitalized children (3,5–7). We recognize the biological origins of behavioral disorders; in this article we favor parsimonious language over precision and adopt the imperfectly contrasted “other” or “physical health” diagnosis to juxtapose with behavioral diagnoses.

We sought to enhance the understanding of the use of behavioral health services for children hospitalized in general and pediatric hospitals in the United States. Retrospective analysis of nationally representative data allowed us to estimate the rates of hospitalizations per capita—that is, the rates per person in the specified population. We looked at children and young adults with a behavioral disorder as a primary or secondary diagnosis and describe systematic variations in these rates by age, race-ethnicity, hospital characteristics, insurance status, and geographic region and estimate additional costs and length of stay when behavioral disorders are comorbid with other primary diagnoses.

Methods

We analyzed the 2012 Kids’ Inpatient Database (KID), part of the Agency for Healthcare Research and Quality’s (AHRQ’s) Healthcare Cost and Utilization (HCUP) project. KID data includes a random sample of 10% of uncomplicated births and 80% of all other pediatric discharges from sampled hospitals in 48 states. KID is designed to produce the most valid and precise estimate of national hospitalizations of U.S. children from birth to age 20 admitted to nonfederal general and children’s hospitals, with use of sampling weights that incorporate hospital characteristics from the American Hospital Association database (4,8). KID includes the principal diagnosis and up to 24 secondary diagnoses, 15 procedure codes, admission and discharge status, patient demographic characteristics, and other characteristics, such as charges and length of stay (4). We used a cost-to-charge ratio provided by HCUP to transform charges into cost. The 2012 U.S. Census data provided denominators for the population (per capita) rates (9). Kaiser Family Foundation data permitted us to provide national estimates of insurance status by age, supporting population-based estimates of hospitalization by insurance status (10). ICD-9-CM codes were used for calendar year 2012, and all diagnosis data were based on these codes. The first diagnosis was considered primary, and all subsequent diagnoses were considered secondary. [A table showing diagnostic classifications by ICD-9-CM codes is included in an online supplement to this article.] We developed the classification scheme from previously published algorithms (3,11) slightly modified for this study as a result of a supplemental review of diagnosis codes for pediatric hospitalizations. All other diagnoses were considered to be other (nonbehavioral) diagnoses. We used a standard algorithm from the Centers for Disease Control and Prevention to identify admissions for childbirth (12).

We analyzed all hospital discharge records of patients ages 0–20 with primary or secondary diagnoses classified as a behavioral disorder, which we sorted into 12 diagnostic categories, with suicide attempts treated as a distinct category [see table in online supplement]. This study was exempted from institutional review board approval and informed consent.

Using standard methods and SAS 9.4, we conducted univariate and bivariate analyses to describe the sample and chi-square tests to compare across categories accounting for sampling weights (13). Proc SurveyLogistic was used to examine racial-ethnic differences after adjustment for the indicated variables.

All numerators were from KID data. Denominators for estimating population rates came from the 2012 U.S. Census estimates, including specified subgroups (age, sex, and race-ethnicity), supplemented by Kaiser Family Foundation data regarding insurance status (10). The denominator to calculate the rates of behavioral diagnoses of peripartum psychiatric disorders per delivery was the number of admissions for childbirth for women ages 12–20 (12).

The study’s unit of analysis was a hospitalization. KID lacks individual identifiers that would allow for analysis by individual, linkages among episodes of care, or determination of whether a hospitalization is a readmission or a transfer.

We estimated differences in cost and length of stay associated with behavioral disorders in the population (ages six to 18) admitted with physical health diagnosis. We enhanced our control of confounding by using propensity score analysis (14,15) to compare these discharges with and without a diagnosis of a behavioral disorder, after adjusting for other covariates: age, gender, race-ethnicity, and diverse clinical variables, including the clinical nature of the primary diagnosis (for which we used categories defined in AHRQ’s clinical classification software) (16). Propensity analysis was confirmed by using standard practices (17,18). [Details of the propensity analysis are available in the online supplement.] All statistical tests reported were two-tailed, with a predetermined level of significance of p<.05.

To improve our comparison of like to like, the analysis included only children discharged home. Our rationale was that hospitalizations resulting in transfer would have underestimated the cost of care (because a transfer will look like a discharge). Furthermore, we expected that patients with primary physical health and comorbid behavioral disorders would be more likely to require transfer to a facility better equipped to handle the behavioral disorders; therefore, the exclusion of these admissions likely biased our findings toward the null, meaning that our estimates are conservative. By not including admissions that resulted in transfer, we likely provide a low estimate of the excess length of stay and cost for children with primary general medical and comorbid behavioral disorders.

Results

There were 6,675,222 pediatric discharges from general or children’s hospitals in 2012: 7.3% had either a primary behavioral diagnosis (253,984 hospitalizations, 3.8%) or a comorbid behavioral disorder diagnosis (228,854 hospitalizations, 3.4%) (Table 1). The rates per 1,000 children ages 0–20 in the U.S. population were 2.9 for hospitalizations with a primary behavioral diagnosis, 2.6 for hospitalizations with a secondary (only) behavioral disorder diagnosis, and 5.5 for hospitalizations with any listed (primary or secondary) behavioral disorder diagnosis. Children under age five constituted a distinct subgroup, including 4,354 admissions for neonatal addiction (three of every 1,000 children ages 0–5 in the U.S. population).

Depression (34%) and other mood disorders (31%) were the most common primary behavioral diagnoses, followed by psychotic disorders (9%) and substance use disorders (7%). The most frequent behavioral diagnoses that were not in the primary diagnosis position (that is, comorbid with a primary diagnosis of another physical health condition) were depression (26%), attention-deficit disorder (ADD) (26%), and substance use disorders (22%). Almost no children were admitted with a primary diagnosis of suicidal ideation or self-injury (.1%); suicidal ideation or self-injury was usually coded as the secondary diagnoses (12% of all discharges with a comorbid behavioral disorder diagnosis) (Table 1). The most common primary diagnosis among patients discharged with a diagnostic category of suicidal ideation or self-injury was poisoning by analgesics.

As expected, hospitalizations varied by age groups (Table 1). Admissions for primary behavioral diagnoses occurred at a rate (per 1,000 children in the U.S. population) of .09 for children ages zero to five, 1.03 for those ages six to 11, 5.31 for those ages 12–18, and 7.41 for those ages 19–20 (p<.001). Adding secondary behavioral disorder diagnoses increased the rates per 1,000 to .56, 2.04, 9.11, and 15.25 admissions, respectively (p<.001).

Other mood disorder was the most common (39%) primary behavioral diagnosis for children ages six to 11 (Table 2). Depression was most common for those ages 12–18 (41%) and those ages 19–20 (28%). Substance use disorder diagnoses peaked at 27.7% of admissions with a primary diagnosis of a behavioral disorder for those age 20; substance use disorder is uncommon before adolescence. Notably, three of four admissions with a primary diagnosis of a behavioral disorder for children ages zero to five (and 96.2% of children admitted before age one) were infants with neonatal addiction.

Peripartum psychiatric disorders among teenage mothers bear scrutiny. Nine of every 1,000 admissions for delivery among females ages 12–20 were for a primary diagnosis of a behavioral disorder and 59 of every 1,000 admissions for delivery among females in this age group had one or more behavioral disorders coded at discharge. The most common behavioral diagnoses for admissions for delivery in this age group were depression (24 of every 1,000 admissions) and substance use disorders (23 of every 1,000).

Psychiatric hospitalizations varied by race and insurance status (Table 1). Overall rates (per 1,000 children in the U.S. population) of admission with primary behavioral disorders for blacks, whites, Hispanics, and Asians were 3.20, 2.89, 1.42, and .73 (p<.001). For admissions with any primary or secondary behavioral diagnosis, the rates per 1,000 child population were similar: 5.88, 5.61, 2.89, and 1.42, respectively. After adjustment for patient age, sex, insurance, rural versus urban county residency, and income, black-white differences were no longer significant, with lower rates for Hispanic and Asian children, compared with white children (p<.001).

Regarding insurance status, admissions with a primary behavioral diagnosis or admissions with any behavioral diagnosis were most common for children with public insurance (2.94 and 5.47 per 1,000 children in the population, respectively), compared with admissions of privately insured children (1.97 and 3.63 per 1,000, respectively) and admissions of uninsured children (.97 and 1.72 per 1,000) (p<.001 for both primary and secondary behavioral diagnoses) (Table 1).

Children admitted with a primary or secondary behavioral diagnosis often had more than one behavioral diagnosis. For example, among children admitted with a primary diagnosis of ADD, 43% had comorbid oppositional disorders; among those admitted with a primary diagnosis of depression, 53% had a secondary diagnosis of self-injury or suicidal ideation. Other mood disorders were comorbid with suicide or self-injury (35%) or substance use disorders (26%), and discharges for posttraumatic stress disorder also had a high rate of suicide or self-injury diagnoses (40%) [see online supplement].

Behavioral disorders were also common secondary diagnoses among children admitted with other primary diagnoses. Table 3 shows the most frequent behavioral comorbidities for children ages six to 18 who were admitted for specific primary diagnoses that are not behavioral. [A table in the online supplement shows the most frequent primary diagnoses that are not behavioral and the most commonly associated behavioral disorders.] Poisoning and epilepsy were the most common physical diagnoses associated with a comorbid behavioral diagnosis (8.5% and 5.3%, respectively, of all hospitalizations with the specified primary diagnosis).

The most common behavioral diagnoses for children hospitalized with other nonbehavioral primary diagnoses were ADD (34%), depression (25%), and anxiety (23%) (Table 3). ADD was most frequently comorbid with epilepsy and asthma. Almost one in five hospitalizations for a physical diagnosis in which depression was a secondary diagnosis was for poisoning (10% by analgesics and 9% by psychotropic agents). Four percent of children with a secondary diagnosis of anxiety were admitted for chemotherapy, and five percent were admitted for general symptoms without a more specific principal diagnosis.

We observed substantial variation in hospitalizations for a primary behavioral diagnosis by hospital size, ownership, location, and teaching status (Table 4). Rates per 100 hospitalizations were higher in large versus small, urban teaching versus rural, and governmental versus private hospitals. Rates of hospitalization (per 1,000 children in the population) for children with a behavioral diagnosis were highest in the West North Central census division (5.2 for hospitalizations with a primary behavioral diagnosis and 8.5 for hospitalization with either a primary or secondary behavioral diagnosis) and lowest in the Pacific census division (1.6 and 3.7, respectively).

Among children admitted with a primary general medical diagnosis, the presence of a behavioral diagnosis was associated with a longer stay, from 3.3 to 4.3 days, and higher costs (mean=$2,813; median=$565; all comparisons p<.001) (Table 5). As predicted, children with behavioral diagnoses were more likely to be transferred to other health care facilities, compared with those who did not have behavioral comorbidity (8.0% versus 2.2%, p<.001).

Discussion

This study found that children were commonly admitted to general and children’s hospitals for a primary diagnosis of a behavioral disorder. Management of such disorders for children in general and children’s hospitals is nearly twice as common when psychiatric comorbidity is taken into account and approaches half a million hospitalizations each year. Study of those with both primary and secondary behavioral health diagnoses led to several anticipated and unanticipated findings.

Children hospitalized for primary diagnoses other than behavioral disorders have longer lengths of stay and higher costs if they have any behavioral comorbidity. These results improve on previous findings (4) by reporting not charges but costs, by propensity matching the hospitalizations with and without behavioral disorders, and by improving the like-to-like comparison by including only hospitalizations for children discharged home (7). Considering our mean estimate of the difference in costs ($2,813) and the number of discharges home of children with comorbid behavioral conditions, the total additional cost in 2012 associated with managing psychiatric comorbidities among hospitalized children was $1.36 billion, not including the unmeasured excess costs for hospitalizations that resulted in transfer to other health care facilities. Our findings may be analogous to findings of higher costs among adults with cardiac diseases complicated by depressive disorders (19).

Behavioral disorders were frequently comorbid with one another. Prior studies that examined comorbidity have typically focused on anticipated combinations (for example, posttraumatic stress disorder and depression, anxiety, or substance use disorders [20]). Our findings extend those from previous studies (3–5,13) by providing empirical estimates of the spectrum of comorbid behavioral disorders, regardless of primary diagnosis. Our findings, such as suicidality comorbid with other mood disorders and posttraumatic stress disorder and anxiety comorbid with cancer chemotherapy, should stimulate innovative clinical investigation. The frequency with which both medical and behavioral health diagnoses co-occurred for a hospitalization suggests a benefit to considering how integrated care might best be delivered in the inpatient setting.

The rate of hospitalizations (per 1,000 children in the population) with a primary or secondary behavioral diagnosis was nearly 50% higher among children with public insurance compared with those with private insurance. Competing and complementary hypotheses could explain our finding. Eligibility for Medicaid is associated with a range of adverse social factors including poverty, single-parent families, and other stresses. For example, the most disabled children covered by Medicaid may receive Supplemental Security Income, and this population generally has higher inpatient utilization than other children (21). Perhaps publicly insured families have less access to resources and support networks that would enable use of alternatives to hospitalization. Alternatively, literature on the effects of biological stress (22,23), toxic stress (24), and other social determinants on health and disease would support poverty as a contributing factor to a higher incidence and prevalence of behavioral health disorders in the lower income population served by Medicaid. In addition, the relative financial advantage of patients with private insurance may result in distinct or even adverse outcomes: stigma, tougher preauthorization screening in order to be hospitalized, or clinician styles that favor other diagnoses over behavioral disorders. Children with private insurance may have better access to psychiatric hospitals that are not included in the KID database. Finally, differential access to medications, outpatient treatment, or day treatment centers might contribute to the large differences that we observed. Given that so large a proportion of U.S. children have public insurance, these differences raise critical questions to consider and address as we design accountable health care for children in this era of transformation. Children with no insurance appeared to be at a disadvantage regarding behavioral health care, and their low rate of admission is disturbing.

One implication of our findings is that there is a disparate financial burden for inpatient mental health care on public insurers compared with private insurers. Behavioral health care was found to be more likely to be provided in public than in private hospitals.

Threefold variations across U.S. Census regions were seen in rates of admissions for primary behavioral diagnoses, and twofold variations were evident for hospitalizations with any behavioral diagnosis. Geographic variation in health services utilization and quality have been well studied for general medical conditions among adults (25) but less so for adults with behavioral disorders (26) and even less for children with behavioral disorders (27,28). Culture and demography may contribute but do not appear sufficient to explain the observed geographic differences. Infrastructure differences in ambulatory care, insurance policies, practice patterns, and availability of beds in psychiatric units and hospitals could be contributing factors and need to be studied.

Behavioral diagnoses were disproportionately prevalent among girls and young women during pregnancy, and pregnant adolescents constituted a substantial proportion of adolescents admitted with any psychiatric diagnosis. Psychiatric hospitalizations for girls and young women admitted for pregnancy, delivery, or postpartum care was 6% of hospitalizations with any behavioral disorder diagnosis, and behavioral disorders were present for 59 of every 1,000 children in the population hospitalized for delivery. This finding underlines the importance of looking beyond principal diagnosis when studying behavioral disorders among children, adolescents, and young adults. There are plausible arguments that psychiatric conditions may constitute a risk of unintended pregnancy. For some adolescents and young women, pregnancy may represent a stressor that leads to a behavioral disorder. We hope that these findings will lead to enhanced resources in services for these adolescents and young women and for research to identify opportunities to reduce psychiatric morbidity among adolescents and neonatal addiction syndrome among their offspring.

The extent and increasing incidence of neonatal addiction are of concern (29–31). Children who are exposed to opioids in utero are at risk of adverse developmental, behavioral health, and educational outcomes (31–33). Because of the widely reported increases in the use of opioids among young people (34), we may anticipate an array of future adverse consequences.

Self-harm was rarely a principal diagnosis for admission in this study, and this finding provides new insights into the extent to which suicidality and self-harm are a part of pediatric hospitalization. About 24% of admissions with a primary or secondary behavioral diagnosis were complicated by suicidality or self-harm. Many such admissions are for treatment of the physical harm caused by an overdose of medication, and other admissions were for a primary psychiatric diagnosis, such as depression or posttraumatic stress. If we seek to use hospitalization as a lens for understanding the impact of suicidality among children and adolescents, it is imperative that future studies consider both primary and secondary diagnoses.

The study had some limitations. The KID database provides nationally representative estimates and was developed for research purposes. As is typical, it lacks individual identifiers, and we had to rely on the codes as recorded. Although ICD-9 codes are recorded by professional abstracters on the basis of clinical information entered in the medical charts and billing incentives typically favor including all diagnoses addressed during the hospitalization, we recognize that coding errors are inevitable. We have no reason to expect specific biases in the data coding that would disrupt our findings. The KID database lacks data from freestanding psychiatric hospitals and federal hospitals and thus underestimates the total frequency and cost of behavioral diagnoses in children. The database did not allow us to differentiate admissions to general hospitals from those to psychiatric units within the sampled hospitals.

Conclusions

Pediatric hospitalizations of children with behavioral disorders to general and children’s hospitals in the United States approached half a million in 2012. The study found that behavioral disorders were common among children admitted for other physical conditions and yielded additional costs of over $1.36 billion per year to these admissions. Behavioral disorders were found to frequently complicate adolescent and young adult pregnancy. Suicide or self-harm was rarely a primary diagnosis for admission; however, children with this diagnosis constituted 24% of all pediatric admissions with a behavioral disorder. The primary diagnosis for admitting children with suicide or self-harm was frequently to treat physical symptoms resulting from self-harm. If they are to yield the kind of information that can improve outcomes for this vulnerable population, studies of inpatient behavioral health care for children and adolescents will need to consider all admissions with primary behavioral diagnoses, as well as hospitalization in which behavioral disorders are secondary to other primary diagnoses.