Factors Associated With Prolonged Mental Health Admissions at US Children’s Hospitals

We performed a multicenter, retrospective cohort study of children aged 5 to 20 years hospitalized with a primary MH discharge diagnosis at 1 of 46 US children’s hospitals between January 1, 2021, and December 31, 2022. Data were obtained from the Pediatric Health Information System (PHIS), which is an administrative and billing database of children’s hospitals affiliated with the Children’s Hospital Association (Lenexa, Kansas). We included children up to 20 years of age to capture patients with medical complexity who may receive care at a children’s hospital past age 18, as well as patients who may have unique barriers to discharge (eg, guardianship concerns). To focus on MH hospitalizations at acute care facilities, we excluded patients billed for a designated psychiatric or rehabilitation bed. We also excluded patients transferred into the hospital to accurately capture the total duration of hospitalization. Figure 1 displays a consort diagram of our included study population.

International Classification of Disease, 10th Edition, Clinical Modification, codes for MH diagnoses were classified using the Child and Adolescent MH Disorders (CAMHD) classification system, which aligns diagnosis codes with the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, psychiatric diagnosis groups.^11,12  To organize our large sample, we combined certain conditions and analyzed the following groups: Anxiety and obsessive compulsive disorders; autism spectrum disorder; bipolar disorders; communication disorders; depressive disorders; developmental, motor, and learning disorders; disruptive, impulse control, and conduct disorders; feeding and eating disorders; intellectual disability; schizophrenia and psychotic disorders; substance-abuse related illnesses; suicidality/self-injury; and trauma-related disorders. Supplemental Table 3 displays the collapsing of CAMHD categories into our diagnostic categories. This study exams all MH conditions classified using the CAMHD system, including feeding and eating disorders. Hospitalizations for feeding and eating disorders may differ in unique ways from other MH hospitalizations, including patient demographics, indication for admission, and discharge criteria, therefore we conducted a sensitivity analysis excluding hospitalizations for feeding and eating disorder. Patients with multiple diagnosis codes within 1 MH diagnostic category were characterized as having 1 MH diagnosis (ie, the total count of MH diagnoses was based on the presence of MH diagnosis codes in separate study diagnostic categories and not total MH diagnosis codes).

The primary outcome was prolonged LOS, defined as the 95th percentile of LOS for all children hospitalized with a primary MH diagnosis, which equated to LOS ≥14 days. Our secondary outcome was hospital cost. Costs were estimated from charges using annual hospital specific cost-to-charge ratios from the Centers for Medicare and Medicaid cost reports.

We included the following patient demographic characteristics: Age, sex, race and ethnicity, and payer. Race and ethnicity data were treated as a social construct, rather than a biologic category. Race and ethnicity data were collected by individual PHIS hospitals per their procedures and then submitted to PHIS. Race and ethnicity categories included non-Hispanic Black, non-Hispanic white, Hispanic, Asian, and other. Included in the other category include Pacific Islander, Native American, and other unspecified. Payer categories included government, private, and other. Clinical characteristics included the CAMHD category of primary MH condition, the number of MH conditions, and the number of complex chronic conditions. Complex chronic conditions were identified using Feudtner et al’s definition, and included medical conditions expected to last at least 1 year and to involve either several different organ systems or 1 organ system severely.¹³  Because of limitations with our data set, we did not include any descriptors for “medical clearance” versus ongoing medical needs.

Hospitalization characteristics included any ICU use and discharge disposition. Discharge disposition was categorized as home, transfer to other medical hospital, transfer to inpatient psychiatric facility or residential facility, and other. Within other were the following discharge dispositions: Deceased, left against medical advice, unknown, and interhospital transfer.

Frequencies with percentages or medians with interquartile ranges described characteristics of the study population and bivariate comparisons were made using χ² or Wilcoxon rank tests. Multivariable models were constructed with generalized estimating equations, clustering on hospital, to estimate the adjusted odds ratios (aORs) and 95% confidence intervals (CIs) for prolonged LOS. All demographic and clinical characteristic predictor variables were statistically significant in bivariate analyses and, therefore, were included in the model. Non-Hispanic white was selected as the reference for race and ethnicity variables, given that this was the largest group and this approach resulted in easily interpretable (>1.0) ORs. Anxiety disorders was selected as a referent for MH diagnoses, to allow for positive ORs among other groups and avoid referent variables that may overlap with other variables (ie, depression and suicidality/self-harm). To describe interhospital variation in prolonged LOS for children hospitalized with a primary MH diagnosis, we calculated the hospital-level adjusted percentage of all MH hospitalizations that were prolonged. All statistical analyses were performed using SAS v.9.4 (SAS Institute, Cary, North Carolina), and P < .05 was considered statistically significant. This study was deemed nonhuman subjects research by the Office of Research Integrity at Children’s Mercy Kansas City.

We hypothesized that patients with an eating disorder and their related goals for hospitalization may have been unique compared with patients with other MH diagnoses. Therefore, we conducted a sensitivity analysis excluding patients with a primary diagnosis of a feeding or eating disorder.

Among 42 654 MH hospitalizations at 46 Children’s Hospitals, 2169 (5.1%) had a prolonged LOS (Table 1). Females accounted for most MH hospitalizations (68.5%). Non-Hispanic white children accounted for 53.8% of hospitalizations, and almost half (48.4%) of those with MH hospitalizations had government insurance. Suicide/self-injury was the most common (37.4%) primary MH condition, followed by depressive disorders (16.6%) and eating disorders (10.9%). Most (60.6%) hospitalizations were associated with 3 or more MH diagnostic categories. Most hospitalizations (80%) had no associated complex chronic conditions, and most hospitalizations (64.5%) were discharged to home. The distribution of LOS followed a positively skewed distribution with an extreme right tail. Although prolonged hospitalizations only accounted for 5.1% of all MH hospitalizations, prolonged MH hospitalizations accounted for 33.3% of MH hospital days (62 628 days) and 26.2% of MH hospitalization costs ($132 428 435).

Prolonged MH hospitalizations had a slightly higher representation of females (70.6% vs 68.4%, P = .037) and a higher representation of children with private insurance (49.1% vs 44.5%, P < .001) when compared with MH hospitalizations with LOS <14 days. The distribution of primary MH diagnoses differed significantly between prolonged MH hospitalizations versus hospitalizations with LOS ≤14 days. Autism spectrum disorders, eating disorders, and schizophrenia spectrum and other psychotic disorders appeared overrepresented in the prolonged LOS group, whereas anxiety disorders; depressive disorders; disruptive, impulse control, and conduct disorders; suicide or self-injury; and trauma and stressor-related disorders were underrepresented. Prolonged MH hospitalizations had a higher proportion of children with 5 or more MH diagnostic categories (29.2% vs 15.7%, P < .001), and children with any complex chronic conditions (46.7% vs 18.5%, P < .001), and were more frequently discharged to home (74.4% vs 64%, P < .001) compared with MH hospitalizations <14 days. Median hospital costs were 7.3 times higher in prolonged MH hospitalizations compared with MH hospitalizations <14 days ($45 526 vs $6247, P < .001).

Compared with patients admitted for anxiety disorders, the primary MH diagnosis with the greatest odds of having a prolonged LOS was eating disorders (aOR 24.81; 95% CI 8.11–75.89); schizophrenia or psychotic disorders (aOR 10.63; 95% CI 4.2–26.89); autism spectrum disorder (aOR 7.07; 95% CI 2.7–18.48); and disruptive, impulse control, and conduct disorders (aOR 6.78; 95% CI 2.43–18.92) (Table 2). Prolonged LOS was also associated with children reported as non-Hispanic Black (aOR 1.69; 95% CI 1.2–2.19), other race and ethnicity (1.59 aOR; 95% CI 1.12–2.26), and Asian race (aOR 1.58; 95% CI 1.13–2.2), compared with non-Hispanic white children. Compared with those with 1 or 2 MH diagnoses, the adjusted odds of prolonged LOS increased with 3 to 4 MH diagnoses (aOR 1.6; 95% CI 1.3–1.97) and 5+ MH diagnoses (aOR 3.72; 95% CI 2.89–4.79). Compared with patients with no complex chronic conditions, the odds of prolonged LOS increased with 1 complex chronic condition (aOR 1.58; 95% CI 1.33–1.89), 2 complex chronic conditions (aOR 2.22; 95% CI 1.59–3.11), and 3+ complex chronic conditions (aOR 6.16; 95% CI 3.82–9.93).

After controlling for demographic and clinical factors, there was significant (P < .001) interhospital variation in the percentage of all MH hospitalizations that had a prolonged LOS (range 0%–22.4%) (Fig 2). Variation in prolonged LOS across hospitals was not associated with volume of MH hospitalizations at each hospital.

Excluding patients admitted for eating disorders moved our main outcome measure (ie, the 95th percentile of LOS) from ≥14 days to ≥10 days. However, our results remained otherwise relatively unchanged. Although some point estimates and CI changed (Supplemental Table 4) no variables (medical complexity, MH complexity, MH diagnoses categories) lost or gained statistical significance (Supplemental Table 5). Additionally, the rank order of strength of association of MH diagnoses remained unchanged (ie, aORs were greatest for autism spectrum disorder followed by bipolar and related disorders and so forth). The lone exception was race and ethnicity, where non-Hispanic Black children were the only group that remained with a higher odd of prolonged LOS (aOR 1.54; 95% CI 1.25–1.9; P < .001), and categorization of Asian race or other race was no longer associated with prolonged LOS.

In this retrospective analysis of children aged 5 to 20 years primarily hospitalized for an MH condition at children’s hospitals, we identified numerous categories of MH diagnoses associated with a prolonged LOS. These findings were robust and unchanged in our sensitivity analysis excluding children hospitalized with a primary diagnosis for a feeding and eating disorder. Children’s number of comorbid MH diagnoses and medical complexity were also associated with prolonged LOS. Collectively, these results can inform in-hospital systems of care and community resources need to minimize prolonged MH hospitalizations at acute care children’s hospitals.

We identified numerous categories of MH diagnoses strongly associated with prolonged LOS that require unique systems of care. Eating disorders; autism spectrum disorder; schizophrenia and psychotic disorders; and disruptive, impulse control, and conduct disorders were all strongly associated with prolonged MH hospitalizations. Eating disorder was the most prevalent and most highly associated condition associated with prolonged MH hospitalizations, which is consistent with recent studies.¹⁴  This result highlights the continued need to increase care capacity for children suffering from eating disorders.^15,16  The association between eating disorders and prolonged MH hospitalization may be influenced by a lack of clear guidelines for the management of malnutrition and hospital discharge criteria for children with eating disorders,¹⁷  and further studies exploring variation in care of eating disorder management and associations with LOS are needed.

Autism spectrum disorder; schizophrenia and psychotic disorders; and disruptive, impulse control, and conduct disorders were also strongly associated with prolonged MH hospitalizations. Children hospitalized primarily for these conditions engage in high levels of challenging behavior (eg, physical aggression, property destruction, and self-injury),^18,19  and such behavior can escalate, influencing discharge disposition and prolong the hospitalization.^8,20–22  Providers such as psychologists and psychiatrists are imperative in the implementation of behavioral and pharmacologic interventions to reduce a child’s overall acuity.^23–25  Behavior analysts are other key providers who identify the underlying cause of behavior and create interventions to (1) avoid and manage behavioral escalations and (2) promote appropriate behavior. Targeted services and behavior plans are associated with reduction in pharmacologic and physical restraints, and improved provider satisfaction, and may decrease hospital LOS.^23,26–28  Decreasing a child’s MH acuity can improve patient and hospital staff outcomes,^25–27  and may impact LOS by:

1. improving compliance with medical care² ;

2. improving behavior such that patients may be safe for discharge home; and/or

3. overcoming “reverse triaging” (ie, practice in which patients with the highest MH acuity are not given priority by admitting psychiatric facilities).²⁹ 

Notably, patients with prolonged LOS were most likely to be discharge to home, which may relate to this reverse triaging phenomenon and lack of inpatient resources versus medical limitations in discharge to home earlier. Despite optimization of inpatient MH supports safe discharge plans frequently require community-based MH resources, which unfortunately are in a critical shortage.⁸  Further studies examining the associations between MH hospitalization LOS with inpatient and community MH resources are needed.

Compared with non-Hispanic white children, we observed that non-Hispanic Black, Asian, and children of other race and ethnicity each had higher odds of prolonged hospitalization, congruent with other pediatric literature describing race- and ethnicity-associated differences in hospital LOS.³⁰  Our study was not designed to assess reasons for this difference, and the influence of factors prehospitalization (eg, differences in access to care, differences in exposure to childhood trauma which may result in more severe MH disease among different racial or ethnic constructs) and/or factors during hospitalization are unclear. Time to diagnosis and treatment in the outpatient treatment of MH conditions are associated with patient race,^31–33  and perhaps race-associated differences in the care of children hospitalized with MH conditions are influencing hospital LOS. Additionally, unconscious bias exists in the evaluation and treatment plans of patients evaluated by psychiatrists, and unconscious bias may have contributed to our findings.³⁴  There may be real or perceived race-associated barriers in posthospitalization care, which could impact hospital LOS. Our study highlights the need for special attention to sociodemographic risk factors when caring for children hospitalized with MH hospitalizations, and the need to consider equity and family-centered care when discharge planning.³⁵ 

We found significant stepwise associations between prolonged MH hospitalizations and number of complex chronic conditions, suggesting that, not only do MH comorbidities prolong medical admissions,³⁶  medical complexity also prolongs MH hospitalizations. Medical complexity likely complicates behavioral and/or psychiatric care and may represent a barrier to safe discharge. Our results highlight the need for continued investigations into the role of pediatric med–psych units (ie, units within medical hospitals dedicated to care for patients with both physical and MH needs). Studies of adult med–psych units suggest that they may be associated with shorter LOS and decreased 30-day readmission.³⁷  Although pediatric med–psych units are becoming more prevalent to match the need for complex pediatric medical and MH care, further data of such units, including cost variance analysis, as well as studies testing the associations of pediatric med–psych units and patient and staff outcomes, are needed.

We found substantial variation in the prevalence of prolonged MH hospitalizations across US children’s hospitals, which persisted after adjusting for factors associated with prolonged MH hospitalizations (eg, volume and types of MH conditions seen at each hospital). Hospital-level variation in the prevalence of MH hospitalizations with a prolonged LOS may indicate differences in discharge criteria or regional options for ongoing postdischarge MH care, be it outpatient or inpatient. Variation may also represent opportunities to decrease LOS through strategic planning of community resources, enhanced MH support in the inpatient setting, and evidence-based care for coexisting medical needs. Understanding systems and resources available at those hospitals with the lowest prevalence of prolonged MH hospitalizations may reveal best practices which can be disseminated across pediatric medical centers.

This study’s findings should be considered in the context of several limitations. Most PHIS-reporting hospitals are academic tertiary or quaternary care hospitals, and our study findings may not generalize to community hospitals where most children are hospitalized. Second, the identification of MH conditions relies on proper documentation and coding practices, which can vary across hospitals. Third, we acknowledge there may be factors that are associated with prolonged MH hospitalizations that were unable to be measured by our study design. For example, severity of challenging behaviors, guardianship barriers, and availability and types of community-based services may be associated with prolonged LOS but were unable to be elucidated utilizing administrative data. Finally, the time frame of our study was after the start of the coronavirus disease 2019 (COVID-19) pandemic. Closure of residential or psychiatric facilities related to COVID-19, and policies related to testing for and/or isolation of children because of COVID-19, may have influenced hospital LOS. It is uncertain if these policies affected children with certain clinical or sociodemographic characteristics more than others, and/or to what degree a comorbid COVID-19 diagnosis affected LOS.

We identified characteristics of patients at US children’s hospitals where expanded investments in community- and hospital-level MH services, such as behavioral therapy and specialized behavioral units, have the potential to improve health outcomes, equity, and reduce LOS and cost of MH hospitalizations.