Pediatric Hospital Utilization During Medical Stabilization for Patients With Eating Disorders Free

For this study, we used data from the Pediatric Health Information System (PHIS), an administrative billing database compiled by the Children’s Hospital Association that includes data from 49 tertiary care pediatric hospitals in the United States.¹⁴  For patients admitted with a primary diagnosis of ED, we examined comorbid diagnoses, LOS, costs, the use of enteral tube feeding, and 30-day readmissions overall and by ED diagnosis. In addition, we examined sociodemographic and clinical factors associated with LOS, costs, and the use of enteral tube feeding in this patient population. Because we used deidentified administrative data in this study, it was considered non-human subjects research by the institutional review board at our institution.

We included inpatient medical admissions for patients aged 4 to 20 years with a primary diagnosis of an ED who were discharged from PHIS hospitals from January 1, 2018 to December 31, 2022 and inpatient medical readmissions through January 31, 2023. Patients with EDs were identified by using principal International Classification of Diseases, Tenth Revision (ICD-10) codes beginning with F50, including AN, ARFID, BN, binge ED (BED), other specified feeding or ED (OSFED), and unspecified feeding and ED (UFED). The full listing of included diagnoses and groupings is available in Supplemental Table 5.

Our analysis was restricted to medical hospitalizations, excluding n = 3204 hospitalizations with psychiatric treatment unit charges based on flags available in PHIS.

We extracted age and sex and administratively collected race and ethnicity, insurance payor, and median household income. Median household income was based on patient ZIP code and divided into 3 categories by using 2016 tertiles of income in the United States.¹⁵ 

We examined comorbid diagnoses using secondary ICD-10 diagnosis codes based on the frequencies within our cohort, as well a priori diagnoses of interest (Supplemental Table 5). Medical diagnoses, such as malnutrition, bradycardia, hypotension, or amenorrhea (among females); psychiatric diagnoses, such as anxiety disorders, depressive disorders, or suicidal ideation or suicide attempt; and neurodevelopmental diagnoses for attention-deficit/hyperactivity disorder (ADHD) or autism spectrum disorder (ASD) were included.

We constructed a composite measure of the signs and symptoms related to severe malnutrition, including bradycardia, dehydration, hypotension, syncope, nutritional deficiency, hypoglycemia, acute organ failure, or amenorrhea such that if any was present, the individual was considered to have markers of severe malnutrition.¹ 

Measures included LOS, total hospital costs, ICU utilization, admission from the emergency department, and 30-day all-cause medical readmission. We used standardized hospital costs provided by PHIS derived from charges converted to costs (direct and indirect) according to hospital-specific ratios of costs to charges and adjusted for geographic variation using the Centers for Medicare and Medicaid Services wage and price index, which was standardized to eliminate between and within-hospital cost variation.¹⁶ 

Enteral tube feeding at any point during admission was identified by using ICD-10 procedure codes and Clinical Transaction Classification (CTC) codes for clinical services, supply, or other transacted services (Supplemental Table 6).

Our study period overlapped with the onset of the COVID-19 pandemic, which has been associated with increased hospitalization volumes for patients with EDs, as well as worsening severity.^17–21  We therefore constructed an indicator based on discharge date for hospitalizations before versus after the onset of the pandemic. Discharges before April 2020 were classified as pre-COVID-19 onset, whereas discharges in or after April 2020 were classified as post-COVID-19 onset.

Discharges with missing or unknown information for sociodemographic factors were retained in the sample. There were no missing data for other clinical variables or hospital utilization.

We report the mean (SD) or median (interquartile range [IQR]) for continuous variables and frequency (percent) for categorical variables. We examined bivariate associations between ED diagnosis, sociodemographic factors, comorbid diagnoses, and hospital utilization accounting for hospital clustering using Cochran–Mantel Haenszel tests for categorical variables and generalized estimating equation regression for continuous variables. We also examined bivariate associations between the discharge time periods before and after the onset of the COVID-19 pandemic and selected sociodemographic and clinical characteristics and hospital utilization. We used adjusted regression models to examine factors associated with LOS using Poisson regression, hospital costs using gamma regression, and the use of enteral tube feeding using logistic regression. The models were adjusted for ED diagnosis, sociodemographic factors, discharge time period (pre- versus post-COVID-19 onset), and comorbid diagnoses by using generalized estimating equations with a robust sandwich estimator to adjust standard errors accounting for hospital clustering with an exchangeable correlation matrix. We report adjusted risk ratios (RRs) and 95% confidence intervals (CIs) for LOS and costs, as well as adjusted odds ratios (95% CI) for enteral tube feeding. All analyses were performed in SAS (version 9.4; Cary, NC) at an α-level of 0.05.

We included N = 10 279 inpatient discharges for patients with EDs across 49 pediatric hospitals. Per hospital, the median number of discharges was 140 (IQR = 221), ranging from 10 to 1494. AN was the most frequent diagnosis (70.9%), followed by ARFID (15.6%), UFED (8.8%), OSFED (3.3%), and binge disorders (1.6%). The monthly volume of discharges overall and by ED diagnosis are presented in the Supplemental Fig 3. The volume of hospitalizations has increased over time, particularly after the onset of the COVID-19 pandemic. This increase has been primarily among patients with AN, although volumes for ARFID and UFED were also higher post-pandemic onset.

Sociodemographic characteristics overall and by ED diagnosis are presented in Table 1. Overall, the average age was 15.1 years (SD = 2.5), ranging from 7 to 20 years among those with AN, 9 to 20 for binge disorders, and 4 to 20 for ARFID, OSFED and UFED. The majority were female (86.6%) and of white non-Hispanic race and ethnicity (70.9%), with private insurance (70.1%). There were significant differences by ED diagnosis for all sociodemographic factors. Patients with ARFID were younger on average, and a lower proportion were female. Nearly two-thirds (63.9%) of discharges occurred after the onset of the COVID-19 pandemic, with the largest increases among patients with AN, ARFID, and UFED (P = .011).

Comorbid diagnoses overall and by ED diagnosis are presented in Table 2. Nearly all (93.4%) admissions had a comorbid medical diagnosis, with malnutrition (82.0%), bradycardia (52.7%), hypotension (31.5%), and dehydration (21.2%) being the most common. There were statistically significant differences by ED diagnosis for all comorbid medical diagnoses. A diagnosis of amenorrhea was present for 17.9% of female patients, most frequently among patients with AN. More than 80.0% of patients had at least 1 physiologic marker of severe malnutrition, and of these, 58.9% had >1 marker. The presence of physiologic markers of severe malnutrition differed significantly by ED diagnosis and were most common among patients diagnosed with AN (P < .001). Overall, 63.8% of admissions had a diagnosis of bradycardia or hypotension, of which 31.7% had both.

More than three-quarters (76.7%) of patients had at least 1 comorbid psychiatric or neurodevelopmental diagnosis with significant differences by ED diagnosis (Table 2). Anxiety disorders were common at 62.9% overall and more common among patients with ARFID (P < .001). Depressive disorders were present for 42.9% of patients overall, less common for patients with ARFID, and more common for those with binge disorders (P < .001). A total of 11.9% of patients had a diagnosis of self-harm, whereas 12.6% had a diagnosis of suicidal ideation or suicide attempt, and these diagnoses were more common among patients with OSFED and binge disorders (P < .001). Neurodevelopmental diagnoses for ADHD or ASD were less common at 8.6% and 3.0% overall. An ADHD diagnosis was more likely among patients with a diagnosis other than AN, whereas ASD was more likely among patients with ARFID and OSFED.

Hospital utilization overall and by ED diagnosis is presented in Table 3. The median LOS was 8.0 days (IQR = 7.0), with longer stays among patients with AN and shorter stays among patients with binge disorders (P = .005). Nearly half of the stays were 7 to 14 days in length, with significant differences in LOS category by diagnosis (P < .001). The majority were admitted from the emergency department (64.9%), whereas ICU admission was rare (1.8%). Enteral tube feeding was used in 18.8% of patients overall and was more common among patients with ARFID and less common for patients with binge disorders (P < .001). Median total hospital costs were $18 099 (IQR = $15 592) with the highest costs among patients with AN and the lowest median costs among those with BN or BED (P = .019). Overall, 8.8% of patients were readmitted within 30 days, with higher readmissions for patients with OSFED and binge disorders (P = .029). There were no in-hospital mortalities. Across all diagnoses, there were a cumulative total of 109 858 bed-days with costs of $289.9 million during the study period.

Select demographic characteristics, comorbid diagnoses, and hospital utilization before and after the onset of the COVID-19 pandemic are presented in Supplemental Table 4. The mean ages were similar between periods (P = .06), although a higher proportion of patients were female after the pandemic onset (87.2% post-pandemic versus 85.4% pre-pandemic; P = .01). There was no difference in the prevalence of any medical comorbid diagnosis (P = .57) or the presence of a diagnosis for any physiologic marker for severe malnutrition (P = .55); however, a slightly higher proportion in the post-pandemic onset period had a diagnosis for malnutrition (82.8% vs 80.7%; P = .006). Post-COVID-19 onset, there was a higher proportion with at least 1 psychiatric or neurodevelopmental diagnosis (78.0% post-pandemic versus 74.4% pre-pandemic; P < .001). There were also differences in hospital utilization after the onset of the COVID-19 pandemic, with longer stays (P = .007), more admissions from the emergency department (P < .001), higher ICU utilization (P = .003), and higher average costs per discharge (P < .001). The utilization of enteral tube feeding (P = .22) and 30-day readmissions (P = .53) were similar before and after the onset of the COVID-19 pandemic.

Results from adjusted regression models examining factors associated with LOS and costs are presented in Fig 1. Adjusting for sociodemographic and clinical factors, there were significant differences in LOS by ED diagnosis, age, race and ethnicity, insurance, comorbid diagnoses, and the use of enteral tube feeding (Fig 1A). Compared with patients with AN, patients with ARFID (RR = 0.88; 95% CI: 0.80–0.98; P = .017), binge disorders (RR = 0.71; 95% CI: 0.61–0.83; P < .001), and UFED (RR = 0.84; 95% CI: 0.76–0.91; P < .001) had shorter stays. Patients in older adolescence or young adulthood (18–20 years) had shorter stays relative to patients in mid-adolescence aged 15 to 17 years, whereas those with public insurance had longer stays. Patients identifying as Asian or Pacific Islander had longer stays compared with those identifying as white non-Hispanic. Comorbid medical diagnoses for malnutrition and physiologic markers of severe malnutrition, as well as the use of enteral feeding during admission, were associated with longer stays. Psychiatric diagnoses, including anxiety disorders, self-harm or suicidal ideation or attempt, or other psychiatric disorders, were also associated with longer stays, whereas neurodevelopmental diagnoses for ADHD or ASD were associated with shorter stays. Results from an adjusted model examining total hospital costs were similar, with significant differences by ED diagnosis, age, insurance, comorbid medical diagnoses, and the use of enteral tube feeding (Fig 1B). In adjusted models, there was no significant difference in LOS or costs for discharges after the onset of the COVID-19 pandemic compared with those before the pandemic onset.

Results from adjusted logistic regression examining factors associated with the use of enteral tube feeding during admission are presented in Fig 2. Patients with ARFID were more likely to receive enteral tube feeding compared with those with AN (odds ratio [OR] = 1.81; 95% CI: 1.41–2.32; P < .001) while those with binge disorders were less likely to receive enteral tube feeding (OR = 0.56; 95% CI: 0.37–0.86; P = .008). Patients aged 4 to 8 or 9 to 11 years were also more likely to receive enteral tube feeding compared with patients in mid-adolescence age (15–17 years). There were no differences by medical comorbid diagnoses for malnutrition or physiologic signs of malnutrition. Psychiatric diagnoses, including anxiety disorder, depressive disorder, self-harm, suicidal ideation or suicide attempt, or other psychiatric disorders, were associated with increased odds of enteral tube feeding, as were neurodevelopmental diagnoses for ADHD or ASD. Discharges after the onset of the COVID-19 pandemic were significantly less likely to receive enteral tube feeding compared with those before the onset of the pandemic (OR = 0.76; 95% CI: 0.65–0.89; P < .001).

In addition, there was significant between-hospital variation in the percentage of ED hospitalizations with any enteral tube feeding, ranging from 0% to 94%, with a median of 15.6%. Nine hospitals used enteral tube feeding for <5% of their ED discharges, whereas 11 hospitals used it for 25% to 50% of ED discharges, and 3 hospitals used it for >75% of ED discharges.

In this study, using data from pediatric tertiary care hospitals in the United States, we examined the sociodemographic and clinical characteristics of individuals hospitalized with EDs and their association with hospital utilization. Consistent with previous studies,^10,21  our findings indicate that medical hospitalizations for patients with EDs are long and costly, with the majority of stays being >1 week. In our cohort, there was a cumulative total of >100 000 bed-days, with hospital costs totaling $290 million. Similar to previous studies since the onset of the COVID-19 pandemic, the majority of these admissions occurred after the pandemic onset, highlighting the marked increases in hospitalizations related to EDs.^17,18  Nearly three-quarters of the hospitalizations were for patients diagnosed with AN, which was associated with longer and more expensive stays on average relative to other diagnoses and consistent with other studies.^9,21  Patients with ARFID were more likely to receive enteral tube feeding during admission, whereas patients with OSFED, BN, or BED were more likely to be readmitted within 30 days. Although the authors of numerous studies have examined health care utilization among patients with EDs, few have focused specifically on hospitalizations for medical stabilization at pediatric hospitals or compared across ED diagnoses.

In our cohort, most admissions had a diagnosis of malnutrition, and 80% had at least 1 marker of physiologic changes associated with malnutrition, including bradycardia, dehydration, hypotension, or acute organ failure, likely because these conditions are drivers for acute inpatient admission.^1,4  Notably, nearly two-thirds of admissions had a diagnosis of bradycardia or hypotension, which are 2 of the key admission criteria outlined in the Society for Adolescent Health and Medicine’s position paper and thus are often criteria supporting and justifying the need for inpatient hospitalization.⁴  Unsurprisingly, the presence of diagnoses suggestive of physiologic instability was associated with significantly longer stays and higher costs, which may be due to longer time to stabilization or the need to remain in the hospital while awaiting placement in specialized ED care (eg, residential or intensive outpatient programs).²¹  Earlier detection of EDs may help prevent secondary outcomes of malnutrition and the need for medical hospitalization;¹  therefore, the authors of future studies should explore novel screening tools, especially for primary care settings.

In addition to increases in the volume of hospitalizations for EDs after the onset of the COVID-19 pandemic, worse severity may be indicative of delayed care early in the disease course, potentially exacerbated by the pandemic.^17–21  We found that hospitalizations after the onset of the pandemic were more likely to have a billed diagnosis for malnutrition, although the prevalence of physiologic markers of severe malnutrition was similar before and after the pandemic onset. Other proxy measures for severity were higher in the post-pandemic onset period, including admission from the emergency department and ICU utilization, which is consistent with previous studies indicating worse severity post-COVID-19 onset.^19,21  Patients admitted after the onset of the pandemic also had longer stays and higher costs in our cohort, which is consistent with previous studies,²¹  although this difference was attenuated after adjusting for sociodemographic and other clinical factors in our analysis.

As has been described previously,^1,6,22  we found a large burden of comorbid mental health conditions in our cohort, with more than three-quarters having at least 1 psychiatric or neurodevelopmental comorbid diagnosis and anxiety and depressive disorders being the most prevalent. Similar to previous studies,^7,13  we found that patients with ARFID had a high burden of anxiety disorders. Individuals with OSFED and binge disorders had a higher prevalence of depressive disorders and self-injury, which may be a contributing factor to higher readmission rates for these diagnoses. Similar to our findings indicating a higher prevalence of psychiatric diagnoses after the onset of the COVID-19 pandemic, recent studies have indicated a worsening mental health burden among patients with EDs, which was likely exacerbated by the pandemic and further complicated treatment approaches.¹⁹  Thus, ED prevention and intervention strategies must address both the ED and comorbid conditions.

To our knowledge, this is the first study to examine enteral tube feeding as an outcome using administrative data to compare across ED diagnoses. Enteral tube feeding was relatively common, at nearly 20% overall, and most likely to occur among patients with ARFID, with comparable prevalence to that in a recent single-center study.²¹  Patients with ARFID were nearly twice as likely to receive enteral tube feeding compared with those with AN, despite a similar prevalence of malnutrition for both diagnoses. Patients with BN or BED were significantly less likely to receive enteral tube feeding compared with patients with AN, which is likely explained by the lower prevalence of malnutrition. Younger patients and those with comorbid psychiatric and neurodevelopmental disorders were also more likely to receive enteral tube feeding, which may be partially due to the overlap with ARFID. Similar to a previous single-center study,²¹  our bivariate analysis indicated similar utilization of enteral tube feeding before and after the onset of the COVID-19 pandemic. However, adjusting for sociodemographic factors and comorbid diagnoses, patients admitted after the pandemic onset were less likely to receive enteral tube feeding than those admitted before the pandemic. The exact mechanism for this finding is unclear, and additional studies are needed to understand changes over time in the use of enteral tube feeding and whether it is related to patient-level factors, census and capacity, or shifting hospital practice.

There was also large variation by hospital in the use of enteral tube feeding, with several hospitals never using enteral tube feeding, although more than one-quarter used enteral tube feeding for at least 25% of their discharges for patients with EDs. This variation may be explained in part by differences in patient acuity or severity impacted by local or regional referral patterns or admitting criteria, although we were unable to explore this further given the lack of detailed clinical data to assess severity at presentation. Importantly, however, care for EDs is not standardized, which likely contributed to the variation we observed and is indicative of differences in approaches to treatment.^1,23  Our findings underscore the need for standardized, evidence-based approaches to medical stabilization within this patient population and appropriate guidelines for the use of enteral tube feeding. In addition, future studies are needed to further explore hospital variation in the use of enteral tube feeding and measures of hospital utilization, including LOS, costs, and readmissions using PHIS or other datasets.

This study has a number of limitations. First, hospitals participating in PHIS may not be generalizable to all hospitals caring for patients with EDs, including other pediatric hospitals, adult tertiary care hospitals, or community hospitals. Second, we used administrative billing data, which may be incomplete with respect to sociodemographic factors, comorbidities, or other clinical characteristics. We identified our cohort of patients with EDs based on the principal ICD-10 diagnosis code for an ED to intentionally restrict to patients hospitalized primarily for an ED, rather than those with a secondary ED diagnosis who may have been hospitalized for another unrelated condition. However, this may underestimate the total population of patients hospitalized with ED at the included hospitals. Third, we were unable to identify psychiatric boarders who were hospitalized on medical floors and were awaiting transfer to psychiatric beds or facilities because this information is not captured by billing codes. There is, therefore, a potential for misclassification among the included discharges that are assumed to be for medical stabilization, although this is unlikely because we excluded discharges or transfers with any billed charge for the psychiatric treatment unit. In addition, readmissions were underestimated because only patients readmitted at the same PHIS hospital were captured. Finally, as noted above, we were unable to ascertain the severity of ED symptomatology at presentation. However, we captured secondary diagnoses that may be indicative of worse severity, including acute organ failure or bradycardia, although these relied on billed diagnoses and may not perfectly capture clinical factors indicative of severity, such as abnormalities in vital signs or percentage of body weight lost. Although we considered secondary diagnoses for common comorbid conditions or medical sequelae,^1,4  the list of included codes was not exhaustive, which may explain why 7% of our cohort did not have any of the specified medical diagnoses. These patients may have been hospitalized for other acute conditions warranting admission (eg, acute food refusal) that are not captured by billing data.⁴  Of note, we were unable to examine the prevalence of refeeding syndrome given the lack of a specific ICD-10 diagnosis code for this condition, which can cause severe, life-threatening metabolic changes among malnourished patients receiving nutritional rehabilitation.^1,24  Additional studies are needed to further contextualize our findings indicating differences in hospital utilization by ED diagnosis, particularly using more detailed measures of severity, including clinical measures, such as percent of body weight lost or vital signs, and patient-reported outcomes using validated scales, such as the ED Examination Questionnaire.

In a diverse cohort of pediatric hospitals in the United States, our findings indicate that hospitalization for patients with EDs are long and costly, annually accounting for >20 000 bed-days and $50 million in costs cumulatively across all the included hospitals. We found substantial differences in hospital utilization, costs, and readmissions by ED diagnosis among patients hospitalized for medical stabilization. AN was the most common diagnosis and was associated with long, costly stays and a high burden of comorbid mental health conditions. To our knowledge, this is the first study to examine utilization during medical hospitalization at pediatric tertiary care hospitals across the spectrum of ED diagnoses. Future studies are needed to further explore variations in utilization and treatment approaches to improve outcomes, facilitate earlier identification of EDs, and prevent readmissions.