Disparities in Management of Acute Gastroenteritis in Hospitalized Children Free

We conducted a single-center retrospective chart review study among children admitted to the pediatric hospital medicine service with AGE. This study took place at a large, freestanding, tertiary children’s hospital in Southern California. The catchment area spans >4500 square miles and encompasses >900 000 children. In 2020, 42.2% of the population living in this area identified as Hispanic.¹⁶  A quarter of the population in the catchment area speaks Spanish, with 28.7% of Spanish-speakers reporting that they speak English less than “very well.”¹⁶  Because of underrepresentation of other marginalized racial groups within the catchment area, our study focused on differences between Hispanic and non-Hispanic children.

We included children aged 2 months to 12 years admitted to the pediatric hospital medicine service between January 2016 and December 2020, with a primary or secondary diagnosis recorded at any time during the hospital encounter of (1) AGE or (2) dehydration in the presence of either feeding intolerance, vomiting, and/or diarrhea using International Classification of Diseases, 10th Revision, codes (AGE [A00-09, B82], dehydration [E86], feeding intolerance [R63.3], vomiting [R11, K92.0], diarrhea [R19.7, K92.1]).

We excluded children without diarrhea at the time of admission on the basis of manual review of the admission note because of a broader differential diagnosis at presentation, often prompting additional diagnostic studies. To account for severity of illness, we excluded any patient admitted to the ICU at any time during admission. We also excluded any patient transferred from an outside facility because of inability to consistently obtain and evaluate laboratory data and treatments rendered before admission. For any child with a readmission within 30 days, only the initial encounter was used for analysis to avoid potential for bias given the likelihood of additional medical evaluation that could occur with a repeat admission. We also excluded children with additional diagnoses that could have warranted either further diagnostic studies or additional medical interventions. Additional diagnoses included children with complex chronic conditions using International Classification of Diseases, 10th Revision, codes defined by Feudtner et al, or those with an additional diagnosis of 1 or more of the following: Coronavirus disease 2019 infection; multisystem inflammatory syndrome in children; complications of AGE including protein-losing enteropathy, bacteremia, or hemolytic uremic syndrome; or other causes of diarrhea and dehydration including constipation with encopresis, appendicitis, inflammatory bowel disease, Henoch-Schönlein purpura, pancreatitis, milk soy protein intolerance, food protein-induced enterocolitis syndrome, malabsorption, severe malnutrition, adrenal insufficiency, and/or intussusception.¹⁷  Two patients in this retrospective study were excluded because of missing ethnicity data. Additionally, those with self-pay or out-of-state insurance were not included in the final analysis because of a low sample number and potential for differences in resource utilization within these groups. Those with hospital duration >4 SDs from the median (>240 hours) were also excluded (Fig 1).

Primary outcome measures were the frequency of clinical pathway use at time of admission, diagnostic studies performed, and medical interventions ordered between Hispanic and non-Hispanic patients. Studies performed and medical interventions ordered initially from the ED or inpatient setting were included because of the influence that ED practices may have on decision-making after the patient is admitted. Clinical pathway use was defined as use of the existing AGE admission order set within the electronic health record (EHR) (Epic Systems Corporation, Verona, Wisconsin). This order set reflects our clinical pathway, which was created in 2007, and is updated biennially and as needed on the basis of pediatric guidelines and evidence-based practices.^3,5  This order set provides options for laboratory testing and management guidance for patients admitted with AGE. From its inception, the pathway has included optional orders for comprehensive and basic metabolic panels, stool studies, IVFs, antipyretics, antiemetics, and lactobacillus. The order set has never included imaging orders. In 2017, blood and stool testing were no longer routinely recommended, and the comprehensive metabolic panel and stool studies were removed.^4,5  Lactobacillus was also removed from the order set in 2019. The current order set includes options for a basic metabolic panel and a separate single point-of-care glucose order, both of which must be individually selected by the provider on the basis of clinical discretion and with guidance from the pathway. It also includes optional orders for IVFs, antipyretics, and antiemetics (Supplemental Information). Providers must actively search for any additional orders not included in the order set. The pathway is used in both the ED and inpatient setting, and includes guidance for electrolyte testing (ie, severe dehydration, intractable emesis), antiemetic use (ie, emesis), and IVF use (ie, severe dehydration, failed oral rehydration).

Completion of common diagnostic studies at any time during the encounter, including stool studies (eg, stool cultures, fecal occult blood, etc), laboratory testing (eg, electrolytes, glucose, urinalysis, etc), and abdominal or pelvic imaging, were abstracted from the EHR. Orders for medical interventions included in the analysis were: IVFs, antiemetics, nonopioid analgesics, opioid analgesics, probiotics, and antispasmodics. Pro re nata orders included whether they were given, and if not given, must have been ordered for at least 6 hours to capture intention to treat.

Secondary outcomes included hospital charges, LOS, and 7- and 30-day–related ED and urgent care returns to our hospital system, defined as any return for nausea, vomiting, diarrhea, feeding intolerance, or abdominal pain.

A data set with eligible cases was created via data query from the EHR. Outcome variables and covariates were obtained by electronic data query and manual chart review. A codebook was developed for the manually extracted variables to standardize where in the chart the data should be obtained and what terms and descriptions were acceptable. Three members of the research team (M.P., M.E., E.M.A.) cross-reviewed a random sample of 5% of the charts to determine reliability of the manually extracted data.¹⁸  Discrepancies were reviewed at 3 intervals throughout the chart review process and consensus was attained.

Clinical characteristics at time of admission were obtained via chart review, including presence, duration, and characteristic of diarrhea (ie, bloody versus nonbloody) and vomiting (ie, nonbloody or nonbilious versus bloody or bilious), presence of nausea, abdominal pain, sick contacts, or other exposures including travel out of the United States, consumption of unpasteurized or uncooked foods, freshwater swimming, camping, and exposure to animals with gastrointestinal transmittable pathogens including reptiles, poultry, or livestock. Fever was defined as a documented temperature of 38.0°C or greater at time of admission, or patient-reported fever at home (≥38.0°C or tactile) within the past 7 days. If no mention of the above clinical characteristics were included in the ED note or admission note, then the characteristic was documented as not present.

Demographic variables including age, sex, race, ethnicity, insurance, language, and median income according to zip code were obtained from the EHR. Race, ethnicity, and language were collected during the registration process by verbal acquisition from staff for each encounter. Language reflected preferred language identified by the caregiver and did not include whether the patient’s language differed from the parent’s preferred language, whether an interpreter was used during the visit, or what language was used during the encounter. In the state of California, Medi-Cal is the Medicaid-funded public insurance program which provides insurance to low-income individuals at little or no cost. Median household income was obtained from 2020 Census data on the basis of patient zip code, which was reported at the time of the encounter.¹⁹ 

Categorical variables were summarized using frequencies and compared between Hispanic and non-Hispanic patients using χ² or Fisher’s exact test. Continuous variables were analyzed using medians with interquartile ranges and Wilcoxon rank-sum tests. Multivariable logistic regression was then performed. Nonparametric continuous variables (eg, LOS, charges) were transformed into quartiles for multivariate analysis to perform logistic regression. Sex and age were included in the models a priori. Variables with bivariate associations of P < .1 with ethnicity were also included in all models; these included insurance status, language, median income, presence of fever, international travel, and sick contact. For all models, any variable with an association (P < .1) with the outcomes of interest were also included. This included presence of abdominal pain, vomiting, bloody diarrhea, and duration of both vomiting and diarrhea. Correlation between several demographic and clinical variables was examined using Pearson’s correlation. To create a parsimonious model, variables that were highly correlated (r > 0.7) with primary independent variables were removed (eg, other exposures). Model fit statistics including Akaike’s information criterion (an estimator of prediction error and quality of models) and −2 log likelihood (goodness of fit) were reviewed to determine the best model fit. This study was reviewed and received approval from the institutional review board.

During the 5-year study period, 512 admissions met inclusion criteria. Patient median age was 28 months (interquartile range 12–63 months), 54.9% were male, 51.6% were Hispanic, and 59.2% were on Medi-Cal. Hispanic patients were less likely to report English as their preferred language (73.9% vs 94.0%, P < .001). There was no difference between Hispanic and non-Hispanic patients in most clinical characteristics, including the presence of bloody diarrhea, vomiting, nausea, or abdominal pain (Table 1).

Bivariate analysis did not reveal any difference in pathway use, defined by the use of an admission order set for AGE, between Hispanic and non-Hispanic patients (Table 2). Despite there being no difference in the frequency of ordering common laboratory tests, Hispanic patients were slightly more likely to have ultrasound (U/S) scans ordered; however, this did not reach a level of significance (37.5% vs 29.4%, P = .054). Hispanic patients were also more likely to receive computed tomography scans than non-Hispanic patients (8.3% vs 3.6%, P = .03).

Medication orders were similar between the 2 groups, including IVF and pain management orders. However, Hispanic patients were less likely to have antiemetics ordered than non-Hispanic patients (73.5% vs 81.9%, P = .02). Despite this discrepancy, administration of antiemetics among all patients did not differ between the 2 groups (57.6% [152 of 264] vs 62.1% [154 of 248], P = .30). In a post-hoc analysis only examining patients with the AGE admission order set used, Hispanic patients still had fewer antiemetics ordered than non-Hispanic patients (79.9% [139 of 174] vs 89.0% [145 of 163], P = .02).

Multivariable analysis was performed, and all models were adjusted for sex, age, insurance status, language, median income, international travel, sick contacts, presence of fever, abdominal pain, vomiting, bloody diarrhea, and duration of both vomiting and diarrhea at time of admission (Table 3). After adjusting for covariates, there was a significant difference in U/S scans performed, with Hispanic patients receiving a larger number of U/S scans than non-Hispanic patients (odds ratio [OR] 1.65, 95% confidence interval [CI] 1.04–2.64); however, there was no longer a difference in computed tomography scans obtained (OR 1.46, 95% CI 0.51–4.23). There remained a difference in antiemetic ordering, with fewer orders placed for Hispanic patients than non-Hispanic patients (OR 0.53, 95% CI 0.29–0.95).

Nonparametric continuous variables were transformed into quartiles for multivariable analysis. After adjusting for covariates, there was no difference in hospital charges quartiles (OR 1.29, 95% CI 0.88–1.89) or LOS quartiles (OR 1.28, 95% CI 0.87–1.87) between Hispanic and non-Hispanic patients.

This single-center retrospective chart review found no difference in laboratory testing and most medical interventions between Hispanic and non-Hispanic patients hospitalized for AGE. However, we found that Hispanic patients received more U/S scans and fewer antiemetic orders, despite no differences in reported nausea, vomiting, or abdominal pain between the 2 groups. Although there are studies demonstrating racial and ethnic disparities among pediatric AGE in the ED setting, those completed in hospitalized children are limited.^14,20  Our study is filling this gap in knowledge and highlighting that inequities also exist in children hospitalized with AGE.

Of the literature that has examined disparities among hospitalized pediatric patients, most were in soft tissue infection, urinary tract infection, bronchiolitis, asthma, diabetes, or pneumonia, with results showing disparities in LOS, hospital cost, and pain control.^21–27  Differences in LOS were hypothesized to be because of delayed care, limited access to outpatient clinics, and comorbidities.^22,28  In contrast to previous studies, we did not find a difference in LOS between Hispanic and non-Hispanic patients after accounting for several demographic and clinical characteristics. Though ethnicity alone did not influence LOS, over half of all Hispanic children reside in low-income households, defined as ≤200% of the federal poverty level.²⁹  Hispanic children living in poverty are more likely to be underinsured and have limited access to health care, which can contribute to delays in seeking medical care.^21,29  This may suggest that previous differences found in LOS are multifactorial and reflect underlying social and structural disparities that put Hispanic children at a disadvantage before even presenting to the hospital setting.

In contrast to previous inpatient studies, we found no difference in pain management; however, we did find differences in symptom management with lower antiemetic ordering for Hispanic patients despite similar frequency of administration. Though it is notable that Hispanic patients eventually received similar amounts of antiemetics than non-Hispanic patients, a discrepancy in intent to treat was still present, which may reflect implicit bias among ordering providers.²²  Clinical pathways can help standardize care, with the goal of reducing bias in health care, but it is important to note that they are not intended to remove provider thought process or judgment. This study highlights that standardizing care through clinical pathways may be beneficial, but alone is not enough to eliminate bias in health care. Previous literature has suggested additional steps for pediatric hospitalists to take to achieve more equitable care for our children. These include reflection of implicit biases, screening for social determinants of health, quality improvement initiatives aimed at reducing disparities in testing and treatment, qualitative and quantitative research aimed to understand the cause of disparities, and advocacy for policies to reduce structural inequities before hospitalization.²²  Our study highlights the need for multisystem interventions to close the gap in care that currently exists and achieve more equitable health care delivery and health outcomes.

Congdon et al evaluated over 30 000 low-acuity ED encounters for children with AGE and found decreased IVF use and fewer admissions for Hispanic patients than non-Hispanic white children, with no increase in 72-hour ED visits. Their study suggests that, rather than undertreating Hispanic patients or other nonwhite patient populations, we may instead be preferentially overtreating white patients.¹⁴  In contrast to this, we found increased use in ultrasound scans in our Hispanic patients compared with non-Hispanic patients. A potential difference in our results could be related to communication barriers between the patient and provider.

Though the impact of language was considered when adjusting our models, historically, this has not been the most accurate data within our EHR, and it is often difficult to differentiate patient’s preferred language from their primary caregiver’s language. It is unclear if language differences in our Hispanic patients may contribute to delays in symptom management or higher rates of imaging studies. For example, parents who have a preferred language other than English may not be able to tell the initial provider that their child is experiencing nausea or abdominal pain, especially if an interpreter is not available. This may require the family to advocate more for their child to receive the same quality of care, and in addition, can affect empathy and the patient–provider partnership. Conversely, communication barriers may lead to overutilizing tests such as imaging as a precaution to not miss a diagnosis.

Despite the additional information garnered about children hospitalized with AGE, there were limitations to this study. Because this was a retrospective chart review, accuracy of the data were limited to what was reported within the EHR. We relied only on what was documented in the medical record, which may have missed certain clinical characteristics that were evaluated such as exposure to sick contacts or international travel. Additionally, we collected clinical characteristics and admission order set pathway use at the time of admission, which may not reflect the longitudinal hospital experience or additional interventions later required.

We were unable to determine clinical reasoning behind specific testing or management; specifically, reasons that may influence provider ordering of antiemetics or obtainment of U/S scans. As previously mentioned, Spanish-speaking families may be underrepresented because of limitations in the accuracy of our EHR language data. Furthermore, in contrast to some previous studies, it was difficult to evaluate racial disparities in the context of ethnicity, because the majority of non-Hispanic patients in our study population identified as either white or other, which was not defined further within the EHR. Additionally, we did not have enough patients who identified as Asian American or Black from whom to make accurate inferences, so they were not separately included in our final analysis. Because of this and the large Hispanic population within our community, we chose to focus on differences in management of Hispanic patients in relation to the rest of the population, which may impact the generalizability of this study in areas with a lower Hispanic population. Multisite studies may garner more information regarding differences in management on the basis of both ethnicity and race, and further increase the generalizability of the study.

Management of children admitted to the hospital with AGE was similar for most common diagnostic testing and interventions. However, ethnic disparities still existed despite pathway use, with Hispanic children receiving more U/S scans and fewer antiemetics orders, even with no differences in abdominal pain, vomiting, or nausea between the 2 groups. Drivers affecting differences in management remain unclear. Future qualitative and/or prospective studies may clarify underlying reasons for differences in management.