Racial and Ethnic Disparities in Length of Stay for Common Pediatric Diagnoses: Trends From 2016 to 2019 Free

We used a retrospective cohort design and the 2016 and 2019 Kids’ Inpatient Database (KID) from the Healthcare Cost and Utilization Project. The KID is the largest publicly available all-payer pediatric inpatient care database in the United States. It features hospital administrative discharge data for children aged <21 years, from a nationally representative sample of hospitals across 46 states. Unweighted, the 2016 KID contains data from ∼3 million pediatric discharges, from which we determined the 10 most common primary medical or surgical diagnoses by frequency of International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes. We excluded hospitalizations for psychiatric diagnoses because we aimed to focus on medical and surgical diagnoses for this study. Newborn care diagnoses were excluded because LOS for newborn admissions can be significantly influenced by maternal LOS. Cancer-related diagnoses were excluded because this patient population is cared for primarily in the tertiary care setting, making LOS data less generalizable. If 2 or more ICD-10-CM codes for the same overarching diagnosis were within the initial top 10 primary diagnoses, those encounters were combined into 1 unifying diagnosis to determine the final top 10 diagnoses and for analysis (eg, bronchiolitis would include bronchiolitis due to respiratory syncytial virus, unspecified, and other specified). For longitudinal analysis, the same 10 diagnoses from 2016 were assessed using the 2019 KID data.

The following procedures were performed for the 2016 cohort and 2019 cohort individually. Descriptive statistics were used to determine frequencies of the top 10 primary diagnoses, and to describe sociodemographic and clinical characteristics of subject encounters. We performed a complete case analysis, because missing data were $≤$7%.²³  For each of the 10 most common primary medical and surgical diagnoses, to assess the association between race/ethnicity and LOS, we fit a generalized linear mixed effects model with a negative binomial distribution. The following covariates comprised the fixed effects and were included in the model to adjust for potential confounding: patient age, sex, severity of illness using the hospitalization resource intensity score for kids,²⁴  household income quartile by zip code, insurance type, hospital region, hospital type, transfer status, and number of diagnoses as a proxy for patient complexity. Standard errors and 95% confidence intervals were adjusted for clustering by hospital through inclusion of a random effect hospital identifier variable. The race data element as defined in the KID includes the following categories: white, Black, Hispanic, Asian American or Pacific Islander, Native American, and other. We collapsed the Asian American or Pacific Islander, Native American, and other groups to form 1 category because of small sample sizes in these groups. The reference group was white, because effects of structural racism should not apply broadly to this group.

For each model, we assessed linearity of the continuous variables and assessed for influence points using Cook’s distance. All statistical analyses were performed using R statistical analysis software version 3.6.0.

For each of the top 10 diagnoses, we used descriptive statistics to describe changes in associations between race/ethnicity and LOS between the 2016 and 2019 data.

Accounting for 267 522 hospitalizations in total, the 10 most common primary diagnoses in 2016 included bronchiolitis, pneumonia, diabetes, dehydration, asthma, sepsis, appendicitis, sickle cell disease, urinary tract infection (UTI), and upper respiratory infection (URI) (Table 1). In 2019, these diagnoses accounted for 225 528 hospitalizations.

Patient demographics were similar in 2016 and 2019 for hospitalizations included in the top 10 diagnoses. Patients were approximately half male and half belonging to Black, Hispanic, and Asian American or Pacific Islander, Native American, and other race/ethnicity. Approximately 60% had public insurance and hospitalizations in urban teaching hospitals, and ∼35% had median household income in the lowest quartile. The median number of diagnoses was 4 and median LOS was 2 days, with 25th to 75th percentile: 1 to 3 days (Table 2).

Analysis of 2016 data revealed statistically significant relationships between race/ethnicity and LOS for 8 of the top 10 diagnoses, excluding sickle cell disease and URI. The diagnoses with significantly longer LOS in 1 or more racial/ethnic group included appendicitis, asthma, bronchiolitis, dehydration, diabetes, pneumonia, sepsis, and UTI. 2019 data showed significant relationships for 7 of the 10 diagnoses, excluding asthma, dehydration, and sickle cell disease. The diagnoses with significantly longer LOS in 1 or more racial/ethnic group included appendicitis, diabetes, pneumonia, sepsis, URI, and UTI (Table 3). Black children hospitalized with bronchiolitis had significantly shorter LOS in both 2016 and 2019 (−3.3%, P < .001 and −3.2%, P < .001, respectively).

In 2016, the largest differences in LOS when compared with white children were seen for Black children with appendicitis (12.4% longer LOS, P < .001), diabetes (11.6%, P < .001), and UTI (9.9%, P < .001), Hispanic children with diabetes (9.2%, P < .001), and Asian American or Pacific Islander, Native American, and other children with diabetes (9%, P < .001). In 2019, the largest differences in LOS were seen for Black children with diabetes (12.3%, P < .001) and appendicitis (9.5%, P < .008), and Asian American or Pacific Islander, Native American, and other children with diabetes (11.7%, P < .001).

Significantly longer LOS persisted from 2016 to 2019 in 4 of 5 diagnoses for Black children, 3 of 5 diagnoses for Hispanic children, and 4 of 7 diagnoses for Asian American or Pacific Islander, Native American, and other children. Although the magnitude of disparities in LOS decreased in some cases from 2016 to 2019, most remained statistically significant (Table 3). For Black children, disparities in LOS decreased in 2019 for appendicitis, sepsis, and UTI; increased for diabetes; and were no longer seen for dehydration. For Hispanic children, disparities decreased for diabetes and sepsis, increased for appendicitis, and were no longer seen for asthma and UTI. For Asian American or Pacific Islander, Native American, and other children, disparities decreased for UTI; increased for diabetes, pneumonia, and sepsis; and were no longer seen for appendicitis, bronchiolitis, and dehydration. Additionally, there were 3 new findings of significantly longer LOS for Hispanic children with pneumonia and URI, and Asian American or Pacific Islander, Native American, and other children with URI. We found no significant leverage points in the multivariable models.

Our study aimed to assess racial and ethnic disparities in LOS and how they change over time for the most common pediatric inpatient diagnoses, as a measure of the effects of structural, institutional, and interpersonal racism, as well as the interventions aimed at addressing inequities. Although previous studies have looked at disparities in specific pediatric diagnoses that are less common and often surgical,^6,7,18–22  we focused on diagnoses for which children are hospitalized most frequently using a large, nationally representative sample. Additionally, we assessed these relationships longitudinally using data from 2016 and 2019. The 2016 KID data showed significantly longer LOS for 1 or more race/ethnicity when compared with white race in 8 of the top 10 primary diagnoses, after adjusting for potential confounding factors. These findings largely persisted in 2019, with improvement in disparities in some instances and worsening in others, as well as 3 new disparities seen. Interestingly, there was only 1 diagnosis for which the data showed a statistically significant decrease in LOS in both years, which was for Black children hospitalized with bronchiolitis.

Appendicitis is a common and frequently studied pediatric surgical diagnosis, and our results are consistent with previous studies that show racial and ethnic disparities in LOS while adding longitudinal changes. Our results showed significantly longer LOS for all groups with appendicitis in 2016, including the largest overall disparity in both years across all diagnoses (12.4% difference for Black children). This difference decreased in 2019 for Black children but grew for Hispanic children. Racial and ethnic disparities in appendicitis are likely related to variations in care delivery, such as undertreatment of pain, delays in diagnosis, lower rates of laparoscopic intervention, and higher rates of perforation.^8–13  A study that specifically analyzed pediatrician implicit attitudes about race and effect on routinely treated pediatric conditions suggests associations between implicit bias and treatment recommendations, including undertreatment of pain in Black children.²⁵  Previous findings of differences in care delivery for pediatric appendicitis and treatment of pain suggest provider biases may play a role in prolonging LOS for other diagnoses in which racial and ethnic disparities are seen, possibly by delaying diagnosis and therefore treatment. It is possible that interventions aimed at addressing these disparities have an unequal impact for Hispanic children who may have other factors contributing to health care inequities such as primary language and immigration status. The only other diagnosis exclusively treated in the inpatient setting for which our study demonstrated a disparity in LOS is sepsis, with the data showing significantly longer LOS for all groups in both years. Longer LOS for Black and Hispanic pediatric patients with severe sepsis has been shown before,¹⁹  and our study corroborates these findings while adjusting for severity of illness, suggesting this disparity in pediatric sepsis is present broadly and remains present over time, despite some improvement for Black and Hispanic children in 2019.

Unlike appendicitis and sepsis, many of the top 10 medical diagnoses we studied are commonly treated in the outpatient setting, such as diabetes. Although other studies have explored disparities in diabetes outcomes in adults,^26,27  our study shows disparately longer LOS for pediatric patients with diabetes in all groups in 2016 and 2019, similar to our findings for sepsis. Black children with diabetes represented 2 of the 4 largest differences in LOS among all included diagnoses in both years (11.6% difference in 2016, 12.3% in 2019). The second largest difference in LOS in 2019 was for Asian American or Pacific Islander, Native American, and other children with diabetes (11.7%). The presence of this disparity in all groups over time is particularly interesting, because diabetes is a chronic condition and its management is dependent on reliable access to medications and medical supplies, as well as routine specialist follow-up and continued reinforcement of diagnosis-related teaching in the primary care setting. These crucial aspects of diabetes care are generally less accessible for historically marginalized racial and ethnic groups who are more likely to live in underresourced areas, and this may contribute to differences in LOS for these patients. For example, a family with difficulties accessing care and medication may require additional diagnosis-related instruction and discharge coordination to assist with these inequities while in the hospital.

Other ambulatory care-sensitive conditions for which we saw racial and ethnic disparities in LOS were asthma, pneumonia, URI, and UTI. Similarly, decreased access to preventive care and inequities in unmet social needs likely contribute to these disparities by complicating and prolonging discharge coordination to ensure a safe and feasible postdischarge outpatient care plan. It is important to note that the disparity initially seen for Hispanic children with asthma was no longer significant in 2019, and there were no disparities in LOS for asthma in other groups. This may suggest that efforts to improve inequities in LOS for pediatric asthma, the most common chronic medical condition in children, are effective. This may include consistent use of asthma action plans as a reference tool for caregivers, asthma educator roles, and asthma-specific visiting nurse services. If so, these interventions may be a helpful blueprint for efforts to address disparities in other diagnoses where they continue to be prevalent, such as diabetes.

For UTI, our analysis found prolonged LOS for all groups in 2016, with minimal change in 2019 for Black children and Asian American or Pacific Islander, Native American, and other children, and resolution of this disparity for Hispanic children in 2019. Previous studies have looked at other demographic and care-related contributors to LOS and outcomes in UTI, but not associations with race and ethnicity.^28,29  This is particularly interesting when considering the 2011 American Academy of Pediatrics clinical practice guideline (CPG) on diagnosis and management of febrile UTI in infants and children, which includes white or non-Black race as a risk factor for UTI. It is possible the inclusion of white race as a risk factor for UTI has led to delays in diagnosis and differences in care delivery that prolong LOS, and potentially increase the likelihood of complications for children of other races. This underscores the importance of clinician understanding that there is no biological basis for clinical outcomes based on race, and supports the call to remove race from this CPG and others.³⁰  Removing race from CPGs is also important to appropriately recognize that it is effects of all types of racism that put particular racial groups at higher risk of certain outcomes, because the inclusion of race suggests certain outcomes are because of race alone.^28–30 

Acknowledging systemic racism as the overarching driver of the social and health inequities seen for historically marginalized populations in the United States is a crucial step in addressing these inequities. The more specific causes of health inequities, such as disparately longer LOS, are likely multifactorial and the effects similarly numerous. For pediatric patients, hospitalization for acute illness has been shown to cause medical traumatic stress and poor quality of sleep while in the hospital, not to mention missed school days and social isolation.^31,32  For parents of hospitalized children, time in the hospital may mean missed work hours, leading to loss of income and increased level of stress. Parents of multiple children may also have difficulties with child care and transportation during hospitalization, leading to additional financial burdens outside costs of the hospitalization itself. It is reasonable to assume that prolonged LOS can exacerbate the financial, emotional, and social burdens related to hospitalization in a population that is more vulnerable at baseline because of the reverberating effects of structural and interpersonal racism.¹  Additionally, LOS is a key driver of cost of hospitalization, so even small disparities in LOS may have significant cost implications on the health care system when they are present across thousands of admissions.^14–17,33 

Strengths of our study include the use of 2 large, nationally representative samples, which allow for robust, longitudinal statistical analysis of >450 000 pediatric admissions. KID data are gathered from a variety of hospital types, and provide the ability to adjust for important clinical and demographic characteristics. A limitation of our study is the level of data granularity within the KID, including the race data element, which combines race and ethnicity, and does not allow for investigation of ethnicity separate from race. The combined category of Asian American or Pacific Islander, Native American, and other race because of low numbers of encounters in those groups also limits our ability to assess for disparities specific to each distinct racial and ethnic group. Similarly, the KID lacks data on the patient’s primary language, which is an important contributor to LOS for patients and families with limited English proficiency.

Furthermore, the KID does not include necessary information to assess for variation in care delivery, such as diagnostic studies performed, treatments given, or timing of such interventions. Therefore, although we can identify disparities in LOS, we cannot further elucidate specific causes of these findings using this data. Lastly, because the KID data are administrative in nature and rely on appropriate application of ICD-10-CM codes, there is likely some human error and misclassification of codes within the data; however, the large sample size and analysis of the diagnoses with the highest frequency (and therefore more commonly coded for) likely mitigate this limitation.

In summary, the KID data showed significantly longer LOS for Black, Hispanic, and Asian American or Pacific Islander, Native American, and other pediatric patients in the majority of the top 10 most common inpatient diagnoses, with disparities present in 2016 largely persisting in 2019. Many of these diagnoses are considered ambulatory care-sensitive conditions, which may suggest that unequal access to preventive care plays a role in these disparities among other inequities in unmet social needs that likely contribute. Inequities in unmet social needs may also increase LOS by prolonging discharge coordination and requiring time for the care team to appropriately address needs such as transportation. For certain diagnoses, such as appendicitis and diabetes, the changes in LOS disparities over time were different for each racial and ethnic group. This may suggest current interventions to address inequities in health care and social needs are inadequately tailored to the unique needs of each group. Further investigation will include mixed-methods research to elucidate specific factors contributing to disparities in LOS. This may include assessing provider biases leading to variation in care delivery, barriers to accessing care, and unmet social needs, as well as elucidating the specific needs of each patient population to better inform health equity policies and interventions. The finding of shorter LOS for Black children with bronchiolitis in both years also warrants further investigation, because this could be indicative of provider biases leading to overtreatment of white children or undertreatment of Black children with bronchiolitis. Lastly, it will be important to further characterize the economic impacts of these disparities across thousands of admissions.