Disparities in Racial, Ethnic, and Payer Groups for Pediatric Safety Events in US Hospitals

This is a population-based, retrospective cohort study. Data were obtained from the 2019 Kids’ Inpatient Database (KID), a sample of pediatric hospitalizations sponsored and maintained by the AHRQ Healthcare Cost and Utilization Project (HCUP) and includes discharges from only 2019. KID data sets are released approximately every 3 years, with data on up to 7.5 million weighted cases. The data set includes public hospitals and academic medical centers and excludes rehabilitation and long-term acute care hospitals. The KID database includes a 10% sample of newborns and 80% sample of other pediatric discharges from 4000 US hospitals. Because KID includes newborn deliveries, >80% of patients are <1 year old. All diagnoses and procedures are coded using the International Classification of Diseases, 10th Revision, Clinical Modification. The 2019 KID included ∼3 million hospitalizations, accounting for 80% of national pediatric hospitalizations from >3900 US hospitals in 49 states plus the District of Columbia. We included patients aged 0 to 17 years old. The Children’s National Hospital institutional review board reviewed the protocol and determined it to be exempt.

AHRQ PDIs were the dependent variables. PDIs included hospital-level measures to detect potential safety problems that occur during a patient’s hospital stay, and incorporated risk adjustments for each indicator. The 7 PDIs include accidental puncture or laceration, iatrogenic pneumothorax, postoperative events (hemorrhage or hematoma, sepsis, and respiratory failure), and blood stream infections (BSIs) (BSIs in neonates and central venous catheter-related BSIs). The PDIs are publicly available (https://qualityindicators.ahrq.gov/Downloads/Modules/PDI/V2021/TechSpecs/PDI_2021_ICD10_techspecs_pdf.zip). Each PDI has a defined unique “at-risk” population that is described in the Technical Reports denominator definitions.²⁷ AHRQ PDIs are not inherently part of KID. PDI analytic methods were applied to the KID database in this study.

Race and ethnicity and payer status were the independent variables. Race and ethnicity is a social construct variable and was selected for this study to identify health disparities and inequities.²⁸ Race and ethnicity data in KID are collected in accordance with hospitals’ policies, which can vary by hospital and state; KID uses a single, aggregate variable, with ethnicity taking precedence over race in hospitals reporting both. Race and ethnicity in KID consists of Asian American and Pacific Islander, Black, Hispanic, Native American, white, other, missing, and unknown. Other includes other races and/or multiracial for included states, and was included in the analysis because it is 1 of the fastest growing groups of children in the United States.²⁹ Insurance status was classified as Medicaid, private insurance, self-pay, Medicare, no charge, and other.

For this analysis, model inclusion was based on sufficient sample size. The final racial and ethnic groups included in the analysis were Asian American/Pacific Islander, Black, Hispanic, white, and other. Patients classified as missing and unknown were excluded because of inability to interpret categories due to variability in the definition by state. Patients classified as Native American were excluded because of an inability to interpret findings due to an inadequate sample size. Patients with Medicare insurance or categorized as self-pay, no charge, and other groups were excluded because of inadequate sample sizes or variability in state definitions.

All analyses were conducted using sampling weights, as per HCUP guidance.³⁰ Unadjusted PDI rates were calculated per 10 000 children at risk, accompanied by 95% confidence intervals. Risk-adjustment varied by PDI type, based on AHRQ severity adjustments, which adjusted for age, gender, age–gender interaction, Diagnosis Related Groups, and comorbidities. None of the risk adjustments provided by AHRQ, however, includes race and ethnicity or payer status. We modified available AHRQ multivariable logistic regression models by adding race and ethnicity or insurance, and present adjusted rates for each stratum, as well as adjusted odds ratios.

For this analysis, the reference groups were the white or private payer groups. We did not select the racial category of white for the reference group as a normative or race-neutral benchmark,^31,32 but rather because they are historically socially advantaged groups.

Separate models were used for race and ethnicity and payer because these variables were highly associated in a subanalysis. Therefore, to better elucidate potential associations, we conducted analyses for primary insurance stratified by race and ethnicity, and for race and ethnicity stratified by payer. Tests of interaction between race and ethnicity and insurance revealed no significant interaction. All statistical analyses were performed with SAS v.9.4 (SAS Institute, Cary, North Carolina), with P values <.05 being considered statistically significant.

For the 5 243 747 hospitalizations analyzed, 48% of patients were female, and almost 80% were <1 year of age (Table 1). With respect to race and ethnicity, 46% of patients were classified as white; 19%, Hispanic; 15%, Black; 7%, other race/multiracial; 5%, Asian American or Pacific Islander; <1%, Native American; and 8%, missing or unknown. For insurance coverage, 47% of patients were insured by Medicaid; 44% by private insurance; 5% were self-pay; 3%, other; 0.3% by Medicare; and 0.1% were missing.

The eligible number of hospitalized children at risk for each PDI (denominator) ranged from 67 264 for neonatal BSI to 3 147 612 for accidental puncture or laceration. Overall PDI rates (per 10 000 discharges) were 331.4 for neonatal BSI (the highest PDI rate), 267.5 for postoperative respiratory failure, 114.9 for postoperative sepsis, 29.5 for postoperative hemorrhage/hematoma, 5.6 for central-line BSI, 3.5 for accidental puncture/laceration, and 0.7 for iatrogenic pneumothorax (the lowest PDI rate). Table 2 shows at-risk denominators and unadjusted rates for each PDI by race and ethnicity and payer status.

Compared with white patients, Black and Hispanic patients had significantly greater odds of 5 and 4 of the 7 PDIs, respectively (Table 3). Black patients had the largest disparity in postoperative sepsis, and Hispanic patients had the largest disparity in postoperative respiratory failure (Fig 1). No disparities were observed for accidental puncture or laceration, and Black and Hispanic children had significant lower adjusted odds of iatrogenic pneumothorax than white children.

Medicaid patients had significantly greater odds of 4 of the 7 PDIs, specifically postoperative events (hemorrhage, sepsis, and respiratory failure) and central venous catheter-related BSIs (Fig 2). Adjusted rates and odds ratios for each PDI with payer analysis are included in Supplemental Table 4.

Stratified analyses (Fig 1 and Supplemental Table 5) demonstrated that, even among privately insured children, racial and ethnic disparities persisted for certain PDIs. Hospitalized Black and Hispanic children had significantly greater odds of 3 PDIs, specifically central-venous catheter-related BSIs for both Black and Hispanic patients, and postoperative hemorrhage for Hispanic patients.

A recent study demonstrated that hospitals with higher safety grades do not consistently deliver that safe care among different patient populations, highlighting that high safety does not associate with equity.²³ Given a potential disconnect between safety and equity, our study sought to identify disparities in pediatric safety events; we found disparities for hospitalized Black and/or Hispanic children in 5 of 7 nationally developed and validated safety metrics (PDIs) and disparities for Medicaid-covered patients were identified for 4 of 7 PDIs. Our findings document national racial, ethnic, and payer disparities in AHRQ PDI measures of pediatric safety. We found that Black patients and Medicaid-insured patients experienced greatest disparities in postoperative sepsis, and Hispanic patients experienced the greatest inequity in postoperative respiratory failure. In addition, Black and Hispanic patients experience inequities with BSIs.

Our findings are consistent with and complement other studies that have focused on different aspects of patient safety for hospitalized children. For example, the Global Assessment of Pediatric Patient Safety Trigger Tools revealed racial and ethnic and socioeconomic disparities in rates of triggers and adverse events for hospitalized children.¹⁴ In this work, triggers were not themselves adverse events, but required chart review to confirm whether an adverse event occurred; triggers included medication triggers (eg, vitamin K administration after warfarin or opiate-related constipation), as well as hospital-care triggers (eg, intravenous infiltration or unplanned transfers).^14,33 Our work with PDIs identified disparities in a different set of safety events and complements findings from these earlier studies to highlight the broad range and depth of disparities in hospitalized children.

Our work is also consistent with findings from other studies that have focused on specific safety events, such as sepsis or central line-associated BSI (CLABSI). Although our study focused on postoperative sepsis, other studies (both single-center and national hospital cohorts) have shown racial and ethnic disparities exist in pediatric sepsis, despite advances in sepsis care that include reductions in sepsis-related mortality in US children.^16,34 In a cohort of children’s hospitals, CLABSI rates were persistently higher in Black and Hispanic versus white children, despite development and implementation of central-line bundle practices known to decrease CLABSI rates.¹⁵ A recent single-center study demonstrated mitigation of disparities in CLABSI rates after quality-improvement (QI) interventions.³⁵ This research documented that collaboration with multidisciplinary stakeholders can create effective, equity-focused QI interventions that reduce infection rates for Black patients.³⁵ QI can thus be an efficacious tool for achieving equity. Accomplishing this goal requires thoughtfully stratifying outcomes to identify and understand disparities before launching targeted QI efforts.³⁶ 

Of note, no disparities were observed in our analysis for accidental puncture or laceration, and Black and Hispanic children had significantly lower adjusted odds of iatrogenic pneumothorax than white children. The AHRQ technical report definition for the iatrogenic pneumothorax PDI focuses on secondary diagnosis of iatrogenic pneumothorax, and excludes nontraumatic iatrogenic pneumothorax present on admission or related to chest trauma, pleural effusion, thoracic surgeries, or transpleural cardiac procedures.²⁷ Lower rates of iatrogenic pneumothorax for Black and Hispanic adult patients have been reported by other investigators.²³ Perhaps there are different drivers of disparities in procedure-related disparities, compared with infection-related disparities, which can be explored in future work. Additional research on safety events for which there are no disparities or lower rates for historically disadvantaged groups may provide valuable insights into how to improve care for patients undergoing these procedures.

Adverse patient-safety events harm children and families, resulting in millions of dollars in direct and indirect medical costs,³⁷ and are likely attributable to a variety of different factors. Our results highlight that racial and ethnic disparities in care go beyond payer differences because stratification by insurance continued to show greater adjusted odds of safety events for certain PDIs. The reasons for these disparities, however, cannot be definitively determined in such a secondary data analysis. It is reasonable, however, to consider several plausible factors, including structural racism in the US health care system, clinician bias, insufficient cultural responsiveness, communication barriers, and/or impaired access to high-quality and timely health care.^38,–42 There are stark examples of structural racism in health care. Recent studies in adults showed that, relative to white patients, Black patients in the United States are more likely to receive care at institutions that have higher rates of adverse safety events and are more likely to experience adverse safety events, even when admitted to the same hospital as white patients.^9,43 Another study demonstrated that hospitals who see a smaller percentage of very low birth weight infants who are Black have higher mortality rates for Black patients, and demonstrated worse overall quality of care at hospitals that service higher populations of historically disadvantaged groups.⁴⁴ Studies have also shown that disproportionately minority-serving hospitals are more likely to have overcrowded emergency departments, reduced access to subspeciality services, less timely interventions, and less chance of providing hospital transfers.^45,46 

Clinician bias, insufficient cultural responsiveness, and communication barriers also can contribute to disparities. Studies have shown both minoritized children and adults who present with fractures to the hospital are less likely to receive opioids and achieve optimal pain reduction,^47,48 and less likely to receive appropriate imaging studies to help make accurate diagnoses.⁴⁹ A recent study demonstrated that even among healthy children, being Black is strongly associated with a higher risk of postoperative complications and mortality, noting that differences in postoperative outcomes may not be explained by racial variation in preoperative comorbidity.⁵⁰ In addition, research indicates that fatigue and cognitive stresses can increase implicit bias in physicians in the emergency department.³⁹ Disparities are additionally associated with lack of cultural responsiveness, as exemplified by research on factors associated with asthma-care quality for Medicaid-insured children, which found that patients at practice sites with the lowest cultural competence scores were significantly more likely to underuse preventive asthma medications and had significantly worse parent ratings of overall quality of asthma care.⁵¹ In addition, communication barriers can lead to a wide variety of adverse outcomes for Hispanic children, including deaths.^52,–55 

Frameworks have been proposed on how to ensure health equity in health care systems.^36,56,57 These frameworks broadly align in recommending a 3-step process for health equity in ongoing quality and safety efforts. Step 1 consists of examining quality and safety metrics by race and ethnicity, primary language, limited English proficiency, and/or proxies for socioeconomic status (insurance coverage and median household income by zip code). Findings from this current study generate data for Step 1 in these frameworks. Step 2 is to partner with wider groups of stakeholders, including community and patient/family partners affected by disparities. These partnerships should focus on qualitatively understanding potential drivers of disparities, with an emphasis on system-related factors. A deeper understanding of these factors could be grounded in safety frameworks such as the Systems Engineering Initiative for Patient Safety model of work system and patient safety, which proposes a more comprehensive definition of structure (including the physical environment, organizational culture, error reporting and analysis, and work design).⁵⁸ Step 3 is to design and test equity-focused QI interventions tailored to disparities populations. Data-monitoring systems developed in Step 1 can then support iterative rapid cycles of improvement.

Our team’s next steps will include a deeper analysis of these disparities and qualitative work to understand potential drivers. On the basis of our findings, postoperative sepsis, postoperative respiratory failure, and BSIs will be the foci. We plan to conduct qualitative interviews, and, ultimately, to design and test interventions to improve health equity in these key areas of preventable harm. We will include elements from disparities frameworks, such as Flores and Ngui³ and McKay and Parente,⁵⁹ to guide our next steps. The framework outlines potential individual and systemic drivers of disparities, including higher prevalence of risk factors (eg, delayed care and limited access to high-quality outpatient care), medical errors of omission or deviations from optimal practice, language barriers, insufficient data monitoring systems, and/or disparate views of what constitutes harm between patients/families and the health system.

Certain study limitations should be noted. Approximately 9% of the sample was missing race and ethnicity data, and there can also be substantial variability in how this information is collected across hospitals.⁶⁰ Of note, KID database reports an aggregate race and ethnicity variable where ethnicity takes precedence over race when hospitals report separate variables. Small sample sizes precluded analyses for important groups known to be historically disadvantaged such as Native Americans. Regression models cannot produce stable risk estimates for small sample sizes. The threshold for this size varies by total sample size, covariables included, and outcome rates. A key limitation of the study findings, therefore, is that small sample sizes did not allow for meaningful analyses for Native Americans and the self-pay group. The data set did not allow for separate analyses for uninsured children. The data set also did not include variables for limited English proficiency or health literacy, known risk factors for patient-safety events.^54,55 Although AHRQ risk adjustment was performed, it is possible that there was collinearity of race and ethnicity and/or payer status with covariates used as risk adjusters in the model, leading to potential overadjustment of the independent variables. This, however, would bias the findings toward the null hypothesis (no differences detected), and yet we were able to identify disparities. In addition, the KID database allows capture of up to 40 diagnostic codes for each encounter, reducing the chance of missed safety events. It is still possible, however, that events might be missed because of limitations in local coding practices and the database extraction process. Although errors in clinical coding of discharge summaries and medical billing records may exist in KID, HCUP uses strict validation criteria, minimizing this potential limitation. Lastly, the analysis was unable to assess clustering by hospital or hospital type (eg, freestanding children’s hospitals) and other confounders associated with the risk of safety events.

Furthering our understanding of disparities in pediatric safety events is a critical, foundational step to ultimately testing and implementing interventions to improve patient safety for historically disadvantaged children and achieve health equity for all children. Disparities in hospital patient-safety events were identified for US Black and Hispanic children, including postoperative hemorrhage and respiratory failure, sepsis, and BSIs. This large, population-based analysis provides the best data to date regarding where future work can be directed to prioritize efforts to better understand these disparities. Future work is needed to identify modifiable factors and design targeted interventions for patients experiencing disparities in pediatric safety events.

Abbreviations

AHRQ

Agency for Healthcare Research and Quality

BSI

blood stream infection

CLABSI

central line-associated blood stream infection

HCUP

Healthcare Cost and Utilization Project

KID

Kids’ Inpatient Database

PDI

pediatric quality indicator

QI

quality improvement