Child Opportunity Index and Changes in Pediatric Acute Care Utilization in the COVID-19 Pandemic

This multicenter, observational study used data from the Pediatric Health Information System (PHIS) database, an administrative and billing database containing encounter-level data from 49 tertiary-care pediatric hospitals across the United States that is maintained by the Children’s Hospital Association (Lenexa, KS). Participating hospitals are located in 27 states and Washington, DC. Hospitals submit encounter-level data, including demographics, ZIP code, medications, and diagnoses based on International Classification of Disease, 10th Revision, Clinical Modification (ICD-10-CM) codes. Data were de-identified at the time of submission, and data quality and reliability were assured by joint efforts between the Children’s Hospital Association and participating hospitals. This study was deemed exempt by the institutional review board of Vanderbilt University Medical Center.

Children aged 0 to 17 years with ED encounters or observation/inpatient admission through the ED from March 15, 2017, to March 14, 2021, were included in the study. To identify the cohort of patients initially presenting for care at a participating hospital, we excluded direct admission and inpatient-to-inpatient transfer encounters. We also excluded 8 hospitals with incomplete data during the study period and encounters with missing or invalid ZIP codes (Supplemental Fig 3). Overall, excluded hospitals had a higher proportion of very low COI patients compared with included hospitals, but there was substantial hospital-level variability across included and excluded hospitals (Supplemental Table 4).

Briefly, the index includes 29 indicators that impact children’s healthy development across 3 domains (education; health and development; social and economic). Each indicator is transformed to a z-score, standardized within and across census tracts, which is used to obtain a weighted average within each domain. The 3 average domain z-scores are then combined into an overall score using similar weighting strategies. Indicator weighting reflects how strongly the variable impacts children’s long-term health and economic outcomes. Child opportunity levels are created by ranking all US census tracts by overall average z-scores and dividing them into quintiles (very low, low, moderate, high, and very high) that represent 20% of the US child population.²³  COI 2.0, typically measured at the census tract level, was mapped to the ZIP code level by the creators of COI and applied to data available in the PHIS database.²⁴ 

The primary exposure in this study was the COVID-19 pandemic time period, defined as March 15, 2020, to March 14, 2021. Outcomes and covariates from the exposure period were compared with the mean annual value across the prepandemic period, defined as rolling years starting March 15, 2017, to March 14, 2018, ending March 14, 2020.

The primary outcome was the relative volume reduction (decrease in ED and inpatient/observation admissions through the ED) within each COI quintile subgroup. Relative volume reduction was specifically defined as (mean annual volume in the prepandemic period–volume in the pandemic period/mean annual volume in the prepandmic period).

To further characterize changes in volume, the secondary outcome was relative volume reduction by COI quintile within clinical diagnosis groups on the basis of 3M All Patient Refined Diagnosis Related Groups (APR-DRGs).²⁵  3M APR-DRG uses a proprietary algorithm to classify claims data into larger clinical groups on the basis of the ICD-10-CM code combinations.

Sociodemographic and clinical variables examined included age, sex, race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, Asian American, other non-Hispanic), presence of a complex chronic condition,²⁶  hospital census region (Northeast, Midwest, South, West), distance from the hospital (in miles), encounter type (ED discharge, admitted through the ED), and primary payer (government, commercial, self-pay, other). Because race and ethnicity are social constructs, these variables were included in the analysis because they are strongly associated with differential opportunity in US society and are highly correlated with COI.²⁷  In PHIS, race and ethnicity are included as 2 distinct variables, which were collapsed into a single variable.²⁸  Hospitals submit race and ethnicity data for each visit according to hospital-specific practices, including parent/guardian self-report at the time of arrival or hospital registration.

Given the large number of ICD-10-CM codes and APR-DRGs, we used a 2-step modified Delphi method²⁹  to establish consensus in further grouping the APR-DRGs into more clinically meaningful groups. Three expert authors (C.F., M.G., and E.F.) classified APR-DRGs into categories to examine changes in presentation to care by diagnosis, and 2 authors (T.R. and H.D.) served as moderators during the process. In round 1, APR-DRGs were independently categorized, primarily on the basis of service line, with the distinction between medical and surgical diagnoses preserved because of the differential impact of early policies suspending elective procedures.³⁰  In round 2, the expert authors and moderators met virtually to “agree” or “disagree” with each APR-DRG grouping. Priority was also placed on distinguishing conditions that were known to show disparities in incidence (eg, trauma/injury, ACSCs),^4,21  or that were identified as experiencing disproportionate volume changes during the COVID-19 pandemic in previous studies (eg, respiratory infection, mental health).^17,31  The “Trauma/Injury’” category was defined using established ICD-10-CM diagnostic and external injury codes (S00–S99, T07, T14–28, T30–34, and T36–T78, *U01–*U03, V01–Y36, Y85–Y87, Y89). Groups were excluded if they did not meet an established criteria of 1000 annual encounters prepandemic. Additionally, APR-DRGs 540–640 relating to obstetrics, 950–952 relating to procedures unrelated to principal diagnosis, and 956 “Ungroupable” were all categorized as excluded from grouping. In the first modified Delphi round, 379 APR-DRGs were grouped, 54% of which met consensus. In round 2, 100% consensus was reached for a total of 23 categories covering 360 APR-DRGs.

We summarized demographic characteristics with frequencies and percentages using χ² tests for categorical data and medians and interquartile ranges using Kruskal-Wallis tests for continuous variables in the overall population and by COI. We performed a 1way χ² goodness-of-fit test to assess if the volume reduction in 2020 compared with the mean of the 3 prepandemic period years was distributed across all COI quintiles equally. All analyses were performed using SAS 9.4 (Cary, NC). P values <.05 were considered statistically significant.

The overall cohort, the prepandemic and pandemic cohorts combined, included 12 138 750 encounters among patients who were most often aged 0 to 2 years (37.0%), non-Hispanic White (36.0%), publicly insured (62.8%), and received care at a hospital in the South (36.8%). The overall cohort had higher representation from the very low COI (30.5%) quintile than the national distribution (by definition, 20% for each quintile).²³  The majority of encounters were ED discharges (88.0%) (Table 1).

During the prepandemic period, the very low COI quintile compared with the very high COI quintile was predominantly younger (38.6% vs 36.9% 0–2 years, P < .001), non-Hispanic Black (46.5% vs 15.1%, P < .001), and had government insurance (80.5% vs 51.7%, P < .001). Compared with the very high COI quintile, the very low COI quintile had a lower rate of complex chronic conditions (5.6% vs 7.4%, P < .001) and a higher frequency of ED discharge encounters (90.8% vs 87.1%, P < .001) (Table 2).

Across the entire cohort, there was a 46.8% reduction in volume in the pandemic period compared with the mean volume during the prepandemic period. With this substantial drop in volume, there was a change across all demographic groups in the cohort seeking care in the pandemic compared with the prepandemic time period (Table 1).

The drop in volume occurred disproportionately among the very low COI quintile (51.1% relative volume reduction) compared with the very high COI quintile (42.8% relative volume reduction), P < .001 (Fig 1).

There was notable variation in the relative volume reduction by COI among each individual hospital, with some hospitals’ relative volume reductions tightly clustered and others experiencing a wide range of volume reduction by COI. Thirty of the 41 included hospitals experienced the greatest relative volume reduction in the very low COI quintile. There was no major outlier driving the overall results (Fig 2).

The largest percentage volume drop occurred among the infectious respiratory (74.8%), infectious gastrointestinal (64.0%), and noninfectious respiratory (63.8%) diagnosis groups. The majority of clinical diagnosis groups demonstrated a significant difference in volume distribution by COI during the prepandemic and pandemic periods (Supplemental Table 5), with a larger relative volume drop among children in the very low COI quintile, as was seen in the overall cohort. The opposite pattern appeared among 2 groups: patients with infectious respiratory (78.8% very high versus 73.8% very low, P < .001) and toxicologic medical (36.4% very high versus 20.3% very low, P < .001) diseases, with a larger relative volume drop among the very high COI quintile (Table 3).

In this observational, multicenter study of acute care utilization at US children’s hospitals from March 2017 to March 2021, we found a 46.8% reduction in hospital utilization during the COVID-19 pandemic compared with the same time period during the 3 previous years. The relative reduction in volume was highest among the very low COI quintile (51.1%) and lowest among the very high COI quintile (42.8%). Overall, this pattern persisted across clinical diagnosis groupings, with the largest drops seen among the infectious gastrointestinal and respiratory groups.

The observed trend in volume reduction is in contrast to our a priori hypothesis that the largest relative volume drop in ED utilization would occur among higher COI groups. Drivers of this disproportionate drop in the lowest COI group may include changes in health care access and other infrastructure during the pandemic, changes in health care-seeking behaviors, and/or changes in the epidemiology of illnesses that especially impact children from families living in low opportunity communities.

One possible explanation is that this pattern may reflect the disparate impact of the pandemic on access to care. Despite the expansion of telehealth services that occurred during the pandemic,^32,33  decreased access to office-based primary care^11,12,30  among higher COI groups may have resulted in increased reliance on the ED to evaluate acute illness, more similar to typical care-seeking pattern of lower COI groups. Alternatively, higher rates of ED visits among pediatric practices with higher telemedicine use suggests that access to telehealth among this population may have facilitated triage and appropriate presentation for emergency care.³⁴  Conversely, children from lower COI neighborhoods may have experienced decreased ability to access acute care because of decreased access to public transportation or having parents and caregivers who were working as essential workers and could not leave their jobs or work remotely.^35–37  Payers have a vested interest in understanding how ambulatory care and telehealth access may have driven changes in expensive ED and inpatient utilization during the pandemic.

Another possible explanation is that individuals in lower COI households were more hesitant to seek care during this time period. A survey study conducted by Macy et al in May 2020 found increased hesitancy to seek care among families who were non-White, non-English speaking, publicly insured, from a neighborhood with a high COVID-19 positivity rate, and from a very low/low COI neighborhood.³⁸  Such avoidance can lead to undervaccination, increased risk of serious illness, and poor chronic illness control. Thus, the drivers of hesitancy and effective interventions to overcome hesitancy should be prioritized.

Both of these hypotheses assume a change in care-seeking behavior rather than a change in true incidence of disease. Alternatively, as suggested by the largest volume decreases among infectious respiratory and gastrointestinal groups in our study and a disproportionate decline in respiratory infections reported in previous studies,^12,17,18,39  better infection control practices such as masking and social distancing may have alleviated need for assessment in the hospital and/or admission among a population that historically relies more heavily on acute care for respiratory and infectious illnesses. Thus, enhanced infection control practices may be beneficial to keep in place long term even after the COVID-19 pandemic to lessen longstanding health disparities.

The greater overall volume decrease in conditions that are traditionally associated with ED utilization among lower COI patients (infectious gastrointestinal conditions, respiratory conditions, etc)^5,6  is 1 likely driver of the higher relative volume drop in the very low COI quintile across the cohort. Interestingly, the infectious respiratory group did experience lower relative volume reduction among the very low COI quintile as we hypothesized. This suggests that families from very high COI neighborhoods may have had greater access to and/or better-quality outpatient or telehealth management of ACSCs.^33,40  Additionally, this may indicate an increased ability to practice social distancing and infection prevention measures among this group. Key interventions for optimizing care of ACSCs should continue to be identified and scaled broadly to lessen such disparities.

There are several limitations of this study. First, the hospitals included in the PHIS database are all tertiary care children’s hospitals; thus, our results may not be representative of trends at all hospitals. Second, we were unable to determine if patients accessed care elsewhere during the study period. Knowing if patients were being seen at other, non-PHIS hospitals or nonhospital-based settings (eg, urgent care, primary care) may reveal different trends in utilization among COI groups. Data regarding primary care, telehealth, and non-PHIS, hospital-based visits would also allow for better understanding of observed trends. Third, examining COI at the ZIP code level may result in misclassification of a small proportion of families into the wrong COI quintile. However, we expect this misclassification to occur equally during the prepandemic and pandemic periods and therefore have minimal impact on our primary outcome: the relative change in volume among these 2 time periods. Finally, this is a descriptive, hypothesis-generating study. The lack of patient-reported and/or qualitative data limits conclusive explanations for the identified trends. All of these limitations represent areas for future study.

The significant decrease in acute care utilization at children’s hospitals during the COVID-19 pandemic compared with 2017 to 2019 occurred disproportionately among patients from very low COI ZIP codes. This was consistent across the majority of clinical diagnosis groupings despite hospital-to-hospital variation in underlying population distribution and volume reduction during the time period. Overall, our results suggest a narrowing in previously described disparities in acute care utilization during this time period. Further study of patient perspectives, specific disease patterns, association between regional COVID-19 disease prevalence and policies, and future trends are needed to better understand the observed patterns in ED/inpatient volume. Understanding the reasons for the observed trends is especially important to ensure that appropriate systemic investments are made to target drivers of the observed care utilization patterns and promote equitable access to care and health outcomes during future surges.

ACSC

ambulatory care sensitive condition

APR-DRG

All Patient Refined Diagnosis Related Group

ED

emergency department

COI

Child Opportunity Index

COVID-19

coronavirus disease 2019

ICD-10-CM

International Classification of Disease, 10th Revision, Clinical Modification

PHIS

Pediatric Health Information System