Distinguishing Children’s Hospitals From Non–Children’s Hospitals in Large Claims Data

Discussion

In this study, we developed a reproducible method to classify hospitals using AHA survey data. This methodology was then validated by using publicly available hospital data as well as health care claims data. Our validations reveal that hospitals in tier A have the most comprehensive pediatric services and have substantially higher pediatric admissions than hospitals in the other tiers, suggesting that these 51 hospitals are CHs. Tier A hospitals are most likely to have level I pediatric trauma centers, conduct organ transplants, and conduct congenital heart surgery and are most likely to have memberships in the CHA, Children’s Oncology Group, and ACS NSQIP-P. Many of the tier B hospitals have similar pediatric services and memberships, but they have a lower proportion of admissions that are pediatric (10.96% in tier B versus 88.95% in tier A). The provision of comprehensive pediatric services and low pediatric admissions suggests that tier B hospitals are general hospitals with children’s units. Therefore, for specific studies and research questions, investigators may elect to consider both Tier A and Tier B hospitals as CHs rather than only considering tier A hospitals as CHs. Although stand-alone children’s hospitals may be directly of interest in some studies, our analysis reveals that limiting the definition of CHs only to those with CHA membership will exclude a significant number of hospitals for which comparable pediatric services are available. tier C and D hospitals have minimal pediatric services, and virtually none of these hospitals are members of the CHA, Children’s Oncology Group, or ACS NSQIP-P. Therefore, we would categorize tier C and D hospitals as NCHs.

We were able to conduct these analyses because the claims data provider that we used, HCCI, provided hospital-level matching on features abstracted from the AHA survey data and from our manual designations of publicly available pediatric services. This level of collaboration with administrative and claims data providers is rare, and health services researchers are left to use proportions of pediatric admissions to reliably differentiate CHs from NCHs. Our study suggests that this practical application of the proportion of pediatric admissions (or discharges) is valid. For example, our study revealed that nearly 90% of admissions at tier A hospitals were pediatric admissions, compared with ∼≤11% at tier B, C, and D hospitals. Thus, distinguishing between hospital types is possible even when hospital-level identifiers are not available. Although possible, we would advise caution of additional stratifications of hospital types, such as tier B versus tier C, by using discharge proportions alone.

Authors of previous studies investigating differences between CHs and NCHs used a variety of strategies to classify hospitals. One commonly used method for determining CH status is participation in the Pediatric Health Information System, which is a pediatric database that includes clinical and resource use data for 49 freestanding CHs that are members of the CHA.^12–16 CHA membership is also commonly used as a method to determine CH status, with >200 US hospitals currently having this membership.^17,18 Another commonly used database is the Healthcare Cost and Utilization Project Kids’ Inpatient Database (KID).^19,20 This database includes pediatric hospital discharge abstracts from 46 states, with all freestanding CHs being classified as CHs and all other hospitals being classified as NCHs. The KID is typically released in 3-year intervals, but the definition used to classify hospitals in the 2012 version dramatically changed compared with the definition used in previous versions, making multiyear comparisons challenging.²¹ Other methods to identify CHs include the availability of pediatric services such as pediatric trauma centers.^17,22

Although commonly used, these previous classification strategies have several limitations. First, the Pediatric Health Information System and KID data sets only include a subset of US hospitals. Furthermore, these strategies are often focused on freestanding CHs and do not granularly account for varying types of pediatric hospitals, such as children’s units within general hospitals. In most studies, children’s units are placed in the NCH category, which may not accurately represent the level of pediatric care provided at these facilities. Furthermore, the lack of a standardized classification method limits cross-study comparisons.

The classification methodology proposed in this study addresses many of the limitations with previous strategies to distinguish CHs from NCHs. First, the AHA survey profiles a universe of >6400 hospitals each year, allowing investigators to easily classify all hospitals in the United States using a single data source. This method may be replicated for all years of the AHA survey. Additionally, the use of a standardized classification strategy will allow for cross-study comparisons. Granular comparisons of care delivery at CHs versus NCHs are needed to inform health care delivery system structure and resource deployment. Trends toward regionalization of care and increasing care complexity at CHs must be critically appraised, with hospitals categorized in a reproducible and validated fashion.²³ Furthermore, the use of up to 4 tiers gives investigators the opportunity to consider varying types of pediatric hospitals, such as freestanding children’s centers and children’s units within general hospitals. Investigators can consider tiers separately or combine them on the basis of specific research questions and analyses.

Although we present a validated strategy to classify hospitals, there are some limitations to this method. Because the tier classifications rely on self-reported data collected by the AHA survey, there is the potential for hospitals to be misclassified. Additionally, there were many hospitals (n = 736) in the AHA survey that had missing data, making it impossible to classify these hospitals into tiers. Although the response rate for the AHA survey is high, the response rate for individual questions is frequently low (<50%).²⁴ To combat the high amounts of missing data in the AHA survey, survey administrators conduct imputation and estimation of some variables on the basis of previous years, but they do not estimate variables related to service offerings, beds, and facilities (ie, variables used to categorize hospitals in this study).²⁴ Despite this limitation, we were able to classify the vast majority of non–specialty hospitals in the United States (n = 3728) and validate the tiers using health care claims data. The percentages and interquartile ranges of pediatric discharges obtained from the HCCI also suggest that the bulk of the unclassified hospitals are NCHs. A final limitation of our proposed methodology is that many administrative and claims data providers strip hospital-level identifiers that facilitate linkage to AHA data. Under ideal circumstances, data custodians will provide linkages with key data elements before providing investigators with fully stripped data for further analyses. If this is not feasible, using proportions of pediatric admissions and discharges and searching for critical services of interest, as proposed in this study, may be possible.