Hospital Access Patterns of Children With Technology Dependence

This retrospective, cross-sectional study follows Strengthening the Reporting of Observational Studies in Epidemiology guidelines.¹⁹  Study data contained no identifiable health information and was made available to facilitate research with informed consent waived and protocol approved by the Institutional Review Board.

We used 2017 inpatient and emergency department encounter datasets from Arizona, Florida, Kentucky, Massachusetts, New York, and Wisconsin. Massachusetts data were obtained directly from its Center for Health Information and Analysis and the balance obtained as state inpatient data and emergency department data from the Healthcare Cost and Utilization Project (HCUP).^20,21  In selecting states, we first identified those with available patient Zip Code information, and then used the US Census Bureau information to obtain a representative mix of size, rurality, population density, hospital referral regions, and health service areas.²²  The resulting sample comprises ∼23% of the US population in states with and without large border communities. To locate hospitals, we matched hospital identifiers to American Hospital Association Annual Survey addresses and used Google Geolocation Services to determine their longitude and latitude.²³  Finally, patient-level Zip Codes were used to assign Childhood Opportunity Index (COI) quintiles (higher COI = greater opportunity).^24,25  COI is a composite measure of 29 neighborhood indicators that affect children’s healthy development, including access to high-quality education, proximity to environmental toxins, health insurance coverage, local employment, and poverty.^24,25 

We selected all encounters involving patients <18 years of age that included International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes indicating the presence of 1 or more of the following technologies: tracheostomy (Z93.0, Z43.0, J95.0 and sub-codes), gastrostomy (Z93.1, Z43.1, K94.2 and subcodes), or ventricular shunt (Z98.2, T85.0, T85.01 and subcodes, T85.02 and sub-odes, T85.03 and subcodes, T85.09 and su-codes, T85.730 and subcodes) to define our cohort of CTD,^8,17  noting that this definition does not include other devices, such as ventricular assist devices or central venous lines.¹²  These technologies and associated codes were adapted from prior studies to maintain consistency with classifications of children with chronic complex conditions, technology-assistance procedures, and technology-dependent pediatric inpatients.^8,17,26  All inpatient admissions and emergency department (ED) encounters in all states were included. ED encounters were those starting and completing within an emergency department. We obtained chronic condition indicator (CCI) and admission type (urgent or emergent versus elective) data.

To determine their capabilities, we calculated the pediatric Hospital Capability Index (pHCI) for all hospitals within each state (see expanded methods).²⁷  Calculation of pHCI provides a continuous metric for comparing hospital pediatric services, applicable to hospitals of all sizes. Capabilities are expressed as pHCI, where 0 refers to hospitals not caring for or always transferring children, <0.25 referring to those that provide limited care for a few conditions, 0.25 to 0.75 covering a range of community hospitals offering varying levels of pediatric service, and values near 1 to referral centers that always admit and seldom transfer children. In previous work, all freestanding, children’s hospitals have returned pHCI’s ≥ 0.75,^1,2,4,6  so we refer to institutions with similar values as “high- capability” to acknowledge similar competencies and referral functions within their hospital systems.

For each encounter, travel distances were estimated as the distance from each hospital to the geographic centroid of the patient Zip Code. Because Euclidean distance is a reasonable surrogate for travel time, geographic proximity was calculated as linear distances using the Haversine formula.^28–30  We identified encounters involving hospital bypass as those where distance to the facility where the encounter occurred was greater than distance to the geographically nearest hospital. When state borders were crossed, we defined the nearest hospital as the closest facility within the destination state.

All analyses were performed using open source data science tools running within Jupyter notebooks using Python 3.7.³¹  We report descriptive statistics for admission and transfer rates, length of stay, and patient demographics with nonnormally distributed data summarized by their medians and interquartile ranges (IQR). Comparisons between encounters involving hospital bypass (verses geographically closest centers) and according to hospital capability (high versus low) were assessed using univariable logistic regression (reporting odds ratios [OR] and 95% confidence intervals [CI]) for categorical variables and the Mann-Whitney U test for continuous variables. Multivariable logistic regression analysis without interactions was performed to estimate the independent association between race, insurance, COI, number of technologies, CCI, and admission type with bypassing of the nearest hospital with adjusted OR (aOR), and 95% CI calculated. Encounters with missing demographic data were excluded from these analyses. Because access patterns may vary regionally, we performed sensitivity analyses for individual states. P values were 2-tailed and statistical significance set at P < .05.

There were 37 108 CTD encounters, including 16 403 (44.2%) ED visits and 20 705 (55.8%) inpatient admissions. Patient demographics and common diagnoses are presented in Table 1 and Supplemental Table 5. Technological dependence included tracheostomy in 17.3% of general encounters, gastrostomy in 75.7%, and ventricular shunt in 24.2%. Linking the American Hospital Association with patient level data identified 543 (64.9%) hospitals that encountered ≥1 CTD, including 525 general acute care hospitals and 18 specialty or rehabilitation hospitals (Fig 1). Of general hospitals with CTD encounters, 34 (6.3%) were high capability centers (pHCI ≥ 0.75), including 7 independent children’s hospitals. The number of high capability centers per state varied widely (range 2–11, Supplemental Tables 6, A–F). Of the remaining 491 general facilities, pediatric services were much more limited (mean pHCI = 0.12, median 0.06, IQR:0.02–0.15). Nearly all encounters (36 574, 98.6%) occurred in 1 of the 525 general hospitals. Most general hospital encounters (29 145, 79.7%) occurred within the 34 high capability centers (including 70% of ED visits and 87.5% of inpatient admissions, Figs 1 and 2). Of these, the 7 independent children’s hospitals accounted for 31.4% of ED visits and 35.0% of inpatient admissions (Fig 1). There were 33 deaths among ED encounters (mortality rate 0.2%) and 226 deaths among inpatient admissions (mortality rate 1.1%) in general hospitals and 27% of all deaths occurred within independent children’s hospitals (mortality rate 0.9%).

Only 20.3% of all encounters occurred outside high-capability centers, with most in ED’s (66.1%). There was some interstate variation with more low capability encounters in Florida and New York versus Arizona and Massachusetts (Supplemental Table 5). Odds of high-capability center encounters were lower for Black versus white CTD (OR 0.77, CI:0.72–0.82), whereas other and unknown races were higher (Table 1). Lower odds for high capability center encounters were observed among patients with Medicaid (OR 0.67 CI:0.63–0.71) versus private insurance and other or self-pay insurance (Table 1). CTD from lower COI quintile neighborhoods had lower odds of visiting high capability hospitals than those from the highest COI quintile (Table 1).

Odds of mortality were similar at high and low capability hospitals, but disposition differed (Fig 3 and Table 2). Overall, although most encounters resulted in routine discharges (77.1% low versus 85.4% high capability hospitals), high capability centers had lower odds of transfer (versus lower capability; OR 0.15, CI:0.13–0.17). For admissions, average length of stay was 1-day greater in high capability hospitals (Table 2).

Travel distance was available for 35 753 (97.7%) of encounters. CTD resided a median 2.8 miles (IQR:1.4–5.0 miles) from their geographically closest hospital but only 4178 encounters (11.7%) involved that site. Instead, most patients traveled nearly 4 times further (median 10.5 miles, IQR:5.0–25.4 miles) for care elsewhere. Median travel distance to a high capability center was 6 miles greater than to other hospitals (P < .001, Table 2). Only 2879 (17.6%) ED visits were to the nearest hospital (median travel distance 8.4 miles, IQR: 3.9–18.5), despite availability of closer institutions (2.7 miles, IQR:1.4–4.7, Supplemental Figure 4).

Sociodemographic characteristics of CTD encounters according to bypass of nearer institutions and distance traveled are presented in Table 3. Black families with CTD had lower odds of bypass compared with white (OR 0.71, CI:0.65–0.77) and traveled the least distance of 4.0 miles per encounter versus 14.3 miles for white. Hispanic CTD and those with unknown race were more likely to bypass (versus white). Hispanic CTDs traveled 6.1 miles and those with unknown race traveled 18.6 miles per encounter (Table 3). Patients with Medicaid insurance had lower odds of bypass than those with private insurance (OR 0.69, CI:0.64–0.75) and CTD residents in the 4 lowest COI quintiles were less likely to bypass a nearer institution compared with the highest quintile COI cohort (P < .001).

When CTD bypassed nearer facilities, they were 10 times more likely to travel to high-capability centers (CI: 9.43–10.8, Tables 2 and 4). When controlling for CCI and admission type the aOR is 8.61 (CI: 7.84–9.45). Overall, median pHCI of bypassed hospitals was 0.05 (IQR: 0.01–0.17) compared with 0.86 (IQR: 0.78–0.91) for the attended center. Moreover, even for encounters where high-capability centers were not involved, CTD bypassed nearer hospitals 63.8% of the time (Table 2). Of these 4707 encounters involving bypass to a low-capability hospital, the vast majority (4260, 90.5%) were to higher capability hospitals (median increase in pHCI = 0.36, IQR: 0.15–0.63). For CTD encounters at the nearest hospital, 77.7% were discharged from the hospital (versus 84.4% in those involving bypass), however encounters involving bypass had much lower odds of transfer (OR 0.25, CI:0.23–0.29, Table 4).

Bypass rates in CTD occurred consistently across all states, and when bypass occurred, high capability hospitals were significantly more likely to be attended, and low capability hospitals had greater odds of transfer. Findings for insurance and COI were similar to overall patterns with patients with Medicare or residents of lower COI areas having lower odds of bypass. Although there was some state-based variation in racial and ethnic access patterns, overall observations were not skewed by single states. Adjusted analyses comparing encounters involving hospital bypass (verses geographically closest centers) controlling for admission type and CCI were similar.

In this multistate study, we observe that children with technology dependence routinely bypass closer hospitals to access care in facilities with higher pediatric capability. Healthcare for CTD is intensely concentrated within high-capability hospitals (representing only 6% of all facilities) but bypass occurs even when accessing other facilities. Overall, CTD traveled almost 4 times further than their nearest facility to reach hospital services and as a result one-third of hospitals never encountered a CTD. This behavior seems warranted since visits to geographically closer and low capability facilities were associated with more interhospital transfers. We observed several sociodemographic differences in bypass, such that Black patients (versus white), publicly-insured (versus privately), and residents from disadvantaged (versus advantaged) neighborhoods were less likely to bypass and more likely to seek care at closer, lower capability facilities.

Our data suggest that families of CTD, and perhaps EMS providers, actively navigate the hospital system.³²  In part, this is unsurprising since it is known that CTD depend heavily on specialized hospitals for care. Here we show that almost 90% of CTD encounters are to high-capability, children’s hospital equivalent facilities. Differential access patterns were also discernable among unspecialized centers with preference shown for more capable, but more distant hospitals when closer facilities offered limited pediatric services. With consolidation of pediatric care,^1,2,5  there are fewer community hospital options and experienced care is less available.³³  Children with complex chronic conditions and their families are therefore under particular pressure to adapt.

Several factors may influence hospital access patterns. Continuity and the availability of specialized resources are of obvious importance for CTD when accessing high-capability centers, but are probably less important when accessing community hospitals. Follow-up after procedures may also play a role, yet procedures in this population are rarely performed outside of specialized centers.²² 

Regionalization offers economies of scale and improved care for uncommon conditions but concentrates care within a few hospitals and reduces experience in first-line facilities. Carried to its extreme, this dynamic risks deskilling of staff, increased transfer rates,^2,4,6,27,34  barriers to care for time-sensitive conditions, and suboptimal outcomes in critical illness.^35–46  As intelligent medical consumers within a consolidating system, families of CTD appear to now self-navigate across remaining hospital options to access needed care. In doing so, they access higher capability institutions where they are less likely to be transferred and more likely to be discharged from the hospital. Whereas additional travel mileage in our sample was not excessive, it could be much greater in rural states or those with fewer high-capability centers. Options may also decline rapidly and distances increase when pediatric services within general hospitals are repurposed following the coronavirus disease 2019 pandemic.^18,47,48  As the landscape evolves, the risk and benefit tradeoff between travel time and distance and pediatric expertise will need to be evaluated.⁴⁹ 

Differential utilization patterns according to socio-demographics could signify inequity in healthcare access, since families who have been made socially vulnerable, economically marginalized, or from diverse racial or ethnic backgrounds could be less able to travel and navigate the healthcare system.^38,40,41,50  Among adults, nongeographic social contextual factors clearly influence hospital utilization patterns and access to quality care.^40,41,51  Here we observed that more Black than white CTD encounters were in lower capability institutions and that Black patients were less likely to bypass local hospitals. We also observed that CTD with public insurance were less likely to bypass and more likely to visit low-capability hospitals than those with private insurance. Finally, although children from lower socioeconomic areas with medical complexity tend to use more services than advantaged,²⁵  we found them less likely to attend high capability institutions or bypass a local center. Taken together, this suggests that patients of different racial, ethnic, and socioeconomic backgrounds may face different barriers and differentially access health systems and, consequently, experience disparate healthcare outcomes. If replicated in larger populations and direct epidemiologic studies, the hospital bypass phenomenon could be helpful to evaluate mediating factors which include, among other things, systemic discrimination, unmeasured social determinants, healthcare relationships, and severity indices.^52,53  Meanwhile, differences in racial and ethnic access patterns between states indicate that relationships between social determinants and access depend on the local social context. Additional, more focused, objective analysis of service allocation with a social determinants of health lens, including between-state comparisons is justified to appropriately assess these factors.^52,53 

These findings have important implications. Children with chronic conditions, and indeed all children, should not be compelled to bypass community facilities to seek appropriate care at ever greater distances. In addition to worsening outcomes for time-sensitive conditions and increasing disparities in access to quality care, hospital bypass degrades competency in front-line facilities and overloads specialized centers with unspecialized demands.⁵⁴  To interrupt this dynamic, community hospitals should embrace pediatric readiness initiatives⁵⁵  and high-capability facilities should partner with community institutions via educational outreach programs, training, and just-in-time supports.⁵⁶  EMS providers could partner with a range of facilities to create pediatric point of entry policies, capability-focused access plans,⁵⁷  and coordinated, noncompetitive, robust interfacility transport systems with pediatric expertise.³⁴  Care teams of CTD may see an opportunity for education surrounding risks and benefits and provide guidance on when to access the closest medical resources. Improved understanding of the socioeconomic factors contributing to bypass may inform optimal and equitable public health planning^{38,40,42,43,50,58–61} . “Bypass” tendencies demand further exploration in other populations and state samples to see if similar effects are documented, and if the same variation according to social determinants of health are identified.^40,41,51  Finally, these findings highlight the importance of incorporating all general acute hospitals in pediatric studies as most encounters are not restricted to “children’s hospitals.”¹⁷ 

This study is subject to all the limitations of using all-encounter administrative datasets, including coding errors, misclassification, missing data, and unavailable variables such as illness severity and readmissions. We attempted to mitigate these effects by using very large datasets from multiple states and confirming our conclusions across hospitals and regions but are mindful that large data studies can sometimes show effects that are both statistically significant and practically inconsequential.⁶²  For several reasons related to the study population, these results should be interpreted with caution. First, we followed a CTD definition used in previous work,¹⁷  but other definitions could yield different results. Second, CTD represent a special population, so conclusions cannot generalize to all children and small numbers prevent full analysis of important variables, such as urban and rural residence, and clinical condition. Third, race is a social construct variably recorded in administrative datasets, here obtained from HCUP Partner organizations with restricted, uniform categorical coding. Fourth, we studied 6 states selected to be representative²²  and comprising almost one quarter of the US population, but findings may not generalize to all states and regions. Similarly, we used the most recent data available and have seen minimal variation in hospital capability and patient behavior over time,²  but access patterns may have changed since the coronavirus disease 2019 pandemic. Also, we used Zip Code centroids and spherical calculations rather than actual travel distances, since these are reasonable approximations in most locales^28–30  but actual travel times may differ and mode of transport was not incorporated. The bypass effect is also likely underestimated in out of state encounters however, this effect will be small as <5% of encounters originated in a different state. Finally, lacking transport information, we were unable to distinguish EMS from family travel decisions.

Most CTD receive care at a small minority of highly specialized hospitals with capabilities equivalent to pediatric referral centers. Those caring for CTD appear to actively navigate their health care systems, bypassing nearby hospitals to reach facilities with higher pediatric capability. When bypassing, CTD are transferred less frequently and more often discharged from the hospital but have longer travel times that may be inappropriate in time-sensitive conditions. Sociodemographic factors may influence this behavior, so its impact on quality, outcomes, and equity of care should be investigated.

CCSR

Clinical Classifications Software Refined

CTD

children with technology dependence

EMS

emergency medical services

HCUP

Healthcare Cost and Utilization Project

HCI

Hospital Capability Index

ICD

International Statistical Classification of Diseases

pHCI

Pediatric Hospital Capability Index

Zip Code

Postal codes from the US Postal Service Zone Improvement Plan