Travel Distances for Interhospital Transfers of Critically Ill Children: A Geospatial Analysis

Abstract

IMPORTANCE:

The U.S. pediatric acute care system has become more centralized, placing increasing importance on interhospital transfers.

OBJECTIVES:

We conducted a geospatial analysis of critically ill children undergoing interfacility transfer with a specific focus on understanding travel distances between the patient’s residence and the hospitals in which they receive care.

DESIGN, SETTING, AND PARTICIPANTS:

Retrospective geospatial analysis using five U.S. state-level administrative databases; four states observed from 2016 to 2019 and one state from 2018 to 2019. Participants included 10,665 children who experienced 11,713 episodes of critical illness involving transfer between two hospitals.

MAIN OUTCOMES AND MEASURES:

Travel distances and the incidence of “potentially suboptimal triage,” in which patients were transferred to a second hospital less than five miles further from their residence than the first hospital.

RESULTS:

Patients typically present to hospitals near their residence (median distance from residence to first hospital, 4.2 miles; interquartile range [IQR], 1.8–9.6 miles). Transfer distances are relatively large (median distance between hospitals, 28.9 miles; IQR, 11.2–53.2 miles), taking patients relatively far away from their residences (median distance from residence to second hospital, 30.1 miles; IQR, 12.2–54.9 miles). Potentially suboptimal triage was frequent: 24.2 percent of patients were transferred to a hospital less than five miles further away from their residence than the first hospital. Potentially suboptimal triage was most common in children living in urban counties, and became less common with increasing medical complexity.

CONCLUSIONS AND RELEVANCE:

The current pediatric critical care system is organized in a hub-and-spoke model, which requires large travel distances for some patients. Some transfers might be prevented by more efficient prehospital triage. Current transfer patterns suggest the choice of initial hospital is influenced by geography as well as by attempts to match hospital resources with perceived patient needs.

KEY POINTS

Question: How far do critically ill children travel from their residence to the first hospital in which they seek care and to the hospital in which they receive critical care services?

Findings: In this geospatial analysis, patients typically present to hospitals near their residence and are transferred relatively large distances. One in five patients are transferred to a second hospital less than five miles further from their residence than the first hospital.

Meaning: The current hub-and-spoke model of pediatric critical care services may provide opportunities for preventing interhospital transfer by optimizing prehospital triage decisions for critically ill children.

RESEARCH IN CONTEXT.

• Most U.S. pediatric emergency department encounters occur in community emergency departments, while most pediatric critical care occurs in specialized pediatric hospitals. As a result, the pediatric critical care system is increasingly centralized and dependent on interfacility transfers.

• At present little is known about the geographic distances undertaken during these transfers, including the relationship between where patients live, where they initially seek care, and where they ultimately seek care.

• If critically ill children frequently initially present to hospitals with nearby pediatric critical care, improving prehospital triage patterns may be an opportunity to reduce transfer burdens, and improving the quality and timeliness of pediatric critical care while reducing costs.

WHAT THIS STUDY MEANS.

• Many patients traveled large distances for care, creating a large burden on patients and their families.

• One in four patients was transferred to a hospital less than five miles further away from their residence than their presenting hospital, suggesting that improved prehospital triage could obviate the need for some transfers.

• As pediatric critical care continues to centralize, efforts to create a more organized interhospital transfer system could lead to improved outcomes and more patient-centered care.

Over time, an ever smaller number of U.S. hospitals are providing inpatient pediatric services (1–4). The result is a pediatric acute care system that is increasingly centralized, with a greater share of pediatric hospitalizations occurring in a smaller share of hospitals (5). This trend is particularly pronounced in critical care: in 2001, 50% of all pediatric critical care occurred in pediatric hospitals, but by 2019, this share was 85% (6). Meanwhile, over 90% of pediatric emergency department (ED) visits occur in general hospitals, creating a mismatch between where most critically ill children initially and ultimately receive care (7). These trends place increasing importance on effective prehospital triage, stabilization of critically ill children at general EDs, and safe interfacility transfer to a specialized pediatric center.

These trends also raise several important issues around the geospatial relationships between where patients live and where they ultimately receive care. For one, long transfer distances may burden families who must attend to their children far from their support systems and places of residence. For two, this hub-and-spoke model may lead to situations in which critically ill children initially present to hospitals with relatively low pediatric capabilities even though a higher capability hospital may be nearby. For these patients, transfer to the higher capability hospital might have been avoidable had the patient simply presented there in the first place. If this occurrence is common, then efforts to optimize prehospital triage could reduce the need for transfers, saving costs and potentially improving outcomes (8). However, if this occurrence is relatively rare, policy approaches might be better focused on improving pediatric capability in general EDs (9).

To better understand these issues, we conducted a geospatial analysis of critically ill children undergoing interhospital transfers in five states. Our goal was to describe the distances between the patient’s residence, the initial presenting hospital, and the hospital in which the patient received critical care. We also aimed to describe the frequency of “potentially suboptimal triage,” which we defined as instances in which patients were transferred to hospitals not significantly further from their residence than their presenting hospital. Finally, we aimed to identify potential factors associated with potentially suboptimal triage, such as rurality, socioeconomic disadvantage, medical complexity, and age.

METHODS

Data Sources

We performed a geospatial analysis using two publicly available administrative databases from the Agency for Healthcare Research and Quality’s Healthcare Cost and Utilization Project (HCUP): the State Inpatient Databases (SID) and State Emergency Department Databases (SEDD). Together these databases capture all inpatient and ED encounters in participating states, enabling comprehensive population-level study of acute care use (10). The SEDD contains stand-alone ED encounters, while the SID contains inpatient encounters as well as ED encounters that result in an admission to the same hospital. We used data from 2016 to 2019 for Florida, Iowa, Nebraska, and Wisconsin, and data from 2018 and 2019 for Maryland. We chose these states and years because they included unique patient identifiers enabling patients to be followed across hospitals, five-digit ZIP codes enabling geolocation of each patient’s residence, and unique hospital identifiers enabling linkage to the American Hospital Association (AHA) Annual Survey Database. Unique patient identifiers are available for 99–100% of all pediatric critical care encounters in all states but Florida, in which these identifiers are only available for approximately 40% of pediatric critical care encounters. We used the AHA survey to obtain latitude and longitude coordinates for each responding hospital. We used the Child Opportunity Index 2.0 project to obtain data on socioeconomic disadvantage specific to child health at the ZIP level (11).

Identifying Interhospital Transfers for Critically Ill Children

Using the SID and the SEDD, we identified episodes of pediatric critical care involving a transfer between at least two nonfederal acute care hospitals, using methods our group has previously deployed (12). Briefly, we first queried the SID and SEDD for all encounters for patients under 18 years old and arranged encounters for each patient in temporal order. We defined episodes of care using adjacent encounters: two encounters for the same patient were linked to form an episode if the second encounter began on the day the first encounter ended or the day after. Second, we identified interhospital transfers, defined as instances in which episodes of care contained encounters from multiple unique hospitals. Third, we excluded episodes that did not include care in an ICU, defined using ICU-specific revenue codes, as listed in Supplemental Appendix 1 (http://links.lww.com/CCX/B428) (6). Finally, we excluded any episodes consisting of encounters in separate hospital entities with a common location since these likely represent movement between different financial entities within the same hospital.

We then made additional exclusions to simplify the analysis and focus on the population of interest. To simplify the analysis, we only studied the first two encounters in an episode. We excluded episodes in which only the first hospital provided critical care to the patient—these episodes indicate that the transfer was a “step down” from critical care rather than a “step up” to a higher level of care. To focus on pediatric critical illness, we excluded any encounter with a primary neonatal or maternal diagnosis using HCUPs Clinical Classifications Software (13); to ensure we excluded neonatal encounters, we also excluded encounters from patients younger than 7 days at admission and encounters that HCUP classified as “newborn” encounters. To exclude patients who were traveling when hospitalized, we excluded any encounter in each state-level database from a patient who did not live in that state or a bordering state.

Geolocation and Distance Calculation

For each transfer, we geocoded three locations: the patient’s residence, the first hospital in which the patient received care, and the second hospital in which the patient received care. We geocoded patient residence based on the patient’s ZIP code using the population-weighted centroid of ZIP codes obtained from the Department of Housing and Urban Development (14). This approach places all patients at the average location of a patient within an area and is more accurate than placing patients at the area’s geometric center (15). We geocoded hospitals using the latitude and longitude coordinates from the AHA survey. We then used the Haversine formula (16) to calculate three distances: the distance from the patient’s residence to the first hospital, the distance from the patient’s residence to the second hospital, and the distance between the two hospitals. We calculated distances rather than drive times due to the frequent use of air transfer for interfacility transport of critically ill children (17, 18) and because distances are a reasonable proxy for drive time (19).

Analyses

We report patient characteristics and travel distances using standard summary statistics. We extracted relevant patient characteristics from the encounter capturing the second hospital stay since these were all inpatient encounters and uniformly reported in the SID. Relevant patient characteristics included demographics such as age, sex, and payer as well as clinical characteristics such as preexisting medical complexity (defined below), Major Diagnostic Category of primary diagnosis, and in-hospital mortality. We used the AHA survey to report the share of ICU encounters that occurred in a hospital with a PICU.

To study potentially suboptimal triage, we calculated the difference in the distance between the patient’s residence and the first hospital and the distance between the patient’s residence and the second hospital. This difference best reflects potentially suboptimal triage by capturing the additional distance a patient would have had to travel to directly reach the second hospital. We defined potentially suboptimal triage as a difference in distances of less than five miles, although we also studied two alternate definitions as sensitivity analyses (a negative difference and a difference of < 10 miles). These results are presented as the proportion of episodes that met the definition of potentially suboptimal triage.

To enrich our descriptive analysis, we then conducted bivariate analyses designed to identify the patient- and area-level factors that might influence triage decisions and thus be associated with potentially suboptimal triage. Because hospitals’ ability to care for a child may vary by the child’s age, we reported findings by patient age group, as defined by the National Institute of Child Health and Human Development (20). We examined medical complexity because families of children with medical complexity are more likely to bypass nearby hospitals to seek care at specialty hospitals (21). We defined medical complexity using the second version of the Pediatric Complex Chronic Conditions (CCCs) Classification algorithm and applying it to the International Classification of Diseases, 10th revision diagnoses coded as present on arrival to the second hospital (22). We then reported our findings by patients who had zero, one, two, or three or more body systems with CCCs, as well as by patients with and without CCCs indicative of technology dependence.

We examined differences in potentially suboptimal triage by area-level urbanicity and socioeconomic disadvantage in light of variation in distances to PICUs and inpatient units by these factors (23, 24). We defined urbanicity for each patient’s residence using HCUPs four-level categorization of the U.S. Department of Agriculture’s county-level Urban Influence Codes (25); we defined socioeconomic disadvantage using the five-level Child Opportunity Level, defined at the ZIP code based on national standards (11). Given presumed differences among states in geography, population density, and transfer networks, we also examined differences in potentially suboptimal triage by state.

We performed hypothesis testing in these bivariate analyses with chi-square tests, with a significance cutoff of 0.05. All analyses were performed using R, Version 4.2.0. This project was deemed exempt from Institutional Review Board review by the University of Pittsburgh because it involved only secondary analyses of existing de-identified data.

RESULTS

Our final cohort included 11,713 care episodes representing 10,655 unique patients. Table 1 presents patient demographics and the clinical characteristics of the ICU-related encounters included in our study cohort. Table 2 reports the distances traveled by critically ill children who require interfacility transfer. The first hospital was a median distance of 4.2 miles from the patient’s residence (interquartile range [IQR], 1.8–9.6). The transfer distance to a second hospital is typically longer (median distance between hospitals, 28.9 miles; IQR, 11.2–53.2 miles). This second hospital is typically much further from the patient’s residence (median distance between patient residence and second hospital, 30.1 miles; IQR, 12.2–54.9 miles).

TABLE 1.

Patient Characteristics

Variable	Value
Episodes	11,713
Patients	10,655
Age, yr	8 (1–14)
Female	5,252 (44.8%)
Payer
Medicaid	7,204 (61.5%)
Private insurance	3,857 (32.9%)
Other	636 (5.4%)
Missing	16 (0.1%)
Systems with PCCCs
0	7,976 (68.1%)
1	2,685 (22.9%)
2	850 (7.3%)
3+	202 (1.7%)
Technology dependent PCCC	267 (2.3%)
ICU encounter in a hospital with a pediatric-specific ICU	10,984 (93.4%)
Major diagnostic category
1: Nervous system	1,903 (16.2%)
4: Respiratory	3,836 (23.7%)
5: Circulatory	635 (5.4%)
10: Endocrine	1,235 (10.5%)
18: Infectious Disease	597 (5.1%)
19: Mental disorders	63 (0.5%)
21 + 24: Injuries, poisoning, multisystem trauma	1,758 (15.0%)
Other	1,686 (14.4%)
In-hospital mortality	287 (2.5%)
Child Opportunity Level of patient’s ZIP
Very high	1,858 (15.9%)
High	2,582 (22.0%)
Moderate	2,966 (25.3%)
Low	2,030 (17.3%)
Very low	2,190 (18.7%)
Missing	87 (0.7%)
Urbanicity of patient’s county
Large metro	4,540 (38.8%)
Small metro	4,547 (38.8%)
Micropolitan	1,359 (11.6%)
Not metropolitan or micropolitan	1,247 (10.6%)
Missing	20 (0.2%)

PCCCs = Pediatric Chronic Complex Conditions.

Data are expressed as n (%) or median (interquartile range).

TABLE 2.

Distances Traveled by Critically Ill Children Requiring Interfacility Transfer

(Miles)	Median (IQR)	Total Range
Patient residence to first hospital	4.2 (1.8–9.6)	0.0–543.0
Transfer distance (i.e., distance from first to second hospital)	28.9 (11.2–53.2)	0.1–451.0
Total distance traveled	35.9 (16.5–62.7)	0.9–683.0
Patient residence to second hospital	30.1 (12.2–54.9)	0.4–552.4
Difference between patient-first hospital distance and patient-second hospital distance	21.6 (5.3–45.8)	–340.1 to 446.8

IQR = interquartile range.

Two thousand eight hundred thirty-eight of the episodes (24.2%) met our definition of potentially suboptimal triage, in which the second hospital was less than five miles further away from the patient’s residence than the first hospital. Figure 1 shows a histogram of differences in distances, demonstrating that potentially suboptimal triage was the modal outcome in critically ill children requiring transfers. Figure 2 shows a scatterplot of distances to each hospital for patients for whom both encounters were within 50 miles of the patient’s residence; this plot demonstrates that many of the patients undergoing potentially suboptimal triage are concentrated within ten miles of their residence.

Figure 1.

Histogram of differences in the distance between a patient’s first hospital and second hospital.

Figure 2.

Scatterplot of distances from patient to first and second hospital (limited to distances < 50 miles).

Figure 3 shows the patient- and area-level factors associated with potentially suboptimal triage. Patient age (Fig. 3A) had no clear relationship with potentially suboptimal triage (p = 0.6). Increasing medical complexity, as measured by number of body systems with pediatric CCCs (Fig. 3B), was associated with a decreased likelihood of undergoing potentially suboptimal triage; 2151 patients (27.0%) without a CCC were potentially suboptimally triaged, with a linear decrease to 27 patients (13.4%) with three or more CCCs being potentially suboptimally triaged (p < 0.001). Socioeconomic disadvantage demonstrated no clear directional relationship with potentially suboptimal triage (Fig. 3C), although it was statistically different among groups (p < 0.001). Potentially suboptimal triage was particularly common for children residing in large metropolitan counties (Fig. 3D), as 1702 (37.5%) of these children were ultimately transferred to a hospital less than five miles further from their residence than their first hospital, compared with 58 (4.7%) of children residing in rural counties (p < 0.001). Supplemental Appendix 2 (http://links.lww.com/CCX/B428) reports the overall frequency of potentially suboptimal triage and subgroup analyses under alternate definitions of potentially suboptimal triage, with similar overall findings. Supplemental Appendix 3 (http://links.lww.com/CCX/B428) reports travel distances and the frequency of potentially suboptimal triage by state, demonstrating inter-state variation in potentially suboptimal triage. Supplemental Appendix 4 (http://links.lww.com/CCX/B428) reports travel distances and the frequency of potentially suboptimal triage by Major Diagnostic Category. Distances were broadly similar across diagnostic groups, although patients with primary cardiac diagnoses were transferred further and the small group of patients with primary mental health diagnoses were transferred shorter distances.

Figure 3.

Frequency of potentially suboptimal triage by subgroup. Subgroups: A = patient age; B = patient medical complexity; C = child opportunity level; D = urbanicity. PCCCs = Pediatric Chronic Complex Conditions.

DISCUSSION

In a geospatial analysis of critically ill children undergoing interhospital transfer, we show that critically ill children most often first present to care near their residence, with three-quarters of children traveling fewer than ten miles to their first hospital. However, transfer distances to receive definitive critical care are much larger, as the median transfer distance was 28.9 miles. These transfers often brought children far from where they live: the median distance from the patient’s residence was approximately 30 miles, and one-quarter of children were admitted to an ICU in a hospital more than 55 miles from their residence.

This hub-and-spoke model highlights a fundamental tension in the modern pediatric acute care system. On one hand, transferring critically ill children to high-volume centers likely results in higher quality care through volume-outcome relationships and concentrated specialty resources (26–28). On the other hand, this higher quality must be weighed against the increased burden on families, who travel longer distances, bear more financial costs, and suffer more family strain (2, 29, 30). The optimal balance of these factors is unknown and may vary by local context. Our findings highlight a need to better understand this balance in order to improve the value of interhospital transfers as well as the quality of care outside of specialty centers.

We also reveal a key opportunity to improve care at the prehospital level by refining triage decisions. One in four patients in our study were transferred to a second hospital less than five miles further away from their residence than the first hospital. Bringing these children directly to their definitive hospitals could improve both the timeliness and effectiveness of care (31). These findings highlight the need for better strategies to identify critically ill children in the field and bring them to hospitals with advanced acute care capabilities. Prehospital decision-making is a complex process involving not just geography and clinical status, but other factors such as preexisting treatment relationships and family experience. Future work should explicitly examine the roles of clinical, geographic, and these equally important factors in how families navigate this process. Because critically ill children may arrive to EDs by both emergency medical services (EMS) and by families (often in consultation with primary care pediatricians), and because the need for timely care may in some circumstances outweigh the need for definitive care, strategies to optimize the choice of the initial hospital must ultimately account for the complex interdependencies between care providers, families, and the unique needs of each individual patient (32–35). Improved decision support through better triage tools and tele-consults may assist in this process as well (36–38).

Potentially suboptimal triage appears to vary according to patient- and area-level factors. While age demonstrates no clear relationship, we found an inverse association of medical complexity with potentially suboptimal triage, consistent with prior work (21). This suggests that families and EMS providers attempt to match hospital resources with patient needs and that further optimizing triage based on measurable clinical factors may be possible. We found no clear association of potentially suboptimal triage with socioeconomic disadvantage but did find that it was significantly more common in urban areas. While the finding itself may be intuitive, the scale of potentially suboptimal triage decisions in urban areas suggests future efforts to improve prehospital triage should be targeted to urban areas. Conversely, the rarity of this problem in rural areas suggests that improving pediatric acute care quality for rural patients will require improvements in both rural hospital quality and the safety of transfers.

More research is needed to realize the promise of a highly functioning pediatric acute care system. At a system level, a deeper understanding of the structure and drivers of interfacility transfers is needed. For example, we do not know how many transfers could have been avoided through higher quality care at patients’ first hospital. To avoid exacerbating known issues of overcrowding in pediatric EDs, any methods to identify critically ill children for direct triage to specialty EDs should be paired with tools to identify children who can safely receive initial care in community settings (39). Geographic analysis to identify areas lacking hospitals highly capable of caring for acutely ill children may help target efforts to improve pediatric capability, such as the National Pediatric Readiness Project (9). In light of state-level variation in the frequency of potentially suboptimal triage, comparative analysis examining the roles of differing hospital system organizations, EMS regulations, and practice patterns among states may reveal best practices that can be applied across states. For example, Iowa and Nebraska are both rural Midwestern states that have meaningfully different rates of potentially suboptimal triage; deeper understanding of the triage, referral, and transfer systems in these states may yield useful insights. As pediatric expertise continues to centralize in referral centers, this system-level work will be increasingly important to ensure critically ill children receive optimal treatment from the moment they seek care.

Our study has several limitations. For one, HCUP data limit our ability to determine patient locations in higher resolution than ZIP codes or directly differentiate interhospital transfers from revisits using ambulance claims. However, our use of population-weighted ZIP code centroids allows for distance measurements that are unbiased on average (15), and our methods for identifying transfers have been reliably used in prior work (12). Although we cannot follow patients within hospitals, transfer to a hospital where a patient receives ICU care typically indicates perceived potential for clinical deterioration rather than transfer for other reasons before an unexpected ICU admission (40). Additionally, without detailed clinical data, we cannot assess the appropriateness of initial hospital choice or perform risk-adjusted analyses of the relationship between triage and outcomes. We believe, however, that this population-level study adds to the existing studies of smaller areas with access to more granular prehospital data (8, 32–34). Finally, our findings in a subset of states may not be applicable to states that have different pediatric acute care referral patterns, geography, or population density. Of specific note, because our data preclude the study of patients transferred across state lines, our findings for Maryland generally do not reflect patients who live in central and southern Maryland, where the nearest specialty center is typically in Washington, DC. That said, these issues were similarly prevalent in all states in the analysis, including those in which this issue is less of a concern (e.g., Florida).

CONCLUSIONS

We performed a geospatial analysis to describe the relationship between the residences of critically ill children and the hospitals that deliver care to these children. We demonstrate that the pediatric critical care system is organized in a hub-and-spoke model and identify areas for improvement in the choice of initial hospital for critically ill children. Improving outcomes for critically ill children in urban areas may require changes in prehospital referral patterns, while increasing the pediatric readiness of general EDs is likely required in rural areas.

Supplementary Material

cc9-6-e1175-s001.pdf ^{(163.9KB, pdf)}