The Extent to Which Geography Explains One of Trauma’s Troubling Trends: Insurance-Based Differences in Appropriate Inter-Facility Transfer

Abstract

Background:

A growing body of literature suggests the persistence of a counterproductive triage-pattern wherein uninsured adults with major-injuries presenting to NTCs are more likely than insured adults to be transferred. Geographic differences are frequently blamed. The objective of this study was to explore geography’s influence on variations in insurance transfer-patterns, asking whether differences in distance and travel-time by road from non-trauma centers (NTCs) to the nearest level 1 or 2 trauma center alter the effect. As a secondary objective, differences in neighborhood socioeconomic disadvantage were also assessed.

Methods:

Adults (16–64 years) with major-injuries (ISS >15) presenting to NTC EDs were abstracted from 2007–2014 state inpatient/ED claims. Differences in the risk-adjusted odds of admission-vs-transfer were compared using mixed-effect hierarchical logistic-regression and spatial-analysis.

Results:

A total of 48,283 adults presenting to 492 NTC EDs were included. Among them, risk-adjusted admission differences based on insurance-status exist (e.g. private-vs-uninsured OR[95%CI]: 1.60[1.45–1.76]). Spatial-analysis revealed significant geographic-variation (p-value<0.001). However, in contrast to expectations, the largest insurance-based discrepancies were seen in less disadvantaged NTCs located closer to larger trauma centers. Stratified analyses comparing the closest-vs-furthest distance, shortest-vs-longest travel-time, and least-vs-most deprived populations agreed, as did sensitivity analyses restricting uninsured transfer patients to those who remained uninsured-vs-subsequently became insured.

Conclusion:

Adults with major-injuries presenting to NTCs were less likely to be transferred if insured. The trend persisted after accounting for differences in access-to-care, revealing that while significant geographic-variation in the phenomenon exists, geography alone does not explain the issue. Taken together, the findings suggest that additional and potentially subjective elements to insurance-based triage disparities at NTCs are likely to exist.

Level of evidence:

Prognostic and Epidemiological, Level III

Consistent with prior literature suggesting the potential emergence of a new form of insurance bias and potential inverse disparity favoring uninsured patients, the results of our study found that adult patients with major injuries initially presenting to non-trauma center (NTC) emergency departments were less likely to be transferred if insured. Surprisingly, the largest insurance-based triage discrepancies were seen in less disadvantaged hospitals located closer to major metropolitan centers with more ready access to higher-level trauma care. Taken together, the findings suggest that additional and potentially subjective elements to insurance-based triage disparities at NTCs are likely to exist.

Introduction

Trauma is a leading cause of death and disability for patients of all ages, costing hospitals and payers more than $400 billion in total lifetime costs each year¹ and resulting in upwards of 18.5 million annual emergency department (ED) visits among adults aged 16–64 years.² For adults with major injuries, defined as an Injury Severity Score (ISS) >15, receipt of care for trauma at larger, more resourced level 1 or 2 trauma centers (Supplemental Table 1)³ is associated with increased survival,^4–7 a finding reflected in current triage guidelines.^8,9 In 2006, the National Study on the Costs and Outcomes of Trauma⁴ highlighted the survival benefit among younger injured patients, setting in motion an ongoing precedent for the preferential referral of severely-injured adults to higher levels of care (level 1 [L1] or level 2 [L2]). Timely inter-hospital transfer after initial evaluation and stabilization at lower-level trauma centers (level 3 or smaller) and non-trauma centers (NTCs) is also thought to improve survival,^4–7 especially in rural areas where immediate access to a L1 or L2 trauma center might not be available.^9,10

Cognizant of these results, regionalized trauma-systems in the United States (US) have been developed over the past 40 years with the goal of facilitating the direct transport or transfer of adult patients with major injuries to designated trauma centers.^11–13 Nevertheless, despite advances in trauma-system organization and a general consensus within the field that adult patients benefit from higher levels of care, a growing body of literature suggests that a counterproductive triage pattern continues to exist wherein uninsured adults with major injuries presenting to NTCs are more likely than insured adults to be transferred.^11–22 Hypothesized explanations for the phenomenon¹⁴ include: residual confounding, economic incentives that make it less likely for (providers at) NTCs to recoup the cost of uninsured patients’ care,^23,24 differential patient preferences that make it less likely for insured patients to want to be transferred,¹¹ unconscious bias among providers that tends to favor retention of racially- or socioeconomically-concordant patients who are more likely to be insured,^13,25,26 lack of knowledge about triage guidelines among some providers practicing at NTCs,²³ and/or a more straightforward lack of viable transfer options for complex patients presenting to rural NTCs.^16,22

The objective of this study was to address the geographic supposition, exploring geography’s influence on the persistence of insurance-based differences in transfer-status among adults with major injuries presenting to NTC EDs by asking whether the distance and travel-time by road from an NTC to the nearest L1 or L2 trauma center²⁷ can explain insurance-based differences in transfer-status. As a secondary objective, differences in neighborhood socioeconomic disadvantage (a regional composite measure of social and structural determinants of health)²⁸ among an NTC’s adult trauma population were also assessed using the average risk-adjusted national Area Deprivation Index (ADI)²⁸ of its severely-injured adult trauma patients. Consideration of neighborhood socioeconomic disadvantage was included in order to determine the extent to which other aspects of where a patient lives beyond immediate access to a larger trauma center might influence the results.

Methods

Data source and study population

State-specific hospital and ED visit claims from 2007–2014 contained with the Agency for Healthcare Research and Quality’s state inpatient (SID) and emergency department (SEDD) databases for Florida, New York, and California (2007–2011, the most recent years of data available) were queried for adult patients aged 16–64 years presenting to NTC EDs with International Classification of Diseases, 9^th revision, Clinical Modification (ICD-9-CM) primary diagnosis codes consistent with trauma (800.x-959.x) and a corresponding ISS >15. Patients with diagnoses of late effects of injury or poisoning (905.x-909.x), superficial injuries (910.x-919.x), foreign bodies (930.x-939.x), and burns (940.x-949.x) were excluded in order to more closely mimic the definition of trauma employed by US trauma registries.²⁹ Information on ISS was calculated based on documented primary/secondary ICD-9-CM diagnosis external cause of injury (E-codes) codes contained within patients’ hospital/ED billing claims. Presentation to an NTC ED was identified based on additional data contained within the American Hospital Association Annual Survey Database to which included hospital/ED visits were linked.

SEDD contains longitudinal information on ED visits that do not result in a hospital admission, including treat-and-release visits and transfers. SID contains longitudinal information on hospital admissions, including patients initially seen in the ED and subsequently admitted to the same hospital for observational (<2 days) or inpatient (≥2 days) stays as well as patients admitted from other hospitals as transfers. Patients admitted from NTC EDs to the same hospital were considered admissions. To account for delayed transfers and recommended stabilization of complex patients, patients initially admitted to an NTC had to remain within the NTC for a period of ≥2 days to be counted as an inpatient admission. Observational stays that were subsequently transferred were counted as transfers, as were patients directly transferred from NTC EDs.

The combination of SEDD and SID together provides longitudinal information on ED and hospital visits from the majority of hospitals/EDs within included states, representing >98.0% of hospital/ED discharges within each state and >24.2% of the total US population. Included states were chosen for their large population size, encompassed geographic urban-rural diversity, and extent of spread across the continental US. In both SID and SEDD, hospital/ED visits were excluded if there were errors in ICD-9-CM coding; missing information on insurance status, discharge disposition, or trauma center level; or if patients were reported to have died prior to arrival (dead on arrival) or while in the ED.

Variable definitions: insurance status, transfer-status, potential confounders, and outcomes

Variations in insurance status (categorized as uninsured, private insurance, Medicare [e.g. due to disability], Medicaid, and other [e.g. insurance through the Department of Defense, TRICARE]) served as the study’s primary explanatory variable. Differences in the odds of admission versus transfer was the primary outcome. Potential confounders accounted for in risk-adjustment included: ISS (continuous), patient age in years on ED presentation (continuous), gender (male versus female), presence of a severe head injury (maximum Abbreviated Injury Scale [AIS] ≥3 versus <3), presence of a severe facial injury (maximum AIS ≥3 versus <3), presence of a severe chest injury (maximum AIS ≥3 versus <3), presence of a severe abdominal injury (maximum AIS ≥3 versus <3), presence of a severe extremity/pelvic injury (maximum AIS ≥3 versus <3), presence of a severe external injury (maximum AIS ≥3 versus <3), mechanism of injury (blunt versus penetrating), Charlson Comorbidity Index (0, 1, 2, or ≥3), race/ethnicity (categorized as non-Hispanic White, non-Hispanic Black, Hispanic, and other), and hospital ownership (categorized as government, not for profit, and for profit). Missing data for potential confounders, where present (0.7% gender, 4.0% race/ethnicity), were assumed to be missing at random and addressed using multiple imputation.

As a sensitivity analysis to account for the potential of uninsured transfer patients being listed based on their expected insurance status in NTC EDs prior to hospitalization, longitudinal data in SEDD and SID were used to follow uninsured transfer patients who were subsequently admitted within the same state forward in time to determine whether during index hospitalization they remained uninsured or became insured either through reporting of existing coverage or enrollment in programs like Medicaid (e.g. hospital presumptive eligibility). Followed patients with all forms of insurance were assessed for differences in outcomes that occurred during index hospital admission or after transfer from NTC EDs, including mortality and major morbidity. Major morbidity was defined as ≥1 of the following calculated using ICD-9-CM diagnosis codes: acute myocardial infarction, Acute Respiratory Distress Syndrome (ARDS), cardiac arrest, cerebrovascular accident, pneumonia, pulmonary embolism, renal failure, sepsis, and septic shock.

Geographic parameters and stratified sub-population definitions

In order to assess for differences in the distance and travel-time by road from an NTC to the nearest L1 or L2 trauma center (even if the trauma center was located in an adjacent state), included NTCs were linked to previously collected data on ZIP code-based ambulance drives.²⁷ Variations in risk-adjusted national ADI were assessed by linking patients’ residential ZIP codes (available in New York and Florida) to their corresponding neighborhood ADI. Hospital-specific values for ADI were taken as the average from among included patients. As a metric, ADI accounts for differences in neighborhood socioeconomic disadvantage, acting as a composite measure of multiple factors including variations in educational distribution, unemployment, poverty, crowding, housing, and access to safe infrastructure for things such as phones, plumbing, and cars.²⁸ Quartiles of each (distance by road in kilometers [km], travel-time by road in minutes [min], and average risk-adjusted national ADI) were calculated and used in stratified comparisons of the closet/lowest/least deprived (Q1) versus the furthest/highest/most deprived (Q4) groups.

Statistical analysis

Differences in insurance status and potential confounders between admitted and transferred patients were compared using descriptive statistics. P-values were omitted given the large sample size. Presence of insurance-based differences in the odds of admission versus transfer were assessed using unadjusted and risk-adjusted mixed-effects hierarchical logistic-regression models with random intercepts for each NTC. Differences in the percentage of insured patients admitted were then mapped alongside variations in the travel-time by road from an NTC to the nearest L1 or L2 trauma center, looking for any apparent patterns. Spatial autocorrelation (i.e. clustering) was assessed using spatial-statistics. Stratified differences were compared using risk-adjusted mixed-effects hierarchical logistic-regression models.

Data analyses were performed using Stata Statistical Software: Release 16.1 (College Station, TX). Two-sided p-values<0.05 were considered significant. The Yale Human Investigation Committee approved the study. Data were reported in compliance with Strengthening The Reporting of Observational Studies in Epidemiology (STROBE) guidelines.

Results

A total of 48,283 adult patients with major injuries presenting to 492 NTC EDs met inclusion criteria. Of these, 81.5% (n=39,342) were admitted, while 18.5% were transferred (n=8.941). Sixteen percent (16.2%) of patients (n=7,831) were initially uninsured. Among uninsured patients, 71.7% (n=5,617/7,831) were admitted. In contrast, 83.7% (n=18,931/22,627) of privately-insured patients, 85.5% (n=6,398/7,485) of patients with Medicaid, 81.1% (n=3,396/4,189) of patients with Medicare, and 85.1% (n=5,000/5,878) of patients with other forms of insurance were admitted.

Differences in demographic characteristics are presented in Table 1. No meaningful differences in ISS, age, or gender were found among adult patients who were admitted versus transferred. However, compared to patients who were transferred, admitted patients were less likely to have severe head injuries (transferred: 65.7%, admitted: 49.8%), more likely to have severe chest injuries (transferred: 26.7%, admitted: 39.4%), more likely to have severe abdominal injuries (transferred: 5.5%, admitted: 11.5%), more likely to have severe extremity/pelvic injuries (transferred: 7.7%, admitted: 18.1%), more likely to have experienced penetrating trauma (transferred: 2.8%, admitted: 7.4%), more likely to have a greater number of pre-existing comorbidities, and more likely to have initially been admitted to a for profit NTC (transferred: 27.3%, admitted: 40.0%).

Table 1.

Differences in demographic characteristics among adult trauma patients who were admitted versus transferred after initial presentation to non-trauma center emergency departments

	Admitted		Transferred		Uninsured transfers on subsequent admission
	39,342	81.5%	8,941	18.5%	Remained uninsured	Became insured	Unknown*
Insurance					n=604, 50.8%	n=584, 49.2%	n=1,026
Uninsured	5,617	14.3%	2,214	24.8%	100.0%	--
Private insurance	18,931	48.1%	3,969	44.4%	--	40.4%
Medicaid	6,398	16.3%	1,087	12.2%	--	26.4%
Medicare	3,396	8.6%	793	8.9%	--	3.7%
Other	5,000	12.7%	878	9.8%	--	29.4%
Injury Severity Score
Median (IQR)	18	17–25	17	16–22	17 (16–21)	17 (17–22)	17 (17–22)
Age
Mean (SD)	43.0	14.5	42.5	14.7	40.6 (12.7)	40.7 (14.0)	38.1 (13.6)
Gender
Male	28,701	73.0%	6,574	73.5%	85.4%	78.4%	82.4%
Female	10,306	26.2%	2,354	26.3%	14.6%	21.6%	17.7%
Imputed (missing)	335		13
Severe head injury
AIS < 3	19,737	50.2%	3,067	34.3%	26.7%	36.0%	30.1%
AIS ≥ 3	19,605	49.8%	5,874	65.7%	73.3%	64.0%	70.0%
Severe facial injury
AIS < 3	39,110	99.4%	8,894	99.5%	--	--	--
AIS ≥ 3	232	0.6%	47	0.5%	--	--	--
Severe chest injury
AIS < 3	23,833	60.6%	6,557	73.3%	81.6%	68.7%	74.4%
AIS ≥ 3	15,509	39.4%	2,384	26.7%	18.4%	31.4%	25.7%
Severe abdominal injury
AIS < 3	34,822	88.5%	8,448	94.5%	96.4%	95.3%	94.3%
AIS ≥ 3	4,520	11.5%	493	5.5%	3.6%	4.7%	5.7%
Severe extremity/pelvic injury
AIS < 3	32,229	81.9%	8,253	92.3%	95.0%	91.7%	92.7%
AIS ≥ 3	7,113	18.1%	688	7.7%	5.0%	8.3%	7.3%
Severe external injury
AIS < 3	39,248	99.8%	8,871	99.2%	--	--	--
AIS ≥ 3	94	0.2%	70	0.8%	--	--	--
Mechanism of injury
Blunt trauma	36,446	92.6%	8,691	97.2%	95.4%	96.9%	94.2%
Penetrating trauma	2,896	7.4%	250	2.8%	4.6%	3.1%	5.8%
Charlson Comorbidity Index
0	30,237	76.9%	8,066	90.2%	93.4%	93.7%	94.7%
1	5,878	14.9%	643	7.2%	6.5%	5.8%	4.3%
2	1,625	4.1%	129	1.4%	0.2%	0.4%	0.7%
≥ 3	1,602	4.1%	103	1.2%	--	0.2%	0.3%
Median income
Q1 - lowest	10,148	25.8%	2,440	27.3%	32.2%	29.8%	33.7%
Q2	9,998	25.4%	2,792	31.2%	33.1%	28.1%	33.9%
Q3	9,478	24.1%	2,002	22.4%	23.1%	24.0%	19.5%
Q4 - highest	8,007	20.4%	1,402	15.7%	11.6%	18.2%	13.0%
Imputed (missing)	1,711		305
Race/Ethnicity
Non-Hispanic White	24,417	62.1%	6,560	73.4%	70.0%	71.2%	65.2%
Non-Hispanic Black	3,921	10.0%	667	7.5%	11.6%	9.4%	11.2%
Hispanic	6,987	17.8%	1,087	12.2%	15.1%	14.9%	18.2%
Other	2,294	5.8%	439	4.9%	3.4%	4.6%	5.4%
Imputed (missing)	1,723		188
Hospital ownership
Government	6,857	17.4%	1,073	12.0%	15.1%	15.9%	13.5%
Not for profit	16,748	42.6%	5,429	60.7%	47.5%	49.4%	54.7%
For profit	15,737	40.0%	2,439	27.3%	37.4%	34.8%	31.8%

Abbreviations: IQR – interquartile range, SD – standard deviation, AIS – Abbreviated Injury Scale

Cell counts and corresponding percentages for n ≤ 10 were omitted in compliance with SID/SEDD reporting standards

p-values omitted due to large sample size

^*Unknown insurance status presumed to be due to a combination of subsequent admission in a different state, death during transfer, potential misreporting of admission dates, and inconsistently reported patient linkage information between SEDD and SID

Association between insurance status and the odds of direct admission

When accounting for clustering of patients within NTCs (Table 2), marked differences in the odds of admission versus transfer based on variations in insurance status were found. For example, compared to uninsured patients, privately-insured patients had odds of admission that were 2.03 times greater (95%CI: 1.87–2.21). Risk-adjustment helped to moderate the effect but did not obviate the difference (OR[95%CI]: 1.60[1.45–1.76]). Similar risk-adjusted results were seen for Medicaid versus uninsured (OR[95%CI]: 2.05[1.81–2.33]), Medicare versus uninsured (1.72[1.49–1.98]), and other forms of insurance versus uninsured (2.10 [1.84–2.40]). Restricting uninsured transfer patients to those known to have remained uninsured (e.g. private-insurance versus uninsured OR[95%CI]: 1.41[1.29–1.54]) versus later become insured (e.g. 1.52[1.39–1.66]) did little to alter the results, nor did further clustering of NTCs within Hospital Referral Regions or Trauma Referral Regions³⁰ in an effort to account for differences among varying trauma systems (two-sided p-value >0.05, results did not change the patient-level fixed-effects).

Table 2.

Odds of admission versus transfer based on variations in insurance status among adult trauma patients initially presenting to non-trauma center emergency departments

	Odds of admission (OR)	95%CI
Unadjusted (random intercept by NTC)
Uninsured	1.00	(reference)
Private insurance	2.03	1.87	2.21
Medicaid	2.13	1.91	2.39
Medicare	2.58	2.28	2.91
Other	2.45	2.18	2.76
Risk-adjusted (random intercept by NTC)
Uninsured	1.00	(reference)
Private insurance	1.60	1.45	1.76
Medicaid	2.05	1.81	2.33
Medicare	1.72	1.49	1.98
Other	2.10	1.84	2.40
Risk-adjusted (random intercept by NTC clustered within HRR/TRR: no change in fixed-effects)
Uninsured	1.00	(reference)
Private insurance	1.60	1.45	1.76
Medicaid	2.05	1.81	2.33
Medicare	1.72	1.49	1.98
Other	2.10	1.84	2.40
Risk-adjusted (random intercept by NTC)
Uninsured who remained uninsured	1.00	(reference)
Private insurance	1.41	1.29	1.54
Medicaid	1.48	1.31	1.66
Medicare	1.78	1.57	2.03
Other	1.69	1.49	1.91
Risk-adjusted (random intercept by NTC)
Uninsured who became insured	1.00	(reference)
Private insurance	1.52	1.39	1.66
Medicaid	1.60	1.42	1.79
Medicare	1.92	1.70	2.18
Other	1.83	1.62	2.07

Abbreviation: NTC – non-trauma center, HRR – Hospital Referral Region, TRR – Trauma Referral Region

Results represent odd ratios (OR) and 95% confidence intervals (95%CI) taken from multi-level mixed-effects logistic regression.

Models were risk-adjusted for differences in patient-level Injury Severity Score (continuous), age (continuous), gender (male versus female), presence of severe head injury, presence of severe facial injury, presence of severe chest injury, presence of severe abdominal injury, presence of severe extremity/pelvic injury, presence of severe external injury, mechanism of injury (blunt versus penetrating), Charlson Comorbidity Index (0, 1, 2, ≥ 3), median income (quartile), race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, other), and hospital ownership (government, not for profit, for profit).

Differences in outcomes as a result of admission versus transfer are presented in Table 3. Compared to adult trauma patients who were transferred, admitted patients were more likely to die (risk-adjusted OR[95%CI]: 2.58[2.04–3.26]). They were also more likely to experience major morbidity, both overall (risk-adjusted OR[95%CI]: 1.73 [1.50–2.00]) as well as among each of the included constituent morbidities (two-sided p-value ≤0.001 for each).

Table 3.

Differences in risk-adjusted outcomes among adult trauma patients who were admitted versus transferred after initial presentation to non-trauma center emergency departments

	Admitted	Transferred		Admitted vs Transferred
	%	%	p-value	Risk-adjusted OR	95%CI
Mortality	7.8%	3.2%	<0.001	2.58	2.04	3.26
Major morbidity	20.7%	13.1%	<0.001	1.73	1.50	2.00
Acute myocardial infarction	1.1%	0.3%	<0.001	3.82	1.94	7.52
ARDS	7.4%	2.3%	<0.001	3.40	2.64	4.39
Cardiac arrest	1.3%	0.5%	0.001	2.64	1.48	4.68
Cerebrovascular accident	1.7%	0.4%	<0.001	4.51	2.57	7.19
Pneumonia	9.7%	3.8%	<0.001	2.74	2.21	3.40
Pulmonary embolism	0.9%	0.2%	<0.001	3.72	1.77	7.85
Renal failure	4.2%	1.0%	<0.001	4.42	3.10	6.29
Sepsis	2.2%	0.5%	<0.001	4.62	2.82	7.58
Septic shock	1.3%	0.3%	<0.001	5.08	2.65	9.72

Abbreviation: ARDS – Acute Respiratory Distress Syndrome

Results represent odd ratios (OR) and 95% confidence intervals (95%CI) taken from multi-level mixed-effects logistic regression.

Models were risk-adjusted for differences in patient-level Injury Severity Score (continuous), age (continuous), gender (male versus female), presence of severe head injury, presence of severe facial injury, presence of severe chest injury, presence of severe abdominal injury, presence of severe extremity/pelvic injury, presence of severe external injury, mechanism of injury (blunt versus penetrating), Charlson Comorbidity Index (0, 1, 2, ≥ 3), median income (quartile), race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, other), and hospital ownership (government, not for profit, for profit).

Geographic mapping and stratified analyses

Results of geographic spatial analysis are presented in Figure 1. While evidence of significant autocorrelation was found (two-sided p-value<0.001), the direction of the resultant association between the percentage of insured patients admitted and travel-time by road to the nearest L1 of L2 trauma center was opposite to what conventional expectations suggest: insured patients were more likely to be admitted when presenting to NTC EDs that were located closer to L1 or L2 trauma centers. On the map, this can be seen when looking at larger metropolitan areas in Florida (e.g. Miami, Tampa, and Jacksonville), New York (e.g. New York City, Buffalo, and Albany), and California (e.g. San Francisco, San Jose, and Los Angeles). Exclusion of NTCs that transferred all severely-injured adult patients (n=28, 4.1% of NTCs), admitted all severely-injured adult patients (n=70, 14.2% of NTCs), or admitted 98–100% of severely-injured adult patients (n=32, 6.5% of NTCs) did little to change the results (Supplemental Figure 1; two-sided p-value<0.001).

Figure 1.

Geographic distribution of included non-trauma centers (N=492; NTCs) and their percentage of insured adult patients with major injuries who were admitted (variation in color) based on differences in the travel-time by road from an NTC to the nearest level 1 or level 2 trauma center, even if the trauma center was located in an adjacent state (variation in size).

Stratified analyses based on quartiles of travel-time (Figure 2A) and distance (Figure 2B) by road to the nearest L1 or L2 trauma center concur, suggesting in risk-adjusted hierarchical modeling results that among patients within 5 minutes of a major trauma center, privately-insured adults with major injuries had 1.89 times the risk-adjusted odds of being admitted compared to uninsured patients (95%CI: 1.48–2.41). Results were similar for NTCs located within 6.0 km by road (3.7 miles) from the nearest L1 or L2 trauma center (OR[95%CI]: 1.84[1.42–2.37]). In contrast, among patients treated at NTCs located >30 minutes by road from the nearest L1 or L2 trauma center, differences based on insurance status were less pronounced (private versus uninsured OR[95%CI]: 1.57[1.32–1.86]). Similarly muted patterns were seen for NTCs located more than 47.5 km (29.5 miles) from the nearest L1 or L2 trauma center (private versus uninsured OR[95%CI]: 1.57[1.33–1.87]).

Figure 2.

Risk-adjusted differences (random intercept by non-trauma center) in the odds of admission versus transfer based on variations in insurance status stratified by: (A) travel-time by road to the nearest level 1 (L1) or level 2 (L2) trauma center (Q1 versus Q4), (B) distance by road to the nearest L1 or L2 trauma center (Q1 versus Q4), and (C) national risk-adjusted Area Deprivation Index (Q1 versus Q4; data from New York and Florida only)

Parallel patterns were seen in sensitivity analyses restricting uninsured transfer patients to those known to have remained uninsured or to have later become insured (Supplemental Figure 2A–2B). Numerical results are presented in Supplemental Table 2. When looking at the percentage of insured patients admitted across quartiles of travel-time (Figure 2A) and distance by road (Figure 2B) to the nearest L1 or L2 trauma center, a similar step-wise pattern emerged.

Geography also played a role based on differences in NTCs’ average patient-level ADI (Figure 2C). Larger in magnitude than differences based on quartile of travel-time or distance by road to the nearest L1 or L2 trauma center, differences in the average neighborhood socioeconomic disadvantage of an NTC’s adult trauma population accounted for upwards of a 27.1% relative increase in the odds of admission versus transfer between insured and uninsured patients. Among NTCs treating privately-insured and uninsured patients living in the least disadvantaged neighborhoods, the relative difference in the risk-adjusted odds of admission versus transfer between the two groups was 1.87 (95%CI: 1.45–2.40). In contrast, among NTCs treating patients living in the most disadvantaged neighborhoods, the relative difference in privately-insured versus uninsured patients’ odds of admission fell to 1.36 (95%CI: 1.17–1.66).

Parallel patterns were again seen in sensitivity analyses restricting uninsured transfer patients to those known to have remained uninsured or to have later become insured (Supplemental Figure 2C). Stepwise decreases in the percentage of insured patients admitted were observed as the quartile of an NTC’s average neighborhood socioeconomic disadvantage increased (Figure 2C).

Discussion

This study of adult trauma patients with major injuries presenting to NTC EDs demonstrates that despite ‘straightforward’ assertions of the role that geography is expected to play in the association between insurance status and transfer,^14,16,22 the reality behind one of trauma’s most troubling triage trends is not simple. In risk-adjusted models that accounted for clustering of patients within NTCs, marked differences in the odds of admission versus transfer were found. Severely-injured patients with any form of insurance were more likely than uninsured patients to be admitted. The trend was not removed by accounting for differences in NTCs’ travel-time and distance by road to the nearest L1 or L2 trauma center but was associated with variations in geography. Intriguingly, in contrast to expectations that rural hospitals with limited referral choice might be driving the insurance-based differences observed,^14,16,22 variations in transfer-status remained significant in all geographic settings and were actually more pronounced in hospitals located closer to major metropolitan areas with more ready access to larger trauma centers. Stratification based on NTC patients’ average neighborhood socioeconomic disadvantage further revealed that the least disadvantaged hospitals tended to exhibit the largest triage differences.

How to interpret these findings in light of the complex interplay of systemic, patient, and hospital factors that underlie the existence of disparities in the US is difficult.^31,32 No simple explanations or singular answers exist. Nevertheless, based on our results, one thing is clear, geography alone does not explain the issue. Accounting for geography’s presence in assessments of insurance status instead revealed a concerning and potentially more insidious issue: the disparity is more pronounced in less disadvantaged hospitals with greater access to higher-level trauma care. This finding, which stands in contrast to conventional expectations, merits investigation as does the overall low rate of transfer among severely-injured trauma patients (even after accounting for the presence of stabilizing short observational stays). Such findings are consistent with prior studies on the topic^11–22 and collectively suggest that at least one additional factor must be at play.

Identifying what those factors are and attempting to disentangle them from the complex array of social determinants that lie at the heart of disparities is beyond the scope of this work, yet in looking at the other proposed mechanisms for insurance-based triage patterns,¹⁴ a few interesting theories persist. We know from the results of this paper that insurance-based differences in the risk-adjusted odds of admission were lessened in rural settings located farther from larger trauma centers. Such a finding could speak to rural practitioners’ greater extent of experience dealing with a broad array of patients and their related ability to quickly discern what they can and cannot handle. With limited alternative options for regionally-available trauma care, differences in patient¹¹ and provider^13,25,26 preferences are less likely to matter as are any potential considerations of economic factors.^23,24 Lack of knowledge about triage guidelines among some practicing providers at NTCs²³ remains a concern as does the potential for residual confounding. Future studies of clinical data where researchers are able to carefully monitor a severely-injured patient’s entire course of care are needed to further explore these issues. Better training and resourcing of NTCs where many severely-injured adult patients appear to receive complex trauma care should also be encouraged to potentially help address this issue and in general improve the outcomes of all trauma patients presenting to all levels of care.

Ultimately, the observation that NTCs in close geographic proximity to L1 or L2 trauma centers are more likely to admit patients with significant injury is concerning. Further research is needed to identify the reasons for this unexpected result, including an understanding of local EMS practices and state regulations regarding field triage to trauma centers. Compliance with field triage guidelines will need to be assessed and potentially modified if a patient cohort with significant injury can be identified that current triage guidelines miss. In responding to these results, regional trauma planners are encouraged to recognize that not all injured patients are transported by EMS—a tendency which potentially explains at least a portion of the population initially presenting to and subsequently being admitted by NTCs. Among such patients, patient preference likely plays a significant role. It is likely that many patients (and potentially local providers at NTCs) do not understand the differences between a trauma center and an NTC. Therefore, meaningful efforts to change the results will likely need to include community (and provider) education aimed at highlighting the improved outcomes experienced by patients with severe injuries treated at designated L1 or L2 trauma centers, focusing on recognition of higher-level trauma centers as ‘centers of excellence’ for complex poly-system trauma care. Regional understandings of local triage patterns will be required to root out the full extent of this issue. For that reason, trauma and emergency services directors are encouraged to look at their own data to determine whether regional education, physician education, or a local public health information campaign would best encourage insured patients with significant injuries to benefit from the expertise of larger trauma centers and/or encourage providers at NTCs to aid in providing more timely access to the right care for the right patient at the right time.

The study has limitations. Most reflect its reliance on administrative data and related lack of nuanced clinical detail. Use of SID/SEDD allowed for a multi-state assessment of geography’s influence on the association between insurance status and the odds of admission versus transfer within a large population of adult patients with major injuries presenting to NTC EDs. Few databases enable such assessment. Inclusive of upwards of 24.2% of the total US population and a wide array of geographic diversity, the results are expected to exhibit a high extent of external validity. Nevertheless, in relying on state-level data, the findings might not be nationally-representative. They are also subject to limitations inherent to the use of multiple state databases, including the potential for variations in reporting. Future research is warranted to ascertain whether trends observed in geographically diverse states with regional access to larger trauma centers, such as Florida, New York, and California, also apply in more consistently rural states without ready access to any form of larger L1 or L2 trauma center. ISS values were retrospectively calculated from ICD-9-CM and E-codes. As such, they are an imperfect proxy of patient severity and a retroactive description of knowledge that, at the time of ED presentation, might not have been known.¹⁴ Despite this, they remain the best methodology available to account for trauma severity in large databases¹¹ and provide a consistent definition with what has already widely been used.^11–22

Consistent with prior literature suggesting the potential emergence of a new form of insurance bias and potential inverse disparity favoring uninsured patients,^11–22 the results of this study found that adult patients with major injuries initially presenting to NTC EDs were less likely to be transferred if insured. The trend persisted after accounting for geographic differences in access to higher-level trauma care, revealing that while significant geographic-variation in the phenomenon exists, geography alone does not explain the issue. Instead, in direct contrast to conventional expectations, the largest insurance-based triage discrepancies were seen in less disadvantaged hospitals located closer to major metropolitan centers with more ready access to higher-level trauma care. Taken together, the findings suggest that additional and potentially subjective elements to insurance-based triage disparities at NTCs are likely to exist.

Supplementary Material

Supplemental Data File (.doc, .tif, pdf, etc.)_1

Supplemental Figure 1. Geographic distribution of included non-trauma centers (N=362; NTCs) and their percentage of insured adult patients with major injuries who were admitted (variation in color) based on differences in the travel-time by road from an NTC to the nearest level 1 or level 2 trauma center, even if the trauma center was located in an adjacent state (variation in size) after removing NTCs that transferred all severely-injured adult patients (n=28, 4.1% of NTCs), admitted all severely-injured adult patients (n=70, 14.2% of NTCs), or admitted 98–100% of severely-injured adult patients (n=32, 6.5% of NTCs)

Supplemental Figure 2. Risk-adjusted differences (random intercept by non-trauma center) in the odds of admission versus transfer based on variations in insurance status stratified by: (A) travel time by road to the nearest level 1 (L1) or level 2 (L2) trauma center (Q1 versus Q4), (B) distance by road to the nearest L1 or L2 trauma center (Q1 versus Q4), and (C) national risk-adjusted Area Deprivation Index (Q1 versus Q4; data from New York and Florida only) among all uninsured, patients who remained uninsured, and patients who subsequently became insured.