Is More Better? Do Statewide Increases in Trauma Centers Reduce Injury-Related Mortality?

Abstract

Objectives

Trauma centers (TCs) are inconsistently distributed throughout the US. It is unclear if new TCs improve care and decrease mortality. We tested the hypothesis that increases in TCs are associated with decreases in injury-related mortality (IRM) at the state level.

Methods

We used data from the American Trauma Society to geolocate every state-designated or ACS-verified TC in all 50 states and DC from 2014–2018. These data were merged with publicly available IRM data from the Centers for Disease Control and Prevention. We used geographic information systems methods to map and study the relationships between TC locations and state-level IRM over time. Regression analysis, accounting for state-level fixed effects, was used to calculate the effect of total statewide number of TC on IRM and year-to-year changes in statewide TC with the IRM (shown as deaths per additional TC per 100,000 population, p-value).

Results

Nationwide between 2014 and 2018, the number of TC increased from 2039 to 2153. IRM also increased over time. There was notable interstate variation, from 1 to 284 TCs. Four patterns in statewide TC changes emerged: static (12), increased (29), decreased (5), or variable (4). Of states with TC increases, 26 (90%) had increased IRM between 2014 and 2017, while the remaining 3 saw a decline. Regression analysis demonstrated that having more trauma centers in a state was associated with a significantly higher IRM rate (0.38, p=0.03); adding new trauma centers was not associated with changes in IRM (0.02, p=0.8).

Conclusion

Having more TC and increasing the number of TC within a state is not associated with decreases in state-level IRM. In this case, more is not better. However, more work is needed identify the optimal number and location of trauma centers to improve IRM.

Level of Evidence:

III, Epidemiologic

BACKGROUND

Traumatic injury represents a significant source of morbidity and mortality in the United States.(1, 2) The United States (US) Health and Human Services Health Resources and Services Administration defines trauma systems as “pre-planned, comprehensive, and coordinated statewide and local injury response network(s)” that are capable of addressing the entire spectrum of trauma, from prevention to treatment and rehabilitation.(3) Trauma centers are specialized hospitals with the capability to provide care of the traumatically injured patient and are typically designated at the state level based on their resources. Research has demonstrated improved mortality outcomes in patients treated at trauma centers compared to non-trauma centers, and data suggest that expeditious transport to a trauma center is critical.(4, 5) Given this, it would seem that increasing the number of trauma centers might improve access to trauma centers and lower mortality from injury, but the evidence suggests that the picture is more complicated.

The US currently lacks a national trauma system, (6–8) and subsequently trauma system design has largely been left to states. As a result, there is significant heterogeneity in the distribution of trauma centers within the United States.(9) In addition, cost and effectiveness limit the number of trauma centers within a system; the monetary cost to develop and sustain a Level I trauma center is estimated at $6.8 million to over $10 million per year.(10, 11) Trauma centers also generate ongoing community costs to maintain the prehospital medical system and legislation and other forms of governance that support them.(12, 13) In addition, spreading trauma patients over many centers can lower patient volumes for each center; indeed, the American College of Surgeons (ACS) requires a minimum annual patient volume for Level I centers to ensure maintenance of expertise.(14, 15) Despite this, new trauma centers continue to open without clear criteria or need. Few studies have examined the consequences of changes in trauma center distribution over time from a population standpoint, instead focusing on regional or other local changes.(16–19)

We sought to describe the distribution in trauma centers in the United States, characterize the trends in the growth vs removal of centers between 2014 to 2017, and identify any relationships in such trends to injury-related mortality (IRM). We hypothesized that having more trauma centers within a state would be associated with lower injury-related mortality. We also hypothesized that increases in trauma centers from year to year would be associated with lower injury-related mortality. We also hypothesized that there might be subgroups of trauma center type (Level I or II vs. other levels) or injury type (homicide or suicide) where the absolute number of trauma centers or increases in number of trauma centers would be associated with lower injury-related mortality.

METHODS

Data Sources and Definitions

We combined two datasets to associate trauma center location with injury-related mortality data within the United States: the American Trauma Society’s Trauma Information Exchange Program (TIEP) and the Centers for Disease Control-Wide-ranging Online Data for Epidemiological Research (CDC WONDER).

Trauma center addresses and characteristics were obtained from the TIEP from 2014–2019. These data describe the location and characteristics of every state-designated or ACS-verified trauma center in the United States. Because not all states use the American College of Surgeons verification system, we utilized state designation level to determine trauma center level. The total number of trauma centers, as well as the total number of Level I and II trauma centers were counted for each state for each year of inclusion. We also calculated a trauma center “delta” for the years 2015–2017, or a measure of change in trauma center number from the year prior. The delta would be 0 if there was no change in trauma center number in a state, −1 if a trauma center closed in that state, or +1 if a trauma center was added.

Data on injury-related mortality (IRM) from 2014–2017 were obtained from the Centers for Disease Control and Prevention Multiple Causes of Mortality database and queried through CDC WONDER, an online public access system. The CDC maintains this annual mortality and population data on a national, state, and county level; data was obtained for the years 2014 to 2017. Mortality data is collected from the Division of Vital Statistics and is categorized by causes of death based on injury intent and mechanism. Mortality due to injury is defined by CDC WONDER and has been extracted from death certificates using 4-digit ICD-10 codes since 1999. Injury intents are categorized by the CDC as unintentional, homicide, suicide, or undetermined. More information about CDC WONDER is available at: https://wonder.cdc.gov/wonder/help/mcd.html.

Our main outcome of interest was injury-related mortality (IRM) rate, defined as the raw number of deaths in a state divided by the total state population per 100,000 individuals (deaths/100,000 population). The CDC calculates IRM for all injuries and subgroups of unintentional, homicide, and suicide, for all states with at least 20 deaths in the subgroup of interest. Due to confidentiality constraints, the CDC suppresses death counts less than 10 for a given category, and rates based on death counts less than 20 are reported as unreliable. For states with mortality counts between 10 and 20, we calculated injury mortality rate as the number of deaths divided by the state population. There was no state with death counts less than ten for any group or subgroup.

For each state (plus the District of Columbia) and each year of the study, the total number of trauma centers, number of state-designated Level I and II trauma centers, injury-related mortality rate, and subgroups of injury intents (homicide, suicide, and unintentional) were merged for all overlapping years of 2014–2017.

Geographic Information Systems Analysis

Using Geographic Information System (GIS) software, (ArcGIS 10.0. Redlands, CA: Environmental Systems Research Institute, Inc., 2010.) trauma centers for all years were geocoded based on their street address with Texas A&M Geoservices. Then, the number of trauma centers was summarized for each state in each study year and joined to the matching IRM data in the overlapping years of 2014–2017. Using this database, we mapped trauma center distribution. and the relationships between trauma center locations and state-level IRM GIS data were assessed to identify patterns of the number of trauma centers within a state, which were described as static, increasing, decreasing, or variable.

Regression Analysis

There are many factors which can affect injury-related mortality rates in a state, and these may differ significantly between states. Examples of these can include laws regarding speed limits, seatbelts, helmets, or firearms; rural-urban mix of state; or trauma system efficiency. In our analysis, we consider these effects as “unmeasured” as we cannot account for these in our dataset. While we would expect unmeasured variables to vary between states, we do not expect these effects to change dramatically within a state over this short 4-year period of examination. Based on this assumption, we chose to use a longitudinal repeated measures regression with “fixed effects,” which adjusts for unmeasured variables within states, assuming they do not vary over time. In this analysis, we use the state as the unit of observation, and the regression, adjusted for unmeasured effects within a state, to provide estimates for the effects of trauma center numbers as the independent variable on IRM as the dependent variable over the 4 years of the study. This way, the effects from the number of trauma centers within a state are compared within states rather than between states. The point estimate result from the regression is then the composite effect estimated over all the states.

The outcome of interest was the IRM rate in a state per 100,000 individuals as described above. We performed two sets of regression models. We analysed total IRM, as well as subgroups of unintentional deaths, suicides, and homicides. We created two model types to characterize the mathematical relationship between trauma centers and state-level IRM. The first set of models (Trauma Center Number) examines the effect of the number of trauma centers on IRM, the IRM subgroups, and a subset using only Level I and II trauma centers. These models examine if states with higher numbers of trauma centers have lower mortality than states with lower numbers of centers, adjusting for within-state effects. The second set of models (Trauma Center Delta) examines the change in number of trauma centers year-to-year and its effect on IRM, the IRM subgroups, and a subset using only Level I and II trauma centers. These models examine the magnitude of change in the number of trauma centers and effect on mortality, adjusting for within-state effects.

As a simplified example, Connecticut had 12 trauma centers in 2014, 11 in 2015 and 2016, and 13 in 2017. The Trauma Center Number models examine Connecticut’s relationship between 12 centers and IRM in 2014, 11 centers with IRM in 2015 and 2016, and 13 centers with IRM in 2017, taking into account the state’s baseline unmeasured factors. The Trauma Center Delta models examine the relationship of a delta of −1 with Connecticut’s 2015 IRM, 0 with Connecticut’s 2016 IRM, and +2 with Connecticut’s 2017 IRM. These analyses were repeated for each IRM subgroup and were also performed examining only Level I and II centers. Results are shown as beta-coefficient (95% CI, p-value). The unit of measure of the beta-coefficient is deaths per 100,000 people.

Analysis was performed using STATA MP, version 16.1 (College Station, TX), and data were analysed as longitudinal panel data using the STATA syntax xtset. Use of these data were deemed exempt as not human subjects research by the Institutional Review Board at MetroHealth Medical Center.

RESULTS

Between 2014 and 2018, the number of state-designated trauma centers increased from 2039 to 2153 (Table 1). There was significant interstate variation where some states designate very large numbers of trauma centers whereas others designate very few. The minimum number of centers in a state was one trauma center (Rhode Island, Vermont), and the maximum was 284 centers (Texas). This variation can be seen easily when mapped across the United States (Figures 1a and 1b), where even adjacent states may have very different philosophies about trauma center allocation and geography. In addition, there is large variation in the distribution of trauma centers by trauma center level designation. For example, Texas, with the highest number of trauma centers (ranging from 280–284 in our study period) had 27–32 Level I or II trauma centers with the remainder as lower-level centers; Vermont had only a single trauma center throughout the study period, an ACS-verified Level I center. Table 1 also shows the range of number and changes in the number of Level I and Level II trauma centers across the United States over the study period.

Table 1.

Total trauma centers by year and state-level changes

	2014	2015	2016	2017
All Trauma Center Levels
Total # in US	2039	2090	2091	2153
Median # per state, [IQR]	26 [10–49]	26 [11–50]	26 [11–50]	32 [13–51]
Range per state	(1–284)	(1–280)	(1–280)	(1–284)
Level I and II
Total # in US	524	529	528	568
Median # per state, [IQR]	7 [3–11]	6 [3–11]	6 [3–10]	6 [3–12]
Range per state	(1–54)	(1–57)	(1–56)	(1–58)

Figure 1a.

Trauma Centers in the United States, 2014–2019. Map of state-designated trauma centers locations in the United from 2014–2019 as registered in the American Trauma Society’s Trauma Information Exchange Program (TIEP). Older years are represented with larger dots. Trauma centers which were added over the study period can be identified as smaller dots on the map.

Figure 1b.

Trauma Centers in the Southeastern United States, 2014–2019. This magnification of the map in Figure 1a demonstrates that adjacent states often have variable distributions of trauma centers.

Four patterns in statewide numbers of trauma centers within states emerged. Trauma centers within a state over the study period were either static (12), increasing (29), decreasing (5), or variable (4). Of the 29 states with trauma center increases, 26 (90%) had increased IRM between 2014 and 2017, while the remaining three (10%) saw a decline. Table 2 demonstrates the trend of IRM across the four years of data. In general, IRM increased over time for all mechanisms of IRM.

Table 2.

Injury-Related mortality by year, by state, per 100,000 people

Injury-Related Mortality	2014	2015	2016	2017
IRM [IQR]	69.8 [69.8–76.3]	74.1 [65.4–81.2]	79.5 [68.6–88.2]	84.5 [68.9–91.0]
Unintentional IRM [IQR]	47.5 [41.1–52.4]	50.1 [44.5–57.2]	55.3 [47.4–60.9]	57.9 [48.8–63.2]
Homicide IRM [IQR]	4.7 [2.8–6.2]	5.1 [3.3–7.0]	5.7 [2.9–7.8]	5.9 [3.0–8.1]
Suicide IRM [IQR]	16.4 [13.9–20.4]	15.4 [12.9–18.7]	15.6 [12.8–18.8]	16.4 [13.9–20.4]

^*IRM=Injury-Related Mortality

Trauma Center Number regression models are presented in Table 3, examining IRM and subgroups related to the total number of trauma centers in a state. IRM increased as the number of trauma centers increased; the interpretation of this is that in a state, each additional trauma center is associated with an increase in IRM of 0.381 per 100,000 (95% CI 0.042,0.720; p=0.028). This increase in mortality is also statistically significant in the unintentional mortality subgroup. When considering only Level I and II centers, these trends are more pronounced – IRM increased 1.324 per 100,000 with each additional Level I or II center (95% CI 0.756,1.892; p<0.001). Unintentional IRM also increased significantly when there were greater number of Level I and Level II trauma centers within the state (1.149 per 100,000; 95% CI 0.700,1.602; p<0.001). Of note, the model R² for these models are low, suggesting that statewide trauma center number is not a very good predictor of IRM as it does not explain the variation in IRM.

Table 3.

Trauma Center Number Regression Models

Model Variable	Coefficient	95% CI	p-value	Model R2
Analysis of All Levels of Trauma Centers
All Injury Related Mortality	0.381	0.042, 0.720	0.028*	0.0401
Homicide Mortality	0.019	−0.009–0.456	0.179	0.0023
Suicide Mortality	0.029	−0.021–0.079	0.255	0.0065
Unintentional Mortality	0.328	0.045–0.611	0.024	0.0313
Analysis of Level I and Level II Trauma Centers
All Injury Related Mortality	1.324	0.756–1.892	<0.001*	0.1562
Homicide Mortality	0.067	−0.032–0.167	0.180	0.0011
Suicide Mortality	0.077	−0.008–0.162	0.075	0.1557
Unintentional Mortality	1.149	0.700–1.602	<0.001*	0.1087

^*p < 0.05.

Trauma Center Delta regression models comparing how changes in trauma centers to IRM are presented in Table 4. In general, changes in the number of trauma centers had no significant effect in any direction on overall mortality. There was, however, a small but significant improvement in homicide mortality rates with an increase in the number of trauma centers within a state, where homicide mortality decreased by 0.031 per 100,000 (95%CI −0.048,−0.013; p=0.001). Further subgroup analysis evaluating only the change in the number of Level I and II trauma centers demonstrated no significant changes except that the suicide IRM increased (0.120 more suicide deaths/100,000 with an additional trauma center; 95% CI 0.052, 0.189; p=0.001). Again, the model R² for these models are extremely low, suggesting that the changing the number of trauma centers in a state is not a reliable predictor of IRM.

Table 4.

Trauma Center Delta Regression Models

Model Variable	Coefficient	95% CI	p-value	Model R2
Analysis of All Levels of Trauma Centers
All Injury Related Mortality	0.021	−0.178, 0.220	0.832	0.0010
Homicide Mortality	−0.031	−0.048, −0.013	0.001*	0.0013
Suicide Mortality	−0.010	−0.039, 0.019	0.498	0.0000
Unintentional Mortality	0.058	−0. 099, 0.215	0.461	0.0025
Analysis of Level I and Level II Trauma Centers
All Injury Related Mortality	0.497	−0.243, 1.018	0.061	0.0003
Homicide Mortality	0.019	−0.063, 0.102	0.641	0.0006
Suicide Mortality	0.120	0.052, 0.189	0.001*	0.0003
Unintentional Mortality	0.367	−0.091, 0.825	0.114	0.0001

^*p < 0.05.

DISCUSSION

By mapping every trauma center in the United States longitudinally, we demonstrated significant heterogeneity in how trauma systems are designed and built nationwide. This can be seen by the maps presented in Figure 1a and 1b. Some systems focus on a small number of centers while others have a large numbers of trauma centers designated as Level III or lower. In addition, our data demonstrate that the number of trauma centers and change in the number of trauma centers in a state generally does not explain variation in IRM. However, the data suggests that there is a small association whereby larger numbers of trauma centers in a state was associated with higher IRM. The only subgroup that appears to benefit from the addition of trauma centers is homicide. These results do not support our initial hypothesis that more trauma centers or adding trauma centers improves IRM.

High quality specialized trauma care is crucial to ensure quality care of the injured patient.(20, 21) There have been many studies examining the importance of trauma center access by examining prehospital transport times, injury location relative to trauma centers, and geographic distance from trauma centers.(22–29) One analysis of geographic trauma center distribution patterns and IRM suggested states with clustered trauma centers have lower IRM compared to states with more geographically dispersed trauma center distributions, although this analysis was limited to Level I and II centers.(9) However, the effects of adding or subtracting trauma centers, and the contribution of the role of lower acuity (Level III/IV/V) centers is poorly understood.(5) Our results demonstrated that adding trauma centers was not associated with overall changes in IRM. The only clear reduction in IRM was seen in the homicide IRM when the total number of trauma centers increased in a year, and this effect was small at a reduction of 0.03 deaths per 100,000 people. Interpreting this another way, 33 new trauma centers would need to open in a state to prevent 1 homicide death per 100,000 people, which is not a cost-effective solution to decreasing homicide deaths.

Geographic distribution of trauma centers is closely related to trauma patient volume at each center. Studies suggest that higher patient volumes are associated with better outcomes due to center-level expertise.(4, 30–32) Adding trauma centers in areas without access may improve access to trauma care; conversely, adding trauma centers in areas without access issues may lead to dilution of expertise and lower patient volumes for those centers.(33, 34) In Ohio, a state-wide analysis demonstrated that a greater number of Level I centers in a county or region was associated with slightly higher mortality compared to counties and regions with only one Level I center.(15) One possible explanation for our findings is that new trauma centers have been created in communities with existing trauma centers and therefore do not reduce trauma mortality. Systematic attention to geography by adding trauma centers to areas with poor access to trauma care might help overcome any existing access issues.

Currently, there is wide heterogeneity in trauma center allocation. One comparison study of trauma centers in the 15 largest US cities demonstrated great variability in trauma center density, population density, and travel distance, suggesting that the needs of an area require consideration for more than mere total population.(35) State designation may allow for some increased control over the design of a trauma system, although this nationwide heterogeneity has led to cost-ineffectiveness.(36, 37) The ACS introduced the Needs Based Assessment of Trauma Systems (NBATS) tool to try to address this by considering factors like population, existing trauma centers, and number of trauma patients in a trauma service area, in addition to a factor called “community support” which includes the degree of city and county backing. (38–40) Unfortunately, the NBATS tool has not been validated and may be too generalizable for practical use. Local or financial pressures, rather than actual patient need, may still dictate whether trauma centers open or close. Indeed, the expansion of insurance coverage in recent years has brought a potential increase in reimbursements for trauma centers.(41) Prior studies from Georgia and California have demonstrated that the NBATS tool underestimates urban trauma center needs and overestimates rural needs.(38, 39) We have demonstrated here that trauma centers are increasing; unregulated growth may be detrimental to patients and their communities if there is no benefit to injury-related mortality.

Our study is limited in its retrospective nature; we cannot draw causative relationships between our data. Additionally, IRM data are limited by the quality of what is reported to the CDC, and many counties either lack complete data or suppress their statistics based on privacy concerns. However, our study uses a population-based approach to examine every state and the District of Columbia, providing a nationwide look at the heterogeneity of the US trauma system. Additionally, we only looked at IRM and did not consider other outcome measures related to traumatic injury; there may be more significant associations in those variables. We were unable to examine local factors that affect injury-related mortality or injury types, and have no data on the types of injuries, individual hospital volumes over time, or individuals who had injuries and survived, which are limit our analysis. In addition, we are limited by the data collected in CDC WONDER, and there may be variability in the identification of trauma patients from state to state. We attempted to overcome some of these limitations by adjusting for state-level confounders by using a fixed effects analysis, which controlled for unmeasured variables in our data. We assumed that policies or other factors in these states did not change significantly over the 4 years of analysis; while this is also a limitation, we believe it to be a reasonable assumption given the slow application of policy changes over time. While there were many limitations to our study, our analysis provides valuable population-level data indicating that opening new trauma centers does not effectively decrease injury-related mortality in a state.

In conclusion, we failed to prove our hypothesis that increases in trauma centers are associated with decreases in injury-related mortality at the state level. In fact, our results demonstrate that trauma centers numbers and injury-related mortality are not strongly related to one another. There is a weak relationship in the opposite direction where states with more trauma centers have higher IRM. Despite this, trauma centers have been increasing over time nationwide. We believe that a deliberate and geographically founded policy approach to trauma center allocation is needed to address specific areas with high injury-related mortality and limited existing access to trauma care.