INTRODUCTION
The geographic distribution of trauma centers relative to injury incident locations determines both the medical care an individual patient receives following injury, and their exposure to trauma center-based injury prevention programs. Under the current system of trauma care regionalization in the United States (US),1,2 90% of the population has timely access to Level I or II trauma center services.3,4 Unfortunately, rural communities, low income populations, and historically underrepresented minority groups are over represented among the 10% of the population without access to trauma center services.3-7 These disparities in access to trauma center services represent one of the greatest limitations of the US emergency medical care system,8 and may contribute to disparities in injury outcomes.9-13
Efforts to expand access to trauma center services typically treat distance as a proxy for prehospital travel time, and locations with estimated travel times under one hour are considered to have access to trauma care.3 Recent studies indicate an effect of trauma center distance independent of prehospital time14,15 and substantial geographic variation in emergency medical service (EMS) arrival times,16 suggesting estimated travel times from scene to hospital are not a sufficient measure of trauma center access or need for services. Based on the relationships between trauma centers and EMS providers in regionalized trauma systems,17 trauma center characteristics (e.g. trauma center designation, hospital ownership) may contribute to the overall prehospital experience.
Need for trauma center services may vary based on the social context at injury event locations. Most injury incidents in the US occur within 10-miles of the patient’s residence,18 and residential communities tend to cluster by socioeconomic status and race.19 Urban trauma center catchment areas tend to be geographically compact and heavily influenced by the characteristics of the community surrounding the hospital.20 These patterns suggest significant variation in the social context at injury incident locations. The social context at injury event locations14 and at trauma centers21 may impact outcomes for all patients, regardless of their individual characteristics. Given known disparities in access to trauma center services,3-7 it is possible that geographic barriers magnify the mortality effect of community-level social measures. Unfortunately, the current approach to trauma system organization does not account for the relationship between social context of injury event locations and need for trauma center services.
This study used data from the Maryland Institute for Emergency Medical Services Systems (MIEMSS) and latent class analyses to develop a profile of spatial characteristics associated with concentrated risk of mortality from traumatic injury, and to examine the role of geographically concentrated risk as a determinant of injury mortality. We hypothesized that injury incident locations would be classifiable based on spatial characteristics, that patient and injury characteristics would vary across injury location classes, and that injury locations classes would be associated with injury mortality.
METHODS
Data Sources, Population, and Setting
This study used data from the 2015 MIEMSS Adult Trauma Registry and associated prehospital records from the MIEMSS eMeds patient care reporting system. Injury records for adult patients (age ≥ 18 years) were eligible for inclusion if the patient was injured in Maryland in 2015, transported by a Maryland-based emergency medical services (EMS) company by ground ambulance or helicopter, and treated at a state designated trauma center. Patients who were not treated at a trauma center but died while under EMS care at the injury scene or in transit were also included. Database linkage, mapping, GIS integration, and multiple imputation for missing data were described in detail in previous work.14 This study was reviewed and approved by the Institutional Review Board at the Johns Hopkins Bloomberg School of Public Health.
Variables
The environmental and community-level variables included in the latent class models were identified through prior analyses.14 Measures of the built environment included distance to the nearest trauma center, characteristics of the nearest trauma center, and land use. Trauma center distance was measured as the Euclidian distance between the scene of the injury event and the nearest trauma center, and categorized as 0-5-miles, 5-10-miles, 10-15-miles, 15-20 miles, or greater than 20 miles. Hospital type was categorized as private Level I/II, public Level I/II, or private Level III. Land use was categorized as residential, transportation, or other. Measures of the social environment included income and median age at the Zip Code Tabulation Area (ZCTA) level. ZCTA per capita income was categorized as less than $20,000, $20,000-$30,000, $30,000-$40,000, or greater than $40,000. ZCTA median age was categorized as less than 30 years, 30-39 years, or 40 of more years. Value ranges for distance, income, and age categories were identified based on visual examination of the distribution of continuous measures. Individual level variables used in the adjusted regression model included age, sex, race and/or ethnicity, injury severity (based on Injury Severity Score and Revised Trauma Score), injury mechanism, Charlson Comorbidity Index, insurance status, and prehospital time, as previously described.14
Analytic Approach
Latent class analyses attempt to identify unmeasured classes within a set of observations based on patterns of measured covariates, and to estimate conditional probabilities of class membership, or the probability of a specific trait or measurement given membership in a specific class. Conditional probabilities can then be used to calculate posterior probabilities, that is, the probability that an individual observation belongs to a specific class given the measured traits. Stata 13 (StataCorp, College Station, TX) was used for all statistical analyses. Latent class models were developed and tested using the LCA Stata Plugin developed by Lanza, et al.22
In this study, latent class analyses were used to identify patterns of injury scene characteristics indicating different location classes. Using complete case data, latent class models were tested with class counts ranging from two to thirteen. Models were assessed based on standard log-likelihood criteria (e.g. Akaike and Bayesian information criteria23), entropy, and visual examination of quantile-quantile plots of the standardized residuals for each model, as described by Wang, et al.24 Together, these criteria were used to identify the point at which inclusion of additional classes no longer improved model fit.
After identifying the best fitting, most parsimonious model, the latent class model was applied to each imputed data set and results were pooled using the guidelines for multiple imputation proposed by Rubin, et al.25 The probability of mortality for each class and conditional probabilities of class membership for each location feature were examined, and posterior probabilities were used to assign each pattern of location features to the most likely latent class.
Using ArcGIS Desktop version 10.2.2 (ESRI, Redlands, CA), a grid with quarter mile squares was overlaid on a map of Maryland, and latent classes were applied to each square based on the observed pattern of geographic characteristics within the square. Qualitative class labels were developed based on conditional probabilities of class membership, visual assessment of the geographic distribution of location classes, and comparison of the location class map with the American Communities Project26 classification of counties in Maryland.
Distribution of individual characteristics were estimated for each class and logistic regression models were used to examine class membership as a determinant of mortality with and without adjustment for individual characteristics. Distributions and regression models were estimated separately for each imputed data set and pooled according to Rubin’s guidelines.25 Standard spatial statistics methods were used to assess residual spatial dependence of the regression models,27 including semivariograms of the standardized residuals at the individual location level and Moran’s I at the ZCTA level.
Sensitivity analyses were used to assess the impact of historic events on injury mortality patterns during the study period. Baltimore City experienced a period of civil unrest in April 2015, which may have impacted EMS response and injury outcomes. To examine the possible effect of the unrest, we performed regression analyses on a subsample of injury incidents excluding locations in Baltimore City. We also compared the periods before and after the unrest using subsamples for March through January 2015 (before) and May through December 2015 (after).
RESULTS
Latent Class Analysis
The study population included 16,082 trauma patients. Most patients were White (52.4%), male (65.8%), younger than age 65 (77.0%), and had blunt injuries (81.9%). Detailed descriptions of the study population and the geographic distribution of injury incidence were previously published.14 The distribution of model fit statistics and quantile-quantile plots of the residuals indicated that the eight class model was the best fitting, most parsimonious model.28 Conditional probabilities and prevalence of class membership are available as online supplemental material (Table S1). The distribution of fatal and nonfatal injuries by location class is illustrated in Figure 1. The geographic distribution of location classes is illustrated in Figure 2. Class 1 locations (3.0%) were labeled as rural, and were most likely at non-residential locations more than 10-miles from a Level III trauma center, in communities with per capita income less than $30,000 and median age greater than 40 years. Class 2 locations (11.2%) were labeled as young suburbs, and were most likely between 5 and 15-miles from a Level I/II trauma center, in communities with per capita income greater than $30,000 and median age between 30 and 40 years. Class 3 locations (9.7%) were labeled as aging suburbs, with locations most likely more than 15-miles from a Level I/II trauma center, in communities with per capita income greater than $40,000 and median age greater than 40 years. Class 4 (10.4%) locations were labeled as low income urban core, and locations were most likely within 5-miles of a Level I/II trauma center, in communities with per capita income less than $20,000 and median age between 30 and 40 years. Class 5 (16.8%) locations, labeled as exurban, were most likely more than 5-miles from the nearest trauma center regardless of designation, and in communities with per capita income between $30,000 and $40,000, and median age greater than 40 years. Class 6 (15.7%) locations were labeled as high income urban core, and locations were most likely within 5-miles of a publicly-owned Level I/II trauma center, and in communities with per capita income between $20,000 and $30,000. Class 7 locations (18.4%) were labeled as urban fringe, and were most likely within 5-miles of a privately-owned Level I/II trauma center, and in communities with per capita income between $20,000 and $30,000. Finally, Class 8 locations (14.8%) were labeled as inner suburbs, and were most likely within 10-miles of a Level I/II trauma center, and in communities with per capita income greater than $30,000.
Figure 1:
Distribution of Fatal and Nonfatal Injuries by Location Class
Figure 2:
Geographic Distribution of Latent Classes of Injury Events in Maryland, 2015
Individual Characteristics
The estimated distributions of individual patient demographic, health, and injury characteristics by location class are presented in Table 1. The case fatality rate was lowest for inner suburb locations (7.0%), and highest for rural (11.6%) and low income urban core (12.9%) locations. Patients injured at low income and high income urban core locations were younger (63.7% and 57.8% younger than 44 years, respectively), more likely to be male (75.3% and 71.7%, respectively), and more likely to be African American (75.7% and 57.5%, respectively) compared to other classes, while exurban, rural, and aging suburb locations had a higher proportion of White/Non-Hispanic injury patients (79.7%, 74.8%, and 65.5%, respectively). Injuries at aging suburb and exurban locations were typically blunt (91.4% and 90.8%, respectively) while low income urban core, high income urban core, and urban fringe locations had the highest incidence of penetrating injury (35.0%, 17.8%, and 13.6%, respectively). People injured at low income urban core locations were least likely to have diagnosed comorbidities (3.0%), while those injured at rural and urban fringe locations were most likely to have comorbidities (9.3% and 6.9%, respectively). Private insurance coverage was highest for rural patients (56.3%), and lowest for low income urban core patients (22.7%). Public insurance and uninsurance rates were highest for high income urban core (44.5% public insurance, 26.2% uninsured) and low income urban core locations (49.7% public insurance, 27.6% uninsured). Mean prehospital time was shortest for low and high income urban core locations (45.1 minutes and 59.2 minutes, respectively), and longest for aging suburb (78.3 minutes) and exurban locations (91.8 minutes).
Table 1:
Estimateda Distribution of Individual Characteristics by Location Class of Injury Incidents in Maryland, 2015
| Rural | Young Suburb | Aging Suburb | Low Income Urban |
Exurban | High Income Urban |
Urban Fringe | Inner Suburbs | |
|---|---|---|---|---|---|---|---|---|
| % (95% CI) | % (95% CI) | % (95% CI) | % (95% CI) | % (95% CI) | % (95% CI) | % (95% CI) | % (95% CI) | |
| Died | ||||||||
| No | 88.4 (85.5, 91.3) | 91.1 (89.4, 92.9) | 92.0 (90.8, 93.2) | 87.1 (85.4, 88.8) | 92.0 (90.8, 93.2) | 92.4 (91.3, 93.4) | 92.5 (91.6, 93.5) | 93.0 (91.8, 94.3) |
| Yes | 11.6 (8.7, 14.5) | 8.9 (7.1, 10.7) | 8.0 (6.8, 9.2) | 12.9 (11.3, 14.6) | 8.0 (6.8, 9.2) | 7.7 (6.6, 8.7) | 7.5 (6.6, 8.4) | 7.0 (5.7, 8.2) |
| Age | ||||||||
| 18-24 | 16.4 (13.0, 19.9) | 16.6 (14.6, 18.5) | 12.2 (10.7, 13.7) | 19.7 (17.7, 21.7) | 14.0 (12.4, 15.6) | 17.0 (15.5, 18.5) | 15.3 (14.0, 16.6) | 15.2 (13.5, 16.9) |
| 25-34 | 19.7 (15.9, 23.6) | 20.1 (17.4, 22.8) | 15.3 (13.6, 16.9) | 29.0 (26.7, 31.3) | 16.1 (14.4, 17.8) | 25.1 (23.4, 26.8) | 20.3 (18.9, 21.7) | 18.7 16.3, 21.2) |
| 35-44 | 13.1 (10.1, 16.1) | 15.3 (13.6, 17.0) | 11.9 (10.5, 13.3) | 15.0 (13.2, 16.8) | 10.6 (9.2, 11.9) | 15.7 (14.1, 17.2) | 12.2 (10.9, 13.4) | 13.0 (11.4, 14.5) |
| 45-54 | 13.8 (10.7, 17.0) | 14.6 (12.7, 16.4) | 14.1 (12.5, 15.7) | 14.7 (12.9, 16.5) | 13.8 (12.2, 15.4) | 14.4 (12.9, 15.87) | 15.7 (14.4, 17.1) | 13.0 (11.5, 14.6) |
| 55-64 | 14.1 (10.5, 17.8) | 13.2 (11.4, 14.9) | 14.9 (13.3, 16.5) | 11.5 (9.9, 13.1) | 15.3 (13.6, 16.9) | 12.8 (11.5, 14.2) | 13.5 (12.3, 14.7) | 11.4 (9.7, 13.0) |
| 65-74 | 10.8 (7.7, 13.9) | 7.8 (6.4, 9.2) | 12.4 (10.9, 13.9) | 4.7 (3.6, 5.8) | 10.4 (8.9, 11.8) | 6.8 (5.8, 7.8) | 8.2 (7.2, 9.2) | 8.8 (7.3, 10.4) |
| 75+ | 12.0 (8.7, 15.3) | 12.5 (10.7, 14.3) | 19.3 (17.5, 21.1) | 5.4 (4.2, 6.6) | 19.9 (18.2, 21.7) | 8.2 (7.1, 9.3) | 14.9 (13.6, 16.2) | 19.9 (17.4, 22.5) |
| Sex | ||||||||
| Male | 63.7 (59.3, 68.2) | 65.3 (62.8, 67.7) | 62.4 (60.2, 64.5) | 75.3 (73.1, 77.5) | 60.1 (57.9, 62.4) | 71.7 (69.9, 73.4) | 64.9 (63.1, 66.6) | 63.1 (60.5, 65.8) |
| Female | 36.3 (31.8, 40.7) | 34.8 (32.3, 37.2) | 37.6 (35.5, 39.8) | 24.7 (22.5, 26.9) | 39.9 (37.6, 42.1) | 28.3 (26.6, 30.1) | 35.1 (33.4, 36.9) | 36.9 (34.2, 39.5) |
| Race/Ethnicity | ||||||||
| White | 74.8 (70.9, 78.7) | 45.1 (39.9, 50.2) | 65.5 (63.0, 68.0) | 18.8 (16.8, 20.8) | 79.7 (77.5, 81.9) | 27.4 (25.6, 29.3) | 60.0 (58.1, 61.8) | 56.3 (50.9, 61.8) |
| African American | 18.4 (14.8, 22.0) | 37.1 (33.5, 40.7) | 20.1 (18.2, 22.0) | 75.7 (73.5, 77.9) | 12.7 (11.0, 14.5) | 57.5 (55.5, 59.5) | 32.1 (30.3, 33.9) | 31.3 (27.8, 34.7) |
| Hispanic | 4.0 (2.2, 5.9) | 9.8 (7.3, 12.2) | 7.0 (5.7, 8.2) | 2.4 (1.6, 3.2) | 3.0 (2.2, 3.8) | 9.7 (8.5, 10.9) | 4.2 (3.4, 4.9) | 6.5 (4.2, 8.7) |
| Other | 2.8 (1.1, 4.5) | 8.1 (6.6, 9.5) | 7.5 (6.2, 8.7) | 3.1 (2.3, 4.0) | 4.6 (3.4, 5.7) | 5.4 (4.4, 6.4) | 3.8 (3.0, 4.5) | 6.0 (4.7, 7.2) |
| Injury Severityb | ||||||||
| Mild | 89.9 (87.1, 92.6) | 90.7 (88.9, 92.5) | 90.7 (89.2, 92.2) | 89.6 (88.0, 91.2) | 90.8 (89.1, 92.6) | 91.8 (90.6, 93.0) | 90.5 (89.0, 92.0) | 91.3 (89.7, 93.0) |
| Moderate | 4.2 (2.3, 6.1) | 4.2 (3.1, 5.2) | 4.4 (3.4, 5.4) | 3.0 (1.5, 4.5) | 3.6 (2.6, 4.5) | 3.4 (2.6, 4.1) | 3.7 (2.9, 4.5) | 3.2 (2.3, 4.1) |
| Severe | 4.9 (2.8, 6.9) | 4.4 (3.3, 5.6) | 4.2 (3.1, 5.2) | 4.9 (3.8, 6.1) | 4.6 (3.5, 5.8) | 3.7 (2.6, 4.7) | 4.6 (3.5, 5.7) | 4.6 (3.4, 5.8) |
| Critical | 1.1 (0.1, 2.0) | 0.8 (0.2, 1.3) | 0.8 (0.4, 1.2) | 2.5 (1.0, 3.9) | 1.0 (0.0, 2.3) | 1.2 (0.7, 1.7) | 1.2 (0.8, 1.7) | 0.9 (0.3, 1.4) |
| Injury Mechanism | ||||||||
| Blunt | 86.9 (83.9, 90.0) | 86.8 (84.9, 88.6) | 91.4 (90.0, 92.8) | 59.6 (57.1, 62.0) | 90.8 (89.4, 92.1) | 76.9 (75.2, 78.5) | 80.1 (78.6, 81.6) | 84.1 (82.3, 85.8) |
| Penetrating | 7.2 (4.8, 9.7) | 7.4 (5.9, 9.0) | 4.9 (3.8, 6.0) | 35.0 (32.6, 37.4) | 5.1 (4.1, 6.1) | 17.8 (16.3, 19.3) | 13.6 (12.4, 14.9) | 10.7 (9.2, 12.1) |
| Both | 1.5 (0.4, 2.6) | 2.4 (1.6, 3.1) | 1.3 (0.8, 1.9) | 3.2 (2.3, 4.1) | 1.7 (1.1, 2.3) | 2.3 (1.7, 2.8) | 3.5 (2.9, 4.1) | 2.5 (1.8, 3.2) |
| Other | 4.3 (2.5, 6.2) | 3.4 (2.5, 4.4) | 2.4 (1.7, 3.1) | 2.3 (1.5, 3.0) | 2.5 (1.8, 3.2) | 3.1 (2.4, 3.8) | 2.8 (2.2, 3.4) | 2.8 (2.0, 3.5) |
| Charlson Index | ||||||||
| 0 | 90.7 (87.9, 93.5) | 95.2 (94.1, 96.4) | 95.1 (94.1, 96.2) | 97.1 (96.2, 97.9) | 93.3 (92.2, 94.5) | 95.8 (95.0, 93.1) | 93.1 (92.2, 94.0) | 95.2 (94.2, 96.2) |
| 1-4 | 9.1 (6.3, 11.8) | 4.3 (3.2, 5.4) | 4.3 (3.3, 5.3) | 2.6 (1.8, 3.4) | 5.9 (4.8, 7.0) | 3.7 (3.0, 4.5) | 6.7 (5.8, 7.6) | 4.7 (3.7, 5.6) |
| 5+ | 0.2 (0.0, 0.7) | 0.5 (0.1, 0.8) | 0.6 (0.2, 0.9) | 0.4 (0.1, 0.7) | 0.8 (0.4, 1.2) | 0.5 (0.2, 0.8) | 0.2 (0.0, 0.4) | 0.2 (0.0, 0.4) |
| Insurance Status | ||||||||
| Private | 56.3 (51.4, 61.1) | 44.1 (41.4, 46.7) | 48.8 (46.3, 51.3) | 22.7 (20.4, 25.0) | 45.9 (43.1, 48.8) | 29.4 (27.5, 31.2) | 38.3 (36.4, 40.3) | 43.3 (41.1, 45.6) |
| Public | 33.3 (28.3, 38.4) | 34.9 (31.8, 38.0) | 37.2 (34.2, 40.2) | 49.7 (47.0, 52.5) | 41.5 (39.0, 44.0) | 44.5 (42.3, 46.6) | 40.5 (38.7, 42.4) | 36.2 (33.9, 38.6) |
| None | 10.4 (7.0, 13.8) | 21.0 (18.5, 23.5) | 14.0 (12.2, 15.9) | 27.6 (24.9, 30.2) | 12.6 (11.0, 14.3) | 26.2 (24.2, 28.2) | 21.1 (19.6, 22.7) | 20.5 (18.3, 22.7) |
| Prehospital time | 68.7 (62.0, 75.4) | 66.9 (62.4, 71.4) | 78.3 (73.3, 83.4) | 45.1 (42.6, 47.5) | 91.8 (86.0, 97.6) | 59.2 (56.1, 62.2) | 54.1 (51.7, 56.6) | 61.5 (58.7, 64.4) |
Estimated distributions based on multiple imputation
Injury severity based on Injury Severity Score (ISS) and unweighted Revised Trauma Score (RTS). RTS was used only when ISS was not available. Mild injury includes ISS 1-9 and RTS 12. Moderate injury includes ISS 10-15 and RTS 11. Severe injury includes ISS 16-24 and RTS 4-10. Critical injury includes ISS ≥ 25 and RTS ≤ 3
Mortality Effects
Figure 3 illustrates the estimated odds ratios (OR) and confidence intervals (CI) from logistic regression of latent class membership on mortality, with and without adjustment for individual patient characteristics. The inner suburb class had the lowest odds of death and was used as the reference for both regression models. In the unadjusted model, odds of death for patients injured at low income urban core locations were nearly twice that of patients injured at inner suburb locations (OR = 1.98, 95% CI: 1.56, 2.51) and odds of death for patients injured at rural locations were 75% greater than those injured at inner suburb locations (OR = 1.75, 95% CI: 1.24, 2.48). After adjustment for individual age, sex, race and/or ethnicity, insurance status, Charlson Comorbidity Index, injury mechanism, prehospital time, and mechanism/time interaction, rural location odds of death were nearly twice that of inner suburb locations (OR = 1.98, 95% CI: 1.36, 2.88), while odds of death were 57% higher for young suburb locations (OR = 1.57, 95% CI: 1.14, 2.17), 36% higher for aging suburb locations (OR = 1.36, 95% CI: 1.04, 1.78), and 38% higher for low income urban core locations (OR = 1.38, 95% CI: 1.04, 1.83).
Figure 3: Mortality Odds Ratios by Location Class for Injury Events in Maryland, 2015.
Inner Suburb is the reference class for odds ratios. Adjusted model included age, sex, race/ethnicity, insurance status, Charlson Comorbidity Index, severity, mechanism, prehospital time, and mechanism/time interaction.
Sensitivity Analyses
Results of the sensitivity analyses are presented in Table 2. The most prevalent location classes observed in Baltimore City were low income urban core (36.3%), high income urban core (22.9%), urban fringe (26.7%), and inner suburb (10.2%). When Baltimore City locations were excluded from the analyses, the unadjusted and adjusted odds ratios for low income urban core location, compared to inner suburb, increased to 7.85 (95% CI: 4.94, 12.48) and 4.19 (95% CI: 2.20, 7.99), respectively. When the analyses were limited to incidents occurring between January and March, the unadjusted effect of low income urban core class was comparable to the effect observed in the total sample (OR = 2.33, 95% CI: 1.13, 4.80). The adjusted effect of low income urban class during the January-March period was comparable to the effect observed in the total sample (OR = 1.33, 95% CI: 0.45, 3.89), but was not statistically significant, likely due to sample size in the three-month period. For the May-December period, the effect of low income urban core was comparable to the effects observed for the total sample (unadjusted OR = 1.91, 95% CI: 1.48, 2.47; adjusted OR = 1.47, 95% CI: 1.03, 2.09).
Table 2:
Sensitivity Analyses for Odds of Injury Mortality in Maryland, 2015
| Unadjusted | Adjusted | |||||
|---|---|---|---|---|---|---|
| Odds Ratio | 95% CI | p | Odds Ratio | 95% CI | p | |
| Excluding Baltimore City | ||||||
| Low Income Urban | 7.85 | 4.94, 12.48 | < 0.001 | 4.19 | 2.20, 7.99 | < 0.001 |
| High Income Urban | 1.05 | 0.79, 1.38 | 0.742 | 1.07 | 0.74, 1.55 | 0.711 |
| Urban Fringe | 0.99 | 0.76, 1.30 | 0.978 | 0.88 | 0.88, 0.52 | 0.657 |
| January-March Onlya | ||||||
| Low Income Urban | 2.33 | 1.13, 4.80 | 0.031 | 1.33 | 0.45, 3.89 | 0.613 |
| High Income Urban | 0.97 | 0.49, 1.92 | 0.926 | 0.95 | 0.36, 2.49 | 0.919 |
| Urban Fringe | 0.96 | 0.49, 1.87 | 0.897 | 0.81 | 0.33, 1.02 | 0.654 |
| May-December Onlya | ||||||
| Low Income Urban | 1.91 | 1.48, 2.47 | < 0.001 | 1.47 | 1.03, 2.09 | 0.044 |
| High Income Urban | 1.10 | 0.84, 1.44 | 0.493 | 1.19 | 0.83, 1.72 | 0.355 |
| Urban Fringe | 1.04 | 0.80, 1.34 | 0.793 | 0.98 | 0.56, 1.69 | 0.930 |
Effect of urban locations examined for January-March and May-December to assess temporal impact of April 2015 civil unrest in Baltimore City.
DISCUSSION
The results of this study suggest that regions with high risk of injury mortality can be identified based on clustering of features of the built and social environments at the injury scene, including trauma center distance and characteristics, land use, community income, and population age. While prior studies examined the built environment and community-level social measures as distinct determinants of injury mortality,3,9,12-14,21 this is the first study to identify patterns of clustering based on incident location characteristics, and to examine location classes as determinants of mortality. Eight distinct classes of injury location were identified in Maryland with variable case fatality rates and distributions of demographic, health, and injury characteristics. Inner suburb locations, marked by high ZCTA income and proximity to Level I/II trauma centers, had the lowest estimated odds of death in both the adjusted and unadjusted regression models while rural, suburban, and low income urban locations had the highest odds of death. The urban and rural location classes identified in this study are consistent with existing classification of geography in the US. The number of suburban classes identified, combined with variation in risk factors and outcomes across these classes, indicates greater geographic variation in injury risk than generally assumed. In addition to identification of previously unclassified location types, the methods used in this study provide an empirical basis for resource allocation, and serve as an important check against implicit bias in our assessment of community-level risk of injury and need for intervention.
Rural, young suburban, aging suburban, and exurban locations all appeared to have some distance-based barriers to trauma center services. Rural locations had the second highest case fatality rate, and the highest odds of death after adjustment for individual characteristics. Estimated odds of death for rural locations increased following adjustment, suggesting that individual characteristics mask some of the risk associated with the location class. This is most likely explained by the low proportion of rural incidents with penetrating injuries, as well as the relatively large proportion of patients with private insurance, despite the low median income associated with class membership. Odds of death for young suburb and aging suburb locations were not significantly different than those for inner suburbs in the unadjusted model, but odds of death for both increased considerably after adjustment for individual characteristics. The effects of both suburb classes were likely masked by low incidence of penetrating injury and high rates of private insurance enrollment, while higher proportions of female and White patients further masked the effect of aging suburb locations. Odds of death were similar for exurban and inner suburb locations despite longer trauma center distances and potential barriers to trauma center services at exurban locations. Individuals injured at exurban locations were more likely to be White, have private insurance, and have blunt injuries, which are all associated with reduced injury mortality risk,29,30 and may mitigate the effect of distance and prehospital time.
Trauma center distances were shortest for urban and inner suburb locations. High income urban core and urban fringe locations were among the lowest risk classes identified based on case fatality rate and adjusted estimates of mortality odds, and were not significantly different than inner suburb locations in terms of mortality. Compared to inner suburb locations, high income urban core and urban fringe locations had higher proportions of critical and penetrating injuries; however, the mortality risk associated with the injury characteristics appears to be mitigated by shorter prehospital times, consistent with the relationship between injury mechanism, prehospital time, and mortality demonstrated in the literature.14,29,30 Low income urban core locations had the highest case fatality rate of all location classes, but the estimated odds of death decreased substantially after adjustment for individual characteristics, suggesting high proportions of individual characteristics associated with increased mortality (e.g. African American race, male sex, penetrating injury29-32) confounded the relationship between location class and odds of death. While adjustment for individual characteristics significantly reduced the effect of low income urban core locations, it was the only urban location class with estimated odds of death significantly greater than inner suburb locations.
Based on the observed patterns in mortality and individual characteristics, location class assignments may be useful when tailoring primary prevention and injury response interventions. For example, odds of death at low income urban core locations were elevated despite proximity of Level I/II trauma centers. Low income urban core locations had high incidence of penetrating injury, suggesting low income urban core locations should be priority targets for violence prevention interventions, community-based training for bystander first aid (such as the American College of Surgeons Stop the Bleed campaign33), and enhanced EMS efforts tailored to treatment of penetrating injuries.
Rural locations were characterized primarily by lack of access to trauma care, with locations in the class generally more than 10-miles from a Level III trauma center. While individuals injured at rural locations had low incidence of penetrating injury and high rates of private insurance coverage, odds of death for rural injury incidents were among the highest observed, potentially due to extended EMS response times.16 This suggests that rural locations would benefit from primary prevention interventions to reduce blunt injuries, as well as bystander first aid training and EMS efforts to support patients with blunt injuries during prolonged prehospital intervals. Trauma centers serving these communities should work closely with prehospital providers and critical access hospitals to implement injury prevention programs and EMS care that respond to the unique needs of rural populations. These concepts align with suggestions from the National Academies of Medicine report “A National Trauma Care System: Integrating Military and Civilian Trauma Systems to Achieve Zero Preventable Deaths After Injury,” which emphasizes the importance of the trauma care continuum, including the high quality prehospital care.34
Limitations
The unique structure of trauma system services35 and the geographic distribution of social and environmental factors related to injury mortality in Maryland limit the generalizability of this study to other settings; however, it may be possible to validate these findings for other states and regions using publicly available data. With limits of generalizability in mind, the results of this study are useful as a starting point for inquiry into the use of spatially defined data as predictors of injury mortality at the population level.
There are concerns for internal and external validity of this study due to historic events during the study period. The study population included injury patients from Baltimore City during and after a period of civil unrest in the spring of 2015, and a subsequent rise in homicide incidence.36 There is considerable overlap between the low income urban core class identified in this analysis and the neighborhoods most impacted by the unrest and increased homicide rate. It is possible that safety concerns for first responders, residual unrest, or other unmeasured factors impacted the delivery of care and injury outcomes in these areas. Low income urban core locations were observed in regions of the state not impacted by civil unrest, and the effect of low income urban core location increased when Baltimore City locations were excluded from analyses, indicating that historic events did not increase the mortality effect associated with low income urban core locations, mitigating concerns regarding internal validity. The unadjusted and adjusted effects of low income urban core were similar for the months before and after the period of civil unrest, which suggests the events of April 2015 did not significantly impact geographic patterns of injury outcomes in the state, despite changes in injury incidence. Homicide incidence rates in Baltimore City have remained elevated in the months following the 2015 unrest,36 supporting the continued relevance of these results for short term practice and policy, despite potential limitations for long term generalizability of the mortality risk attributed to low income urban core locations.
Finally, this study was limited to injury patients transported to trauma centers by EMS, did not include patients transported by private vehicle, those transported to non-trauma centers without transfer to a trauma center, or those who died at the injury scene before EMS arrival. Private vehicle transport is most common among patients with penetrating trauma injured at urban locations,37 while injury deaths at non-trauma centers or prior to EMS arrival at the incident scene are most likely to occur among patients with critical injuries and/or significant barriers to trauma care. Exclusion of these patients may attenuate our results towards the null.
Conclusion
This study indicates that clusters of spatially defined risk factors for injury mortality are identifiable using latent class analyses, and that the population injury patients varies across location classes. Regression analyses suggest the protective benefit of some location classes may compensate for the effect of individual patient characteristics, while the same individual patient characteristics may mask the effect of location in other regions. The results of this study may be useful for planning and implementation of prevention and treatment efforts, including targeted efforts to reduce injury incidence, and identification of regions in need of additional prehospital or trauma center resources.
Public health practitioners at the state and local levels may be able to supplement the models used in this study with additional data not available to researchers to support nuanced tailoring of prevention and treatment efforts. Researchers should attempt to replicate this study in other regions, both to enhance our understanding of geographic concentration of injury risk factors, and to identify additional environmental and social factors that may improve our ability to identify regions with elevated injury mortality risk. Policymakers and practitioners should consider community level measures of income and age when making decisions regarding the allocation and designation of trauma centers.
Supplementary Material
Acknowledgements
This study was supported by an Agency for Healthcare Research Quality training grant (T32HS000029) and by the Johns Hopkins Center for Injury Research and Policy (William Haddon Jr. Fellowship).
We thank Dr. William Thompson, John New, Shanae Williams, Tefferra Alemayehu, and the Maryland Institute for Emergency Medical Services Systems for their support of the project.
Conflicts of Interest and Sources of Funding:
M. P. Jarman’s work on this study was funded by a National Research Service Award from the Agency for Healthcare Research Quality (T32HS000029) and by the Johns Hopkins Center for Injury Research and Policy (William Haddon Jr. Fellowship). She is currently supported by a research grant from the Center for Orthopaedic Trauma Advancement ("Treatment of injury in the United States: Projections for an ideal system of trauma care").
E. R. Haut is the primary investigator of a grant from the Agency for Healthcare Research and Quality (AHRQ) (1R01HS024547-01, “Individualized Performance Feedback on Venous Thromboembolism Prevention Practice”). He is also a co-investigator for grants from the National Institutes of Health National Heart, Lung, and Blood Institute (NIH-NHLBI) (1R21HL129028-01A1, “Analysis of the impact of missed doses of venous thromboembolism prophylaxis”) and The Henry M. Jackson Foundation, Uniformed Services University of the Health Sciences (HU0001-14-0038, “Validation of Training for Cricothyroidotomy”). Dr. Haut is the primary investigator of two contracts with The Patient-Centered Outcomes Research Institute (PCORI) (CE-12-11-4489, “Preventing Venous Thromboembolism: Empowering Patients and Enabling Patient-Centered Care via Health Information Technology” & DI-1603-34596, “Preventing Venous Thromboembolism (VTE): Engaging Patients to Reduce Preventable Harm from Missed/Refused Doses of VTE Prophylaxis”), and co-investigator for the PCORI contract titled “A Randomized Pragmatic Trial Comparing the Complications and Safety of Blood Clot Prevention Medicines Used in Orthopaedic Trauma Patients” (PCS-1511-32745). Dr. Haut receives royalties from Lippincott, Williams, Wilkins for a book - "Avoiding Common ICU Errors."
For remaining authors, no conflicts were declared.
Footnotes
Conference Presentations:
Presented at the 31st Annual Meeting of the Eastern Association for the Surgery of Trauma, January 9-13, 2018 in Orlando, FL
Contributor Information
Molly P. Jarman, Department of Health Policy and Management, Johns Hopkins Bloomberg School of Public Health Center for Surgery and Public Health, Brigham and Women’s Hospital.
Elliott R. Haut, Department of Surgery, Johns Hopkins School of Medicine.
Frank C. Curriero, Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health.
Renan C. Castillo, Department of Health Policy and Management, Johns Hopkins Bloomberg School of Public Health.
References
- 1.Eastman AB, Mackenzie EJ, Nathens AB. Sustaining a coordinated, regional approach to trauma and emergency care is critical to patient health care needs. Health affairs (Project Hope). Dec 2013;32(12):2091–2098. [DOI] [PubMed] [Google Scholar]
- 2.National Academy of Sciences, Division of Medical Sciences Committee on Trauma and Committee on Shock. Accidental Death and Disability: The Neglected Disease of Modern Society. Washington, DC: National Academies Press; 1966. [PubMed] [Google Scholar]
- 3.Branas CC, MacKenzie EJ, Williams JC, Schwab CW, Teter HM, Flanigan MC, Blatt AJ, Revelle CS. Access to trauma centers in the United States. JAMA. Jun 1 2005;293(21):2626–2633. [DOI] [PubMed] [Google Scholar]
- 4.Carr BG, Branas CC, Metlay JP, Sullivan AF, Camargo CA Jr. Access to emergency care in the United States. Ann Emerg Med. Aug 2009;54(2):261–269. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Hsia R, Shen YC. Possible geographical barriers to trauma center access for vulnerable patients in the United States: an analysis of urban and rural communities. Arch Surg. Jan 2011;146(1):46–52. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Hsia RY, Shen YC. Rising closures of hospital trauma centers disproportionately burden vulnerable populations. Health Affairs (Project Hope). Oct 2011;30(10):1912–1920. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Haider AH, Weygandt PL, Bentley JM, Monn MF, Rehman KA, Zarzaur BL, Crandall ML, Cornwell EE, Cooper LA. Disparities in trauma care and outcomes in the United States: a systematic review and meta-analysis. J Trauma Acute Care Surg. May 2013;74(5):1195–1205. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Epstein SK, Burstein JL, Case RB, Gardner AF, Herman SH, Hirshon JM, Jermyn JW, McKay MP, Mitchiner JC, Sullivan WP, et al. The National Report Card on the State of Emergency Medicine: evaluating the emergency care environment state by state 2009 edition. Ann Emerg Med. Jan 2009;53(1):4–148. [DOI] [PubMed] [Google Scholar]
- 9.Peek-Asa C, Zwerling C, Stallones L. Acute traumatic injuries in rural populations. Am J Public Health. Oct 2004;94(10):1689–1693. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Coben JH, Tiesman HM, Bossarte RM, Furbee PM. Rural-urban differences in injury hospitalizations in the U.S., 2004. Am J Prev Med. Jan 2009;36(1):49–55. [DOI] [PubMed] [Google Scholar]
- 11.Esposito TJ, Maier RV, Rivara FP, et al. The impact of variation in trauma care times: urban versus rural. Prehosp Disaster Med. 1995 Jul-Sep 1995;10(3):161–166; discussion 166-167. [DOI] [PubMed] [Google Scholar]
- 12.Cubbin C, LeClere FB, Smith GS. Socioeconomic status and injury mortality: individual and neighbourhood determinants. J Epidemiol Community Health. Jul 2000;54(7):517–524. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Winkleby MA, Cubbin C. Influence of individual and neighbourhood socioeconomic status on mortality among black, Mexican-American, and white women and men in the United States. J Epidemiol Community Health. Jun 2003;57(6):444–452. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Jarman MP, Curriero FC, Pollack KM, Haut EH, Castillo RC. Associations of distance to trauma care, community income, and neighborhood median age with rates of injury mortality. JAMA Surgery. 2017; In Press. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Brown JB, Rosengart MR, Billiar TR, Peitzman AB, Sperry JL. Distance matters: Effect of geographic trauma system resource organization on fatal motor vehicle collisions. J Trauma Acute Care Surg. Jul 2017;83(1):111–118. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Mell HK, Mumma SN, Hiestand B, Carr BG, Holland T, Stopyra J. Emergency Medical Services Response Times in Rural, Suburban, and Urban Areas. JAMA Surgery. Jul 19 2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Rotondo MF, Cribari C, Smith RS, American College of Surgeons Committee on Trauma. Resources for the Optimal Care of the Injured Patient: 2014. 2014; https://www.facs.org/~/media/files/quality%20programs/trauma/vrcresources.ashx. Accessed December 1, 2017. [Google Scholar]
- 18.Haas B, Doumouras AG, Gomez D, et al. Close to home: an analysis of the relationship between location of residence and location of injury. J Trauma Acute Care Surg. Apr 2015;78(4):860–865. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.United States Census Bureau. Census Data Mapper. 2017; https://www.census.gov/geo/maps-data/maps/datamapper.html. Accessed December 1, 2017.
- 20.Chang D, Cornwell EE 3rd, Phillips J, Baker D, Yonas M, Campbell K. Community characteristics and demographic information as determinants for a hospital-based injury prevention outreach program. Arch Surg. Dec 2003;138(12):1344–1346. [DOI] [PubMed] [Google Scholar]
- 21.Haider AH, Ong'uti S, Efron DT, Oyetunji TA, Crandall ML, Scott VK, Haut ER, Schneider EB, Powe NR, Cooper LA, et al. Association between hospitals caring for a disproportionately high percentage of minority trauma patients and increased mortality: a nationwide analysis of 434 hospitals. Arch Surg. Jan 2012;147(1):63–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Lanza ST, Dziak JL, Huang L, Wagner AT, Collins LM. LCA Stata Plugin Users' Guide (Version 1.2). University Park: The Methodology Center, Penn State. 2015. [Google Scholar]
- 23.Karen L Nylund TA, Bengt O Muthen. Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study. Struct Equ Modeling. 2007 2007;14(4):535–569. [Google Scholar]
- 24.Wang C, Brown C, Bandeen-Roche K. Residual diagnostic for growth mixture models. J Am Stat Assoc. 2005;100(471). [Google Scholar]
- 25.Rubin DB. Multiple imputation for nonresponse in surveys. New York: ;: Wiley; 1987. [Google Scholar]
- 26.Chini D. American Communities Project. 2017; http://americancommunities.org/. Accessed December 1, 2017. [Google Scholar]
- 27.Jerrett M, Gale S, Kontgis C. Spatial modeling in environmental and public health research. Int J Environ Res Public Health. Apr 2010;7(4):1302–1329. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Jarman MP. Concentrated Trauma Risk: Social and Environmental Determinants of Injury Mortality. Baltimore, MD: Health Policy and Management, Johns Hopkins University; 2017. [Google Scholar]
- 29.McCoy CE, Menchine M, Sampson S, Anderson C, Kahn C. Emergency medical services out-of-hospital scene and transport times and their association with mortality in trauma patients presenting to an urban Level I trauma center. Ann Emerg Med. Feb 2013;61(2):167–174. [DOI] [PubMed] [Google Scholar]
- 30.Swaroop M, Straus DC, Agubuzu O, Esposito TJ, Schermer CR, Crandall ML. Pre-hospital transport times and survival for Hypotensive patients with penetrating thoracic trauma. J Emerg Trauma Shock. Jan 2013;6(1):16–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Sorenson SB. Gender disparities in injury mortality: consistent, persistent, and larger than you'd think. Am J Public Health. Dec 2011;101 Suppl 1:S353–358. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Mostafa G, Huynh T, Sing RF, Miles WS, Norton HJ, Thomason MH. Gender-related outcomes in trauma. J Trauma. Sep 2002;53(3):430–434; discussion 434-435. [DOI] [PubMed] [Google Scholar]
- 33.Butler FK. Stop the bleed. Strategies to enhance survival in active shooter and intentional mass casualty events. The Hartford Consensus. A major step forward in translating battlefield trauma care advances to the civilian sector. J Spec Oper Med. Winter 2015;15(4):133–135. [PubMed] [Google Scholar]
- 34.Berwick DM, Downey AS, Cornett E, National Academies of Sciences Engineering and Medicine (U.S.). Committee on Military Trauma Care's Learning Health System and Its Translation to the Civilian Sector. A National Trauma Care System: Integrating Military and Civilian Trauma Systems to Achieve Zero Preventable Deaths After Injury. Washington, DC: The National Academies Press; 2016. [PubMed] [Google Scholar]
- 35.Maryland Institute for Emergency Medical Services Systems. 2013-2014 Annual Report. Baltimore, MD. 2015. [Google Scholar]
- 36.George J. Shoot to kill. Baltimore Sun. September 30, 2016, 2016. [Google Scholar]
- 37.Wandling MW, Nathens AB, Shapiro MB, Haut ER. Association of Prehospital Mode of Transport With Mortality in Penetrating Trauma: A Trauma System-Level Assessment of Private Vehicle Transportation vs Ground Emergency Medical Services. JAMA Surgery. Sep 20 2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Butts JA, Roman CG, Bostwick L, Porter JR. Cure violence: a public health model to reduce gun violence. Annu Rev Public Health. Mar 18 2015;36:39–53. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.