Supplemental Digital Content 1: Validation of prediction model for volume of severely injured patients discharged from non-trauma centers (NBATS Criterion #4) and volume of severely injured patients discharged from Level I/II trauma centers (NBATS Criterion #6).
Methods
The volume of severely injured patients discharged from Level I/II trauma centers and from non-trauma centers was observed for 125 out of 324 Trauma Service Areas (TSAs) in the 48 contiguous states and the District of Columbia. Level I/II and non-trauma center patient volumes in unobserved TSAs were
predicted based on associations between patient volumes in observed TSAs, population characteristics derived from the United States Census Bureau American Community Survey and TSA hospital resources derived from the American Hospital Association Annual Survey.
For NBATS Criterion #4, the volume of severely injured patients was modeled as the absolute number of non-trauma center discharges from hospitals other than Level I, II, or III centers in each region. For NBATS Criterion #6, the volume of severely injured patients was modeled as the departure (positive or negative) from the expected patient volume defined in the NBATS rubric. Per the NBATS rubric, the expected number of Level I/II discharges was calculated as 500 x the number of Level I and II centers in each region. Departure from the expected volume was then calculated by subtracting the absolute number of severely injured patients discharged from Level I/II centers in the region from the expected volume.
Multivariable linear regression models predicting patient volumes for both criteria were developed with a training set that included a random sample of 1/3 of the observed TSAs, and variables were selected using a combination of forward and backward step-wise regression with an inclusion threshold at p ≤ 0.10.
Variables were removed from consideration if they were rejected by both the forward and backward approach. This process was repeated until a single concordant set of include variables was reached.
Variables considered in the model development process were:
- TSA median trauma center travel time
- Number of inpatient beds at TSA hospitals with emergency departments - Number of Level I/II trauma centers in the TSA
- Number of Level III trauma centers in the TSA
- Proportion of the TSA population who routinely commute by motor vehicle - Proportion of TSA population without health insurance
- Proportion of TSA population at/below federal poverty level - Proportion of TSA population with race other than White - Proportion of TSA population who were male
- Median household income for TSA population - Median age of TSA population
- Total residential population within TSA
Once the final model was selected, it was used to predict Level I/II and non-trauma center patient volumes for the 2/3 of observed TSAs not included in the training sample. We then used predicted volumes to determine the predicted NBATS score/group for each criterion, and validated the ability of the model to predict NBATS criteria scores by examining the Spearman Rank Correlation between observed and predicted criterion scores as well as visual examination of scatter plots for the observed versus predicted patient volumes.
Results: NBATS Criterion #4
Variables selected for the final model predicting NBATS Criterion #4 (volume of severely injured patients at non-trauma centers) included inpatient beds at hospitals with emergency departments, number of Level I/II and Level III trauma centers, total TSA population, proportion of the TSA population without health insurance and proportion of TSA population at/below poverty. Coefficients and confidence intervals for the associations between these variables and the number of severely injured patients
discharge from non-trauma centers are presented in Table A1. Predicted NBATS criterion scores were highly correlated with observed criterion scores (r = .80, p < 0.001).
Table A1: Parameters from predictive model for non-trauma center patient volume
Predictor Variable β (95% CI) P
Emergency department beds 0.07 (0.03, 0.10) 0.001 Level I/II Centers -50.7 (-76.3, -25.1) <0.001 Level III Centers -15.7 (-22.2, -9.2) <0.001
% TSA uninsured 785.4 (193.9, 1376.8) 0.010
% TSA at/below poverty -575.1 (-1154.4, 4.2) 0.052 TSA population 0.0002 (0.0001, 0.0002) <0.001
Figure A1: Scatter plot of observed and predicted patient volume at non-trauma centers
Results: NBATS Criterion #6
Variables selected for the final predictive model included inpatient beds at hospitals with emergency departments, number of Level I/II and Level III trauma centers, total TSA population, and proportion of TSA population at/below poverty. Coefficients and confidence intervals for the associations between these variables and the number of severely injured patients discharged from Level I/II trauma centers are presented in Table A2. Predicted NBATS criterion scores were highly correlated with observed criterion scores (r = 0.84, p < 0.001).
Table A2: Parameters from predictive model for Level I/II trauma center patient volume
Predictor Variable β (95% CI) P Level I/II Centers -454.7 (-533.3, -376.1) <0.001
Level III Centers -18.5 (-38.1, 1.1) 0.064
TSA population 0.0004 (0.0002, 0.0005) <0.001
Figure A2: Scatter plot of observed and predicted patient volume at Level I/II centers