A population-based comparison of CIREN and NASS cases using similarity scoring

Abstract

The Crash Injury Research and Engineering Network (CIREN) provides significant details on injuries, and data on patient outcomes that is unavailable in the National Automotive Sampling System (NASS). However, CIREN cases are selected from specific Level I trauma centers with different inclusion criteria than those used for NASS, and the assertion that a given case is similar to the population of NASS cases is often made qualitatively. A robust, quantitative method is needed to compare CIREN to weighted NASS populations. This would greatly improve the usefulness and applicability of research conducted with data from the CIREN database. Our objective is to outline and demonstrate the utility of such a system to compare CIREN and NASS cases. This study applies the Mahalanobis distance metric methodology to determine similarity between CIREN and NASS/CDS cases. The Mahalanobis distance method is a multivariate technique for population comparison. Independent variables considered were total delta V, age, weight, height, maximum AIS, ISS, model year, gender, maximum intrusion, number of lower and upper extremity injuries, and number of head and chest injuries. The technique provides a unit-independent quantitative score which can be used to identify similarity of CIREN and NASS cases. Weighted NASS data and CIREN data were obtained for the years 2001-2005. NASS cases with Maximum AIS 3 resulted in a subset of 1,869 NASS cases, and 2,819 CIREN cases.

Figure 1

Schematic of ways to think about NASS/CDS and CIREN databases.

Figure 2

Simple example illustrating Mahalanobis distance. Male 1 and 2 are equidistant from the average in Cartesian coordinate space, but Male 2 is much further from the average and would have a higher DIST or less similarity.

Figure 3

Distribution of cases by Maximum AIS – NASS/CDS 2005 versus CIREN. NASS is composed of lower severity cases.

Figure 4

Distribution of MAIS3+ cases – NASS/CDS 2005 versus CIREN. NASS/CDS and CIREN are roughly comparable.

Figure 5

Normalized CIREN to NASS Means

Figure 6

Probability density versus Mahalanobis distance. The NASS and CIREN means are similarly distributed, with a greater proportion of CIREN cases clustered about the mean than for NASS cases.

Figure 7

Principal Component plots for NASS and CIREN data showing dataset overlap. The first 2 principal components explain 39.0% of the variability in the datasets. A subset (dark black line) of CIREN data has been chosen which lies within the area containing NASS data.

Figure 8

Mahalanobis Distance versus Number of Injuries to the Head, Chest, Upper, and Lower Extremities

Figure 9

Mahalanobis distance versus Total Delta V, Maximum Intrusion, and Model Year

Figure 10

Mahalanobis distance versus Maximum AIS and ISS

Figure 11

Mahalanobis Distance versus Age, Weight, and Height