On Scene Injury Severity Prediction (OSISP) Algorithm for Truck Occupants

Transcription

1 On Scene Injury Severity Prediction (OSISP) Algorithm for Truck Occupants Ruben Buendia 1,2,3 Stefan Candefjord 1,2,3 Helen Fagerlind 2,4 András Bálint 2,4 Claudia Wege 2,5 Bengt Arne Sjöqvist 1,2,3 2 SAFER Vehicle and Traffic Safety Centre at Chalmers, Sweden 4 Department of Applied Mechanics Chalmers University of Technology , Gothenburg, Sweden 1 Department of Signals and Systems Chalmers University of Technology , Gothenburg, Sweden 3 MedTech West Sahlgrenska University Hospital Röda Stråket 10 B , Gothenburg, Sweden 5 Volvo Group Trucks Technology (GTT) , Gothenburg, Sweden Corresponding author buendia@chalmers.se These authors contributed equally to this work. They are co-first authors. March 16, 2015 Abstract Objective: The aim of this study is to develop an On Scene Injury Severity Prediction (OSISP) algorithm for truck occupants using only Mechanisms of Injury that are feasible to assess at the scene of accident. The purpose of developing this algorithm is to use it as basis for a field triage tool used in traffic accidents involving trucks. In addition the model can be valuable for recognizing important factors for improving triage protocols used in Sweden and possibly in other countries with similar traffic environments and prehospital procedures. Methods: The scope is adult truck occupants involved in traffic accidents on Swedish public roads registered in the Swedish Traffic Accident Data Acquisition (STRADA) database for calendar years 2003 to STRADA contains information reported by the police, and medical data on injured road users treated at emergency hospitals. Using data from STRADA, two OSISP multivariate logistic regression models for deriving the probability of severe injury, here defined as having an injury severity score (ISS) > 15, were implemented for light and heavy trucks, i.e. trucks with weight up to 3500 kg and kg, respectively. A 10-fold cross validation procedure was used to estimate the performance of the OSISP algorithm in terms of the area under the receiver operating characteristic curve (AUC). Results: The rate of belt use was low, especially for heavy truck occupants. The OSISP models developed for light and heavy trucks achieved cross-validation AUC of 0.81 and 0.74, respectively. AUC values obtained as the models were evaluated on all data without cross-validation were 0.87 for both light and heavy trucks. The difference in the AUC values with and without use of cross-validation indicates 1

2 overfitting of the model, which may be a consequence of relatively small datasets. Belt use stands out as the most valuable predictor in both types of trucks; and type of accident and age are important predictors for light trucks. Conclusions: The OSISP models achieve good discriminating capability for light truck occupants and a reasonable performance for heavy truck occupants. The prediction accuracy may be increased by acquiring more data. Belt use was the strongest predictor of severe injury for both light and heavy truck occupants. There is a need for information campaigns and/or other means to encourage truck occupants to always wear seat belt. Keywords: triage; prehospital care; logistic regression; postcrash; traffic safety; trucks Introduction Improving the prehospital care process is fundamental for decreasing mortality and mitigate injury for trauma patients (Murad, Larsen, and Husum, 2012). Minimizing the delay to definitive treatment has been shown to decrease mortality substantially (Haas et al., 2010). A key to achieve this is to take early correct decision on optimal treatment and where to transport the patient; patients with severe injury should be taken to a trauma center, having the expertise to treat major injury (Haas et al., 2010; MacKenzie et al., 2006). The prehospital personnel s most important decision support is the triage protocol. It facilitates identifying patients with severe injury, while using health care resources efficiently by recognizing patients not likely to be in need of urgent and specialized care. Schoell et al. (2014) stated that improvements in triage accuracy is nowadays the most promising research field to continue reducing fatal and severe injuries for motor vehicle crashes. Developing and maintaining a highly accurate triage protocol requires detailed and updated knowledge about what type of traffic accidents are most dangerous. Statistics about mortality are rather well established but there are few publications studying how injury severity is linked to Mechanisms of Injury (MOI). In this paper the term MOI is used to refer to the characteristics of the crash and the involved vehicle(s) and occupant(s) such as Posted Speed Limit (PSL), whether the airbag was deployed, and the occupant s sex and age. A retrospective study including data of around a million trauma patients concluded that using physiologic and anatomic criteria alone results in undertriage, and strongly supported the use of MOI in triage systems for trauma (Brown et al., 2011). Hence, the importance of MOI lie in their potential to predict occult injuries to reduce undertriage. The aim of this study is to develop an On Scene Injury Severity Prediction (OSISP) algorithm for truck occupants using only MOI that are feasible to assess at the scene of accident. The purpose of developing the model is to use it as basis for field triage for traffic accidents involving trucks. In addition it can be valuable for recognizing important factors for improving triage protocols used in Sweden and possibly in other countries with similar traffic environments and prehospital procedures. In order to gain understanding of how modern triage systems are designed, the Guidelines for Field Triage of Injured Patients Recommendations of the National Expert Panel on Field Triage (Sasser et al., 2012) used in US and the Rapid Emergency Trauma and Triage System (RETTS) (Widgren and Jourak, 2011), which is the most widespread protocol in Sweden, are reviewed. In both systems the first step is based on physiological criteria, such as pulse and breathing rate, and the second step is based on anatomical criteria of identified injuries such as type of fractures. If any of the criteria in the first two steps is fulfilled the patient is given the highest priority level (red in RETTS). Next, in step three, MOI are assessed. In 2

3 RETTS, if any of the MOI inclusion criteria are met the patient is given the second highest priority level (orange). This differs from the US protocol where patients that fulfill MOI criteria should be handled as stated in (Sasser et al., 2012): Transport to a trauma center, which, depending upon the defined trauma system, need not be the highest level trauma center. Focusing on the MOI that apply for motor vehicle crashes some criteria are shared between both triage systems but differences are also found. Both systems include death in the same car and occupant being ejected from the vehicle. Rollover accident was previously a criterion in the US protocol but it has been removed in its two last versions. A similar criterion still in use in RETTS is person trapped in overturned vehicle. Other criteria in RETTS is deployment of airbag. These have no direct equivalent criteria in the US system, which recommends to measure intrusion of the occupant compartment to assess the forces having acted upon the occupant(s). Furthermore, the US expert panel recommends to utilize vehicle telemetry data (transmitted from vehicle to dispatch centre) consistent with high risk of injury when available (Sasser et al., 2012). Such data were used in the development of the URGENCY algorithm (Augenstein et al., 2003) for identifying severe crashes and a similar algorithm for the OnStar system (Kononen, Flannagan, and Wang, 2011). It was found that seat belt use, direction and location of impact, and delta-v i.e. the total change in velocity are important predictors of severe injury that can be recorded by telemetry systems integrated in vehicles (incorporating sensors such as accelerometers and gyroscopes to measure impact forces). However, the use of telemetry systems and access to their data is currently limited in Sweden and Europe, thus algorithms that can predict risk of severe injury as a function of crash characteristics that can be quickly assessed at the scene of accident are required. MOI included in the RETTS system has some important drawbacks such as considering airbag deployment which has a weak association to probability of severe injury (Buendia et al., 2015). But most importantly it does not consider unbelted occupant that was proven to be by far the strongest predictor of severe injury. Moreover, although MOI considered in RETTS are applicable to trucks, they do no not distinguish between car and trucks, in spite of the fact that truck safety may be different from car safety. This is the first study of accidents in Sweden that evaluates how MOI influence the risk that truck occupants have of sustaining severe injury using a probabilistic model. Materials and Methods Data Selection The scope of this study is adult truck occupants in traffic accidents registered in the Swedish Traffic Accident Data Acquisition (STRADA) database for calendar years 2003 to STRADA is the Swedish Transport Administrations national information system for traffic accidents occurring on the Swedish road network. It contains information reported by the police, and medical data on injured road users treated at emergency hospitals reporting to STRADA. The police visits the scene of an accident and reports the circumstances whereas the hospitals determine the injuries the patient sustains and their degree of severity. By combining data from two sources the STRADA system can provide more comprehensive information on the circumstances and consequences of road traffic accidents. The hospital reports focus on the individual level, i.e. the patient, as opposed to the police reports that focus on the circumstances of the accident, which may involve multiple casualties. The database is contained in a Microsoft Access database file. More information about STRADA can be found in (Howard and Linder, 2014). There are over injured people registered in STRADA from 2003 to In the database every 3

4 accident has a unique accident number. That accident number is shared between the police report and the hospital report and is the link between accident and patients. Only adult truck occupants, i.e. age 18 years, having both reports available in STRADA, i.e. the hospital report for the occupant and the police report of the corresponding accident, are considered in this study. The time span considered was from 2003, when the police started to systematically report to STRADA, until and including accidents reported in After this selection process the data was divided into light trucks, with total weight up to 3500 kg and heavy trucks with a total weight over kg. These divisions follow the Swedish classification and were used used in order to address potential dependence of truck occupant protection on the weight of the truck. Medium weight trucks with a total weight between 3500 kg to kg were not considered because there were only 111 cases containing just five severely injured subjects, thus there was insufficient data to construct a prediction model with a reasonable discrimination power. The final sample sizes were 2775 light truck occupants injured in 2608 accidents, and 922 heavy truck occupants injured in 903 accidents. Injury severity assessment and dichotomization In order to classify occupants as severely injured or not the Injury Severity Score (ISS) (Baker and O Neill, 1976) was used. ISS is based on the classification of the severity of each injury according to the Abbreviated Injury Scale (AIS) (AAAM, 2005). We used the recommended threshold for triage of ISS > 15 (Sasser et al., 2012), i.e. a patient is considered severely injured if ISS > 15 and not severely injured if ISS 15. For the light truck occupants 80/2775 (2.9 %) were severely injured, whereas 37/922 (4.0 %) of the heavy truck occpants were severely injured. Variables included in the model The dependent variable is whether the patient is classified as severely injured or not. All predictors that were included in the model are detailed in Table 1. Data Analysis All statistical calculations were performed with IBM SPSS Version 22. Univariate chi-squared tests of association were used to compute p-values under the null hypothesis of no association between the predictor and the dependent variable. Next multivariate analyses were performed. Logistic regression modelling was used to discriminate between severe and non-severe injury and compute the odds ratio (OR) of severe injury for each predictor. Logistic regression is a maximum-likelihood method that is commonly used in studies of traffic accidents, see e.g. (Harrell, 2001; Schiff, Tencer, and Mack, 2008; Augenstein et al., 2003; Kononen, Flannagan, and Wang, 2011). A binary logistic regression model fits the log odds (logit) by a linear function of the predictors (Equation 1). ( ) P (Y = 1) ln 1 P (Y = 1) = β 0 + n β k x k (1) Y is the dependent variable, Y = 1 for severe injury and Y = 0 for non-severe injury. P denotes the probability of severe injury. The OR for the k:th predictor are e βk. Results of logistic regression modelling are expressed as adjusted OR, with corresponding 95 % confidence intervals (CI). Because of the overall low k=1 4

5 proportion of severe injuries in this study (< 5 %) the OR can be considered a reasonable approximation of the relative risk. The area under the receiver operator characteristic (ROC) curve (AUC) was used to measure the classification performance of the model. The ROC shows the power in terms of sensitivity and specificity for prediction of severe injury for different cut-off values of P (Y = 1) (Equation 1). The cut-off value determines the trade-off between sensitivity and specificity; increasing the sensitivity (identifying more patients with severe injury) is at the cost of decreasing the specificity (more false positives, i.e. more patients that are not severely injured are predicted to have severe injury). Using the OSISP algorithm in the field will require finding a suitable value for this cut-off. This is best determined by the management for health care trauma systems and is outside the scope of this study. A 10-fold cross validation (CV) procedure was used to estimate the performance of the OSISP algorithm on unseen data, in terms of ROC and AUC. The dataset was divided into ten randomized folds with approximately equal number of casualties. One fold at a time was left out, a model was derived using Equation 1 on data from the remaining nine folds. The model was then validated by classifying the observations in the left out fold. This procedure was repeated for all folds, i.e. performed ten times. Results The results will be presented as follows. For each group, i.e. light and heavy truck occupants, the influence of belt use and age on the probability of sustaining severe injury are shown. Then the results for univariate analyses is shown. Finally the two binary logistic regression models for light and heavy trucks are derived. For each predictor in these models, statistical significance (p-values), OR and 95 % CI are shown. In order to test the performance the ROC, AUC and corresponding CI with and without use of CV are presented. Belt use and age versus severe injury Figure 1 demonstrates a decreasing number of subjects with increasing age. Figure 2 shows the rate of belt use, which is higher in light trucks (79 %) than in heavy trucks (55 %). The frequency of severe injury is substantially lower for belted occupants than for unbelted/unknown. For both truck types, the frequency of severe injury is higher in the group where belt use was not reported (status unknown) than for unbelted occupants. Figure 2 also shows that for belted occupants the frequency of severe injury is lower for heavy trucks than for light trucks. The same relation holds for those cases where belt use was unknown; whereas unbelted occupants have a slightly higher frequency of severe injury for light trucks. Figure 3 shows belt use for each age group for heavy trucks. Univariate analysis Table 2 shows the following values for each predictor: degrees of freedom, p-value and proportion of severely injured patients associated with each level of the predictor. Logistic regression models Tables 3 and 4 show the results of the multivariate model for occupants in light and heavy trucks, respectively. The ROC curves are shown in Figure 4 and AUC values are presented in Table 5. The models achieve CV 5

6 AUC of 0.81 for light trucks and 0.74 for heavy trucks. Belt Use stands out as the most valuable predictor for both models. Type of accident and age are important predictors in the case of light trucks. Discussion Age and belt use The number of subjects decrease with age in both types of trucks, see Figure 1. There is a clear increase of the frequency of severe injury with age for light truck occupants, and a similar pattern can be observed for heavy trucks although there are too few subjects in the two categories with highest age to estimate their respective frequency (Figure 1). No severe injuries were observed for occupants in heavy trucks of age 66 years. The absence of severe injuries in this group is probably due to that it includes only 19 subjects with a higher proportion of belt use (Figure 3), 12/19 subjects were belted. Increased proportions of severely injured can be observed for light truck occupants > 55 years of age. This result is consistent with the recommended age partition at 55 years by the expert panel on field triage (Sasser et al., 2012). It is also consistent with the strong association between age and severe injury reported in several studies (MacKenzie et al., 2006; Champion et al., 2005; Tavris, Kuhn, and Layde, 2001) and with the results of similar study performed on car occupants (Buendia et al., 2015). For heavy truck occupants the age group 46 years to 55 years experienced a quite high proportion of injuries (Figure 2), a fact for which we currently have no explanation. Interestingly, a higher overall proportion of severely injured was found for heavy truck occupants (4.0 %) than for light truck occupants (2.9 %). This result was unexpected because the heavy trucks high weight and the placement of the occupant compartment high above the wheelbase is expected to provide good protection for the occupants. Furthermore, an even lower probability (2.0 %) of severe injury was found in the parallel study on cars (Buendia et al., 2015). The best explanation for these discrepancies is the large difference in compliance of wearing the seat belt. The rate of belt use for the subjects in the dataset was 75 % in light trucks and only 55 % in heavy trucks, which can be compared to 94 % in cars (Buendia et al., 2015). Figure 2 shows that for a large proportion of patients belt use status was unknown, and that this group has an even higher rate of severe injury than unbelted. Figure 2 shows that for belted occupants heavy trucks have lower frequency of severe injury (0.79 %) than light trucks (1.23 %), which can be compared to frequency for cars at 1.5 % found in (Buendia et al., 2015). The percentage of severe injury among unbelted occupants is slightly higher for heavy trucks than for light trucks. However, if the belt status unknown and unbelted were merged heavy trucks would have a lower proportion of severe injury than light trucks. This is consistent with our belief that heavy trucks are safer than light trucks and cars, and that trucks are safer than cars. Predictors In both types of trucks belt use is the strongest predictor of severe injury. This conclusion is analogous to the results for car occupants in. Elderly occupant, i.e. aged > 55 years, was a strong predictor for both types of trucks. Type of accident was a very strong predictor for light truck occupants, with head-on accidents having the highest OR, which is consistent with the analysis of cars (Buendia et al., 2015). Note that tram/train accidents were so few that they are not considered in this discussion. Although type of accident was not a strong predictor for heavy truck occupants, it is remarkable that single accidents was by far the most common type as well as the most dangerous. Volvo Trucks (2013) showed that single accidents in rural environments 6

7 is the most common accident producing severe injury, representing 50 % of the total number of severely injured cases. In the present study the proportion was much higher, 84 %. It is also a remarkable finding that, for heavy truck occupants, head-on accidents have associated a much lower severe injury probability than rear-end accidents. We have no good explanation for this, maybe there is a data selection bias due to that occupants in head-on accidents are more prone to seek/receive hospital care because the accident appear to be more severe as compared to rear-end accidents. Airbag deployment was unknown in a large number of cases in both types of trucks. Nevertheless, the models show a higher probability of severe injury in trucks that were not equipped with airbag. This result in itself is insufficient to show the efficiency of airbags because of the presence of potential confounding factors, e.g. that trucks without airbag are typically older models and presumably provide less degree of protection than newer models. Sex, traffic environment (whether the accident happened in an urban or rural area) and PSL were not strong predictors for any type of truck. Although sex may not be a strong predictor it is remarkable that women have smaller probability of being severely injured than men. This result differ from the one reported for cars, where male occupants have a smaller probability of severe injury (Kononen, Flannagan, and Wang, 2011; Buendia et al., 2015). The lower risks for women is only partially explained with belt use as similar percentages to the average were found, i.e. 82 % of women versus 79 % of men were belted for light trucks, and 61 % of women versus 55 % of men were belted for heavy trucks. Another possible reason is that women drive more safely and are producing lower impact crashes. The injury severity coding in STRADA was changed from AIS 1990 to AIS 2005 in 2007, see (Howard and Linder, 2014), and this change generally resulted in a decrease of the coded AIS values for certain injuries. A statistically significant reduction of injury severity was found in the univariate test. However, the period of the accident ( or ) was not a statistically significant predictor in the logistic regression models. Furthermore, according to the OR in the multivariate model, after adjusting for every predictor, the probability of severe injury even increased from 2007 for light truck occupants (Table 3). These results differ from those for car occupants described in (Buendia et al., 2015). In that study, the difference in the probability of severe injury of car occupants produced by the AIS coding change was found substantial in both univariate and multivariate analyses. OSISP models for trucks The OSISP model developed for light trucks achieved CV AUC of To put this performance into perspective, Harrell (2001) stated that a model with an AUC > 0.80 has a good discriminating capability. The discriminating capability achieved for heavy trucks is lower but still performs reasonably well with a CV AUC of AUC values obtained when the complete dataset was used for evaluating the classification performance, i.e. for the full model without CV, were 0.87 for both types of trucks. The difference in the AUC values with and without CV indicates overfitting of the model. This may be a consequence of the small datasets available for this study, i.e and 922 subjects whereof only 73 and 37 were severely injured for light and heavy trucks, respectively. In particular, overfitting is more severe for heavy trucks since the dataset is smaller, which is reflected in a larger discrepancy between evaluation of the full model and CV than for light trucks. Overfitting may be reduced by removing the less important predictors or by merging predictor levels with similar OR. However, we decided to include all predictors and levels to allow for a fair comparison between light trucks, heavy trucks and cars (Buendia et al., 2015). Since new cases are continuously added to 7

8 the STRADA database, future analyses based on more data may suffer less from overfitting. The models were developed to be used as basis for a field triage tool. For achieving the highest triage accuracy possible, overfitting will need to be reduced in future analyses. Prospective evaluations where the final algorithm is tested in the field will be needed to assess the performance of the OSISP algorithm and compare it to the performance of current triage protocols. Limitations of the study We consider the main limitation to be the size of the dataset. This causes overfitting and may be the reason behind the lack of statistical significance of certain predictors. In this study belt use, elderly occupant and type of accident yielded p < 0.05 for light truck occupants, and only belt use yielded p < 0.05 for heavy truck occupants. This could be compared to the results for car occupants in Buendia et al. (2015) in which all predictors considered in this study, except for airbag, yielded p < Another limitation is that medium weight trucks were not included due to that very few accidents were available. According to the Swedish classification of trucks, heavy trucks are those that exceed 3500 kg, which require a special driver s license. Heavy trucks are divided according to weight up to kg, referred to as medium weight trucks in this study, and above kg, referred to as heavy trucks in this study. Volvo Trucks (2013) reported similar results in many aspects for accidents of heavy and medium weight trucks, but they also found several important differences. Therefore, the results obtained in this paper for heavy truck occupants do not necessarily hold for occupants of medium weight trucks. Future studies are needed to develop OSISP algorithms for medium weight trucks. A number of limitations are related to the underlying database STRADA. Importantly, belt use in STRADA s hospital report is self-reported by the patient and may possibly be influenced by a perceived need to conform with societal expectations or a fear of liability issues. Furthermore, the requirement of having the hospital report available in STRADA means that only those truck occupants who were admitted to a hospital were included; therefore, the probability of severe injury is actually the conditional probability of severe injury given transportation to a hospital, which may produce a data selection bias. This approximates the unconditional probability of severe injury if the probability of transportation to a hospital is close to one, which may be the case for the target population in which the triage tool is intended to be used. Future prospective evaluations of the OSISP algorithm, i.e. field test in prehospital settings that include the actual accidents handled by ambulance personnel, is needed to test the accuracy for the relevant population. Not all hospitals in Sweden report to STRADA; according to Howard and Linder (2014) the number of hospitals in the system has gradually increased from 29 hospitals in 2003 to 68 hospitals in 2012 out of the 80 hospitals in Sweden in total. This means that the geographical distribution of accidents in the sample may have varied over the 11-year period considered in this study. Finally, STRADA does not provide detailed information about some presumably important variables regarding the accident. Information about the angle of impact and delta-v could potentially have improved model accuracy (Kononen, Flannagan, and Wang, 2011). Lack of information about compartment intrusion and drunk driving may decrease the predictive capability of the model as well as the accuracy of OR. Vehicle information about the truck such as model year and weight were not considered due to the perceived difficulty of specifying these variables at the scene of accident. In future OSISP algorithms vehicle telemetry data could provide such information. 8

9 Conclusion For the purpose of developing future complementary tools for field triage two OSISP multivariate logistic regression models that predict the probability of severe injury (ISS > 15) based on crash characteristics that can be quickly assessed at the scene of accident were implemented for light and heavy trucks, respectively. These models achieve good performance for light truck occupants (CV AUC = 0.81) and a reasonable performance for heavy truck occupants (CV AUC = 0.74). The predictive power is expected to increase in the future when the models are further optimized to reduce overfitting and based on larger datasets. Belt use was the strongest predictor of severe injury for occupants in both light and heavy trucks. The rate of belt use was low, in particular for heavy truck occupants. This appears to be the cause of a relatively high proportion of truck accidents producing severe injury, despite that safety of trucks is considered high. There is a need for information campaigns, technical innovations and other means to increase the compliance of wearing seat belt. Acknowledgements This work has been carried out at SAFER - Vehicle and Traffic Safety Centre at Chalmers, Sweden. References AAAM Abbreviated Injury Scale. Association for the advancement of automotive medicine, Barrington, IL, USA. Augenstein, Jeffrey, Elana Perdeck, James Stratton, Kennerly Digges, and George Bahouth Characteristics of crashes that increase the risk of serious injuries. In Annu Proc Assoc Adv Automot Med, Vol Baker, Susan P., and Brian O Neill The Injury Severity Score. The Journal of Trauma: Injury, Infection, and Critical Care 16: Brown, Joshua B., Nicole A. Stassen, Paul E. Bankey, Ayodele T. Sangosanya, Julius D. Cheng, and Mark L. Gestring Mechanism of Injury and Special Consideration Criteria Still Matter: An Evaluation of the National Trauma Triage Protocol. Journal of Trauma-Injury Infection and Critical Care 70: Buendia, Ruben, Stefan Candefjord, Helen Fagerlind, András Bálint, and Bengt Arne Sjöqvist On Scene Injury Severity Prediction (OSISP) Algorithm for Car Occupants. Submittted to Accident Analysis & Prevention. Champion, H.R., JS Augenstein, AJ Blatt, B. Cushing, KH Digges, MC Flanigan, RC Hunt, LV Lombardo, and JH Siegel New tools to reduce deaths and disabilities by improving emergency care: URGENCY software, occult injury warnings, and air medical services database. In Proceedings of the 19th International Technical Conference on Enhanced Safety of Vehicles (NHTSA sponsored), Washington, DC,. Haas, Barbara, David Gomez, Brandon Zagorski, Therese A. Stukel, Gordon D. Rubenfeld, and Avery B. Nathens Survival of the Fittest: The Hidden Cost of Undertriage of Major Trauma. Journal of the American College of Surgeons 211:

10 Harrell, Frank E Regression modeling strategies, with applications to linear models, survival analysis and logistic regression. Springer Science and Business Media, Inc., New York. Howard, Christian, and Astrid Linder Review of Swedish experiences concerning analysis of people injured in traffic accidents. Swedish National Road and Transport Research Institute (VTI) VTI notat 7A-2014: Kononen, Douglas W, Carol A C Flannagan, and Stewart C Wang Identification and validation of a logistic regression model for predicting serious injuries associated with motor vehicle crashes. Accident Analysis & Prevention 43: MacKenzie, E.J., F.P. Rivara, G.J. Jurkovich, A.B. Nathens, K.P. Frey, B.L. Egleston, D.S. Salkever, and D.O. Scharfstein A national evaluation of the effect of trauma-center care on mortality. New England Journal of Medicine 354: Murad, Mudhafar K, Stig Larsen, and Hans Husum Prehospital trauma care reduces mortality. Ten-year results from a time-cohort and trauma audit study in Iraq. Scand J Trauma Resusc Emerg Med 20 (13): Sasser, SM, RC Hunt, M Faul, D Sugerman, WS Pearson, D Theresa, MM Wald, et al Guidelines for Field Triage of Injured Patients - Recommendations of the National Expert Panel on Field Triage, MMWR 61: Schiff, Melissa A, Allan F Tencer, and Christopher D Mack Risk factors for pelvic fractures in lateral impact motor vehicle crashes. Accident Analysis & Prevention 40: Schoell, Samantha L., Andrea N. Doud, Ashley A. Weaver, Jennifer W. Talton, Ryan T. Barnard, R Shayn Martin, J Wayne Meredith, and Joel D. Stitzel Development of a Time Sensitivity Score for Frequently Occurring Motor Vehicle Crash Injuries. Journal of the American College of Surgeons. Tavris, Dale R, Evelyn M Kuhn, and Peter M Layde Age and gender patterns in motor vehicle crash injuries: importance of type of crash and occupant role. Accident Analysis & Prevention 33: Volvo Trucks European Accident Research and Safety Report Widgren, Bengt R., and Majid Jourak Medical Emergency Triage and Treatment System (METTS): A New Protocol in Primary Triage and Secondary Priority Decision in Emergency Medicine. Journal of Emergency Medicine 40:

11 Table 1: Definitions and descriptions of the model variables. For each variable level the proportion of the casualties having that characteristic is shown for light and heavy trucks, respectively. Variable Definition Level Light Heavy Description Belt Belt use Unknown 12 % 17 % Whether the occupant used seat belt when accident occurred Unbelted 8.7 % 28 % Belted 79 % 55 % Airbag Airbag Deployment Unknown 34 % 50 % Whether the airbag (if present) was deployed Undeployed 39 % 45 % Deployed 26 % 1.6 % No Airbag 2.0 % 2.9 % Type Type of accident according to the hospital report in STRADA Turning 4.4 % 1.7 % Turning accident Intersection 13 % 2.8 % Collision (head on or side) with another vehicle in an intersection Head-on 11 % 13 % Frontal collision with another vehicle outside an intersection Overtaking 2.2 % 1.3 % Overtaking accident Single 34 % 61 % Single vehicle collides with stationary object or departs from road Tram/Train 0.5 % 1.2 % Collision with a tram or a train Rear end 28 % 15 % Vehicle impacts another vehicle from behind WLA 2.8 % 0.8 % Collision with a wild life animal (WLA), in most cases a moose Other 4.2 % 3.8 % Other type of accident PSL Posted speed limit Unknown 12 % 12 % Maximum speed allowed where the accident occurred 30 km/h 1.1 % 0.5 % 40 km/h 0.7 % 0.3 % 50 km/h 19 % 12 % 60 km/h 1.2 % 0.3 % 70 km/h 30 % 26 % 80 km/h 4.4 % 5.5 % 90 km/h 18 % 29 % 100 km/h 3.9 % 2.7 % 110 km/h 9.7 % 12 % 120 km/h 0.9 % 1.2 % Location Elderly Location of accident Victim over or under 55 years old Unknown 8.8 % 8.0 % Whether accident occurred in urban or rural environment, i.e. Urban 27 % 16 % inside or outside a population center. Circumvallation roads are Rural 64 % 77 % defined as urban % 83 % Age 55 years old is used as threshold. The mean age was 39 > % 17 % years and 8 months (correct? same info as cars paper). Sex Sex of occupant Male 82 % 92 % Male or female occupant Female 18 % 8 % Period Calendar years % 26 % In January 2007 the injury coding system was changed from % 74 % AIS-1990 to AIS-2005, causing a substantial effect towards lower levels of injury. Number of subjects Heavy trucks Light trucks Frequency severely injured (%) > 76 Age (years) 0 Figure 1: Number of subjects (bars, left hand side axis) and frequency of severely injured (lines with markers, right hand side axis) per age group for heavy and light truck occupants, respectively. 11

12 Proportion (%) Heavy trucks Light trucks Unknown Unbelted Belted Frequency severely injured (%) Figure 2: The proportion of belt use (bars, left hand side axis) and its relation to the frequency of severely injured (lines with markers, right hand side axis) for heavy and light truck occupants, respectively Proportion (%) Age (years) Figure 3: The proportion of belt use per age group for occupants in heavy trucks. 12

13 Table 2: Results for univariate analyses. For each predictor: degrees of freedom (DF), p-value (p) and proportion (P) of severely injured patients associated with each variable level for light and heavy truck occupants, respectively. Variable DF p (light) p (heavy) Level P (light) P (heavy) Belt 2 < 10 4 < 10 4 Unknown 11 % 9.6 % Unbelted 6.2 % 6.9 % Belted 1.2 % 0.8 % Airbag 3 < Unknown 5.4 % 5.4 % Undeployed 1.1 % 2.6 % Deployed 2.2 % 0.0 % No Airbag 3.6 % 3.7 % Type 8 < Turning 0.8 % 0.0 % Intersection 1.7 % 0.0 % Head-on 9.2 % 0.8 % Overtaking 0.0 % 0.0 % Single 3.0 % 5.5 % Tram/Train 27 % 9.1 % Rear end 0.9 % 3.0 % WLA 1.3 % 0.0 % Other 4.3 % 0.0 % PSL Unknown 3.5 % 3.7 % 30 km/h 0.0 % 0.0 % 40 km/h 0.0 % 0.0 % 50 km/h 1.6 % 7.4 % 60 km/h 0.0 % 0.0 % 70 km/h 3.1 % 5.1 % 80 km/h 1.7 % 2.0 % 90 km/h 4.4 % 2.6 % 100 km/h 0.9 % 0.0 % 110 km/h 3.3 % 3.7 % 120 km/h 4.0 % 9.1 % Location Unknown 3.3 % 0.0 % Urban 2.0 % 4.9 % Rural 3.2 % 4.3 % Elderly % 3.8 % > % 5.1 % Sex Male 3.1 % 4.1 % Female 2.0 % 2.9 % Period % 5.9 % % 3.4 % 13

14 Table 3: Logistic regression model for light trucks. The variable level used as reference is shown as the last level given (for predictors with two levels) or within parenthesis (for predictors with more than two levels). DF = degrees of freedom. Variable β DF p-value OR (e β ) [95 % CI] Belt (Unbelted) 2 < 10 4 Belted < [0.07, 0.28] Unknown [0.60, 2.6] Airbag (Not deployed) Deployed [0.61, 3.0] No Airbag [0.81, 19] Unknown [0.89, 4.0] Type (Head-on) 8 < 10 4 Turning [0.01, 0.63] Intersection [0.08, 0.56] Overtaking Single < [0.13, 0.44] Tram/Train [0.50, 12] Rear end < [0.04, 0.22] WLA [0.009, 0.57] Other [0.12, 0.95] PSL (Unknown) km/h km/h km/h [0.15, 1.22] 60 km/h km/h [0.37, 2.03] 80 km/h [0.1, 2.48] 90 km/h [0.70, 3.96] 100 km/h [0.1, 7.19] 110 km/h [0.62, 4.81] 120 km/h [0.24, 22.53] Location (Urban) Rural [0.88, 3.0] Unknown [0.67, 4.3] > 55 / [1.3, 3.7] Female/Male [0.35, 1.5] / [0.59, 1.9] Constant

15 Table 4: Logistic regression model for heavy trucks. The variable level used as reference is shown as the last level given (for predictors with two levels) or within parenthesis (for predictors with more than two levels). DF = degrees of freedom. Variable β DF p-value OR (e β ) [95 % CI] Belt (Unbelted) 2 < 10 4 Belted < [0.03, 0.29] Unknown [0.57, 3.4] Airbag (Not deployed) Deployed No Airbag [0.73, 92] Unknown [0.72, 4.6] Type (Head-on) Turning Intersection Overtaking Single [0.95, 57] Tram/Train [0.32, 130] Rear end [0.45, 49] WLA Other PSL (Unknown) km/h km/h km/h [0.72, 11] 60 km/h km/h [0.48, 5.6] 80 km/h [0.10, 11] 90 km/h [0.20, 2.9] 100 km/h km/h [0.19, 4.0] 120 km/h [0.10, 17] Location (Urban) Rural [0.62, 4.5] Unknown > 55 / [0.67, 3.9] Female/Male [0.18, 4.0] / [0.32, 1.5] Constant

16 1 0.8 Sensitivity Heavy trucks, full Heavy trucks, 10-fold CV Light trucks, full Light trucks, 10-fold CV specificity Figure 4: ROC curves for full models and 10-fold cross-validation (CV). Table 5: AUC values for the multivariate models. Model AUC [95 % CI] Heavy trucks 0.87 [0.83, 0.91] Heavy trucks, 10-fold CV 0.74 [0.65, 0.82] Light trucks 0.87 [0.83, 0.91] Light trucks, 10-fold CV 0.81 [0.75, 0.87] 16