Development of a Comprehensive Approach for Serious Traffic Crash Injury Measurement and Reporting Systems (2021)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

7 This report is organized in the following way. First, we start with a review of injury coding systems and injury severity metrics, leading to the recommendation of a specific medical- diagnosis-based metric for use by states. This review will include a comparison of crash-report- based and medical-diagnosis-based metrics using crash data. Second, we present the results of the state survey of data systems and linkage activities, which give context to the current condition of datasets and linkage programs that a roadmap would address. Third, we discuss near-term solutions to measuring serious injuries in crashes, focusing on sampling programs that could be implemented at the state level. Fourth, we present a roadmap to linkage. In this section, we discuss ways in which the comprehensive linkage goal of the project can be facilitated. We also discuss “choice points” where decisions made earlier may facilitate or fail to facilitate future linkages. 3 Measuring Serious Injury 3.1 Injury Classification Systems To understand the processes used to identify serious injuries, it is necessary to first consider the available injury coding systems. The primary purpose of an injury coding system is to identify specific types of injuries. Ranking injury severity is not necessarily an explicit purpose of injury coding, but ultimately, a ranking system must be imposed in order to identify serious injuries as a distinct class. The three major injury coding systems relevant to the traffic crash domain are the KABCO scale, the Abbreviated Injury Scale (AIS) (Gennarelli & Wodzin, 2005), and the International Classification of Disease, Clinical Modification (ICD-CM) system (WHO, 1992). Each of these typologies have different scale qualities, may be found in different databases, and have different advantages and disadvantages for use. KABCO. The KABCO scale is used by police officers on the scene of a crash to judge the general injury severity level of each occupant. The scale was developed by the National Safety Council and is recommended in the MMUCC guidelines for crash data (DOT, 2012). In general, K is Killed, A, B, and C are injuries of decreasing severity, and O is property-damage only. One of the problems with KABCO is that different states use different definitions of the A, B, and C injury codes. The MMUCC 3rd edition (DOT, 2008) and the American National Standards Institute D16.1-2007 (ANSI, 2007) recommended A for Incapacitating Injury, B for Non-Incapacitating Injury, and C for Possible Injury. Most, but not all states have used these definitions, and the lack of universal consistency may create problems in the usage of this scale across different jurisdictions. In 2012, the 4th revision of MMUCC tried to standardize KABCO usage with new definitions: A for Suspected Serious Injury, B for Suspected Minor Injury, and C for Possible Injury. KABCO does not characterize or “type” injuries. The primary advantage of KABCO is that it is available in the police report database of virtually every state. KABCO is strictly an injury ranking system and not an injury classification system. Each crash-involved party is given a single scale score for his/her apparent overall injury severity level. As such, there is very little that can be done to glean additional information from this scale regarding injury typology. AIS. The AIS was developed by the Association for the Advancement of Automotive Medicine (Gennarelli & Wodzin, 2005) for the purpose of coding injury types and injury severity, based upon an in-hospital clinical assessment. The AIS system assigns a unique numeric code to each specific injury type and each code is associated with a severity score

8 ranging from 1 (minor) to 6 (life threatening). Coding is done by coders trained specially on the AIS lexicon using existing medical records. With AIS, both injury coding and injury severity ranking are embodied in a single system. However, each individual anatomical injury is coded separately. Thus, to identify seriously injured crash victims, it is still necessary to combine a patient’s injury severity scores into a single person-level metric. There are a number of systems for doing this that will be described later. International Classification of Disease, Clinical Modification (ICD-CM). In hospital administrative databases, injuries (as well as diseases, and other aspects of medical typologies) are coded using the International Classification of Disease, Clinical Modification (ICD-CM) system (WHO, 1992). The ICD-CM is a general-purpose classification system for diagnoses of all health conditions and includes codes for both the nature of the injury and causes of injury. Coding is done by trained medical coders who work from hospital records. Medical coders must pass an exam to become a Certified Professional Coder or Registered Health Information Technician(RHIT) and must have background in anatomy and physiology as well as the coding system itself. ICD-CM is widely used in clinical and health research settings, and is commonly found in hospital and trauma databases. The U.S. is currently in transition between using ICD-9- CM and a newer revision, ICD-10-CM. Unlike AIS, ICD-CM does not include an explicit ranking of injury severity. To be used to identify seriously injured crash victims, ICD-CM must either be mapped to AIS or some other ranking system must be imposed on the coded injuries to assess severity. Both of these approaches will be discussed below. 3.2 Severity Metrics Each of the three aforementioned coding (or ranking) systems is often manipulated to provide an abridged injury severity metric that suggests the presence of serious vs. non-serious injury in a patient. Severity metrics are typically intended to reflect increasing threat to life, with higher severity scores associated with higher probability of mortality. Severity can also be associated with risk of long-term disability, though the difficulty of obtaining long-term follow- up data has limited studies of this association. Table 2 summarizes the key injury severity scoring systems based on the three injury coding systems described above. The table includes calculation and cutpoints that have been found in the literature. Details are given in the text that follows.

9 Table 2 Summary of Injury Severity Metrics and Characteristics Injury Coding System Injury Severity Metric Calculation Common Cutpoint(s) for Serious Injuries KABCO KABCO All occupants with K or A-injury severity rating on police accident report KA AIS MAIS Highest AIS severity score of all injuries 3+ ISS Sum of squares of highest injury severity in each of three different, most-injured body regions 16+ (9+ is also used) NISS Sum of squares of three highest injury severities regardless of body region 16+ (9+) ICD ICISS Product of Survival Risk Ratios (SRRs) of each individual injury SRRs = <0.90 has been used, but no standard established mSRR Worst (minimum) Survival Risk Ratio among injury diagnosis codes No standard found TMPM Regression model designed to predict mortality outcome No standard found Other LoS Length-of-Stay in the hospital, measured in days 4 days has been used; no established standard Sentinel Diagnosis Presence of any of an agreed-upon list of diagnoses No standard KABCO-Based. Each of the three aforementioned coding (or ranking) systems are often manipulated to provide an abridged injury severity metrics that suggests the presence of serious vs. non-serious injury. Researchers and practitioners using only police-reported information and KABCO typically use K plus A (KA) to characterize transportation crash injuries resulting IRTAD, 2011]. K, A, B and C (KABC) are also sometimes used to identify all injured occupants [e.g., NHTSA, 2010]. AIS-Based. Because AIS is designed to characterize injury types and has an embedded severity scale for each specific injury, AIS has been used to develop several injury severity metrics that quantify the multiple injuries that may be experienced by the occupant of a transportation crash. The most common AIS-based metric is a single maximum AIS (MAIS) across all body regions that are coded for injuries. MAIS is often used as a measure of overall injury level for an occupant, and MAIS of 3 or greater (MAIS 3+) is commonly used as the cutoff for defining serious injury (e.g., IRTAD, 2011). Since the maximum AIS severity score does not distinguish between patients with several serious injuries to different body regions and those with more localized injury, Baker et al. (1974) developed the Injury Severity Score (ISS). The ISS is the sum of the squares of the most

10 severe AIS scores in each of three different body regions. The highest possible ISS is 75. An ISS cutoff of 16+ has been used to define seriously injured occupants (e.g., AACN expert panel). The New Injury Severity Score (NISS) is similar to ISS, but is computed as the sum of squares of the three most severe injuries, regardless of body region (Osler, Baker & Long, 1997). ICD-CM-Based. Because of its widespread use in hospital settings, there is a great deal of interest in assessing injury severity using ICD-CM. However, the ICD-CM lexicon only characterizes injury and does not assess severity. One approach used to transform injury descriptions to measures of severity is to map ICD-CM to AIS codes and use AIS-based severity scoring. A private software product called ICDMAP™ translates ICD-9-CM codes to AIS 90 (1998 revision). A more up-to-date application is the ICDPIC mapping procedure developed for use with STATA©, a common statistical package. Injury severity estimates, based upon ICDPIC manipulations of ICD-9-CM, correlate well with hand-calculated AIS provide by trained coders, but these estimates tend to produce a slight but systematic underestimate of true injury severity (Fleischman, Mann, Wang et al., 2012). Haas et al. (2012) published an evaluation of their mapping from ICD-10 to AIS 1998. The translation produced 57% agreement in overall MAIS and 83% agreement in identifying patients with MAIS 3+ injuries. Agreement in ISS was also evaluated, with promising results (87% of cases resulted in a difference of ≤10 points). Although the mapping is to an older version of AIS, the authors discuss the potential to map to the AIS 2005 revision. Zonfrillo et al. (2015) reported on a mapping between ICD-10-CM (as well as ICD-9- CM) and general AIS categories of AIS 3+, AIS<3, or indeterminable. While less nuanced than a mapping to the full AIS scale, this approach allows mapping to the key cutoff and the definition of serious injury recommended in this report. They report that 27% of ICD-10-CM codes and 17% of ICD-9-CM were rated indeterminable. An alternative to mapping from ICD-CM to AIS is to use an ICD-CM based severity metric. Because ICD does not include an explicit ranking of injury, ICD-based scoring must introduce severity through another means. One approach relies on computed SRRs for certain classes of ICD-based diagnoses. The SRR is the proportion of patients with the diagnosis code who survive out of all the patients with that code. Thus, the SRR allows categorical injury codes to be ranked with respect to survival (i.e., severity). The SRR approach presents some challenges for widespread adoption of ICD-based metrics for measuring serious in crashes. Notably, SRRs are database-specific and thus affected by the patient population and treatment protocols of the hospital(s) that contribute to the database (Cryer, 2006). Efforts are being made to standardize SRRs, based on the NTDB (Meredith et al, 2003). The ICD-9-CM Injury Severity Score (ICISS; Osler, et al., 1996) is the product of all SRRs for a patient’s injuries. The calculation is meant to represent the joint probability of survival given the particular group of injuries, and is the most widely used ICD-based injury severity metric. One of the issues with ICISS is that SRRs calculated from all patients with the diagnosis (“traditional SRR”) may not represent the independent probability of survival for each injury. Meredith et al. (2003) reported that ICISS using SRRs calculated only from patients with a single diagnosis (“independent SRR”) is preferable to ICISS using traditional SRRs. Another issue with ICISS and mSRR is the lack of a well-established criterion cut point. An ICISS of <0.90 (>10%

11 chance of mortality) was used by Newgard et al. (2010), but most of the work on ICISS has focus on discrimination performance without consideration of specific cut point choice. Other approaches use a patient’s worst (minimum) SRR rather than the product of all SRRs (Meredith, Kilgo & Osler, 2003). This approach simplifies calculation based on SRRs, relative to ICISS, but has the same problem with a lack of agreed-upon criterion. An alternative approach that is gaining momentum is the regression-based approach embodied in the Trauma Mortality Prediction Model (TMPM) developed by Glance et al. (2009) using ICD-9-CM codes. The TMPM is based on a regression model developed using mortality as the outcome measure and thus its performance at predicting mortality is, not surprisingly, better than other measures (Haider et al., 2012). An additional advantage is that the model does not depend on SRRs and therefore can be standardized for use in different hospitals. Other Metrics. In their report on severity metrics, the International Road Traffic Accident Database (IRTAD) (2011) evaluated four candidate metrics, of which two fall outside the categories described above. One was Length-of-Stay in the hospital (LoS), and the other was the presence of a sentinel diagnosis. Hospital Length-of-Stay is simply the number of days a patient is admitted to the hospital for treatment of crash-related injuries. It is a proxy for injury severity rather than a direct measure, but is used in a number of countries where anatomically-based measures are not available. Sentinel diagnosis is simply the presence of one or more of a list of specific diagnoses that are selected because of their association with a high probability of hospital admission. These diagnoses can be identified using either AIS codes or ICD-CM codes. “Serious Injury” is then defined as the presence of any of the selected sentinel diagnosis codes and “No Serious Injury” is defined as the absence of all such codes. There are other prominent injury severity ranking systems that rely on physiologic parameters to calculate (i.e., Revised Trauma Score [RTS] and Trauma and Injury Severity Score [TRISS]). However, these measure are more likely to characterize the physiological stability of a patient and may not correlate well with an anatomical assessment of injury severity. In addition, there are many other injury severity metrics based on AIS and ICD codes. There is a large literature on measuring injury severity and tying such metrics to survival at hospital discharge. The measures we have selected are the most-studied and most-recognized of available metrics. New developments in measurement of injury severity may bring newer ones to the forefront, but at this time, it makes the most sense to focus on measures that have been studied and used the most. The major advantage of linkage between crash and medical outcome datasets is that if a new metric becomes state-of-the-art, the data will be available to re-compute performance metrics for crashes in the future relatively easily. 3.3 Evaluation Criteria We selected three criteria on which to evaluate injury severity metrics for use in measuring road safety. Our first criterion is the ability of the metric to predict outcome, specifically survival. The second criterion is availability of data. Given the condition of datasets and data linkages in the U.S. at this time, it is important to discuss the issue of data availability and the impact it will have on states’ ability to assess serious injuries using the recommended metric. Finally, the third criterion is ease of use. Predicting Outcome. A good injury severity metric should be calibrated to an agreed- upon outcome. Survival to the time of hospital discharge is most commonly used for this

12 purpose. A more serious injury will result in a greater threat to life, and a good injury severity metric should reflect that. MAIS 3+ has been shown to be a good predictor of in-hospital mortality (Meredith et al., 2002; Kilgo, Osler & Meredith, 2003). In contrast, because of the lack of data linked to in- hospital mortality, “A” injury (from KABCO) has not been included in investigations of this relationship. However, “A” injury has been shown in several studies to be only moderately associated with MAIS 3+ injury (e.g., Farmer, 2003; Compton, 2005). While this is not a direct test of the relationship between “A” and mortality, it does suggest that good performance is not likely. A larger problem with “A” is the fact that judgments are made by police officers rather than medical professionals, and that “A” is a global measure of injury severity for a single person. Even if correspondence were perfect, “A” would make an effective, straightforward, and readily available measure of overall injury severity, but would leave no further options for analysis of specific injury types (e.g., head injury). It should be noted that several researchers (e.g., Cryer, 2006) as well as readers of our interim reports have pointed out that attention should also be paid to threat-of-disability as an outcome. In particular, costs to states from injury-related disability can be quite high in the long run, even compared to the costs of treating serious injuries in general. However, in the context of recommending a definition of serious injury to address MAP-21 requirements, the most critical distinction is between medical-outcome-based metrics and police report-based metrics. Using threat-to-life to judge the quality of a serious injury metric will strongly favor a medical- outcome-based metric such as MAIS 3+. Since this requires linkage to medical data, other metrics that are better tied to adverse long-term outcome can be used as well. Data Availability. Data availability is a significant issue in the U.S. at the state level. In state crash databases, only KABCO is readily available. State hospital discharge datasets, where available (not all states have a statewide trauma registry), include ICD-9-CM (and soon ICD-10- CM) codes, and sometimes AIS codes. ICD-9-CM codes can also be translated into AIS codes as described previously. Although state trauma registries generally code whether a patient was in a motor vehicle crash (MVC) and include detail on the driver and occupants, the data are not linked to highway variables, crash configuration, and vehicle damage that would help assess details of the highway system in a state with respect to safety. As of this writing, FHWA is proposing to require only a total count of serious injuries in crashes and the ratio of total serious injuries to vehicle miles traveled at the state level (FHWA, 2014) for MAP-21 reporting. These values can be calculated using only a state hospital discharge or trauma database. However, to tie serious injury to implementation of countermeasures, linked information from the state crash database is required. Current crash data limits analysis of the safety-related performance of roadway, vehicle, and behavior interventions to KABCO-based measures of serious injury based on police reports. However, data linkage between crash and hospital datasets can make AIS- and ICD-based options viable. Detailed discussion of linkage methods will be addressed later in this report. However, efforts to promote good linkage can dramatically change the relative merits of different severity metrics for use in the U.S.. Linkage from crash databases to EMS and hospital databases makes measures of injury severity readily available to analysts and makes AIS or ICD- based metrics viable. These injury codes, in turn, are tied to crash, vehicle, roadway, and environment characteristics that must be accounted for by state DOTs in evaluating the safety performance of their highway systems with respect to serious injuries.

13 Ease of Use. Ease of use is included here because it facilitates widespread use of a new metric that may be unfamiliar to many highway safety planners. Employees in state agencies who are responsible for compiling performance metrics are typically not trained in statistics. Metrics that require complex calculation are both difficult to compute correctly and difficult to understand. Although calculations can be automated, understanding is still critical to the process. Ease of use favors MAIS and “A” over others. MAIS is a simple maximum severity of all coded injuries. Both of these are easy to calculate when the appropriate data are available. ICD- based metrics require more sophisticated analysis as well as further work to standardize them on a national level. Some efforts have been made to standardize the calculations (NCHS, 2004), but the general approach will always be more complex than other methods. 3.4 Conclusions A variety of injury severity metrics, including AIS-based and non-AIS-based metrics were reviewed in Flannagan et al. (2012). In weighing the relative merits of different metrics, much of the crash community has embraced MAIS 3+ as the preferred measure of serious injury. The IRTAD Group recommended using MAIS 3+ as an international standard definition of serious injury in their 2011 Annual Report (IRTAD, 2011). In addition, the European Commission declared in March 2013 that MAIS 3+ should be the metric used in EC countries for this purpose (see European Commission High-Level Group, 2012). It is clear that the ideal combination of available data and demonstrated predictive value are not currently available in most states. This would require linkage from crash to injury outcome. If serious-injury-based metrics in MAP-21 are to be calculated soon, there must be some attention given to near-term solutions. These might include adaptations of “A” injury (e.g., calibration using state trauma databases) or estimation based on sampling of EMS/hospital records for some crash cases. Once high-quality linkage between crash and hospital datasets has been achieved, and mapping from ICD-10-CM to AIS 2005 is available (as through Zonfrillo et al., 2015), any ICD- or AIS-based metric becomes viable. At this point, usability combined with predictive value favors MAIS 3+. This metric consistently performs well at predicting survival, and it is easily calculated. Although ICISS may outperform MAIS 3+, its complexity makes it less appealing as the primary metric for measuring serious injuries in crashes. Because linked crash datasets will contain ICD codes, more sophisticated ICISS- and SRR-based analyses can always be conducted by statistical experts. However, until methods are developed and provided to states to automate the calculation of ICISS, MAIS 3+ will be the easiest and most reliable metric for serious injury. Based on our review, we recommend adopting MAIS 3+ as the primary measure of serious injury for use in MAP-21 performance metrics. Although data linkage needed to use MAIS 3+ to evaluate safety countermeasures has not yet been established in most states, the time has come to move to a medical-outcome-based metric. Diagnosis made by medical professionals is more accurate and reliable than the general injury severity impression of police. Taking the step of defining serious injury in terms of medical diagnosis motivates data linkage, which in turn allows for a much richer and more precise understanding of the relationship between crash, vehicle, and occupant characteristics and injury outcome. The remainder of this report addresses the many practical implications of requiring a medical-outcome-based measure of serious injury.