Chapter 5: Injury Severity Scoring and Outcomes Research

Injuries have long been classified in terms of severity. The world’s oldest known surgical document, the Edwin Smith Surgical Papyrus (ca. 17th century BC), classified 48 traumatic injuries from ancient Egyptian battlefields and construction sites as successfully treatable, possibly curable, or untreatable.¹ Such predictions about patient outcomes, and attempts to quantify the severity of injury, are today the realm of more than 50 published injury severity scores, scales, and models.

Injury severity scoring quantifies the risk of an outcome following trauma and provides metrics based on elements of clinical acumen and statistical modeling to describe aspects of the patient condition post-injury. The primary outcome of interest is commonly survival or a measure of morbidity (eg, complications) or resource use (eg, hospital or intensive care unit [ICU] length of stay [LOS]), but may be any primary endpoint of interest (eg, compliance to an evidence-based procedure).

The injury scores and models discussed here are generally applied using retrospectively collected data. They are therefore suitable for case mix adjustment or risk prediction in groups of patients but are not appropriate for individual patient predictions to guide clinical management decisions. They have a host of applications including quality improvements in patient care, trauma systems and health care delivery, injury prevention initiatives, and epidemiological studies of injury. Patient-specific clinical prediction applications are more simple, requiring decision rules to categorize patients into low-risk/high-risk “bins” and present less sophisticated statistical demand.

This chapter provides a concise description of injury severity scoring and modeling, with particular emphasis on application to outcomes research and quality improvement. Because there can be confusion about the precise meaning of terminologies used in this process, a few definitions are warranted to distinguish the qualitative from the quantitative.

Scores: Injury severity scores place numerical values on injury descriptions, thus enabling standardized communication about the injury state. Scores should be as objective as possible and their assignment should be supported by inter- and intrarater reliability studies.

Scales: Scales are a ranking, based on consensus, assessment, or mathematics and statistics, of risk or severity. Examples include the Glasgow Coma Scale (GCS), which ranges from 3 (deep coma) to 15 (full consciousness) and the Abbreviated Injury Scale (AIS) score, which generally ranks severity from 1 (minor) to 6 (virtually unsurvivable, Table 5-1).

Scales can be

• Nominal: the values of the score are not numerical and have no order

• Ordinal: the values of the score are not numerical but have an order (eg, increasing or decreasing injury severity)

• Interval: the values of the score are numerical and the distance between the levels of the score is consistent both within and across the system of scoring

Models: Modeling is defined here as the use of scoring or scaling systems to create processes for determining probabilities of risk for specific outcomes (eg, mortality, morbidity). The process of modeling is a means of combining or integrating multiple attributes, for example, multiple injuries, or anatomy and physiology. Examples include logistic regression modeling to formulate coefficients of risk for specific elements within a scale, or risk associated with individual group diagnosis. Injury severity models quantify the risk of an outcome after trauma, for both clinical and research purposes.

Outcomes research: Outcomes research is defined as a method of creating “empirically verified information” to better understand how variables in the real-world setting (from injury to treatment) affect a wide range of outcome variables (from mortality to satisfaction with care).²

Risk adjustment: Risk, or case-mix, adjustment entails isolation of the effects to be studied from the effects of other “noisy” factors that can influence the outcome. Risk adjustment is essential for appropriate outcomes analysis. In injury outcomes research, severity scores are essential for risk adjustment because they allow accurate comparisons among disparate patient populations with varied degrees of risk. The goal is to compare populations with similar degrees of injury so that other risk factors (eg, time to treatment, mechanism of injury, injury prevention, and mitigation equipment) may be properly isolated to examine their relationship to particular outcomes. Risk adjustment might be as simple as defining classes of a variable to stratify risk groups or as complicated as using a risk adjustor in a multivariable regression model.³

Collecting reliable, accurate, and timely data on injury care is central to injury research and surveillance, and is the key to reducing the societal burden of injuries. Ideally, routinely collected data would be linked throughout the care continuum from prehospital care through community care. However, in most health care systems, injury datasets are limited to trauma registries and hospital discharge (administrative) databases.

Trauma Registries

One of the earliest civilian trauma registries was the Illinois Trauma Registry, which was started to benchmark and track improvements in trauma center implementation in a small geographic area.⁴ However, it was preceded by an “accidental” trauma registry comprising the Wound Data and Munitions Effectiveness Team (WDMET) data collected in Vietnam in 1967–1969. About 15 years later, the WDMET database was rescued from destruction by a young naval officer and has provided a basis for our understanding of combat injury and future combat trauma registries.⁵ The next significant step was the Major Trauma Outcome Study (MTOS),⁶ which collected data from more than 100 hospitals in North America and Canada, and serves as a model for all trauma registries established since the mid-1990s around the world.

Trauma registries are designed to provide information for quality improvement activities within trauma systems. They contain data on patient demographics, injury circumstances, prehospital care and transport (in some cases), emergency department (ED) and in-hospital interventions, anatomic injury, physiological measurements on arrival, complications, outcomes, and patient destinations. Some trauma registries also include information on comorbidities. The information is generally coded from patient charts by trained data abstractors and injuries are usually described using the AIS lexicon. One of the most complete trauma registries in the world is the Victorian State Trauma Registry (VSTR) in Australia, which provides 3-, 6-, and 12-month follow-up data on functional status for injury patients using an opt-out consent strategy that would not be possible in most jurisdictions.⁷

Trauma registries can be limited to a single provider but are most useful for quality improvement when they contribute to regional, state, or national registries. Trauma registries with a national reach include the US National Trauma Data Bank (NTDB), Canadian National Trauma Registry (NTR), UK Trauma Audit & Research Network (TARN), and the German TraumaRegister Deutsche Gesellschaft für Unfallchirurgie (TR-DGU). Because of its circumstances and integration of military and civilian medical systems, the Israeli Defense Forces (IDF) and Israeli trauma centers have a database that captures both civilian and military injuries, including those resulting from terrorist incidents. Trauma registry inclusion/exclusion criteria vary, with some registries including all injury admissions and ED fatalities, and some applying inclusion criteria based on injury severity, trauma team activation, or LOS.

The most common use of trauma registries is for benchmarking, whereby provider performance is monitored over time or compared to an internal or external standard to identify hospital outliers. Trauma registries are also used for resource allocation planning, injury prevention, and outcome research. Trauma registries most commonly used for research are the NTDB (238 PubMed citations in 2015), VSTR (45 citations), North Carolina Trauma Registry (14 citations), and the Québec Trauma Registry (QTR, 13 citations).

The major advantage of trauma registries is the fact that they include detailed clinical data, essential for accurate benchmarking and outcomes research. Disadvantages include (1) the fact that they are based on retrospectively collected patient chart data subject to data quality issues, (2) they only cover the index acute-care admission for injury, and (3) they are not population-based. Indeed, trauma registries are generally limited to patients treated in designated trauma centers, which can represent a small proportion of regional injury admissions, particularly in exclusive trauma systems. However, trauma registries based on mandatory data collection in inclusive trauma systems such as the VSTR or the QTR generally offer high population coverage (>90% of major trauma admissions in the QTR).⁸ Another important limit of trauma registries is the investment required in terms of efforts and resources. In 2004, the Victorian State Trauma Outcomes Registry and Monitoring (VSTORM) group estimated the cost of the VSTR at approximately A$100 (US $84) per patient.⁹ Despite the fact that this represents a fraction of a trauma patient’s health care costs, it is usually perceived as an onerous financial burden and is a major barrier to trauma registry implementation and maintenance.

Hospital Discharge Data

Hospital discharge data are routinely collected in most high-income countries to meet administrative and reimbursement needs. These datasets generally include patient demographics, injury diagnoses, comorbidities, surgeries, complications, and injuries, coded using the International Classification of Diseases (ICD) lexicon. Advantages of hospital discharge data include population coverage of all hospital admissions and the fact that patients can be tracked through the system via a unique identifier, enabling researchers to obtain information on transfers, readmissions, and pre/post-injury use of health care services. Disadvantages include the lack of clinical information, the absence of injury severity scores, and omission of ED fatalities. In addition, discharge data are often geared to billing and may not include all injury diagnoses. Hospital discharge datasets are most useful when they can be linked to trauma registry data. Systems that enter shared data simultaneously into both databases (as in the QTR) are the most efficient.

Prehospital Data

A major challenge of data acquisition in trauma systems relates to prehospital clinical data. Few trauma registries capture these data, which, because of the emergency nature of prehospital intervention, are particularly hard to acquire and accurately record. For example, most prehospital providers do not ascertain patients’ GCS scores except by gestalt. Further, although they record interventions, abstracting and converting the information into a database is particularly challenging. This is an important hurdle because around half of all civilian trauma deaths occur in the prehospital period. This percentage is greatly increased in low- and middle-income countries (LMICs). Military data reveal that at least 90% of combat deaths occur prior to reaching a medical treatment facility (MTF)¹⁰ and yet the 75th Ranger Regiment instituted a process of complete capture of prehospital data for the decade they were involved in Afghanistan and Iraq.¹¹ This is a fine example of prehospital data collection that influences the process of care, the speed with which it is rendered, and subsequent reduction in preventable deaths.

Other Data Sources

Other useful sources of injury data include emergency medical services (EMS) systems, police reports, rehabilitation facility databases, physician billing records, and community care databases. Unfortunately, there are considerable barriers to linking these data sources, such as lack of unique identifiers, data-protection acts, and lack of harmonization across datasets.¹²

The US Department of Transportation National Highway Traffic Safety Administration (NHTSA) houses two large injury databases related to vehicular crashes, based on linkage of multiple data sources. The Fatal Accident Reporting System (FARS) is a near-census population of deaths from vehicular injury in the United States, which contains data from police reports, EMS, and EDs. The National Accident Sampling System (NASS) is a sample of vehicle crash patient data obtained from prehospital and hospital sources throughout the United States. Both of these databases, although imperfect, have been populated for decades and are a valuable resource for tracking, among other information, prehospital and in-hospital deaths from vehicle crashes over the years (Fig. 5-1).

Data Quality

Trauma registries generally have rigorous data quality assurance programs to ensure the accuracy of data extraction from medical charts. However, the quality of information in the medical charts themselves remains an important problem. Examples of common data quality issues in trauma registries include underreporting of complications, underestimation of injury severity for early deaths, and missing data on physiological parameters such as the GCS and respiratory rate (RR). Many clinicians ignore these data quality issues and mistakenly believe that large numbers of patients in a dataset ensures quality. However, whatever the sample size, systematic variations in data quality across providers can lead to selection or information bias, whereby observed differences in performance across injury groups is due to differences in coding practices rather than quality of care. These data quality problems can be partly addressed in the analysis phase through sensitivity analyses and strategies to simulate missing data (eg, multiple imputation). However, analytical solutions will never replace high-quality data, and every effort should be made to ensure the accuracy of collected data.

Trauma researchers must be familiar with injury scoring schemes to accurately risk-adjust injury group comparisons in an attempt to isolate the effects of an independent predictor variable on a dependent outcome variable. Four types of severity scores are typically used: (1) anatomic injury scores, (2) physiological scores, (3) comorbidity scores, and (4) combinations of the three and other factors. The scores vary considerably in complexity of calculation and ease of use. Injury severity score selection should be based on a clear sense of what one wants to measure and why, and a good understanding of their strengths and limitations.

Anatomic Injury Classification Systems, Scoring Systems, and Models

Anatomic injury classification systems, scores, and models describe the site of injury and extent of damage. Many anatomic scores have been proposed in the literature, but this review is limited to general scores, scales, and models that have gained practical acceptance. We will not review the huge number of specialty classification and grading systems such as the AO Foundation/Orthopaedic Trauma Association (AO/OTA) Classification of Fractures and Dislocations. Most general scoring algorithms are designed to predict mortality (Table 5-2) and are not specifically validated for other outcomes, such as LOS or functional status, although moderate correlations may exist. The majority of scores are based on either the trauma-specific AIS coding classification or the more general ICD-9-CM taxonomy. Injury scoring and modeling systems are continuously being revised, tested, and compared to each other, and no consensus has been reached on which is best. It is likely that the optimal scoring system is contextual and therefore depends on the research question at hand and on the population under study.

Anatomic Injury Classification Systems

Abbreviated Injury Scale (AIS)

The AIS was first conceived as a system to define the type and severity of injuries arising from aircraft and motor vehicle crashes.¹³ It began with a dictionary of 73 injuries in 1971 and now has more than 2000, adding to its precision but also to its cost and complexity. Major revisions to the AIS occur every few years, (eg AIS85, AIS90)^14,15 with a new update in 2015. To calculate AIS scores, medical records are transcribed into specific codes that capture individual injuries. The AIS is a proprietary classification system requiring specialized training for coding personnel. Because of this, AIS is not captured at every hospital.

AIS divides the body into nine regions: head, face, neck, thorax, abdomen, spine, upper extremities, lower extremities, and external. The actual AIS code consists of two numerical components separated by a decimal point. The first is a six-digit injury descriptor “pre-dot” code, which classifies the injury by region, type of anatomic structure and specific structure injured, and level of injury. The “post-dot” component is a severity score ranging from 1 (minor) to 6 (virtually unsurvivable), as shown in Table 5-1. AIS severity scores are consensus-derived assessments assigned by a group of experts. The maximum AIS (MAIS), which is the highest AIS severity among all of a patient’s injuries, is often used to quantify a patient’s injury severity. This score is highly correlated with mortality but ignores information on concomitant injuries.¹⁶

AIS has many intrinsic shortcomings, including anatomically incorrect body regions, inability to code combat injuries,^17,18 and inconsistent scaling of severity across body regions.¹⁹ For example, an AIS 3 in the abdomen is associated with a higher probability of mortality than an AIS 3 in the chest or head. Even when performed by experts, AIS coding has low interrater reliability and 65–82% of AIS codes are not used, even in the largest registries.²⁰ Further, the limitations of the AIS are carried forward and multiplied when they are used as the basis for models (eg, the Injury Severity Score [ISS], discussed below).²¹ Redactions of AIS are now available for both civilian and military injury databases.

International Classification of Diseases

Owned by the World Health Organization, the International Classification of Diseases (ICD) is not injury specific, but is a general, all-purpose diagnosis taxonomy for all health conditions. It is well over a century old and is currently in its 11th clinical modification (ICD-11). In most countries, the tenth clinical modification, ICD-10, is being used.²² However, in the United States, ICD-9-CM is the standard for coding hospital diagnoses including injuries.²³ In the ICD-9 lexicon, ICD codes exist for more than 10,000 medical conditions, about 2000 of which are physical injuries (the block of ICD-9-CM codes from 800.0 to 959.9 encompasses traumatic injuries). For trauma researchers, AIS codes are generally preferred over ICD-9 because of their greater specificity of injury description (the pre-dot classification) and the availability of an injury severity classification (post-dot code).

Organ Injury Scale

In 1987, the American Association for the Surgery of Trauma (AAST) introduced the Organ Injury Scale (OIS).²⁴ The goal of the scale was not to predict outcomes, but to standardize the descriptive language of injuries to improve communication between trauma surgeons and other physicians. Like the AIS, the OIS provides an ordinal scale to each level of organ disfigurement, ranging from grade 1 (relatively minor) to grade 5 (likely fatal). As such, it emulated similar efforts by orthopedic surgeons to classify fractures that go back decades. These scales exist for 32 organ and body region systems.^{25,26,27,28,29,30} The OIS is rarely used in modern injury outcome research.

Barell Injury Diagnosis Matrix

The Barell Injury Diagnosis Matrix is a matrix of ICD-9-CM codes for classifying injury diagnosis by type and anatomic region into “injury profiles.”³¹ The Barell Matrix enables epidemiologic, management, and clinically oriented analysis by serving as a standard for case-mix comparison and characterization of injury patterns. The matrix consists of 12 columns based on ICD-9-CM sequence representing nature of injury (eg, fractures, amputations), and rows with varying levels of detail relating to body region (eg, head and neck, traumatic brain injury [TBI]). Using this matrix, data can be aggregated into summary reports that indicate injury distribution, enabling descriptive comparisons.

Anatomic Injury Scoring System Models

Current anatomic injury models are based on the AIS or the ICD, and may also be based on injury scores derived by expert consensus (AIS injury severity score) or derived empirically (probabilities of mortality calculated using large injury databases). AIS-based consensus models include the Injury Severity Score (ISS), the New Injury Severity Score (NISS), and the Anatomic Profile (AP). AIS-based empirical models include the Trauma Registry Abbreviated Injury Score (TRAIS) and the Trauma Mortality Prediction Model (TMPM). ICD-based models are all derived empirically and include the ICD Injury Severity Score (ICISS) and its derivatives, as well as the OIS.

Injury Severity Score

In 1974, Baker et al first posited a multi-injury score by introducing the Injury Severity Score (ISS).³² Injuries in each AIS region are given an AIS score and the highest AIS scores in the three most severely injured regions are squared and summed to form the ISS. The ISS range is an ascending scale of severity from 1 (least severe) to an assigned 75 (originally designated as unsurvivable). AIS 6 automatically translates to ISS 75.

ISS correlates with mortality and remains for researchers the most widely used anatomical scoring system. However, ISS has many limitations.³³ Its association with the probability of mortality is neither smooth nor monotonic. Further, it only considers one injury in each body region and thus ignores important injury information, especially in penetrating and ballistic injury (Table 5-3).^18,34 Because of these shortcomings, we continue to believe ISS should be retired and replaced by one of the more modern injury models now available (see below).

New Injury Severity Score

The New Injury Severity Score (NISS) was formulated by Osler et al to address ISS shortcoming in quantifying multiple occurrences of serious injuries within the same body region.³⁶ NISS is the sum of the squares of the three most severe AIS severities, regardless of body region and, as with the ISS, AIS 6 automatically translates to NISS 75. This permutation offers a slight predictive advantage but has several of the same shortcomings as the ISS.

Anatomic Profile

The Anatomic Profile (AP), developed by Sacco et al, uses AIS severity scores with an adjustment for differences across body regions.¹⁸ Three modified components are weighted to form a single scalar based on anatomic location of all serious injuries (AIS >3). The AP overcomes many modeling defects inherent in ISS and NISS and avoids the excessive complexity and nontransportability in scores such as the TMPM.³⁷ The fact that the AP has not supplanted the ISS, however, is probably a reflection of the general lack of understanding of injury severity modeling in the injury research community.

ICD Injury Severity Score

The ICD Injury Severity Score (ICISS), created by Osler et al, took an empirical estimation approach to injury severity scoring with the formulation of ICD-9 survival probability (P_s).³⁸ P_s is an ICD-9 code-specific estimate of the survival probability associated with that particular injury. For a set of patients, P_s for a particular injury code is the number of patients who survive that injury divided by the number of patients who display the injury. The traditional ICISS is calculated as the product of P_s for as many as 10 injuries, and ranges from 0 (unsurvivable) to 1 (high likelihood of survival). Other versions of the ICISS include P_s of the worst injury, and independent P_s calculated on patients with isolated injuries. A similar approach was used in the 1970s, with each ICD code being given a conditional and definitive P_s related to the presence of other injuries.³⁹

ICISS offers several advantages over other anatomic scores. First, because it is based on the ICD coding lexicon, it can be used in any clinical setting, including smaller centers that typically do not perform AIS coding. Second, unlike the consensus-derived AIS severity scores, ICISS’ empirical approach enables powerful statistical estimates of injury-specific survival if the patient population used to derive them is large enough. Consequently, unlike ISS and NISS, ICISS is a smooth, if nonlinear, function of mortality.

ICISS does, however, have limitations. First, although it resembles an overall probability, ICISS can only be considered a scalar because P_s is often “contaminated” by patients with multiple injuries. Independent P_s can be calculated as mentioned earlier, but these are not available for all codes because many injuries rarely occur in isolation.⁴⁰ Second, P_s is database-specific and the degree to which it is applicable in injury populations with different patient characteristics remains uncertain.⁴¹ Finally, because empirically based injury severity scores are based on observed all-cause mortality, the ICISS is not an independent measure of anatomical injury but inherently incorporates physiological reserve and physiological reaction to injury, which vary across injury diagnoses.

Trauma Registry Abbreviated Injury Score

ICD-9 codes are nominal, meaning they are unordered, qualitative categories not ranked by severity. If one ignores the AIS severity score, AIS codes can also be treated nominally, taking advantage of their specificity in injury classification. As such, AIS injury descriptor codes can be used to calculate P_s, similar to the ICISS. The Trauma Registry Abbreviated Injury Score (TRAIS) is the product of AIS-derived P_s. Kilgo et al showed that ICISS and TRAIS behave similarly in a large group of patients coded both ways and that TRAIS predicts mortality more accurately than its AIS counterparts ISS, NISS, and AP.⁴² Although ICISS and TRAIS are derived from two different coding systems, they behave similarly in terms of their association with mortality. This suggests that empirical approaches might obviate the inherent structure of the coding systems.

Trauma Mortality Prediction Model

The Trauma Mortality Prediction Model (TMPM) is also based on P_s associated with AIS codes, generated using NTDB data. It adds a level of complexity as it is calculated as a weighted sum of coefficients from two probit models of mortality: one based on AIS pre-dot codes and one on body region injured.

ICD-MAP

To address the lack of an injury severity scoring mechanism in the ICD classification system, MacKenzie and colleagues developed an algorithm to convert ICD-9-CM codes to AIS90 codes.⁴³ The algorithm has more recently been updated to map ICD-10 codes to AIS98 codes.⁴⁴ Mapped AIS codes can be used to calculate AIS consensus-based severity scores including the ISS, NISS, and AP. This algorithm has proven useful for quantifying injury severity in administrative databases but mapping is not achieved for all ICD codes, and the performance of mapped AIS injury severity scores for predicting mortality is significantly inferior to directly coded AIS severity scores and ICD-based scores (ICISS).^16,45

Trauma clinicians, outcomes researchers, and hospital administrators may ask which of these approaches is best. There is no consensus, and many publications each year continue to debate this question.

Several large studies, including ones by Sacco et al and Meredith et al, compared these anatomic scores in terms of their ability to predict mortality.^16,46 Both studies found that AP and ICISS better discriminate survivors from nonsurvivors than ISS, NISS, and the ICD-MAP versions of ISS, NISS, and AP. A surprising finding was that MAIS performed better than its multi-injury counterparts ISS and NISS. Based on this result, Kilgo et al showed that the patient’s worst injury, regardless of coding lexicon (ICD-9 or AIS) or estimation approach (AIS severity-consensus or empirical P_s), was a better predictor of mortality than multi-injury scores.⁴² Harwood et al. reported that the NISS was better than the ISS and equivalent to the MAIS in the prediction of mortality in blunt trauma patients.⁴⁷

However, many of these studies have not accounted for other risk factors inherently incorporated into empirical scores (age, comorbidities, physiological reaction). Studies that do not account for these risk factors when comparing consensus-based and empirical scores unfairly disadvantage the former.⁴⁸ As mentioned earlier, each of these scores has strengths and limitations, and the choice of score should be adapted to the outcome, data, and study population at hand.

Physiological Scoring Systems and Models

Anatomic injury unleashes a set of physiological and bio-chemical consequences that require modulation and mitigation as part of the patient’s treatment. The concept of integrating physiology into injury severity modeling recognizes the dynamic and time-dependent changes to physiological status following injury. A patient with a ruptured spleen who is seen within 1 hour of injury might be normotensive and, when appropriately and promptly treated at a trauma center, will have a very low risk of dying. A patient with the identical anatomic injury who is seen 4 hours later, for whatever reason, may present with a systolic blood pressure (SBP) of 60 mm Hg and a significant risk of death despite availability of the same system of in-hospital care. The hospital should not be penalized if the patient dies (but perhaps the system should). Integration of physiological parameters on arrival with other data is therefore essential for accurate case mix adjustment and outcome prediction in injury research.

Early death following injury occurs as a result of central nervous system (CNS) impairment, hemorrhage, or respiratory causes or combinations of causes. Clinical markers, including RR, SBP, base deficit, and reaction to stimuli/state of consciousness are important prognosticators of outcome and are routinely used in clinical management. However, unlike anatomic injuries and preexisting comorbidities, which are fixed at the time of hospital admission, physiological parameters are ever-changing, both spontaneously and in response to therapy. Thus, it is necessary to obtain a “snapshot” of physiological status at one point in time, usually immediately upon ED or trauma center arrival. The main physiological scores currently in use in injury research include the GCS and the Revised Trauma Score (RTS).

Glasgow Coma Scale (GCS)

The GCS was first proposed by Teasdale & Jennett as a means to directly triage brain-injured patients and to monitor postoperative craniotomy patients.^49,50 The GCS was subsequently integrated into the Trauma Score/Revised Trauma Score (RTS) and Triage Score to describe level of consciousness without using subjective terms such as “semicomatose” and “lethargic” to classify head injury following trauma.⁵¹ GCS measures brain function via three components: (1) motor (GCS-M), (2) verbal (GCS-V), and (3) eye opening (GCS-E), each with ordinal characterizations of severity (Table 5-4). The scale ranges from 3 (completely unresponsive) to 15 (fully conscious) and the GCS score has been shown to be strongly correlated with survival.⁵² The motor component alone has been shown to be almost as powerful as the full GCS score,⁵³ and could theoretically replace the full GCS score for predicting mortality. However, the verbal and eye components discriminate noncomatose patients and are thus valuable for predicting nonfatal outcomes.⁵⁴

Revised Trauma Score

The Trauma Score, later updated to the Revised Trauma Score (RTS), was designed by Champion et al as an approach to combining clinical and observational physiological data into one score.^55,56

Two forms of the RTS exist, one for triage (Triage-RTS) and one for outcomes evaluation and risk adjustment (RTS). Both are based on the GCS, SBP, and RR (Table 5-5). The Triage-RTS is calculated by summing the coded values for each of the three variables, and ranges between 0 and 12. The RTS equation for outcomes evaluation computes indexed values of GCS, SBP, and RR by weighting their coded value with logistic regression coefficients and summing them.

The RTS ranges from 0 to 7.84, with lower scores translating into more physiological derangement. RTS correlates strongly with mortality⁵⁷ and remains important in injury scoring through its contribution to the TRISS model (see below). Studies have also shown that the combined use of SBP and GCS-M are just as effective at predicting patient survival as the RTS.⁵⁸ Disadvantages of the RTS include the fact that coefficients are based on MTOS data and have not been updated, and that categories of the GCS, SBP, and RR used to calculate the RTS often have very sparse data.⁵⁷

Comorbidity Scoring Systems

Injury outcomes research has long recognized the importance of comorbidities to patient risk and outcomes. For that reason, comorbidities were integrated into the American College of Surgeons Committee on Trauma (ACS-COT) field triage decision scheme developed by Champion.⁵⁹ Morris et al, among others, identified several preexisting conditions that worsen prognosis following trauma, most notably liver cirrhosis, chronic obstructive pulmonary disease (COPD), congenital coagulopathy, diabetes, and congenital heart disease.⁶⁰ Morbid obesity has now been added to this list.⁶¹ The incorporation of preexisting conditions into injury severity models is difficult because so many potential comorbidities exist, each of which may occur with variable severity. Further, many are relatively rare and may be inconsistently recorded.

Specific comorbidity adjustments, such as the Charlson Comorbidity Index (CCI), which are widely used in other disciplines,⁶² are frequently used in injury severity models in attempts to enhance their predictive abilities. Results, however, have been poor.^63,64 This may be because such scores are not adapted to acute injury populations.⁶⁴ Indeed, the CCI, a weighted sum of 17 preexisting conditions, is based on coefficients derived in a population of cancer patients using a Cox proportional hazards model and is therefore clearly not appropriate for injury admissions. The number of Charlson comorbidities has been shown to predict injury mortality as well as the CCI.⁶⁴ Other approaches include using the presence of individual comorbidities or classes of conditions (ICD-9-CM ranges) in risk-adjustment methods or simply using patient age as a surrogate for comorbidities. With the aging of trauma populations, comorbidity and multimorbidity will increase, and accounting for these factors in injury research will become increasingly important. Efforts should therefore be made to develop an injury-specific comorbidity index.

Combat Injury: A Special Case

Since the addition of descriptors for coding penetrating injuries with the AIS 1985 edition, researchers have had a tool for evaluating both blunt and penetrating injuries. The descriptors of penetrating injuries included in AIS versions since 1985 describe low-kinetic-energy injuries treated in civilian trauma centers and hospitals. Subsequent iterations have been used to code military combat injuries as well.⁶⁵ These codes, however, did not adequately describe commonly seen penetrating combat injuries such as multiple and massive soft-tissue fragment wounds; high velocity penetration; blast overpressure injuries (mutilating or nonmutilating), and/or bilateral and multiple injuries^17,18 that result from explosive devices including improvised explosive devices (IEDs), which account for 55–75% of combat injuries.¹⁰ They also did not account for, nor were they designed to, some of the injury phenomena associated with mass casualty incidents, for example, crush injuries in earthquake disasters.

Abbreviated Injury Scale—Military Edition (AIS-Mil)

To address these issues, a committee of military physicians was formed to work with the International Injury Scaling Committee (IISC) of the Association for the Advancement of Automotive Medicine (AAAM) to propose guidelines for developing a version of the AIS specifically for coding combat injuries. These physicians represented all three services of the US military (Army, Navy/Marines, Air Force) as well as a spectrum of medical specialties relevant to combat casualty care including emergency medicine and trauma, orthopedic, neurosurgery, and general surgery. AIS 2005-Military is used for coding of all injuries in the three combat trauma registries: (1) the DoD Trauma Registry (DoDTR, formerly the Joint Theatre Trauma Registry [JTTR] based in San Antonio, TX; (2) the Navy/Marine Combat Trauma Registry (CTR) Expeditionary Medical Encounter Database (EMED) based in San Diego, CA; and (3) the Mortality Trauma Registry (MTR) based at the Office of the Armed Forces Medical Examiner (OAFME) in Rockville, MD. These registries also code in AIS 2005 (civilian version) and AIS 1998 for future comparisons with civilian trauma registry data.

Development of AIS 2005-Military coincided with the revision efforts that would culminate in publication of the civilian AIS 2005, which included additional expanded descriptors for orthopedic trauma based on the OTA scale and expanded bilateral injury codes, particularly for vessel injuries. The same consensus model used in determining changes to each injury description by the IISC was used to determine AIS 2005-Military scores.

Military Combat Injury Scale

Despite revisions that culminated in the AIS-Military, numerous combat injuries such as those caused by explosive devices still could not be coded or adequately described. Trying to adapt AIS was only moderately successful. Therefore, the Military Combat Injury Scale (MCIS)²⁰ was drafted by a large panel of military and civilian experts. First, a more anatomically correct and militarily relevant set of body regions was developed (head and neck, torso, arms, legs, multiple), five combat severity levels were determined (minor through likely lethal), and combat-relevant injury descriptions were tabulated.

Using these new body regions, severity levels, and injury descriptors, a five-digit MCIS coding scheme was developed and 269 codes were assigned. Digit 1 indicates injury severity; digit 2 indicates body region; digit 3 indicates the type of tissue involved; and digits 4 and 5 together indicate the specific injury when combined with digits 1, 2, and 3. This coding scheme allows for injuries to the skull and brain to be identified separately from injuries to the face or neck, and for injuries to the chest, abdomen, and pelvis to be separately identified despite being assigned to the same body region. It also allows for identification of unilateral or bilateral injuries, right or left for specific injuries, and easy identification of junctional-area vascular injuries.

Military Functional Incapacity Scale

The Military Functional Incapacity Scale (MFIS) was developed at the request of military personnel to correlate immediate functional impairment with MCIS injury severity for ground troops, and later for shipboard environments.²⁰ The ground operational requirement is based on the ability of an injured combatant to (1) communicate, (2) move, and (3) fire a weapon. The MFIS was developed as an ascending scale of functional impairment with four levels, as follows: (1) able to continue mission, (2) able to contribute to sustaining mission, (3) lost to mission, and (4) lost to military.

MFIS levels of incapacity were linked directly to MCIS injury severity. MCIS Severity 1 injuries are not associated with immediate functional incapacity and casualties are able to continue with the mission; MCIS Severity 2 injuries usually result in immediate functional impairment with the potential for the casualty to contribute to the mission; and MCIS Severity 3, 4, or 5 injuries require medical treatment—casualties who sustain one or more of these injuries are lost to the mission or to the military. Specific Army and Navy scales have been developed.

Combination Injury Severity Models

Combination injury severity models attempt to combine some or all of the three concepts of risk described by MacKenzie⁶⁶: (1) pre-injury physiological reserve (eg, age, comorbidities), (2) physiologic status of the injured patient (eg, GCS, RR, SBP), and (3) anatomic injury severity (eg, ISS, NISS, MAIS, ICISS).

Injury severity models are generally used to quantify patient case mix or to perform adjusted comparisons across injury groups. To date, most injury severity models have been based on mortality but models based on nonfatal outcomes (eg, complications, readmissions, resource use), have more recently been proposed.^67,68,69 The most common use is probably for institutional benchmarking.

The most well-known injury severity model is the Trauma Injury Severity Score (TRISS).⁷⁰ Documented limitations of TRISS (discussed in more detail below) include the fact that age is modeled in just two categories, assuming equivalent mortality risk in all patients 55 years of age or older, it does not account for comorbidities, and has other limitations inherent to the ISS and RTS (discussed above).⁷¹ Further, the original coefficients are more than 20 years old, although they have been recently revised.

In an attempt to address these limitations, many other injury mortality prediction models have been proposed. A Severity Characterization of Trauma (ASCOT) includes age modeled in five categories and uses the AP instead of the ISS.⁷² The Harborview Assessment for Risk of Mortality (HARM) is based uniquely on hospital discharge data and includes anatomic injury descriptors (ICD), mechanism, comorbidities, and injury intent.⁷³ The Trauma Risk Adjustment Model (TRAM) includes complications and transfer status, and employs flexible modeling techniques to use all information on continuous covariates (ie, age, GCS, RR, SBP), and to preserve their nonlinear associations with mortality.⁷⁴ Several studies have compared the predictive accuracy of injury severity models, and results indicate that the most complex models offer significantly better predictive accuracy and change the results of trauma center benchmarking analyses.^16,45,48 However, these more complex models have trouble supplanting TRISS.

To address the problem of limited injury data available in LMICs, the Kampala Trauma Score (KTS) has also been proposed. Originating in Uganda, the KTS is a “simplified composite of the RTS and the ISS and closely resembles… TRISS,”⁷⁵ and is calculated along a descending scale of severity, ie, 5 to < 11 (severe), 11–13 (moderate), and 14–16 (mild).

Development of Injury Severity Models

Development of an injury severity model implies rigorous statistical methods in line with guidelines proposed for prediction models.⁷⁶ The purpose of the model must be clearly defined, the model should be derived on a large sample of (representative) patients subject to the highest standards of care, the choice of potential risk factors should be based on literature review and expert opinion in line with a conceptual model, and the model should allow for nonlinear associations with the outcome (eg, the probability of mortality does not increase linearly with age or the ISS and associations with SBP and RR are nonmonotonic). Both the internal and external validity of the model should be evaluated. Some injury severity models published to date have been evaluated in terms of apparent performance (discrimination and calibration on the sample used for derivation) but relatively few have been subject to rigorous validation.

Evaluating Predictive Models

The performance of models is evaluated according to their capacity to accurately predict the outcome of interest. Injury severity models based on binary outcomes (eg, mortality, readmission, complications) are generally based on the logistic regression model. The predictive accuracy of logistic models is evaluated by calculating measures of discrimination and calibration.

Model Discrimination

Model discrimination describes the accuracy of the model for distinguishing between survivors and nonsurvivors, and is generally measured using the area under the Receiver Operating Characteristic (ROC) curve (AUC). This area varies between 0 and 1, where 0.5 indicates a model that discriminates no better than chance alone (noninformative) and 1 indicates a model that discriminates perfectly. Discrimination depends on the frequency of the outcome but, unlike calibration, tends to be relatively stable from one population to another. For example, injury severity models generally have excellent discrimination for predicting mortality (AUC >0.9),^74,77 and good discrimination for complications (AUC = 0.807),⁶⁹ but poor discrimination for unplanned readmission (AUC = 0.65).⁶⁷ This indicates that that baseline risk (physiological reserve, physiological parameters on arrival, and anatomical injury severity) explain mortality well but that complications and unplanned readmissions are explained to a greater extent by other factors such as quality of care. Discrimination is usually considered to be more important than calibration because it cannot generally be improved by modeling strategies.

Model Calibration

Model calibration (or goodness of fit) indicates how well the model fits the data or how closely model risk estimates approximate observed event rates across different levels of risk. Good model calibration is dependent on the data at hand and can, to a large extent, be ensured by appropriate model specification, respecting clinically plausible associations between each independent variable and the outcome of interest.

Calibration is often quantified using the Hosmer–Lemeshow (HL) statistic,⁷⁸ based on the difference between observed and predicted probabilities of the outcome of interest in prespecified risk groups. The HL statistic has several limitations, including the fact that it is sensitive to sample size (a large, statistically significant value does not necessarily indicate poor model fit), is dependent on the risk groups used (deciles or other), and cannot be compared over different patient samples.^76,79,80,81

Calibration should therefore also be evaluated using other strategies, the most useful of which is Cox’s calibration curve. This curve is based on plotting predicted against observed probabilities of the outcome, thus providing a global impression as to how the model fits the data, and enabling the analyst to identify areas where the fit is problematic. The intercept a and slope b of the calibration curve, which should be as close to a = 1 and b = 1 as possible, are useful summary indicators of calibration.⁸²

Models can also be evaluated in terms of explanatory power using, among others, r-squared adapted to binary outcomes, the Akaike Information Criterion (AIC),⁸³ and the Brier score.⁸⁴ AIC is one of several information criteria that can be used in model evaluation^85,86 but is the preferred choice because the underlying concepts emphasize expected predictive accuracy in new data.^86,87 In AIC, information loss is incurred when a model is substituted for the true model. Smaller AIC values mean less loss, so the model with the smallest AIC is the best choice in model comparisons. Large values do not necessarily indicate extreme information loss, however, because AIC increases with sample size. This is one reason that AIC can only be used comparatively within a given study. One caution regarding AIC is that it can be very sensitive to small effects and may favor unnecessarily complex models.

The Brier score is a promising alternative because its decomposition yields some insight beyond a simple misfit summary.^88,89,90 The U-test derived by regressing outcomes on risk estimates is another option that may have greater sensitivity to model discrepancies.⁹¹ Ready availability in computer programs has been one reason for the wide use of the HL test, but improvements in data analysis packages have made the other options more readily available.

Model Validation

Because the performance of predictive models tends to be overoptimistic in the sample used to derive them, predictive models should be validated in the study population from which they were derived (internal validation or temporal validation) and in a completely independent sample (external validation).

The internal validity of a model may be evaluated using split-sampling, cross-validation, or bootstrapping. In split-sampling, the model is derived on a random sample of the study population (eg, 2/3) and it is validated by fitting the same model to remaining observations and calculating metrics of discrimination and calibration. In cross-validation, the sample is split in k samples of equal size. The model is repeatedly derived on one or several subsamples and its predictive accuracy evaluated on the remaining subsamples. In bootstrapping, the whole sample is used to derive the model and it is validated on repeated random samples drawn from the original sample. The advantage of split-sampling is that the validation sample is theoretically independent from the derivation sample (although in practice it has the same characteristics as it is a random sample). However, bootstrapping has been found to be equivalent to split-sampling and is generally preferred because it uses all observations to derive the model, thereby increasing model precision.⁹² The temporal validity of the model can then be evaluating by fitting the model to data collected in the same population at a different time. If the model has acceptable internal and temporal validity, model performance should then be evaluated on a completely independent sample (external validity).

Summary

Current documented limitations do not invalidate the available injury severity models. Indeed, empirical validation studies provide strong evidence that all available models yield risk estimates of acceptable accuracy for groups of patients. The ongoing concerns are how to determine which model is best and how to improve available models. Several trends in recent modeling efforts provide initial answers to both questions. Models that reduce the weight given to secondary injuries relative to primary injuries,⁹³ incorporate interactions between injuries, and utilize better body region information are examples of promising directions for improving the accuracy of outcome predictions.^94,95,96

Multilevel modeling and methods that smooth the risk function (eg, spline regression, fractional polynomials) demonstrate directions for analytic refinement.^97,98,99,100 Data simulation techniques such as multiple imputation improve the feasibility of adding physiologic variables to the current anatomic/demographic models.¹⁰¹ The growing access to extensive databases, improvements in analytic tools, and increased sophistication of substantive models lead to a straightforward conclusion: Today’s models are good; tomorrow’s will be better.

Injury outcomes research aims to improve our understanding of the determinants of optimal injury outcomes with the ultimate goal of reducing the societal burden of injury. Patient outcomes were at one point focused solely on survival but contemporary injury outcomes research has integrated nonfatal outcomes including measures of morbidity and resource use. We are also gradually moving away from a predominate focus on quantitative outcome measures toward more qualitative and subjective measures such as health-related quality of life, chronic functional impairment, and quality-adjusted life years (QALYs). These changes reflect a trauma community that has begun to embrace the World Health Organization’s definition of health as a “state of complete physical, mental, and social well-being and not merely the absence of disease or infirmity.”¹⁰² The objective of this section is to provide an overview of the basic elements of injury outcomes research.

Basics of Outcomes Research

The first step in successful research is the rigorous elaboration of a strong research protocol in collaboration with methodological experts and key stakeholders. There is an increasing trend toward registering and publishing study protocols.¹⁰³ Systematic literature reviews are also quickly becoming an essential preliminary step to a well-designed research project.¹⁰⁴

According to Kane,¹⁰⁵ outcomes research comprises the following five essential steps:

1. Define a research question

2. Develop a conceptual model

3. Identify the critical dependent and independent variables

4. Identify appropriate measures for each

5. Develop an analysis plan

One example of injury research that demonstrates the challenges of conducting studies in injury populations and has defeated many over the past decades is acute resuscitation. The basic steps to outcomes research, including a systematic review of the literature, analysis of retrospective cohort data, pilot data collection, and progression to a randomized controlled trial, are essential to answering the complex research questions that arise in this field. Significant challenges include obtaining sufficient sample sizes (participation from multiple institutions is required), randomizing patients in the acute care phase, and defining appropriate endpoints. The insistence on 30-day endpoints by the US Food and Drug Administration (FDA) when the vast majority of deaths occur within the first 3 to 12 hours following injury represents a significant barrier to interpretable research results, and increases cost and risk of confounders.

Research Question

Central to the success of a research project is the translation of a research question into carefully thought out research objectives and hypotheses. Rigorous research requires significant financial and human resources. Scientific ethics therefore require a demonstration that addressing the research question will significantly advance current knowledge. To do so, the researcher must be able to show, usually through a systematic literature review, that the question has not been adequately addressed in the past and that the research has the potential to have a positive direct or indirect impact on patient outcome. One important aspect in defining the research question is to establish whether the intention is to predict an outcome from a series of independent variables or to explain the association between one or more exposures and the outcome of interest. The injury severity models mentioned above (eg, TRISS, ASCOT) were derived as predictive models to calculate predicted probabilities of mortality for benchmarking purposes, but injury outcomes research often aims to explain the association between an exposure variable (eg, intervention) and an outcome by estimating a measure of association (eg, odds ratio). Predictive and explicative research goals require very different analysis plans. The derivation and validation of predictive models has been covered earlier in this chapter. In this section, we will concentrate on explicative modeling of injury data.

Conceptual Model

Each step of outcomes research is critical, though none more so than refining a research question through the elaboration of a conceptual model. Development of a conceptual model applies to both predictive and explicative research goals but is particularly important for the latter. Prior to building the conceptual model, investigators should carefully define the target population for the study, that is, the population to which they wish to generalize results. For example, in the study of the effect of surgical delay on complications for long-bone fracture fixation, the target population may be all patients 16–64 years of age admitted to a not-for-profit acute care center in a high-income country and requiring long-bone fracture fixation. The conceptual model should be elaborated as a causal diagram using directed acyclic graphs¹⁰⁶ (DAG, Fig. 5-2) and based on information from literature review and expert opinion. DAGs will provide the foundation for understanding how the association of interest between exposure variable(s) and outcome(s) is influenced by confounding, mediating, and moderating variables. In the example below (Fig. 5-2), the association between trauma center care (exposure) and mortality (outcome) is of primary interest. Physiological reserve, physiological reaction, and anatomical injury severity are confounders, interventions are mediators, and transfer from another hospital is a moderator.

Study Sample

Prior to collecting information on the variables of interest, one must identify the study sample that will be used to address the research question. Careful choice of the study sample with clear inclusion and exclusion criteria in line with the target population is important for the external validity of a study. Ideally, the study population would be a random sample of the target population but this is rarely possible. A convenience sample is therefore often used. For example, when studying the association between surgical delay for long-bone fracture fixation and complications, data availability may lead us to include only patients eligible for participation in the trauma registry when ideally we would include all admissions for long-bone fracture fixation within a health system. The consequences of the choice of study sample on the generalizability of results should be carefully thought out and discussed.

Outcome/Dependent Variables

The dependent variable represents the outcome of interest that we want to predict or explain with independent variables in the model under study. According to the quality of health care model proposed by Donabedian,¹⁰⁷ outcomes that should be evaluated to improve patient care include mortality, adverse events, unplanned readmission, resource use, quality of life, and ability to function in daily activities. A 2013 systematic review identified 14 nonfatal outcomes used to evaluate injury care.¹⁰⁸ The most common were complications (35/40 studies) and hospital/ICU LOS (34 studies). Only three studies evaluated ability to function in daily activities, four evaluated unplanned readmission, and none used quality of life.

Outcome variables are frequently dichotomous in injury research because they represent the presence or absence of a health state (eg, mortality, complications). Depending on the design of the study, the outcome may be measured in terms of prevalence, incidence proportion, or incidence rate. For example, in a transversal study on community-acquired infection, the outcome may be measured as a prevalence (eg, presence or absence of infection on arrival). In a longitudinal study on hospital-acquired infection, the outcome may be measured as an incidence proportion or incidence rate (development of the infection over the hospital stay). In injury research based on registry/hospital discharge data, retrospective cohort studies are common and hospital outcomes are generally measured in terms of incidence proportions because events of interest often occur shortly after injury and time to event is of little interest. Incidence rates are common in chronic disease populations and may be more appropriate than incidence proportions for longer-term injury outcomes where loss to follow-up is an issue. Incidence proportions should be based on evaluation of outcome over a fixed period of time (eg, 30-day mortality) but information on postdischarge events is rarely available. Previous research has shown that hospital mortality, 30-day in-hospital mortality, and 30-day in-hospital and postdischarge mortality agree well, at least for patients less than 65 years of age.¹⁰⁹ However, the same is unlikely to be true for complications, which are a common reason for unplanned hospital readmission.¹¹⁰

The choice of outcome variable and the timing of outcome evaluation should be carefully adapted to the research question and target population. Hemorrhagic shock and/or brain injury are responsible for 90% of deaths following injury and 80% of these deaths occur in-hospital within 4 hours.^111,112 Therefore, as mentioned above, in resuscitation research, 4- or 24-hour mortality may be more appropriate primary endpoints than 30-day mortality. The latter includes deaths due to comorbidities and late effects of injury (eg, complications), which may confound intervention evaluations. As mentioned earlier, the inclusion of prehospital deaths, which comprise more than 50% of all injury deaths,¹¹³ represents a significant advantage when systems of care or prehospital interventions are being evaluated.^114,115,116

Independent Variables

If the goal is to predict outcome, independent variables will be risk factors of the outcome, with consideration for the availability of information in populations intended for the prediction model. If the goal is to explain the association between an exposure(s) and outcome, independent variables will include the exposure variables of interest and covariates that influence the exposure(s)-outcome association(s).

Covariates

In an explicative design, covariates are independent variables that are known to influence the exposure-outcome association but whose relationship to the outcome is not of primary interest. They may be confounders, mediators, or modifiers and will be identified by the conceptual model (eg, Fig. 5-2) and confirmed empirically during the analysis stage.

Confounders

Confounding variables are risk factors of the outcome that are also associated with the exposure (Fig. 5-2). Not adjusting for confounder variables in the statistical model will lead to a biased estimate of the exposure-outcome association. For example, if we evaluate the association between treatment in a designated trauma center (exposure) and mortality (outcome) and patients treated in a trauma center are sicker than patients treated in nondesignated hospital, not accounting for injury severity would lead to an underestimation of the beneficial effect of trauma centers on injury mortality. The measure of association between trauma center designation and mortality (eg, odds ratios or relative risks) would thus suffer from confounding bias. In injury research, potential confounders are often identified among risk factors of the outcome that commonly cover the three concepts of baseline risk described by MacKenzie: (1) physiological reserve (eg, age, gender, comorbidities), (2) physiological reaction to injury (eg, GCS, SBP, RR), and (3) anatomic injury severity (mechanism of injury, body regions injured, AIS- or ICD-based severity scores).¹¹⁷ Randomized double-blinded, controlled clinical trials should not suffer from confounding bias because risk factors are evenly distributed across exposure groups so the confounder-exposure association in the study sample is null.

Mediators

Mediator variables are caused by exposure and in turn cause the outcome. For example, in the association between trauma center care and mortality, acute care (in-hospital) interventions would act as mediators. All of the effect of the exposure on the outcome may pass by the mediator or only part of the effect. In outcomes research, we usually aim to evaluate the total effect of exposure on outcome. It is therefore important not to adjust for mediator variables, otherwise we will obtain an estimate of the direct effect only: an underestimation of the total effect. In the example, if we adjusted for in-hospital interventions, we would underestimate the effect of trauma center care on mortality. The distinction between mediator and confounding variables is critical to valid injury research.

Modifiers

Effect modifiers are variables that modify the association of interest. For example, we may expect trauma center care to reduce mortality for major trauma but to have no effect for minor trauma. If this is the case, injury severity would be said to modify the trauma center-mortality association. In outcomes research, potential effect modifiers are identified à priori and usually only one or two are specified because effect modification implies estimating the exposure-outcome association for each category of the modifying variable, therefore increasing necessary sample size and the probability of a type I error (rejecting the null hypothesis when it is true). Note that effect modification is not a bias but rather a natural phenomenon, which is present even under randomization. Global estimates of the exposure-outcome association are valid even under effect modification. Frequent effect modifiers in injury research are age (pediatric, adult, geriatric), injury severity, type of injury (eg, isolated TBI, multiple blunt, penetrating), and transfer status. Stratifying analyses by injury type is particularly informative due to the heterogeneous nature of injury. For example, the pathophysiological sequence to death in patients with isolated head injury is completely different from that for hemorrhagic shock; estimating a global exposure-mortality association for both patient populations may therefore not be very meaningful.

Analysis Plan

Injury outcomes research is widely based on generalized linear models. The scale of measurement for the outcome variable will determine which model should be used. For outcomes measured on a continuous scale (eg, LOS, costs), a linear model will generally be appropriate. However, variables such as LOS or costs are right-skewed (many patients have an LOS <7 days and a few have LOS >90 days) so when sample sizes are small (ie, <100 observations per independent variable), log-linear models or gamma models may be more appropriate.¹¹⁸ Sensitivity analyses should be conducted to evaluate the influence of outliers (eg, truncate LOS at the 95% percentile) and model specification (eg, linear, log-linear, or gamma models) on study results. For ease of interpretation, simpler models should be used instead of more complex models when similar results are obtained.

For injury outcomes measured on a dichotomous scale (eg, mortality, readmission, complications), logistic or log-binomial models are commonly used. Logistic models are easier to use because convergence is rarely a problem (the logit of a proportion is not bounded whereas the log of a proportion has a maximum of 0). However, odds ratios (logistic model) overestimate relative risks (log-binomial model) and this overestimation increases as the outcome becomes more common. For this reason, log-binomial models are generally preferred when the outcome occurs in more than 10% of patients.¹¹⁹ The convergence problems encountered in log-binomial models can generally be overcome by using a Poisson model with robust variance estimates.¹¹⁹ Finally, if long-term outcomes are of interest (eg, mortality up to 12 months post injury), incidence rates may be more appropriate than incidence proportions, in which case the Cox proportional hazards model is usually the model of choice.

Once the analysis model has been chosen, particular attention should be paid to the specification of independent variables. This should be based on à priori hypotheses about the functional relationship between each independent variable and the outcome using literature review and expert opinion. This stage is essential if we are to build a model with good face validity and to respect the postulate of a linear relationship between the independent and dependent variable, required for most parametric regression models. If this postulate is not respected for exposure variables, associations may be underestimated or missed and for confounding variables, the exposure-outcome association may be subject to residual confounding, which can lead to over- or underestimation of the association(s) of interest. In injury research, most continuous independent variables do not have a linear relationship to mortality. For example, for general injury admissions, mortality risk is stable for patients 16–54 years of age and increases exponentially thereafter.¹²⁰ For SBP and RR, the change in mortality risk is not monotonic because risk increases both for patients who are hypotensive/hyperventilate and for those who are hypertensive/hyperventilate. These variables can therefore not be modeled using simple linear terms but should be specified using dummy variables on categories or more complex functions such as fractional polynomials or splines.¹²¹

All of these modeling strategies have advantages and disadvantages. Dummy variables on categories are easy to understand, operationalize, and present. However, it is difficult to identify appropriate cut-points, categories with small sample sizes and nonintuitive step changes in risk. Fractional polynomials (FPs) or splines offer more flexibility and therefore better model fit, but are complex and difficult to present. Simulations suggest that covariables modeled with dummy variables on at least four categories offer equivalent confounding control to FPs or splines.¹²² However, FPs or spline functions are interesting tools for graphically describing the functional relationship between the exposure variable(s) and the outcome. One often overlooked point is that when scores based on probabilities (eg, ICISS) are modeled as independent variables in a logistic (log-binomial) model, they should be logit (log)-transformed to preserve the postulate of linearity.¹²³

As discussed above, an explanatory model contains exposure variable(s) and all potential confounders of the exposure-outcome association, commonly all risk factors of the outcome. However, when sample size is an issue, a more parsimonious model may be required. This may involve a manual backward model selection approach, whereby potential confounders are removed from the model one at a time, starting with those that have the weakest association with the outcome (highest p-value). Covariables are reintroduced into the model if their removal leads to a change of more than 10% in the exposure-outcome association estimate (eg, OR, RR).¹²⁴

Sample Size Guidelines

Sample size and power analysis can quickly become extremely complex with many hypothetical parameters to estimate, particularly for multivariable models and further still for multilevel models. However, several rules of thumb based on simulations are available to simplify the sample size estimation process. In general, for logistic, log-binomial, or Cox models, 10 outcome events per independent variable are necessary to obtain estimates with acceptable bias and precision.¹²⁵ For example, in a study sample of 1000 patients with 10% mortality, not more than 10 independent variables should be included in the model (dummy variables on categories count!). More recent research has suggested that five outcome events per independent variable are sufficient.¹²⁶ For linear models, suggested sample size requirements are 10–25 observations per independent variable.¹²⁷ In injury outcomes research based on specific pathologies (eg, patients in hemorrhagic shock), studies based on single institutions will never reach sufficient statistical power to be meaningful. Collaborative multicenter research is thus essential to achieving sufficient sample size and representivity to verify research hypotheses rather than simply generate them.

Multilevel Models

The generalized linear models described above are only valid under the assumption that outcome probabilities are independent. In many injury research contexts, this assumption is not respected. Common examples include study samples in which the same patient can be counted more than once for different injury events, samples that include several hospitals or trauma systems (cluster design), or studies in which the outcome variable is measured repeatedly over time on each patient (repeated measures design). If traditional methods are used to analyze such study samples, effect estimates (eg, odds ratio, relative risk) will be accurate but variance estimates will generally be too low (increased chance of rejecting the null hypothesis when it is true).

The first example is not generally a problem in injury research because repeat visits for different injury events usually represent less than 10% of injury populations. Cluster samples are problematic, however, and generally require multilevel modeling techniques to correctly estimate variance.^100,128,129 Multilevel models are a simple extension of the traditional models described above, which allow for estimation of different intercepts and/or slopes for different clusters (eg, hospitals). These models can be useful for simultaneously modeling patient-level effects such as age or injury severity, and hospital-level effects such as volume or designation level. Generalized estimating equations are commonly used to account for intraindividual dependence in outcomes in repeated measures designs.¹³⁰

Other Statistical Models for Injury Research

The traditional explicative models described above may not control well for confounding, especially with small sample sizes. Several new methods have been proposed to address this problem, including propensity scores and instrumental variables.

Propensity scores are obtained by modeling the exposure variable on all confounding variables and obtaining a score that is the probability of being exposed given confounders.¹³¹ This score can then be used for matching, or simply included in the exposure-outcome model as a covariate to control for confounding. Propensity scores have been used to evaluate the hypothesis that ICP monitoring leads to a reduction in mortality in severe TBI patients¹³² and to evaluate the advantage of helicopter transport over ground EMS in the same patient population.¹³³ The propensity scores approach is based on the same confounder information as a traditional model so cannot theoretically reduce residual confounding. However, it does offer an advantage in terms of statistical power, particularly when the exposure is common and the outcome is rare.

Instrumental variables are intended to act as a proxy for unmeasured confounders.¹³⁴ Including them in the analysis model should therefore reduce residual confounding if they are chosen carefully. For example, Prada and colleagues used the proportion of resident population served by helicopter ambulance services (at the state level) as an instrumental variable for evaluating the influence of Level I trauma center treatment on return to work¹³⁵ and Haas and colleagues used county-level trauma center transport rate to evaluate the benefit of direct trauma center transport in a regional trauma system.¹³⁶

The major problem with the methods described so far is that they are based on simple unidirectional associations with a single outcome. To improve our understanding of the influence of trauma systems and their components on multiple short- and long-term outcomes, more sophisticated statistical models are required, including computer endpoints. Examples of statistical methods that can better account for the multidimensional time-varying associations in injury research include marginal structural models,¹³⁷ structural equational models,¹³⁸ and microsimulation models.¹³⁹ These models are promising for gaining a better understanding of complex causal patterns and for evaluating the potential impact of modifying interventions on outcomes. However, all of these methods rely on the availability of high quality, multidimensional data.

Common Pitfalls and Shortcomings in Injury Outcomes Research

As in any research domain, injury research is vulnerable to bias, which should be considered in study design, analytical approaches, and interpretation of results. The following are some of the more common challenges of injury outcomes research:

• Evaluating long-term outcomes. Studies based on injury cohorts are often subject to high loss to follow-up over time as patients are more exposed to precarious living conditions than in chronic disease populations. They are also less likely to comply with treatment instructions or follow-up visits.

• Obtaining patient consent. Obtaining consent from the patient or family in a critical care situation is often very difficult. Ways to circumvent this problem include deferred and community consent.

• Multiple admissions due to transfer. Acute injury care often implies multiple hospital admissions due to transfer, particularly in an inclusive trauma system. Trauma registries and hospital-based cohort studies often fail to incorporate information on multiple admissions. This can lead to an underestimation of mortality and complication rates, and resource use including interventions. In many systems, linkage between registry- or hospital-specific databases and hospital administrative data enables investigators to track patients through multiple admissions.¹⁴⁰

• Survivor bias. Injury research based on nonfatal outcomes is complicated by survivor bias.^141,142 For example, LOS is not fully observed (right censored) in fatalities. Therefore, if deaths are included in group comparisons of mean LOS, the latter will be underestimated in groups with higher mortality. If deaths are excluded from analyses, groups with lower mortality will have a high proportion of critically injured patients (those who did not die) and therefore higher mean LOS. In both scenarios, analysis will unfairly favor groups with higher mortality rates. Survivor bias may lead to biased measures of associations between interventions and outcomes, as fatalities may die before they can receive the intervention. It also leads to the detection of more injuries, comorbidities, and complications in survivors than in fatalities. Analytical techniques such as multiple imputation of unobserved outcomes in deaths (eg, LOS, costs), marginal structural models, and competitive risks analyses can be used to address these problems.

• Heterogeneity in injury coding. The validity of intergroup comparisons in injury research is dependent on homogeneous injury coding across groups. Injury coding is particularly susceptible to heterogeneity across data-collection sites, due to local coding conventions and over time due to changes in coding conventions. More obvious examples of the latter include changes in AIS versions; injury severity scores calculated with AIS2008 are higher on average than injury scores calculated with AIS1990.¹⁴³ In such a situation, analyses should be stratified by time period, and temporal comparisons should be conducted with caution. For comparisons between trauma centers or systems, analysts should be aware of inter-site/system differences in coding conventions and perform appropriate sensitivity analyses.

• Missing data on physiological parameters. The GCS and RR are frequently used as covariates for risk adjustment in injury research. However, they are missing in 15–50% of patient files (and thus in trauma registries) due to prehospital sedation, and/or intubation or lack of evaluation in patients with minor extracranial injury.^102,144 Excluding these variables from analysis may lead to confounding bias whereas excluding patients with missing physiological data may lead to selection bias and reduces statistical power. Possible solutions to this problem that have been proposed include using just the motor component of the GCS (can be evaluated in sedated/intubated patients) and simulating missing data using imputation techniques.¹⁴⁵ Perhaps the most promising technique proposed to date is multiple imputation (MI), whereby multiple plausible values are generated by simulation for each missing data value based on an imputation model that includes all independent and dependent variables to be used in the analysis model.¹⁴⁵ This technique takes account of the uncertainty surrounding the real data value in subsequent variance estimates. Much work has been done to evaluate the validity of MI for missing physiological data in injury research and simulations suggest that when the MI model is specified correctly, MI leads to valid effect estimates.^146,148 However, these solutions will never be as good as complete data and should not be used as a replacement for efforts to improve data quality.

• Missing anatomical data. Without autopsy records, anatomical data for early deaths are not fully available; in many jurisdictions, autopsies are not performed.¹⁴⁸ This results in an underestimation of injury severity in early deaths, which leads to information bias in the evaluation of associations between exposure(s) and mortality.

Sensitivity Analyses

Because no observational study is exempt from selection, information, or confounding bias, sensitivity analyses should be used to evaluate the extent of the influence of these biases and any other analytical assumptions on study results. For example, in evaluating the influence of trauma center care on mortality, sensitivity analyses may consist of (1) excluding deaths within 6 hours of arrival to account for underestimation of injury severity in early deaths, (2) restricting analysis to patients aged less than 55 years to assess the potential influence of underreporting comorbidities (cofounder), (3) excluding deaths transferred-in with GCS3 on arrival to address the problem of interfacility transfer for organ donation, and (4) simulating missing GCS data to evaluate the influence of missing physiological data. These sensitivity analyses will not provide definitive answers but will give an idea of the potential influence of bias on study results that can be used in formulating study conclusions.

Comparing quality metrics within or across trauma centers or systems is central to the process of quality improvement. Benchmarking in injury care was initiated with the TRISS model, which has been used for decades to inform trauma care quality improvement activities worldwide. The TRISS model is based on a comparison of the number of expected and observed deaths (usually within a hospital) using the W and Z statistics.¹⁴⁹ The W statistic represents the number of unexpected survivors per 100 patients (observed deaths-expected deaths/100) and the Z statistic is a measure of how likely such a difference was to have occurred by chance alone.

Alternatively, the TRISS model can be used to calculate hospital standardized mortality ratio (SMR), the ratio of the number of observed to expected deaths.¹⁵⁰ Hospitals with a statistically significant W or Z statistic or SMR are considered to have statistically lower or higher mortality than the standard population. Under the original TRISS model, the standard population was the MTOS cohort. Hospitals therefore compared their injury mortality to that of approximately 80,000 patients admitted to 139 North American hospitals between 1982 and 1987.⁶ More recently, coefficients have been generated using the NTDB, allowing comparison with a more modern injury cohort.¹⁵¹ Under the TRISS model, comparisons are adjusted for age, ISS, and RTS, as described above.

Despite its important contribution to trauma care quality, as mentioned above, TRISS has several documented limitations including lack of adjustment for comorbidities and transfer status, consideration of age in just two categories, and the fact that performance statistics (W, Z, or SMR) cannot be used for interhospital comparisons.¹⁵² Perhaps the most important limitation is the standard used for comparison. The MTOS cohort is now outdated and the NTDB cohort used to generate published coefficients is based on voluntary participation in the national registry, with nonuniform inclusion criteria and nonstandardized coding practices. These limitations have been addressed in more recent benchmarking models.

Injury benchmarking has now moved beyond TRISS mortality calculations, and the injury research community has embraced the quality of care model introduced in 1966 by Donabedian.² Donabedian’s model is based on the evaluation of structure, process, and outcome (Fig. 5-3) where structure refers to physical characteristics of the health care provider (eg, presence of closed ICU), process refers to the use of evidence-based clinical interventions (eg, reduction of major joint dislocation within 1 hour), and outcome refers to the result at the end of the care process (eg, mortality, morbidity, resource use). According to this model, improvements in structure lead to improvements in compliance with evidence-based processes, which in turn have a positive impact on patient outcome.

Benchmarking under Donabedian’s model implies the derivation and validation of a series of quality indicators (QIs) that cover the three quality domains according to the following steps: (1) identify structures and processes of care, and patient outcomes on which health care providers should be evaluated; (2) develop a risk-adjustment strategy, when appropriate; (3) establish a benchmark to which providers can be compared; and (4) evaluate the internal and external validity of QI.

Step 1. The goal is to identify structures and processes of care that have a positive impact on patient outcome and measures of outcome that appropriately represent the burden of injury in terms of mortality, morbidity, and resource use. These should be identified using a systematic literature review and a formal expert consensus process. This step is crucial if benchmarking results are to be used to improve care; previous research has provided evidence that there is little correlation between certain quality metrics and actual quality of care.

Step 2. Structure or process QI rarely require case-mix adjustment because they are based on the careful definition of providers/patients eligible for each structure/process (eg, all Level I trauma centers should have closed ICUs and all patients with GCS <9 on arrival should be intubated in the ED). However, appropriate risk adjustment is central to the process of benchmarking outcomes. Development of outcome QI therefore implies the additional step of deriving and validating an injury severity model, as discussed above.

Step 3. The benchmark can be based on (1) a consensus-based standard (eg, all patients with major trauma should have an ED stay <1 hour); (2) an empirical, internal standard (eg, global mean in the system under evaluation); or (3) an empirical, external standard (global mean in an external population, such as the NTDB used for the TRISS model). The external standard has the potential to provide valuable information for quality control programs but should be based on an injury population with similar characteristics, subject to exemplary quality of care and high quality data.

Step 4. The evaluation of QI validity implies providing evidence of (1) face validity (QIs appear to measure quality), (2) content validity (QIs contain all elements necessary to measure quality), (3) construct validity (QIs correlate with other measures of quality), (4) predictive validity (QIs predict patient outcome or future performance), and (5) discriminant validity (QIs discriminate between providers). The selection of QIs through literature review and expert consensus should ensure both face and content validity. Construct validity can be evaluated by assessing the correlation between QIs measuring the same quality construct (structure, process, outcome). Predictive validity can be evaluated by assessing the correlation between structure-process QIs and process-outcome QIs (under Donabedian’s model, structure influences process which, in turn, influences outcome). Predictive validity should also be verified by assessing forecasting properties, ie, care provider results on QIs in one time period correlated with those in a following period. Finally, discriminant validity can be evaluated by assessing whether a QI detects statistically significant variation in performance across providers. Once the internal validity of QIs have been demonstrated using all these metrics, their external validity should be evaluated in a new, independent patient sample.

Several research groups have proposed using QIs to evaluate injury care. In 1993, the ACS-COT introduced 22 audit filters to evaluate trauma care in terms of clinical processes.¹⁵⁴ More recently, Stelfox and colleagues performed a systematic review to identify evidence-based QI followed by an international expert consensus study according to UCLA-RAND guidelines.¹⁵⁵ They generated a list of 31 QIs that experts consider should be used to evaluate the quality of injury care.^155,156 Inspired by this work, Moore, Stelfox, and colleagues derived and validated a comprehensive tool to evaluate the quality of acute injury care. The tool includes (1) a composite QI for evaluating trauma center structure including elements of commitment, trauma program, and procedural protocols based on ACS accreditation criteria¹⁵⁷; (2) 15 clinical process QIs as well as composite process QIs based on compliance with best practice clinical processes^158,159; and (3) outcome QIs to evaluate mortality, complications, unplanned readmission, and hospital LOS.^67,68,160 The face, content, construct, predictive, and discriminant validity of these QIs has been demonstrated but their external validity has yet to be evaluated.

Investigators have provided evidence that their injury care quality evaluation tool reproduces the structure-process-outcome associations of Donabedian’s conceptual model.¹⁶¹ In a mature, inclusive trauma system, trauma centers with higher scores on the composite structure QI also had higher scores on the composite process QIs, and centers with high scores on the process QIs also had lower risk-adjusted mortality, complications, readmission rates, and mean LOS. In addition, a significant positive correlation was observed among all QIs on outcomes; hospitals with lower risk-adjusted mortality also tended to have lower rates of readmission and complications and lower mean LOS.

Finally, many trauma centers/systems have subscribed to the Trauma Quality Improvement Program (TQIP) benchmarking system, a for-profit organization that receives and processes institutional data to produce benchmarking results for mortality (SMR), LOS (proportion of patients with excess LOS), major complications, and selected clinical processes, based on the NTDB framework.¹⁶² TQIP uses extensive data quality-control mechanisms to account for heterogeneity in coding practices across participating hospitals but is still faced with the challenge of variations in injury coding (AIS90/AIS05/AIS08/ICD). Details on how TQIP models were derived have been published,¹⁶³ but information on model coefficients and validation of risk adjustment models is unavailable.

International Injury Care Comparison

Benchmarking within a trauma system is an essential part of quality improvement activities but international comparisons across health care systems provide a much greater opportunity to improve injury care and reduce the global burden of injury. There is clear evidence that trauma systems improve mortality, morbidity, and functional outcome following injury^{141,164,165,166} but it is unclear which aspects of trauma systems lead to optimal patient outcomes. Many countries have or are in the process of adopting a trauma system model but trauma systems worldwide have very different structures and levels of integration. International comparisons of injury care and injury outcomes therefore provide a unique opportunity to identify which components of trauma systems improve patient outcomes and which offer no advantage. Gabbe and colleagues have made considerable progress in international comparisons of trauma systems by comparing mortality and functional injury outcomes in Victoria, Australia to those in the United Kingdom¹⁶⁷ and Hong Kong.¹⁶⁸

In addition, using the Utstein template to develop a pan-European trauma registry, Ringdal et al now have tools to compare injury outcomes across Europe.¹⁶⁹ However, to truly improve our understanding of how differences in structures and processes of care across systems influence patient outcomes, efforts should be made to build an international injury dataset including high-, middle-, and low-income countries. We also need to develop a standardized methodology to benchmark injury structures, processes, and outcomes on an international scale. A group of international injury researchers is currently working toward this goal. The first steps are to conduct a systematic review of international injury comparisons, to compare population-based injury outcomes using WHO data, and to establish a framework for an international trauma registry.

Challenges of Benchmarking

Many health care administrators and policymakers are moving toward making providers accountable for their performance by publishing benchmarking results (eg, health care report cards) or linking them to financial incentives such as pay for performance. However, rather than stimulating performance improvement, research has shown that the threat of institutional stigmatization or losing resources often leads to data gaming (eg, overestimation of injury severity, underreporting of complications), refusal of high-risk patients, and concentration on quality elements that are measured at the detriment of those that are not.¹⁷⁰ In addition, systematic variation in results across providers may be due to provider-level quality of care that can be modified by the provider, but may also be due to system-level problems or poor data quality. It has therefore been suggested that benchmarking activities should be used on an informative basis to identify potential problems and implement solutions as part of a quality loop (plan-implement-evaluate).¹⁷¹ Future research should work toward developing QIs across the trauma care continuum to evaluate both trauma systems and trauma centers. This implies building datasets to track patients throughout the health care system and evaluate long- as well as short-term outcomes.

The MTOS of the late 1980s⁶ was the first to compare institutional outcome of trauma patients while controlling for case mix. This process is now instantiated in trauma systems throughout the world and despite some hiccups has resulted in the NTDB, which contains data on more than 4 million patients, and the ACS TQIP, which has more than 200 subscriber institutions. These approaches have been vastly improved upon by taking into account structure and process as well as outcome, as espoused by Lynne Moore.¹⁵⁷ This Donabedian structure-process-outcome approach to quality improvement in trauma care has been tested most significantly in Canada, where it has been shown that institutions with the strongest structural measures and best processes have the best outcomes.¹⁵⁹ In the United Kingdom, TARN has developed processes by which international and UK institutional comparisons for trauma care can be made.¹⁷² The UK results are publically available (https://www.tarn.ac.uk/Content.aspx? ca=15).¹⁷³ If a hospital is a consistent outlier, the problem resolution goes to the institutional level where structure, process, and outcome are evaluated together with the influence of individual surgeon, systems, and teams; as well as availability of resources.

To improve our understanding of the complex associations underlying the burden of injury, we need to improve the quality and coverage of injury data, employ more sophisticated analytical methods, improve the knowledge translation-to-action cycle of research results, and look toward comparing outcomes across health care systems in high-, middle-, and low-income countries.

Accurate injury research depends on high quality data. Trauma registries and other injury data sources worldwide suffer from similar data quality problems. Some of these problems, such as missing physiological data and underreporting of complications, have been documented and potential solutions, such as data imputation, have been proposed. However, to better understand the impact of poor data quality on injury outcomes research, we need to improve our understanding of the presence and extent of injury data quality problems and the impact of data quality on outcomes research. In addition, we need to develop standardized methods to measure, monitor, and benchmark injury data quality to drive data quality improvements.

Injury research is currently based largely on the acute (hospital) phase of care in trauma centers. Research has shown that the introduction of trauma systems within a public health care model leads to better short-term patient outcomes. However, we need to improve our understanding of which components of trauma systems drive optimal patient outcomes. To do so, population-based data from all components of the care continuum including prevention, prehospital, emergency, acute care, rehabilitation, and community care are needed. We need to be able to track all injured patients through the health care system and obtain information on long-term outcomes including quality of life and functional status. Some of these data are already available in many systems but data-protection acts and lack of unique identifiers often prevent database merging. Fortunately, fast progress is being made in methodology related to the acquisition, linkage, and treatment of big data, which should lead to breakthroughs in injury data coverage in the near future.^174,175,176

Obtaining high-quality, high-coverage injury data represents a much greater challenge than employing appropriate analytical techniques, yet injury outcomes research is fraught with inappropriate statistical models. Despite our preference for simpler statistical models, traditional regression analysis is unlikely to be sufficient to fully understand the complex nature of trauma systems. Future research therefore needs to look toward more sophisticated analytical methods such as causal inference and microsimulation models to better understand the determinants of optimal injury outcomes. In addition, whereas data coverage and quality may be improved, no observational study is exempt from selection, information, and confounding bias. We thus need to develop methods to evaluate the influence of such biases on outcomes research through sensitivity and bias analysis.

Efforts should also be made to improve translation of research results into quality of care improvements as part of the quality loop. This is the essence of the public health model. Outcomes research must therefore be based on a comprehensive integrated and end-of-grant knowledge-translation strategy.¹⁷⁷ Practitioners, policymakers, decision makers, and patient/family representatives should be involved in all phases of research projects and results should be distributed to all stakeholders, not just in the form of scientific articles but as policy briefs, clinical guides, and decision rules. We should also work to strengthen international collaborations to pool resources, avoid duplicating research projects, and generate results with maximum impact.

Injury severity scoring and biometric modeling remain fertile research fields. These include biometric markers and models for head injury and impending shock states, as well as a variety of biochemical markers being considered to augment models for identifying at-risk patients, including lactate, bicarbonate, and tissue oxygenation. Ultimately, these processes of risk evaluation of trauma patients will improve the general strategy and link in with therapeutic innovations, which include improved resuscitation therapies and energy-directed cellular-level therapies of the future.

Fundamental to the process of evaluating innovation and the inexorable progress toward improving quality of care is a firm foundation of understanding of the power and limitations of injury severity scoring systems and biological, biometric models.