Injury severity and probability of survival assessment in trauma patients using a predictive hierarchical network model derived from ICD-9 codes

The 100 most cited articles in artificial intelligence related to orthopedics

Background

This bibliometric study aimed to identify and analyze the top 100 articles related to artificial intelligence in the field of orthopedics.

Methods

The articles were assessed based on their number of citations, publication years, countries, journals, authors, affiliations, and funding agencies. Additionally, they were analyzed in terms of their themes and objectives. Keyword co-occurrence, co-citation of authors, and co-citation of references analyses were conducted using VOSviewer (version 1.6.19).

Results

The number of citations of these articles ranged from 32 to 272, with six papers having more than 200 citations The years of 2019 (n: 37) and 2020 (n: 19) together constituted 56% of the list. The USA was the leading contributor country to this field (n: 61). The most frequently used keywords were "machine learning" (n: 26), "classification" (n: 18), "deep learning" (n: 16), "artificial intelligence" (n: 14), respectively. The most common themes were decision support (n: 25), fracture detection (n: 24), and osteoarthrtitis staging (n: 21). The majority of the studies were diagnostic in nature (n: 85), with only two articles focused on treatment.

Conclusions

This study provides valuable insights and presents the historical perspective of scientific development on artificial intelligence in the field of orthopedics. The literature in this field is expanding rapidly. Currently, research is generally done for diagnostic purposes and predominantly focused on decision support systems, fracture detection, and osteoarthritis classification.

Artificial intelligence and machine learning for hemorrhagic trauma care

Artificial intelligence (AI), a branch of machine learning (ML) has been increasingly employed in the research of trauma in various aspects. Hemorrhage is the most common cause of trauma-related death. To better elucidate the current role of AI and contribute to future development of ML in trauma care, we conducted a review focused on the use of ML in the diagnosis or treatment strategy of traumatic hemorrhage. A literature search was carried out on PubMed and Google scholar. Titles and abstracts were screened and, if deemed appropriate, the full articles were reviewed. We included 89 studies in the review. These studies could be grouped into five areas: (1) prediction of outcomes; (2) risk assessment and injury severity for triage; (3) prediction of transfusions; (4) detection of hemorrhage; and (5) prediction of coagulopathy. Performance analysis of ML in comparison with current standards for trauma care showed that most studies demonstrated the benefits of ML models. However, most studies were retrospective, focused on prediction of mortality, and development of patient outcome scoring systems. Few studies performed model assessment via test datasets obtained from different sources. Prediction models for transfusions and coagulopathy have been developed, but none is in widespread use. AI-enabled ML-driven technology is becoming integral part of the whole course of trauma care. Comparison and application of ML algorithms using different datasets from initial training, testing and validation in prospective and randomized controlled trials are warranted for provision of decision support for individualized patient care as far forward as possible.

Machine Learning in the Prediction of Trauma Outcomes: A Systematic Review

STUDY OBJECTIVE

Machine learning models carry unique potential as decision-making aids and prediction tools for improving patient care. Traumatically injured patients provide a uniquely heterogeneous population with severe injuries that can be difficult to predict. Given the relative infancy of machine learning applications in medicine, this systematic review aimed to better understand the current state of machine learning development and implementation to help create a basis for future research.

METHODS

We conducted a systematic review from inception to May 2021, using Embase, MEDLINE through Ovid, Web of Science, Google Scholar, and relevant gray literature, for uses of machine learning in predicting the outcomes of trauma patients. The screening and data extraction were performed by 2 independent reviewers.

RESULTS

Of the 14,694 identified articles screened, 67 were included for data extraction. Artificial neural networks comprised the most commonly used model, and mortality was the most prevalent outcome of interest. In terms of machine learning model development, there was a lack of studies that employed external validation, feature selection methods, and performed formal calibration testing. Significant heterogeneity in reporting was also observed between the machine learning models employed, patient populations, performance metrics, and features employed.

CONCLUSION

This review highlights the heterogeneity in the development and reporting of machine learning models for the prediction of trauma outcomes. While these models present an area of opportunity as an ancillary to clinical decision-making, we recommend more standardization and rigorous guidelines for the development of future models.

Age-related mortality in abusive head trauma

BACKGROUND

Abusive head trauma (AHT) peaks during early infancy and decreases in toddler years. Infants and toddlers experience different injuries, possibly impacting the risk of mortality. We aimed to evaluate the association of age with mortality.

METHODS

We conducted a retrospective study of AHT hospitalizations in 2000, 2003, 2006, 2009, and 2012 from the Kid's Inpatient Claims Database. An accidental head trauma cohort was included to hypothesize that the association between age and mortality is unique to abuse. A nested multivariable logistic regression was used to perform the analysis.

RESULTS

Children aged 2 years to 4 years experienced higher mortality than those younger than 2 years (22% vs. 10%, p < 0.0001; adjusted odds ratio [OR], 1.6; 95% confidence interval [CI], 1.1-2.2). The presence of subarachnoid hemorrhage (OR, 1.9; 95% CI, 1.3-2.9), cerebral edema (OR, 4.0; 95% CI, 2.9-5.4), and retinal hemorrhage (OR, 1.9; 95% CI, 1.5-2.5) were associated with an increase risk in mortality. Children younger than 2 years experienced more fractures and hemorrhage (subdural, subarachnoid, retinal) while children aged 2 years to 4 years encountered more internal injuries and cerebral edema.In children with accidental head trauma, those aged 2 years to 4 years have a lower mortality compared with those younger than 2 years (OR, 0.4; 95% CI, 0.3-0.6). Among children younger than 2 years, AHT and accidental trauma had comparable risk of mortality (OR, 0.9; 95% CI, 0.6-1.3). However, among those aged 22 years to 4 years, AHT had a higher risk of mortality than accidental trauma (OR, 3.3; 95% CI, 2.1-5.1).

CONCLUSION

There is a considerable risk of mortality associated with age at diagnosis in children with AHT.Children younger than 2 years and those aged 2 years to 4 years present with different types of injuries. The high risk of mortality in the children aged 2 years to 4 years is unique to AHT. Efforts should be made to increase awareness about the risk of mortality and identify factors that can aide in a timely accurate diagnosis.

LEVEL OF EVIDENCE

Prognostic and epidemiological study, level III.

Comparison of manual and computer assigned injury severity scores

BACKGROUND

The study objective was to compare the ISS manually assigned by hospital personnel and those generated by the ICDPIC software for value agreement and predictive power of length of stay (LOS) and mortality.

METHODS

We used data from the 2010-2016 trauma registry of a paediatric trauma centre (PTC) and 2014 National Trauma Data Bank (NTDB) hospitals that reported manually coded ISS. Agreement analysis was performed between manually and computer assigned ISS with severity groupings of 1-8, 9-15, 16-25 and 25-75. The prediction of LOS was compared using coefficients of determination (R²) from linear regression models. Mortality predictive power was compared using receiver operating characteristic (ROC) curves from logistic regression models.

RESULTS

The proportion of agreement between manually and computer assigned ISS in PTC data was 0.84 and for NTDB was 0.75. Analysing predictive power for LOS in the PTC sample, the R²=0.19 for manually assigned scores, and the R²=0.15 for computer assigned scores (p=0.0009). The areas under the ROC curve indicated a mortality predictive power of 0.95 for manually assigned scores and 0.86 for computer assigned scores in the PTC data (p=0.0011).

CONCLUSIONS

Manually and computer assigned ISS had strong comparative agreement for minor injuries but did not correlate well for critical injuries (ISS=25-75). The LOS and mortality predictive power were significantly higher for manually assigned ISS when compared with computer assigned ISS in both PTC and NTDB data sets. Thus, hospitals should be cautious about transitioning to computer assigned ISS, specifically for patients who are critically injured.

Machine Learning for Predicting Outcomes in Trauma

To date, there are no reviews on machine learning (ML) for predicting outcomes in trauma. Consequently, it remains unclear as to how ML-based prediction models compare in the triage and assessment of trauma patients. The objective of this review was to survey and identify studies involving ML for predicting outcomes in trauma, with the hypothesis that models predicting similar outcomes may share common features but the performance of ML in these studies will differ greatly. MEDLINE and other databases were searched for studies involving trauma and ML. Sixty-five observational studies involving ML for the prediction of trauma outcomes met inclusion criteria. In total 2,433,180 patients were included in the studies. The studies focused on prediction of the following outcome measures: survival/mortality (n = 34), morbidity/shock/hemorrhage (n = 12), hospital length of stay (n = 7), hospital admission/triage (n = 6), traumatic brain injury (n = 4), life-saving interventions (n = 5), post-traumatic stress disorder (n = 4), and transfusion (n = 1). Six studies were prospective observational studies. Of the 65 studies, 33 used artificial neural networks for prediction. Importantly, most studies demonstrated the benefits of ML models. However, algorithm performance was assessed differently by different authors. Sensitivity-specificity gap values varied greatly from 0.035 to 0.927. Notably, studies shared many features for model development. A common ML feature base may be determined for predicting outcomes in trauma. However, the impact of ML will require further validation in prospective observational studies and randomized clinical trials, establishment of common performance criteria, and high-quality evidence about clinical and economic impacts before ML can be widely accepted in practice.

Improving life expectancy: A 'broken neck' doesn't have to be a terminal diagnosis for the elderly

Background

Elderly patients with cervical spine fractures require optimal care. Treatment with a cervical collar or halo instead of surgical fixation may increase mortality. This investigation intends to describe the life expectancy after injury and evaluate the impact of surgical intervention on mortality.

Methods

Patients ≥65 years, with traumatic cervical spine fractures without cord injury were identified in the 1995-2009 California Office of Statewide Health and Planning database. Those with halo placement or surgical spine fixation were identified. Primary outcome was death, studied at the initial admission, 30 days, 1 year, and the entire study period. Univariate and multivariate regressions were performed to identify predictors of death. Kaplan-Meier survival curves were used to describe life expectancy after injury.

Results

10 938 patients were identified. Mortality rate was 10% during the initial admission, 28% at 1 year and 50% during the entire study period. A halo was placed in 14% of patients and 12% underwent surgical fixation. Mortality rates during the initial admission were 11% for patients without an intervention, 7% with halo placement and 6% with surgical fixation; at 1 year, these increased to 30%, 26% and 19%, respectively. At 1 year, more than one in four patients above 75 years of age will die.At 1 year spine fixation, female gender and admission to a trauma center predicted a lower risk of death at 1 year (OR 0.59, 0.68; p<0.001 and OR 0.89; p=0.02, respectively). Having a complication, fall mechanism, and traumatic brain injury (OR 1.84, 1.33, 1.37; p<0.001, respectively) were predictors of a higher risk of death. Halo use had no impact on death at 1 year (OR 0.98; p=0.77).

Discussion

Mortality rates after cervical spine fracture in the elderly is high. Surgical fixation is associated with improved survival; remaining true after adjusting for age and comorbidities; suggesting that surgical fixation may improve outcomes in the elderly.

Level of evidence

Level IV.

Trauma scores and trauma outcome measures

Objective measurement and quantification of injury severity are necessary for triage, performance evaluation and research. In order to evaluate interventions, outcomes must also be compared. While this can be done using hospital stay or mortality, these will fail to detect subtle differences. Impact of injury on health can be quantified using a variety of scoring systems. Trauma scoring and outcome measurement have grown increasingly complex in recent years and are likely to become more so in the future.

Establishing an injury indicator for severe paediatric injury

BACKGROUND

Routinely gathered injury data, such as hospitalisations, may be subject to variation from sources other than injury incidence. There is a need for an indicator that defines severe injury, which may be less vulnerable to fluctuations due to changes in care policies. The purpose of this study was to identify International Classification of Diseases-10 codes associated with severe paediatric injuries and to specify and validate a severe paediatric injury indicator.

METHODS

Two data sets that included the ISS and the survival risk ratio were used to produce a list of diagnoses to define severe paediatric injury. The list was sent to trauma surgeons who classified each code as severe enough or not severe enough to require care in a trauma centre. The indicator was fully specified, then validated by using a different data set to validate the codes in a real-world situation.

RESULTS

Sixty diagnoses were identified as representing severe paediatric injury. Following specification, the indicator was applied to an existing comprehensive data set of paediatric injuries. The decline in hospitalisation of paediatric injuries was significantly steeper for severe than non-severe injuries, suggesting that factors related to the decline in this trauma subset are unlikely to be related to changes in access or other components of trauma care delivery.

CONCLUSIONS

This indicator can be used for the evaluation of trends in severe paediatric trauma and will help identify populations at risk. This research may inform policies and procedures for referrals of severe childhood injury to appropriate levels of care.

Adult trauma scoring models: a critical review

Since the development of the TRISS (Revised Trauma Score and Injury Severity Score) trauma scoring model several alternative models have been developed. Trauma scoring systems can be broadly categorized into anatomically based injury severity models, e.g. NISS (New Injury Severity Score) and ASCOT (A Severity Characterization of Trauma) or data driven models, e.g. ICISS (Insternational Classification of Diseases Injury Severity Score). Trauma scoring models using death/survival as the outcome measure can either be developed using logistic regression or a neural network approach. Assessment of the worth of a model is most commonly performed using receiver operating curve analysis or the Hosmer-Lemeshow statistic. Both of these statistical methods have their inherent weaknesses when applied to trauma scoring model development. This article aims to review four trauma scoring models and to discuss the limitations of the statistical methods used to assess the worth of these models.

Evaluation of a Consensus-Based Criterion Standard Definition of Trauma Center Need for Use in Field Triage Research

Research on field triage of injured patients is limited by the lack of a widely used criterion standard for defining trauma center need. Injury Severity Score (ISS) >15 has been a commonly used outcome measure in research for determining trauma center need that has never been validated. A multidisciplinary team recently published a consensus-based criterion standard definition of trauma center need, but this measure has not yet been validated. The objective was to determine if the consensus-based criterion standard can be obtained by medical record review and compare patients identified as needing a trauma center by the consensus-based criterion standard vs. ISS >15. A subanalysis of data collected during a 2-year prospective cohort study of 4,528 adult trauma patients transported by EMS to a single trauma center was conducted. These data included ICD-9-CM codes, treatment times, and other patient care data. Presence of the consensus-based criterion standard was determined for each patient. ISS was calculated based on ICD-9-CM codes assigned for billing. The consensus-based criterion standard could be applied to 4,471 (98.7%) cases. ISS could be determined for 4,506 (99.5%) cases. Based on an ISS >15, 8.9% of cases were identified as needing a trauma center. Of those, only 48.2% met the consensus-based criterion standard. Almost all patients that did not meet the consensus-based criterion standard, but had an ISS >15 were diagnosed with chest (rib fractures (100/205 cases)/pneumothorax (57/205 cases), closed head (without surgical intervention 88/205 cases), vertebral (without spinal cord injury 45/205 cases), and/or extremity injuries (39/205 cases). There were 4,053 cases with an ISS <15. 5.0% of those with an ISS <15 met the consensus-based criterion standard with the majority requiring surgery (139/203 cases) or a blood transfusion (60/203 cases). The kappa coefficient of agreement for ISS and the consensus-based criterion standard was 0.43. We determined that the consensus-based criterion standard could be identified through a medical record review. Use of the consensus-based criterion standard for field triage research will more accurately identify injured patients who need the resources of a trauma center when compared to ISS.

Hospital care of childhood traumatic brain injury in the United States, 1997-2009: a neurosurgical perspective

OBJECT

The goal in this paper was to study hospital care for childhood traumatic brain injury (TBI) in a nationwide population base.

METHODS

Data were acquired from the Kids' Inpatient Database (KID) for the years 1997, 2000, 2003, 2006, and 2009. Admission for TBI was defined by any ICD-9-CM diagnostic code for TBI. Admission for severe TBI was defined by a principal diagnostic code for TBI and a procedural code for mechanical ventilation; admissions ending in discharge home alive in less than 4 days were excluded.

RESULTS

Estimated raw and population-based rates of admission for all TBI, for severe TBI, for death from severe TBI, and for major and minor neurosurgical procedures fell steadily during the study period. Median hospital charges for severe TBI rose steadily, even after adjustment for inflation, but estimated nationwide hospital charges were stable. Among 14,932 actual admissions for severe TBI captured in the KID, case mortality was stable through the study period, at 23.9%. In a multivariate analysis, commercial insurance (OR 0.86, CI 0.77-0.95; p = 0.004) and white race (OR 0.78, CI 0.70-0.87; p < 0.0005) were associated with lower mortality rates, but there was no association between these factors and commitment of resources, as measured by hospital charges or rates of major procedures. Increasing median income of home ZIP code was associated with higher hospital charges and higher rates of major and minor procedures. Only 46.8% of admissions for severe TBI were coded for a neurosurgical procedure of any kind. Fewer admissions were coded for minor neurosurgical procedures than anticipated, and the state-by-state variance in rates of minor procedures was twice as great as for major procedures. Possible explanations for the "missing ICP monitors" are discussed.

CONCLUSIONS

Childhood brain trauma is a shrinking sector of neurosurgical hospital practice. Racial and economic disparities in mortality rates were confirmed in this study, but they were not explained by available metrics of resource commitment. Vigilance is required to continue to supply neurosurgical expertise to the multidisciplinary care process.

Early death and late morbidity after blood transfusion of injured children: a pilot study

BACKGROUND/PURPOSE

Early postinjury death after packed red blood cell (pRBC) transfusion is attributed to uncontrolled hemorrhage and coagulopathy. The adverse immunomodulatory effects of blood transfusion are implicated in subsequent morbidity. We hypothesized that injured children requiring pRBC transfusion demonstrate patterns in outcome similar to those observed in adults.

METHODS

Our prospectively collected trauma registry was queried for demographics, treatment, and outcome (2006-2009). Outcomes of children who received pRBC transfusion were compared with those of age- and Injury Severity Score (ISS)-matched children who did not receive pRBC transfusion by both univariate and multivariable analysis.

RESULTS

Eight percent (43/512) of injured children received a pRBC transfusion: 20 early and 23 late. The likelihood of pRBC transfusion increased with increasing ISS (ISS <15, 2%; ISS 16-25, 17%; ISS >25, 72%). One-half of injured children who received an early pRBC transfusion died; however, most deaths were because of central nervous system injury. Both ventilator and intensive care unit days were increased in children who received pRBC transfusion as compared with those who did not.

CONCLUSION

Early pRBC transfusion is associated with a high mortality in children. Late blood transfusion is associated with worse outcomes, although this relationship may not be causal. This pilot study provides evidence of an association between pRBC transfusion, morbidity, and mortality among injured children that warrants refinement in larger, prospective investigations.

CK/CK-MB ratio as an indirect predictor for survival in polytraumatized patients

BACKGROUND

Accurate assessment of injury severity is critical for decision making related to the prevention, triage, and treatment of several injured patients. Early estimation of mortality risk of critically injured patients is mandatory for adequate therapeutic strategies. Current risk stratification relies on clinical diagnosis and scoring systems. In our study, we hypothesized whether a simple laboratory test, the CK/CK-MB ratio, could help improving risk prediction in severely traumatized patients.

METHODS

In a 9-year period, 328 nonselected trauma patients were included in our retrospective study at a Level I Trauma Center up to September 2002. Data for this study were obtained from our computerized trauma database, established in September 1992.

RESULTS

In our study population, we could show a negative correlation between Injury Severity Score (ISS) and leukocytes. A positive correlation was detected for liver enzymes and CK-MB. The correlation between ISS and Na(+) was significant. No correlation between ISS, K(+), and Hb/Ht could be observed. Exitus was associated with ISS, alteration in thrombocytes, CK, CK-MB, CRP, Crea, and Na(+).

CONCLUSION

In our study population, CK-MB levels showed a significant correlation with overall surveillance in polytraumatized patients. In our opinion, this might suggest that CK-MB levels could be taken as an indirect predictor for survival. Our findings need to be proven in further prospective clinical trials.

Does mechanism of injury predict trauma center need?

OBJECTIVE

To determine the predictive value of the mechanism-of-injury step of the American College of Surgeons Field Triage Decision Scheme for determining trauma center need.

METHODS

Emergency medical services (EMS) providers caring for injured adult patients transported to the regional trauma center in three midsized communities over two years were interviewed upon emergency department (ED) arrival. Included was any injured patient, regardless of injury severity. The interview collected patient physiologic condition, apparent anatomic injury, and mechanism of injury. Using the 1999 Scheme, patients who met the physiologic or anatomic steps were excluded. Patients were considered to need a trauma center if they had nonorthopedic surgery within 24 hours, had intensive care unit admission, or died prior to hospital discharge. Data were analyzed by calculating positive likelihood ratios (+LRs) and 95% confidence intervals (CIs) for each mechanism-of-injury criterion.

RESULTS

A total of 11,892 provider interviews were conducted. Of those, one was excluded because outcome data were not available, and 2,408 were excluded because they met the other steps of the Field Triage Decision Scheme. Of the remaining 9,483 cases, 2,363 met one of the mechanism-of-injury criteria, 204 (9%) of whom needed the resources of a trauma center. Criteria with a +LR ≥ 5 were death of another occupant in the same vehicle (6.8; CI: 2.7-16.7), fall >20 feet (5.3; CI: 2.4-11.4), and motor vehicle crash (MVC) extrication time >20 minutes (5.1; CI: 3.2-8.1). Criteria with a +LR between >2 and <5 were intrusion >12 inches (4.2; CI: 2.9-5.9), ejection (3.2; CI: 1.3-8.2), and deformity >20 inches (2.5; CI: 1.9-3.2). The criteria with a +LR ≤ 2 were MVC speed >40 mph (2.0; CI: 1.7-2.4), pedestrian/bicyclist struck at a speed >5 mph (1.2; CI:1.1-1.4), bicyclist/pedestrian thrown or run over (1.2; CI: 0.9-1.6), motorcycle crash at a speed >20 mph (1.2; CI: 1.1-1.4), rider separated from motorcycle (1.0; CI: 0.9-1.2), and MVC rollover (1.0; CI: 0.7-1.5).

CONCLUSION

Death of another occupant, fall distance, and extrication time were good predictors of trauma center need when a patient did not meet the anatomic or physiologic conditions. Intrusion, ejection, and vehicle deformity were moderate predictors. Key words: wounds and injury; triage; emergency medical services; emergency medical technicians; predictors; mechanism of injury; trauma center.

Developing a diagnosis-based severity classification system for use in emergency medical services for children

OBJECTIVES

Lack of adequate risk adjustment methodologies has hindered the progress of emergency medicine health services research. The authors hypothesized that a consensus-derived, diagnosis-based severity classification system (SCS) would be significantly associated with actual measures of emergency department (ED) resource use and could ultimately be used to examine severity-adjusted outcomes across patient populations.

METHODS

A panel of subject matter experts used consensus methods to assign severity scores (1 = lowest severity to 5 = highest severity) to 3,041 ED International Classifications of Diseases (ICD), 9th revision, diagnosis codes. SCS scores were assigned to ED visits using the visit diagnosis code with the highest severity. We tested the association between the SCS scores and measures of ED resource use in three data sets: the Pediatric Emergency Care Applied Research Network Core Data Project (PCDP), the National Hospital Ambulatory Medical Care Survey (NHAMCS), and the Connecticut state ED data set.

RESULTS

There was a significant association between the five-level SCS and all six measures of resource use: triage category, disposition, ED resource use, Current Procedural Terminology Evaluation and Management (CPT E&M) codes, ED length of stay, and ED charges within the three ED data sets.

CONCLUSIONS

The SCS demonstrates validity in its strong association with actual ED resource use. The use of readily available ICD-9 diagnosis codes makes the SCS useful as a risk adjustment tool for health services research.

The role of trauma scoring in developing trauma clinical governance in the Defence Medical Services

This paper discusses mathematical models of expressing severity of injury and probability of survival following trauma and their use in establishing clinical governance of a trauma system. There are five sections: (i) Historical overview of scoring systems--anatomical, physiological and combined systems and the advantages and disadvantages of each. (ii) Definitions used in official statistics--definitions of 'killed in action' and other categories and the importance of casualty reporting rates and comparison across conflicts and nationalities. (iii) Current scoring systems and clinical governance--clinical governance of the trauma system in the Defence Medical Services (DMS) by using trauma scoring models to analyse injury and clinical patterns. (iv) Unexpected outcomes--unexpected outcomes focus clinical governance tools. Unexpected survivors signify good practice to be promulgated. Unexpected deaths pick up areas of weakness to be addressed. Seventy-five clinically validated unexpected survivors were identified over 2 years during contemporary combat operations. (v) Future developments--can the trauma scoring methods be improved? Trauma scoring systems use linear approaches and have significant weaknesses. Trauma and its treatment is a complex system. Nonlinear methods need to be investigated to determine whether these will produce a better approach to the analysis of the survival from major trauma.

Comparison of the 1999 and 2006 trauma triage guidelines: where do patients go?

BACKGROUND

In 2006, the Centers for Disease Control and Prevention (CDC) released a revised Field Triage Decision Scheme. It is unknown how this modified scheme will affect the number of patients identified by emergency medical services (EMS) for transport to a trauma center.

OBJECTIVES

To determine the change in the number of patients transported by EMS who meet the 2006 scheme, compared with the 1999 scheme, and to determine how the scheme change would affect under- and overtriage rates.

METHODS

The EMS providers in charge of care for injured adult patients transported to a regional trauma center in three mid-sized cities were interviewed immediately after completing transport. All injured patients were included, regardless of severity. The interview included patient demographics, vital signs, apparent anatomic injury, and the mechanism of injury. Included patients were then followed through hospital discharge. The 1999 and 2006 scheme criteria were each retrospectively applied to the collected data. The numbers of patients identified by the two schemes were determined. Patients were considered to have needed a trauma center if they had nonorthopedic surgery within 24 hours, were admitted to an intensive care unit (ICU), or died. Data were analyzed using descriptive statistics including 95% confidence intervals.

RESULTS

EMS interviews were conducted for 11,892 patients and outcome data were unavailable for one patient. The average patient age was 48 years; 51% of the patients were men. Providers reported bringing 54% of the enrolled patients to the trauma center based on their local trauma protocol. Medical record review identified 12% of the enrolled patients as needing a trauma center. Use of the 2006 scheme would have resulted in 1,423 fewer patients (12%; 95% confidence interval [CI]:11%-13%) being identified as needing a trauma center by EMS providers (40%; 95% CI: 39%-41% versus 28%; 95% CI: 27%-29%). Of those patients, 1,344 (94%) did not actually need the resources of a trauma center, whereas 78 (6%) actually needed the resources of a trauma center and would have been undertriaged.

CONCLUSION

Use of the 2006 Field Triage Decision Scheme would have resulted in a significant decrease in the number of patients identified as needing the resources of a trauma center. These changes reduced overtriage while causing a small increase in the number of patients who would have been undertriaged.

Predicting ICU survival: a meta-level approach

Background

The performance of separate Intensive Care Unit (ICU) status scoring systems vis-à-vis prediction of outcome is not satisfactory. Computer-based predictive modeling techniques may yield good results but their performance has seldom been extensively compared to that of other mature or emerging predictive models. The objective of the present study was twofold: to propose a prototype meta-level predicting approach concerning Intensive Care Unit (ICU) survival and to evaluate the effectiveness of typical mining models in this context.

Methods

Data on 158 men and 46 women, were used retrospectively (75% of the patients survived). We used Glasgow Coma Scale (GCS), Acute Physiology And Chronic Health Evaluation II (APACHE II), Sequential Organ Failure Assessment (SOFA) and Injury Severity Score (ISS) values to structure a decision tree (DTM), a neural network (NNM) and a logistic regression (LRM) model and we evaluated the assessment indicators implementing Receiver Operating Characteristics (ROC) plot analysis.

Results

Our findings indicate that regarding the assessment of indicators' capacity there are specific discrete limits that should be taken into account. The Az score ± SE was 0.8773± 0.0376 for the DTM, 0.8061± 0.0427 for the NNM and 0.8204± 0.0376 for the LRM, suggesting that the proposed DTM achieved a near optimal Az score.

Conclusion

The predicting processes of ICU survival may go "one step forward", by using classic composite assessment indicators as variables.

Early markers of acute respiratory distress syndrome development in severe trauma patients

PURPOSE

The aim of the study was to identify early risk factors for development of acute respiratory distress syndrome (ARDS) in severe trauma patients.

MATERIALS AND METHODS

This was a prospective observational study of 693 severe trauma patients (Injury Severity Score >or=16 and/or Revised Trauma Score <or=11) in 17 hospitals in a Spanish region of 8 million inhabitants from July 2002 to December 2002.

RESULTS

Acute respiratory distress syndrome developed in 6.9% of patients who were more severely ill with higher APACHE II (P < .001) and Injury Severity Score (P = .002) scores vs patients not developing ARDS. Acute respiratory distress syndrome development was associated (P < .001) with fractures of femur (International Classification of Diseases, Ninth Revision [ICD-9] codes 820, 821), tibia (ICD-9 code 823), humerus, and pelvis, with a number (>or=2) of long bone fractures, and with chest injuries (rib/sternal fracture [ICD-9 code 807] and hemo/pneumothorax [ICD-9 code 860/861]). Patients with ARDS required more colloids (P = .005) and red blood cell units (P = .02) than patients without ARDS during the first 24 hours. Multivariate analysis showed that ARDS was related to chest trauma diagnosis (ICD-9 code 807) (odds ratio [OR], 3.85), femoral fracture (OR, 3.16), APACHE II score (OR, 1.05), and blood transfusion during resuscitation (OR, 1.32).

CONCLUSIONS

Risk of ARDS development is related to the first 24-hour admission variables, including severe physiologic derangements and specific ICD-9-classified injuries. Blood transfusion may play an independent role.

Índices de trauma em pacientes submetidos à laparotomia

OBJETIVO: O trauma é um problema de saúde pública de enormes proporções. Constitui-se na principal causa de óbitos na população jovem. O Major Trauma Outcome Study (MTOS) é um estudo descritivo e retrospectivo da gravidade das lesões e evolução dos pacientes, considerado como o maior arquivo contemporâneo de informações descritivas de traumatizados. O objetivo do presente estudo é comparar o cálculo retrospectivo do New Injury Severity Score (NISS) com o Injury Severity Score (ISS) já calculado prospectivamente, utilizando o Trauma and Injury Severity Score (TRISS) e uma simples modificação deste índice, denominado de NTRISS (New Trauma and Injury Severity Score), e também comparar esta população submetida à laparotomia com os pacientes do MTOS. MÉTODO: Foram estudados 1.380 pacientes adultos traumatizados e submetidos à laparotomia na Disciplina de Cirurgia do Trauma da Unicamp, em Campinas, durante um período de oito anos. Os dados avaliados foram: demográficos, causa do trauma (fechado ou penetrante, ferimento por projétil de arma de fogo ou arma branca), estado fisiológico na admissão (RTS), diagnóstico anatômico de lesões (ATI, ISS e NISS), probabilidade de sobrevida utilizando o TRISS e o NTRISS, e a evolução do paciente (sobrevivência ou óbito). Foram utilizadas as estatísticas Z e W, inicialmente descritas por Flora, a fim de comparar a predição de óbitos ou sobreviventes com o estudo controle (MTOS). RESULTADOS: A maioria dos pacientes (88,3%) era do sexo masculino e jovem (média de idade de 30,4 anos). O ferimento por projétil de arma de fogo foi o mecanismo de trauma mais freqüente, com 641 casos (46,4%). Quatrocentos e trinta pacientes (31,2%) sofreram trauma fechado. As médias do ATI, ISS e NISS foram, respectivamente, de 12,3, 17,6 e 22,1. A taxa global de mortalidade foi de 16,8% e os pacientes com trauma contuso tiveram a maior mortalidade (29,3%). O NISS identificou melhor os sobreviventes e óbitos se comparado ao ISS, obtendo-se uma maior especificidade com o NTRISS. Foi observado um número significativamente menor de sobreviventes do que o esperado pelo estudo basal, com Z -16,24 com o TRISS e Z -9,40 se aplicado o NTRISS. Variações no valor da estatística W para cada paciente mostraram uma diferença no número de óbitos equivalente a 7,89 mais casos de óbito do que o esperado pelo MTOS, por 100 pacientes tratados, ao se empregar o TRISS, enquanto que estes valores foram reduzidos para 5,14 utilizando-se o NTRISS. CONCLUSÕES: Os métodos utilizados para cálculo da probabilidade de sobrevivência apresentaram limitações, particularmente nesta população com predomínio dos traumas penetrantes. O NISS, com o seu derivado NTRISS, foi o escore que obteve uma melhor predição de sobrevivência se comparado com o ISS. Os resultados obtidos com o TRISS e NTRISS foram estatisticamente piores do que os do MTOS, porém este processo de monitorização destes pacientes traumatizados tem sido importante para assegurar uma condição continuada de controle de qualidade.

[Artificial neural networks. Theory and applications in anesthesia, intensive care and emergency medicine]

Artificial neural networks (ANN) are constructed to simulate processes of the central nervous system of higher creatures. An ANN consists of a set of processing units (nodes) which simulate neurons and are interconnected via a set of "weights" (analogous to synaptic connections in the nervous system) in a way which allows signals to travel through the network in parallel. The nodes (neurons) are simple computing elements. They accumulate input from other neurons by means of a weighted sum. If a certain threshold is reached the neuron sends information to all other connected neurons otherwise it remains quiescent. One major difference compared with traditional statistical or rule-based systems is the learning aptitude of an ANN. At the very beginning of a training process an ANN contains no explicit information. Then a large number of cases with a known outcome are presented to the system and the weights of the inter-neuronal connections are changed by a training algorithm designed to minimise the total error of the system. A trained network has extracted rules that are represented by the matrix of the weights between the neurons. This feature is called generalisation and allows the ANN to predict cases that have never been presented to the system before. Artificial neural networks have shown to be useful predicting various events. Especially complex, non-linear, and time depending relationships can be modelled and forecasted. Furthermore an ANN can be used when the influencing variables on a certain event are not exactly known as it is the case in financial or weather forecasts. This article aims to give a short overview on the function of ANN and their previous use and possible future applications in anaesthesia, intensive care, and emergency medicine.

Use of admission Glasgow Coma Score, pupil size, and pupil reactivity to determine outcome for trauma patients

BACKGROUND

Determination of nonsurvival in trauma patients is difficult because valid prognostic indicators are lacking. It was hypothesized that patients presenting with a Glasgow Coma Score (GCS) of 3 as well as fixed and dilated (FD) pupils do not have a reasonable chance of survival.

METHODS

From 1999 through 2001, adult trauma patients (age, >14 years) admitted with a GCS of 3 were reviewed. Patients receiving paralytic agents before initial assessment were excluded from analysis. Fixed and dilated pupils were defined as being 4 mm or more in diameter bilaterally and nonreactive to light. In this study, the FD patients were evaluated for survival, resuscitative measures, surgical procedures, length of hospital stay, and organ donation. The non-FD patients were evaluated for survival and length of hospital stay.

RESULTS

Of the 137 patients evaluated with a GCS of 3, 104 had FD pupils and 33 did not. In the FD group, there were no survivors. On arrival, 28 (37.3%) of the patients were declared dead, and no further interventions were undertaken. Of the 76 patients (62.7%) who underwent further resuscitation, which included 53 surgical procedures, 30 died in the resuscitation bay, 39 within 24 hours, 4 within 48 hours, 2 within 72 hours, and 1 on day 6. There were 18 (23.7%) organ donors. Of the 33 patients without FD pupils, 11 (33%) survived to discharge (mean hospital stay, 21.4 days). Of the 22 nonsurvivors (67%), 10 died in the resuscitation bay, 8 within 24 hours, 1 within 48 hours, 1 on day 4, and 2 on day 6.

CONCLUSIONS

Patients presenting with a GCS of 3 and FD pupils have no reasonable chance for survival. A significant percentage of these patients can be salvaged for organ donation. This information should be used in deciding to pursue aggressive resuscitation efforts and in discussing prognosis with family. Patients with a GCS of 3 who are not FD should be aggressively resuscitated because many of these patients survive to discharge.

Bullet trajectory predicts the need for damage control: an artificial neural network model

BACKGROUND

Effective use of damage control in trauma hinges on an early decision to use it. Bullet trajectory has never been studied as a marker for damage control. We hypothesize that this decision can be predicted by an artificial neural network (ANN) model based on the bullet trajectory and the patient's blood pressure.

METHODS

A multilayer perceptron ANN predictive model was developed from a data set of 312 patients with single abdominal gunshot injuries. Input variables were the bullet path, trajectory patterns, and admission systolic pressure. The output variable was either a damage control laparotomy or intraoperative death. The best performing ANN was implemented on prospectively collected data from 34 patients.

RESULTS

The model achieved a correct classification rate of 0.96 and area under the receiver operating characteristic curve of 0.94. External validation showed the model to have a sensitivity of 88% and specificity of 96%. Model implementation on the prospectively collected data had a correct classification rate of 0.91. Sensitivity analysis showed that systolic pressure, bullet path across the midline, and trajectory involving the right upper quadrant were the three most important input variables.

CONCLUSION

Bullet trajectory is an important, hitherto unrecognized, factor that should be incorporated into the decision to use damage control.

Comparison of artificial intelligence techniques with UKTRISS for estimating probability of survival after trauma. UK Trauma and Injury Severity Score

BACKGROUND

The development of TRISS was principally a search for variables that correlated with outcome. It is not known, however, if linear statistical models provide optimal results. Artificial intelligence techniques can answer this question and also determine the most important predictor variables.

METHODS

An artificial neural network, using 16 anatomic and physiologic predictor variables, was compared with the latest United Kingdom version of TRISS model.

RESULTS

Both methods were 89.6% correct, but TRISS was significantly better by the area under the receiver operating characteristic curve (0.941 vs. 0.921, p < 0.001). The artificial neural network, however, was better calibrated to the test data (Hosmer-Lemeshow statistic, 58.3 vs. 105.4). Head injury, age, and chest injury were the most important predictors by linear or nonlinear methods, whereas respiration rate, heart rate, and systolic blood pressure were underused.

CONCLUSION

Prediction using linear statistics is adequate but not optimal. Only half the predictors have important predictive value, fewer still when using linear classification. The strongest predictors swamp any nonlinearity observed in other variables.

Motor vehicle crashes in New York State: importance of accounting for emergency department deaths when assessing differences in in-hospital mortality by level of care

BACKGROUND

Motor vehicle crashes (MVCs) are one of the leading causes of death in the nation and in New York State, particularly among younger adult males. It is important to study how to reduce mortality from MVCs.

METHODS

Hospitalized victims of motor vehicle crashes in the 1994-1995 New York State Trauma Registry were identified for the study. A statistical model was used to calculate risk-adjusted mortality rates for groups of hospitals constituting each level of care (regional trauma center, area trauma center, noncenter). Levels of care were also compared with respect to the location of deaths in the hospital (emergency department, inpatient), and the time between emergency department admission and death for patients dying in the hospital.

RESULTS

The risk-adjusted mortality rate for MVCs in patients in regional centers was higher, although not significantly higher (6.91%; 95% confidence interval [CI], 6.18%-7.70%) than for area centers (5.53%; 95% CI, 4.43%-6.82%) or for noncenters (5.83%; 95% CI, 4.70%-7.15%). However, regional centers admitted seriously injured trauma patients from the emergency department much more quickly than other levels of care. Whereas only 18% of all in-hospital deaths occurred in emergency departments of regional centers, the comparable percentages for area centers and noncenters were 39% and 46%, respectively. Also, 43% of all deaths in regional centers occurred within 24 hours of presentation to the emergency department, compared with 15% in area centers and 21% in noncenters.

CONCLUSION

Risk-adjusted inpatient mortality rates for victims of MVCs may not yield a fair comparison of performance for different levels of care or for different hospitals because of differences in how quickly emergency department patients are admitted to the hospital. A more equitable way to assess hospital mortality rates may be to include emergency department deaths in addition to inpatient deaths.

Insurance status and admission to hospital for head injuries: are we part of a two-tiered medical system?

Previous studies have shown an association between insurance status and use of resources for inpatient care. We sought to assess whether insurance status influences decisions regarding the evaluation and treatment of head injured patients in the emergency department (ED). Head injured patients were identified from ED data from 4 hospitals reporting to the Kentucky Emergency Medical Services Information System. Multiple regression analysis using admission, ED length of stay, and ED charges as outcome variables was then performed. From 216,137 ED visits there were 8,591 (4%) head injured patients identified from the database. Eliminating those with revisits, transfers to another hospital in the database, and isolated facial lacerations, there were 3,821 cases. Controlling for age, hospital, race, primary diagnosis, and indicators of severity of the injury, insurance status was significantly associated with hospital admission. Those uninsured were the least likely to be admitted (OR 0.41; 95% CI (0.31, 0.50), whereas those with public insurance had an intermediate probability (OR 0.50 95% CI (0.37, 0.68) as compared with those with private insurance. Similarly, ED charges were lower for Medicaid patients than insured patients ($880) and tended to be slightly lower for uninsured patients ($1,043) than insured patients ($1,141) (P =.001). Length of stay in the ED was shorter for publicly insured patients (179 minutes) than uninsured (186 minutes) and privately insured patients (192 minutes) (P =.001). The extent of evaluation and admission for head injured patients is associated with insurance status. This creates a dual standard of care for patients. Practitioners should work to standardize the evaluation of patients independent of paying status.

Harborview assessment for risk of mortality: an improved measure of injury severity on the basis of ICD-9-CM

BACKGROUND

There have been several attempts to develop a scoring system that can accurately reflect the severity of a trauma patient's injuries, particularly with respect to the effect of the injury on survival. Current methodologies require unreliable physiologic data for the assignment of a survival probability and fail to account for the potential synergism of different injury combinations. The purpose of this study was to develop a scoring system to better estimate probability of mortality on the basis of information that is readily available from the hospital discharge sheet and does not rely on physiologic data.

METHODS

Records from the trauma registry from an urban Level I trauma center were analyzed using logistic regression. Included in the regression were Internation Classification of Diseases-9th Rev (ICD-9CM) codes for anatomic injury, mechanism, intent, and preexisting medical conditions, as well as age. Two-way interaction terms for several combinations of injuries were also included in the regression model. The resulting Harborview Assessment for Risk of Mortality (HARM) score was then applied to an independent test data set and compared with Trauma and Injury Severity Score (TRISS) probability of survival and ICD-9-CM Injury Severity Score (ICISS) for ability to predict mortality using the area under the receiver operator characteristic curve.

RESULTS

The HARM score was based on analysis of 16,042 records (design set). When applied to an independent validation set of 15,957 records, the area under the receiver operator characteristic curve (AUC) for HARM was 0.9592. This represented significantly better discrimination than both TRISS probability of survival (AUC = 0.9473, p = 0.005) and ICISS (AUC = 0.9402, p = 0.001). HARM also had a better calibration (Hosmer-Lemeshow statistic [HL] = 19.74) than TRISS (HL = 55.71) and ICISS (HL = 709.19). Physiologic data were incomplete for 6,124 records (38%) of the validation set; TRISS could not be calculated at all for these records.

CONCLUSION

The HARM score is an effective tool for predicting probability of in-hospital mortality for trauma patients. It outperforms both the TRISS and ICD9-CM Injury Severity Score (ICISS) methodologies with respect to both discrimination and calibration, using information that is readily available from hospital discharge coding, and without requiring emergency department physiologic data.

An artificial neural network as a model for prediction of survival in trauma patients: validation for a regional trauma area

BACKGROUND

To develop and validate an artificial neural network (ANN) for predicting survival of trauma patients based on standard prehospital variables, emergency room admission variables, and Injury Severity Score (ISS) using data derived from a regional area trauma system, and to compare this model with known trauma scoring systems.

PATIENT POPULATION

The study was composed of 10,609 patients admitted to 24 hospitals comprising a seven-county suburban/rural trauma region adjacent to a major metropolitan area. The data was generated as part of the New York State trauma registry. Study period was from January 1993 through December 1996 (1993-1994: 5,168 patients; 1995: 2,768 patients; 1996: 2,673 patients).

METHODS

A standard feed-forward back-propagation neural network was developed using Glasgow Coma Scale, systolic blood pressure, heart rate, respiratory rate, temperature, hematocrit, age, sex, intubation status, ICD-9-CM Injury E-code, and ISS as input variables. The network had a single layer of hidden nodes. Initial network development of the model was performed on the 1993-1994 data. Subsequent models were generated using the 1993, 1994, and 1995 data. The model was tested first on the 1995 and then on the 1996 data. The ANN model was tested against Trauma and Injury Severity Score (TRISS) and ISS using the receiver operator characteristic (ROC) area under the curve [ROC-A(z)], Lemeshow-Hosmer C-statistic, and calibration curves.

RESULTS

The ANN showed good clustering of the data, with good separation of nonsurvivors and survivors. The ROCA(z) was 0.912 for the ANN, 0.895 for TRISS, and 0.766 for ISS. The ANN exceeded TRISS with respect to calibration (Lemeshow-Hosmer C-statistic: 7.4 for ANN; 17.1 for TRISS). The prediction of survivors was good for both models. The ANN exceeded TRISS in nonsurvivor prediction.

CONCLUSION

An ANN developed for trauma patients using prehospital, emergency room admission data, and ISS gave good prediction of survival. It was accurate and had excellent calibration. This study expands our previous results developed at a single Level I trauma center and shows that an ANN model for predicting trauma deaths can be applied across hospitals with good results

Effect of prehospital advanced life support on outcomes of major trauma patients

OBJECTIVE

Determine whether prehospital advanced life support (ALS) improves the survival of major trauma patients and whether it is associated with longer on-scene times.

METHODS

A 36-month retrospective study of all major trauma patients who received either prehospital bag-valve-mask (BVM) or endotracheal intubation (ETI) and were transported by paramedics to our Level I trauma center. Logistic regression analysis determined the association of prehospital ALS with patient survival.

RESULTS

Of 9,451 major trauma patients, 496 (5.3%) had either BVM or ETI. Eighty-one percent received BVM, with a mean Injury Severity Score of 29 and a mortality rate of 67%; 93 patients (19%) underwent successful ETI, with a mean Injury Severity Score of 35 and a mortality rate of 93%. Adjusted survival for patients who had BVM was 5.3 times more likely than for patients who had ETI (95% confidence interval, 2.3-14.2, p = 0.00). Survival among patients who received intravenous fluids was 3.9 times more likely than those who did not (p = not significant). Average on-scene times for patients who had ETI or intravenous fluids were not significantly longer than those who had BVM or no intravenous fluids.

CONCLUSION

ALS procedures can be performed by paramedics on major trauma patients without prolonging on-scene time, but they do not seem to improve survival.

Severity of injuries due to accidental fall across the life span: a retrospective hospital-based study

This retrospective study investigated injuries due to accidental fall across the life span for which 19,593 patients were admitted to the Emergency Unit of the Groningen University Hospital during the period 1990 through 1997. 64% of the accidental falls were found for those in the age range between 10 years and 59 years; however, the proportion of accidental falls with regard to other causes of injuries by age group were the highest in youngsters (infants up to 9 years old) and in elderly persons (over 60 years old), 43% or more of these patients having falls with injuries. The clinically treated patients had on the average a statistically greater Injury Severity Score (7.2) than the outpatients (2.4). The highest percentages of medically treated inpatients were the patients of 60 years and over. Their mean ISS score was about the same for elderly inpatients, but the percentage of clinical treatment increased with age as well as the mortality. 30% of the injuries were found in the lower extremities and 30% in the upper extremities. Bone fracture was statistically significantly the major (36%) injury followed by contusion (20%). 34% of the accidental falls occurred at home, and statistically significantly more females, 50 years of age and older, were injured than males.

Validation of the International Classification of Diseases 10th Edition-based Injury Severity Score (ICISS)

OBJECTIVE

To compare the predictive power of International Classification of Diseases 10th Edition (ICD-10)-based International Classification of Diseases 9th Edition-based Injury Severity Score (ICISS) with Trauma and Injury Severity Score (TRISS) and ICD-9CM-based ICISS in the injury severity measure.

METHODS

ICD-10 version of survival risk ratios was derived from 47,750 trauma patients from 35 emergency centers for 1 year. The predictive power of TRISS, the ICD-9CM-based ICISS and ICD-10-based ICISS were compared in a group of 367 severely injured patients admitted to two university hospitals. The predictive power was compared by using the measures of discrimination (disparity, sensitivity, specificity, misclassification rates, and receiver operating characteristic curve analysis) and calibration (Hosmer-Lemeshow goodness-of-fit statistics), all calculated by logistic regression procedure.

RESULTS

ICD-10-based ICISS showed a lower performance than TRISS and ICD-9CM-based ICISS. When age and Revised Trauma Score were incorporated into the survival probability model, however, ICD-10-based ICISS full model showed a similar predictive power compared with TRISS and ICD-9CM-based ICISS full model. ICD-10-based ICISS had some disadvantages in predicting outcomes among patients with intracranial injuries. However, such weakness was largely compensated by incorporating age and Revised Trauma Score in the model.

CONCLUSION

The ICISS methodology can be extended to ICD-10 horizon as a standard injury severity measure in the place of TRISS, especially when age and Revised Trauma Score were incorporated in the model. For patients with intracranial injuries, the predictive power of ICD-10-based ICISS was relatively low because of differences in the classifying system between ICD-10 and ICD-9CM.

Accounting for intubation status in predicting mortality for victims of motor vehicle crashes

BACKGROUND

Two of the important predictors of mortality for trauma patients are the Glasgow Coma Scale and the respiratory rate. However, for intubated patients, the verbal response component of the Glasgow Coma Scale and the respiratory rate cannot be accurately obtained. This study extends previous work that attempts to predict mortality accurately for intubated patients without using verbal response and respiratory rate.

METHODS

The New York State Trauma Registry was used to identify 1994 and 1995 victims of motor vehicle crashes (MVCs). For the subset of patients who were not intubated, we developed two statistical models to predict mortality: one did not contain verbal response or respiratory rate, and the other contained a predicted verbal response. These were compared with a model that did include verbal response and respiratory rate. We also compared the predictive abilities of the first two models for all MVC patients (intubated and nonintubated) and determined the extent to which intubated patients were at increased risk of dying in the hospital after having adjusted for other predictors of mortality.

RESULTS

For nonintubated patients, the statistical model without verbal response and the model with predicted verbal response had slightly better discrimination and worse calibration than the model that included verbal response and respiratory rate. Predicted verbal response did not improve the strength of the model without verbal response. For all MVC patients (intubated and nonintubated), predicted verbal response was not a significant predictor of mortality when used in combination with the other predictors. Intubation status was a significant predictor, with intubated patients having a higher probability of dying in the hospital than patients with otherwise identical risk factors.

CONCLUSION

Inpatient mortality for intubated MVC patients can be accurately predicted without respiratory rate or verbal response. There appears to be no need for predicted verbal response to be part of the prediction formula, but intubation status is an important independent predictor of mortality and should be used in statistical models that predict mortality for MVC patients.

Predictors of mortality in adult patients with blunt injuries in New York State: a comparison of the Trauma and Injury Severity Score (TRISS) and the International Classification of Disease, Ninth Revision-based Injury Severity Score (ICISS)

BACKGROUND

The purpose of this study was to determine the statistical model that best predicted mortality from blunt trauma using a contemporary population-based database.

METHODS

1994-1995 New York State Trauma Registry data for patients with blunt injuries were used to predict mortality using three statistical models: (1) the original Trauma and Injury Severity Score (TRISS) model based on Major Trauma Outcome Study data, (2) a new TRISS model whose coefficients were derived using New York data, and (3) the International Classification of Disease, Ninth Revision-based Injury Severity Score (ICISS) with predicted survival values obtained from the Agency for Health Care Policy and Research's Health Care Utilization Project. The models were compared with respect to discrimination (using the C statistic) and calibration (using the Hosmer-Lemeshow [H-L] statistic). In addition, the models were tested to see how well they predicted outcomes for each of the three mechanisms of blunt injury.

RESULTS

The ICISS model had a significantly higher C statistic (0.878) and a better H-L statistic (29.38) for predicting mortality for all adult patients with blunt injuries. The original TRISS model had very poor calibration (H-L = 687.38). None of the three models predicted mortality accurately for victims of motor vehicle crashes or victims of low falls. When separate models were developed for all motor vehicle crashes, low falls, and other blunt injuries, the ICISS and New York TRISS models both fit well, although the calibration was marginal in most cases. The ICISS model had a statistically significantly higher C statistic for other blunt injuries and for motor vehicle crashes. The New York TRISS model had better calibration for low falls.

CONCLUSIONS

The ICISS has promise as an alternative to TRISS, but many more comparative studies need to be undertaken using updated TRISS coefficients. Models should also be developed for mechanisms of injury, not just for blunt and penetrating injuries.

Mild traumatic brain injury from motor vehicle accidents: factors associated with return to work

OBJECTIVES

To describe return to work (RTW) for motor vehicle accident (MVA) survivors with mild traumatic brain injury (MTBI) and to examine relationships between RTW and injury severity, cognitive impairment, social interaction, discharge disposition, and sociodemographics.

DESIGN

Inception cohort assessed within 1 month of injury and at follow-up 6 to 9 months (mean = 7.4) after injury, for comparisons on outcome of RTW.

SETTING

Tertiary care center in Toronto (time 1); at home for follow-up.

PARTICIPANTS

Fifty patients with MTBI resulting from MVA who were consecutively admitted during a 20-month period ending April 1994. Thirteen of 63 eligible patients refused consent or were lost to follow-up. Mean age was 31; 62% were men.

ELIGIBILITY CRITERIA

(1) patients had been working; (2) they had no history of head injury, neurologic disease, or psychiatric illness requiring hospitalization; and (3) they had no catastrophic impairment from accident.

MAIN OUTCOME MEASURE

Return to work (at premorbid or modified level).

RESULTS

Of the 42% who returned to work, 12% resumed their premorbid level of employment and 30% returned to modified work. There were significant differences (p<.05) between the groups in level of social interaction, premorbid occupation, and discharge disposition. On one test of cognitive functioning the difference was at p = .06.

CONCLUSION

Social interaction, jobs with greater decision-making latitude, and discharge home were positively related to RTW for this population. Cognitive impairment within the first month was not a reliable indicator of RTW potential.

A neural network approach to the diagnosis of morbidity outcomes in trauma care

This paper introduces the application of artificial neural networks to trauma complications assessment. The potential financial benefits of improving on trauma center diagnostic specificity in complications assessment are illustrated and the operational feasibility of the use of diagnostic neural models across institutions is discussed. A prototype neural network model is described, which, after training, succeeds in diagnosing the complication of sepsis in victims of traumatic blunt injury. Its diagnostic performance with 100% sensitivity and 96.5% specificity is accomplished with test data from a regional trauma center. The model is further shown to have correctly detected, during training, incorrectly coded data. The potential this suggests, for parsimonious database scrubbing through the use of neural network models, is discussed.

Lethal abdominal gunshot wounds at a level I trauma center: analysis of TRISS (Revised Trauma Score and Injury Severity Score) fallouts

BACKGROUND

The TRISS methodology (composite index of the Revised Trauma Score and the Injury Severity Score) has become widely used by trauma centers to assess quality of care. The American College of Surgeons recommends including negative TRISS fallouts (fatally injured patients predicted to survive by the TRISS methodology) as a filter to select patients for peer review. The purpose of this study was to analyze the TRISS fallouts among patients with lethal abdominal gunshot wounds admitted to a level I trauma center.

STUDY DESIGN

All patients categorized as TRISS fallouts admitted from January 1995 through December 1996 were analyzed.

RESULTS

During the study period, 848 patients with abdominal gunshot wounds were admitted. Of the 108 patients with any sign of life on admission who subsequently died, 39 (36%) were TRISS fallouts. The patients were largely young (mean age, 29 years) and male (87%), received rapid transport (mean scene time, 11 minutes), and had an attending-led trauma-team response (<5 minutes, 87%). Major vascular (80%) and multiple intraabdominal injuries (90%) predominated. The mean Penetrating Abdominal Trauma Index was 40.3. The mean TRISS probability of survival was 89%. The peer-review process deemed the deaths to be nonpreventable in 38 patients (97%) and potentially preventable in one patient (3%).

CONCLUSIONS

"TRISS fallouts" were predominantly patients who died despite receiving rapid prehospital transport, rapid senior-level trauma-team response, and surgical intervention for a serious complex of injuries. We conclude that without regional adjustment of coefficients used to predict the probability of survival, the TRISS methodology is of limited use in patients with abdominal gunshot wounds.

Neural network assessment of perioperative cardiac risk in vascular surgery patients

Neural networks were developed to predict perioperative cardiac complications with data from 567 vascular surgery patients. Neural network scores were based on cardiac risk factors and dipyridamole thallium results. These scores were converted into likelihood ratios that predicted cardiac risk. The prognostic accuracy of the neural networks was similar to that of logistic regression models (ROC areas 76.0% vs 75.8%), but their calibration was better. Logistic regression overestimated event rates in a group of high-risk patients (predicted event rate, 64%; observed rate 30%; n=50, p<0.001). On a validation set of 514 patients, the neural networks still had ROC similar areas to those of logistic regression (68.3% vs 67.5%), but logistic regression again overestimated event rates for a group of high-risk patients. The calibration difference was reflected in the Hosmer-Lemeshow chi-square statistic (18.6 for the neural networks, 45.0 for logistic regression). The neural networks successfully estimated perioperative cardiac risk with better calibration than comparable logistic regression models.

Comparison of the Injury Severity Score and ICD-9 diagnosis codes as predictors of outcome in injury: analysis of 44,032 patients

INTRODUCTION

Appropriate stratification of injury severity is a critical tool in the assessment of the treatment and the prevention of injury. Since its inception, the Injury Severity Score (ISS) has been the generally recognized "gold standard" for anatomic injury severity assessment. However, there is considerable time and expense involved in the collection of the information required to calculate an accurate ISS. In addition, the predictive power of the ISS has been shown to be limited. Previous work has demonstrated that the anatomic information about injury contained in the International Classification of Diseases Version 9 (ICD-9) can be a significant predictor of survival in trauma patients. The goal of this study was to utilize the San Diego County Trauma Registry (SDTR), one of the nation's leading trauma registries, to compare the predictive power of the ISS with the predictive power of the information contained in the injured patients' ICD-9 diagnoses codes. It was our primary hypothesis that survival risk ratios derived from patients' ICD-9 diagnoses codes would be equal or better predictors of survival than the Injury Severity Score. The implications of such a finding would have the potential for significant cost savings in the care of injured patients.

METHODS

Data for the test population were obtained from the SDTR, which contains data from 1985 through 1993 from five participating hospitals. Four data sources were utilized to estimate the expected survival rate/mortality rate for each ICD-9 code in the SDTR. These were (1) the SDTR patients themselves, (2) the North Carolina State Hospital Discharge Database, (3) the North Carolina Trauma Registry Database, and (4) the Agency for Health Care Policy Research's Health Care Utilization Project Database. Each of these data sources was separately utilized to develop a survival risk ratio (SRR) for each ICD-9 diagnoses code. The SRR was calculated by dividing the number of survivors for patients with each ICD-9 code by the total number of all patients with the particular ICD-9 diagnoses code. The four groups of SRRs derived from our four data sources were used as predictors of survival and the ability of the SRRs to predict survival was compared with the predictive power of the ISS using measures of accuracy, sensitivity, specificity, and receiver operator characteristic curves.

RESULTS

During the years 1985 through 1993, complete data were available for analysis on 44,032 patients. Of these, 2,848 patients died during their hospitalization (6%). Survival risk ratios were calculated for each of the diagnoses in the data base. Logistic regression, using the SAS System for statistical analysis, was used to assess the relative predictive power of the ISS and the survival risk ratios derived from the ICD-9 diagnoses codes from each of the four data bases. The analyses demonstrated that the regression models using the SRRs were generally as good or better than ISS as predictors of survival. The predictive power of the SRRs derived from the SDTR data, the North Carolina Trauma Registry data and the Health Care Utilization Report data were the best. In a subsequent analysis, the SRR values and the ISS were added to the patient's age and the revised Trauma Scores to create new predictive models in the mode of TRISS methodology. The analyses again indicated that the models using SRRs had as good or better predictive power than the model using the ISS.

CONCLUSIONS

The present study confirms previous work showing that survival risk ratios derived from injured patients' ICD-9 diagnoses codes are as good as or better than ISS as predictors of survival.

The impact of demographics, injury severity, and trauma type on the likelihood of survival in child and adolescent trauma patients

OBJECTIVE

The purpose of this study was to examine the influence of demographics, clinical factors, and injury types on the likelihood of survival.

DESIGN

This cross-sectional analysis was restricted to persons younger than 18 years, treated in trauma centers in 1990.

MATERIALS

A total of 3,540 individuals from the 1990 Virginia Statewide Trauma Registry was included in the analysis.

METHODS AND MEASUREMENTS

The outcome variable was the patient's survival likelihood after injury. The independent variables included Injury Severity Score (ISS), demographics (age, gender, and race), and injury types -- motor vehicle collision (MVC) or gunshot wound (GSW). Correlation coefficients were obtained from the study variables. Logistic regression evaluated the effect of injury severity, controlling for demographics and type of injury.

MAIN RESULTS

Three variables (ISS, GSW, and MVC) exerted significant effects of survival. Individuals with more severe injuries were more likely to die than their counterparts. Patients with gunshot wounds and motor vehicle injuries were more likely to die than those who had other injuries. After adjusting for demographics and injury type, injury severity (beta = -0.323) was found to be the most influential predictor of survival.

CONCLUSIONS

The overall findings suggest a need for hospitals to collect data routinely for calculating injury severity for the management and treatment of injured patients.