An accurate prediction of outcomes following injury is an important aspect of triage and clinical management of injured patients. Beyond clinical care, outcome prediction is essential for trauma system monitoring, resource allocation, quality improvement, and research. Injury severity scores attempt to predict outcomes by reducing the myriad complexities of a clinical situation to a single number. This chapter provides an overview of the most widely used injury severity scores, statistical considerations, and potential approaches to better predictions.

Fitting a log binomial regression model using standard software can result in numerical difficulties because its functional form does not restrict the estimated probabilities to values not exceeding unity. The common approaches to resolve the issue are to introduce a constraint to limit the results of each iteration. However, if the ML solution lies on the boundary of the allowable parameter space, some fitted probabilities are equal to unity (named as boundary vector). The common approaches have trouble dealing with those boundary vectors and can only reach somewhere before the ML solution. Previously a remedy has been proposed, but without the details necessary to implement it. Here we provide these details, including formulae for estimating the covariances essential to implement the method, an explanation of inter-dependency between coefficient estimates, and a proof that the method can be implemented in general. Code written for R implements choice of fitting algorithms, finding appropriate starting values, identifying the covariate vector(s) with a fitted probability of unity, and strategies for covariate ordering to handle the issues caused by the common values in two distinct boundary vectors. The model-fitting results are compared with two alternative methods by simulation and example data.

Foreseeing the outcome for injured patients has always been a concern of physicians, patients, and patient's families. The 20th century saw outcome prediction evolve from simply a guess to a mathematically defined probability of dying. Typically, this is done by using the information contained in past experience with many patients to create a mathematical model that estimates the future outcome for any given future patient. Of course, many different models can be created, so the process of finding the most accurate and yet parsimonious model requires data, guesswork, some mathematical intuition, and a fair bit of luck.

Background: The association between education and health has been observed globally for a multitude of health measures. Variation in county-level COVID-19 burden in the United States (U.S.) provides an opportunity to identify community pandemic risk factors. We describe the influence of secondary education on county-level COVID-19 cases and deaths in the U.S.Methods: County-level environmental, social, behavioral, economic, political, regulatory, and health characteristics were obtained. Covariates used for modeling were selected from a candidate list through negative binomial stepwise forward selection with backwards elimination, as well as clinical judgement for prediction of COVID-19 cases and deaths. Counterfactual and bootstrap analyses estimated the impact of secondary education on COVID-19 mortality.Findings: High school dropout rate is one of the strongest predictors of COVID-19 cases (IRR 1.13, 95% CI 1.07; 1.18) and deaths (IRR 1.17, 95% CI 1.10; 1.26). Counterfactual analysis reducing the county-level high school dropout rate to the lowest in the nation resulted in a predicted 14.9% decrease in median county COVID-19 deaths per 100,000 (95% CI 6.98%; 22.82%). This hypothetical improvement in secondary education translated to approximately 57,680 preventable deaths throughout 2020 (95% CI: 30,654; 84,705).Interpretation: Adjusting for covariates, secondary education is among the strongest predictors of early county vulnerability to COVID-19 in the U.S. A reduction in high school dropout rate predicted a near 15% decrease in county COVID-19 mortality. Improving secondary education could strengthen the effectiveness of a community's response to emerging threats such as a pandemic.Funding: None

Amato, Stas MD; Ranney, Stephen MD; Benson, Jamie BS; Hosmer, David W. PhD; Osler, Turner MD, FACS Author Information

Dunne, Emma BS; Amato, Stas MD; Murphy, Serena MD; Benson, Jamie BS; Osler, Turner MD, FACS; Hosmer, David PhD; An, Gary C. MD, FACS; Malhotra, Ajai K. MBBS, FACS Author Information

Attrition is common in longitudinal studies and can lead to bias when the missingness pattern affects the distributions of analysed variables. Characterisation of factors predictive of attrition is vital to longitudinal research. Few studies have investigated the factors predictive of attrition from childhood cohorts with large-scale loss to follow-up. Methods to remove potential bias are available and have been well studied in scenarios of short intervening periods between contact and follow-up. Less is known about the performance of such techniques when there is a large initial loss of participants after a long intervening period. The Australian Schools Health and Fitness Survey (ASHFS) was conducted in 1985 when participants were school children aged 7–15 years. The first follow-up occurred 20 years later with substantial loss of participants: 80% were traced, 61% enrolled and provided brief questionnaire information, 47% provided more extensive questionnaire information and 28% attended clinics. Factors associated with attrition were examined and two common techniques, multiple imputation (MI) and inverse probability weighting (IPW) were used to determine the potential for correcting the bias in the estimate of the association between self-rated fitness and BMI in childhood. Attrition from childhood to adulthood was found to be influenced by the same factors that operate in adult cohorts: lower education, lower socio-economic position and male sex. Attrition patterns varied by the stage of follow-up. Estimated childhood associations biased by adulthood attrition were able to be corrected using MI, but IPW was unsuccessful due to a lack of completely observed informative variables.

The development of rapid point-of-care tests for HIV infection has greatly reduced the problem of failure to return for test results. Test manufacturers are now developing test kits that can test for two or even three diseases at the same time, multiple-disease test kits. This study reports on the sensitivity and specificity of HIV tests when included on multi-disease test kits. 1029 participants were recruited from 2011 to 2014. HIV test kit sensitivities ranged from 91.1 to 100%, and the HIV test kit specificities from 99.5 to 100%. The two HIV kits which used oral fluid instead of blood performed well.

BACKGROUND Outcome prediction models allow risk adjustment required for trauma research and the evaluation of outcomes. The advent of ICD-10-CM has rendered risk adjustment based on ICD-9-CM codes moot, but as yet no risk adjustment model based on ICD-10-CM codes has been described. METHODS The National Trauma Data Bank provided data from 773,388 injured patients who presented to one of 747 trauma centers in 2016 with traumatic injuries ICD-10-CM codes and Injury Severity Score (ISS). We constructed an outcome prediction model using only ICD-10-CM acute injury codes and compared its performance with that of the ISS. RESULTS Compared with ISS, the TMPM-ICD-10 discriminated survivors from non-survivors better (ROC TMPM-ICD-10 = 0.861 [0.860–0.872], ROC [reviever operating curve] ISS = 0.830 [0.823–0.836]), was better calibrated (HL [Hosmer-Lemeshow statistic] TMPM-ICD-10 = 49.01, HL ISS = 788.79), and had a lower Akaike information criteria (AIC TMPM-ICD10 = 30579.49; AIC ISS = 31802.18). CONCLUSIONS Because TMPM-ICD10 provides better discrimination and calibration than the ISS and can be computed without recourse to Abbreviated Injury Scale coding, the TMPM-ICD10 should replace the ISS as the standard measure of overall injury severity for data coded in the ICD-10-CM lexicon. LEVEL OF EVIDENCE Prognostic/Epidemiologic, level II.

Background: Agreement regarding indications for vena cava filter (VCF) utilization in trauma patients has been in flux since the filter's introduction. As VCF technology and practice guidelines have evolved, the use of VCF in trauma patients has changed. This study examines variation in VCF placement among trauma centers. Materials and methods: A retrospective study was performed using data from the National Trauma Data Bank (2005-2014). Trauma centers were grouped according to whether they placed VCFs during the study period (VCF+/VCF-). A multivariable probit regression model was fit to predict the number of VCFs used among the VCF+ centers (the expected [E] number of VCF per center). The ratio of observed VCF placement (O) to expected VCFs (O:E) was computed and rank ordered to compare interfacility practice variation. Results: In total, 65,482 VCFs were placed by 448 centers. Twenty centers (4.3%) placed no VCFs. The greatest predictors of VCF placement were deep vein thrombosis, spinal cord paralysis, and major procedure. The strongest negative predictor of VCF placement was admission during the year 2014. Among the VCF+ centers, O:E varied by nearly 500%. One hundred fifty centers had an O:E greater than one. One hundred sixty-nine centers had an O:E less than one. Conclusions: Substantial variation in practice is present in VCF placement. This variation cannot be explained only by the characteristics of the patients treated at these centers but could be also due to conflicting guidelines, changing evidence, decreasing reimbursement rates, or the culture of trauma centers. (C) 2019 Published by Elsevier Inc.

Firearm violence in the United States knows no age limit. This study compares the survival of children younger than five years to children and adolescents of age 5–19 years who presented to an ED for gunshot wounds (GSWs) in the United States to test the hypothesis of higher GSW mortality in very young children. A study of GSW patients aged 19 years and younger who survived to reach medical care was performed using the Nationwide ED Sample for 2010–2015. Hospital survival and incidence of fatal and nonfatal GSWs in the United States were the study outcomes. A multilevel logistic regression model estimated the strength of association among predictors of hospital mortality. The incidence of ED presentation for GSW is as high as 19 per 100,000 population per year. Children younger than five years were 2.7 times as likely to die compared with older children (15.3% vs 5.6%). Children younger than one year had the highest hospital mortality, 33.1 per cent. The mortality from GSW is highest among the youngest children compared with older children. This information may help policy makers and the public better understand the impact of gun violence on the youngest and most vulnerable Americans.

Accurately predicting the outcomes of injury is an important aspect of the clinical and administrative management of injured patients. But are we close to defining the perfect predictor? Is a perfect predictor even possible?

BACKGROUND:The Trauma Audit and Research Network (TARN) in the UK publicly reports hospital performance in the management of trauma. The TARN risk adjustment model uses a fractional polynomial transformation of the Injury Severity Score (ISS) as the measure of anatomical injury severity. The Trauma Mortality Prediction Model (TMPM) is an alternative to ISS; this study compared the anatomical injury components of the TARN model with the TMPM.METHODS:Data from the National Trauma Data Bank for 2011-2015 were analysed. Probability of death was estimated for the TARN fractional polynomial transformation of ISS and compared with the TMPM. The coefficients for each model were estimated using 80 per cent of the data set, selected randomly. The remaining 20 per cent of the data were used for model validation. TMPM and TARN were compared using calibration curves, measures of discrimination (area under receiver operating characteristic curves; AUROC), proximity to the true model (Akaike information criterion; AIC) and goodness of model fit (Hosmer-Lemeshow test).RESULTS:Some 438 058 patient records were analysed. TMPM demonstrated preferable AUROC (0·882 for TMPM versus 0·845 for TARN), AIC (18 204 versus 21 163) and better fit to the data (32·4 versus 153·0) compared with TARN.CONCLUSION:TMPM had greater discrimination, proximity to the true model and goodness-of-fit than the anatomical injury component of TARN. TMPM should be considered for the injury severity measure for the comparative assessment of trauma centres.

IntroductionReadmission following hospital discharge is both common and costly. The Hospital Readmission Reduction Program (HRRP) financially penalizes hospitals for readmission following admission for some conditions, but this approach may not be appropriate for all conditions. We wished to determine if hospitals differed in their adjusted readmission rates following an index hospital admission for traumatic injury.Patients and MethodsWe extracted from the AHRQ National Readmission Dataset (NRD) all non-elderly adult patients hospitalized following traumatic injury in 2014. We estimated hierarchal logistic regression models to predicted readmission within 30 days. Models included either patient level predictors, hospital level predictors, or both. We quantified the extent of hospital variability in readmissions using the median odds ratio. Additionally, we computed hospital specific risk-adjusted rates of readmission and number of excess readmissions.ResultsOf the 177,322 patients admitted for traumatic injury 11,940 (6.7%) were readmitted within 30 days. Unadjusted hospital readmission rates for the 637 hospitals in our study varied from 0% to 20%. After controlling for sources of variability the range for hospital readmission rates was between 5.5% and 8.5%. Only 2% of hospitals had a random intercept coefficient significantly different from zero, suggesting that their readmission rates differed from the mean level of all hospitals. We also estimated that in 2014 only 11% of hospitals had more than 2 excess readmissions. Our multilevel model discriminated patients who were readmitted from those not readmitted at an acceptable level (C = 0.74).ConclusionsWe found little evidence that hospitals differ in their readmission rates following an index admission for traumatic injury. There is little justification for penalizing hospitals based on readmissions after traumatic injury.

Introduction The United States (US) leads all high income countries in gunshot wound (GSW) deaths. However, as a result of two decades of reduced federal support, study of GSW has been largely neglected. In this paper we describe the current state of GSW hospitalizations in the US using population-based data. Patients and methods We conducted an observational study of patients hospitalized for GSW in the National (Nationwide) Inpatient Sample (NIS) 2004 −2013. Our primary outcome is mortality after admission and we model its associations with gender, race, age, intent, severity of injury and weapon type, as well as providing temporal trends in hospital charges. Results Each year approximately 30,000 patients are hospitalized for GSW, and 2500 die in hospital. Men are 9 times as likely to be hospitalized for GSW as women, but are less likely to die. Twice as many blacks are hospitalized for GSW as non-Hispanic whites. In-hospital mortality for blacks and non-Hispanic whites was similar when controlled for other factors. Most GSW (63%) are the result of assaults which overwhelmingly involve blacks; accidents are also common (23%) and more commonly involve non-Hispanic whites. Although suicide is much less common (8.3%), it accounts for 32% of all deaths; most of which are older non-Hispanic white males. Handguns are the most common weapon reported, and have the highest mortality rate (8.4%). During the study period, the annual rate of hospitalizations for GSW remained stable at 80 per 100,000 hospital admissions; median inflation-adjusted hospital charges have steadily increased by approximately 20% annually from $30,000 to $56,000 per hospitalization. The adjusted odds for mortality decreased over the study period. Although extensively reported, GSW inflicted by police and terrorists represent few hospitalizations and very few deaths. Conclusions The preponderance GSW hospitalizations resulting from assaults on young black males and suicides among older non-Hispanic white males have continued unabated over the last decade with escalating costs. As with other widespread threats to the public wellbeing, federally funded research is required if effective interventions are to be developed.

Ordinal regression models are used to describe the relationship between an ordered categorical response variable and one or more explanatory variables. Several ordinal logistic models are available in Stata, such as the proportional odds, adjacent-category, and constrained continuation-ratio models. In this article, we present a command (ologitgof) that calculates four goodness-of-fit tests for assessing the overall adequacy of these models. These tests include an ordinal version of the Hosmer–Lemeshow test, the Pulkstenis–Robinson chi-squared and deviance tests, and the Lipsitz likelihood-ratio test. Together, these tests can detect several different types of lack of fit, including wrongly specified continuous terms, omission of different types of interaction terms, and an unordered response variable.

Among substance abusers in the US, the discrepancy in the number who access substance abuse treatment and the number who need treatment is sizable. This results in a major public health problem of access to treatment. The purpose of this study was to examine characteristics of Persons Who Use Drugs (PWUDs) that either hinder or facilitate access to treatment. 2646 participants were administered the Risk Behavior Assessment (RBA) and the Barratt Impulsiveness Scale. The RBA included the dependent variable which was responses to the question "During the last year, have you ever tried, but been unable, to get into a drug treatment or detox program?" In multivariate analysis, factors associated with being unable to access treatment included: Previously been in drug treatment (OR=4.51), number of days taken amphetamines in the last 30days (OR=1.18), traded sex for drugs (OR=1.53), homeless (OR=1.73), Nonplanning subscale of the Barratt Impulsiveness Scale (OR=1.19), age at interview (OR=0.91), and sexual orientation, with bisexual men and women significantly more likely than heterosexuals to have tried but been unable to get into treatment. The answers to the question on "why were you unable to get into treatment" included: No room, waiting list; not enough money, did not qualify, got appointment but no follow through, still using drugs, and went to jail before program start. As expected, findings suggest that limiting organizational and financial obstacles to treatment may go a long way in increasing drug abuse treatment accessibility to individuals in need. Additionally, our study points to the importance of developing approaches for increasing personal planning skills/reducing Nonplanning impulsivity among PWUDs when they are in treatment as a key strategy to ensure access to additional substance abuse treatment in the future.

We examine three approaches for testing goodness of fit in ordinal logistic regression models: an ordinal version of the Hosmer-Lemeshow test (), the Lipsitz test, and the Pulkstenis-Robinson (PR) tests. The properties of these tests have previously been investigated for the proportional odds model. Here, we extend the tests to two other commonly used models: the adjacent-category and the constrained continuation-ratio models. We use a simulation study to assess null distributions and power. All three tests work well and can detect several types of lack of fit under both the adjacent-category and constrained continuation-ratio models. The and Lipsitz tests are best suited to detect lack of fit associated with continuous covariates, whereas the PR tests excel at detecting lack of fit associated with categorical covariates. We illustrate the use of the tests with data from a study of aftercare placement of psychiatrically hospitalized adolescents. Based on the results here and previous research, we can make a joint recommendation for testing goodness of fit in proportional odds, adjacent-category, and constrained continuation-ratio logistic regression models: because the tests may detect different types of lack of fit, a thorough assessment of goodness of fit requires use of all three approaches.

IMPORTANCE:The GCS was created forty years ago as a measure of impaired consciousness following head injury and thus the association of GCS with mortality in patients with traumatic brain injury (TBI) is expected. The association of GCS with mortality in patients without TBI (non-TBI) has been assumed to be similar. However, if this assumption is incorrect mortality prediction models incorporating GCS as a predictor will need to be revised. OBJECTIVE:To determine if the association of GCS with mortality is influenced by the presence of TBI. DESIGN/SETTING/PARTICIPANTS:Using the National Trauma Data Bank (2012; N=639,549) we categorized patients as isolated TBI (12.8%), isolated non-TBI (33%), both (4.8%), or neither (49.4%) based on the presence of AIS codes of severity 3 or greater. We compared the ability GCS to discriminate survivors from non-survivors in TBI and in non-TBI patients using logistic models. We also estimated the odds ratios of death for TBI and non-TBI patients at each value of GCS using linear combinations of coefficients. MAIN OUTCOME MEASURE:Death during hospital admission. RESULTS:As the sole predictor in a logistic model GCS discriminated survivors from non-survivors at an acceptable level (c-statistic=0.76), but discriminated better in the case of TBI patients (c-statistic=0.81) than non-TBI patients (c-statistic=0.70). In both unadjusted and covariate adjusted models TBI patients were about twice as likely to die as non-TBI patients with the same GCS for GCS values<8; for GCS values>8 TBI and non-TBI patients were at similar risk of dying. CONCLUSIONS:A depressed GCS predicts death better in TBI patients than non-TBI patients, likely because in non-TBI patients a depressed GCS may simply be the result of entirely reversible intoxication by alcohol or drugs; in TBI patients, by contrast, a depressed GCS is more ominous because it is likely due to a head injury with its attendant threat to survival. Accounting for this observation into trauma mortality datasets and models may improve the accuracy of outcome prediction.

Algebraic relationships between Hosmer-Lemeshow (HL), Pigeon-Heyse (J(2)), and Tsiatis (T) goodness-of-fit statistics for binary logistic regression models with continuous covariates were investigated, and their distributional properties and performances studied using simulations. Groups were formed under deciles-of-risk (DOR) and partition-covariate-space (PCS) methods. Under DOR, HL and T followed reported null distributions, while J(2) did not. Under PCS, only T followed its reported null distribution, with HL and J(2) dependent on model covariate number and partitioning. Generally, all had similar power. Of the three, T performed best, maintaining Type-I error rates and having a distribution invariant to covariate characteristics, number, and partitioning.