Traumatic brain injury (TBI) is a major cause of disability, morbidity, and mortality. The effect of the acute respiratory distress syndrome and acute lung injury (ARDS/ALI) on in-hospital mortality after TBI remains controversial.
To determine the epidemiology of ARDS/ALI, the prevalence of risk factors, and impact on in-hospital mortality after TBI in the United States.
Retrospective cohort study of admissions of adult patients >18 years with a diagnosis of TBI and ARDS/ALI from 1988 to 2008 identified through the Nationwide Inpatient Sample.
During the 20-year study period, the prevalence of ARDS/ALI increased from 2% (95% confidence interval [CI], 2.1%–2.4%) in 1988 to 22% (95% CI, 21%–22%) in 2008 (P < .001). ARDS/ALI was more common in younger age; males; white race; later year of admission; in conjunction with comorbidities such as congestive heart failure, hypertension, chronic obstructive pulmonary disease, chronic renal and liver failure, sepsis, multiorgan dysfunction; and nonrural, medium/large hospitals, located in the Midwest, South, and West continental US location. Mortality after TBI decreased from 13% (95% CI, 12%–14%) in 1988 to 9% (95% CI, 9%–10%) in 2008 (P < .001). ARDS/ALI-related mortality after TBI decreased from 33% (95% CI, 33%–34%) in 1988 to 28% (95% CI, 28%–29%) in 2008 (P < .001). Predictors of in-hospital mortality after TBI were older age, male sex, white race, cancer, chronic kidney disease, hypertension, chronic liver disease, congestive heart failure, ARDS/ALI, and organ dysfunctions.
Our analysis demonstrates that ARDS/ALI is common after TBI. Despite an overall reduction of in-hospital mortality, ARDS/ALI carries a higher risk of in-hospital death after TBI.
Traumatic brain injury (TBI) is the major cause of disability, morbidity, and mortality among individuals in the United States and developed countries.1 The acute respiratory distress syndrome (ARDS)2 or the less severe acute lung injury (ALI) are the result of insults to the lung and are characterized by inflammation of the lung parenchyma leading to impaired gas exchange resulting in hypoxemia and abnormal lung physiology.3 Current definitions of ARDS/ALI applied to research and bedside patient care are based on the American-European Consensus Conference (AECC) on ARDS, published in 1994.4
The ARDS/ALI has a tremendous impact on public health with an estimated in-hospital mortality of 39% and an economic burden for society of more than 3 million hospital-days per year.5 Additional epidemiological studies in other cohorts of primarily medical and surgical populations have reported variable incidence rates.6–12 Similarly, small-cohort studies after TBI have shown a prevalence of ARDS/ALI of 8% to 31%13,14 but have confirmed an inconsistent effect of ARDS/ALI on in-hospital mortality.13,14
The purpose of this study was to assess the national trends of ARDS/ALI in the United States after admission for TBI, with specific examination of its prevalence, risk factors, and effect on outcome by using a robust administrative dataset.
PATIENTS AND METHODS
Retrospective cohort study of the Nationwide Inpatient Sample (NIS). The NIS is a database maintained as part of the Healthcare Cost and Utilization Project of the Agency for Healthcare Research and Quality.15 The institutional review board of our hospital exempted this analysis from full review.
Data Sample and Definition of Outcome Variables
Hospitalizations of patients ≥18 years with a primary diagnosis of TBI were identified by querying the database between 1988 and 2008. Age, sex, and race were obtained from the NIS records. The International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) was used to identify admissions of TBI patients16–19 (see Table, Supplemental Digital Content 1, http://links.lww.com/NEU/A475). Those who developed ARDS/ALI during hospital admission were identified through querying secondary diagnoses. Our methods for selecting ARDS/ALI were derived from Reynolds et al8 and Thomsen and Morris9 (see Table, Supplemental Digital Content 1, http://links.lww.com/NEU/A475). Querying all diagnosis fields based on the appropriate ICD-9-CM codes identified comorbidities20,21 and in-hospital complications (see Table, Supplemental Digital Content 1, http://links.lww.com/NEU/A475).20–23 The outcome was in-hospital mortality.
Cases with ARDS/ALI were patients admitted to a sea-level hospital in 2011 with ARDS/ALI8,9 ICD-9-CM codes in their discharge records. The controls were patients admitted immediately before or after each identified patient with ARDS/ALI who were included if there were no ARDS/ALI8,9 codes in their discharge records. We identified date of onset of hypoxemia, calculated PaO2/FiO2, verified the presence of infiltrates, and documented pulmonary artery occlusion pressures (PAOPs) and/or results of echocardiography. ARDS/ALI was deemed to be present when the AECC definition was met.4
National estimates were calculated according to accepted guidelines for the accuracy of NIS- Healthcare Cost and Utilization Project data, that is, using the provided discharge weights and accounting for stratification and clustering effects. Differences between weighted national and prevalence estimates were quantified with the Z-score test. Continuous data are presented as medians and interquartile ranges and categorical data are presented as proportions and 95% confidence intervals (CIs). The χ2 test or the Wilcoxon rank tests were used to determine differences in demographics, hospital characteristics, comorbidities, in-hospital complications, procedures, and outcomes. The multivariate analysis proceeded in 2 stages. In the first stage, significant risk factors were identified from the candidate variables. Multivariable logistic regression modeling was used to calculate odds ratios and 95% CI. In all multivariate analyses, all factors of interest were included, and parsimonious models were found by systematically removing the least significant factor and recalculating the model. In the second stage, potential hospital effects were assessed with generalized estimated equations (GEEs).24 For the validation study, we calculated: sensitivity, specificity, positive predictive value (PPV), and receiver operator characteristic curve. The analysis was conducted by using Structured Query Language (SQL) and the LME-4 package (version 0.99) in the R-programming language for statistical computing (version 2.11, http://cran.r-project.org) and commercially available statistical software (SPSS-IBM Version 19.0 for Macintosh). Statistical significance was judged when P < .05.
During the 20-year study period, there were more than 750 million hospitalizations in the United States. The demographic characteristics and coexisting conditions in the population of admissions of patients with TBI with and without ARDS/ALI are shown in Table 1.
During the 20-year study period, there were 987 305 admissions of patients with TBI of which 209 342 corresponded to a diagnosis of ARDS/ALI for a total prevalence of 21.2% (95% CI, 21.0%–21.4%). The prevalence of ARDS/ALI after TBI increased from 2% (95% CI, 2.1%–2.4%) in 1988 to 22% (95% CI, 21%–22%) in 2008 (P < .001) (Figure 1).
Independent predictors of ARDS/ALI are shown in Table 2. Diabetes mellitus and cancer were protective for the development of ARDS/ALI. In-hospital complications, such as sepsis, cardiovascular dysfunction, renal dysfunction, and hematological dysfunction, were found to be associated with ARDS/ALI after admission for TBI. Hospital characteristics such as nonrural setting, medium/large size, and Midwest, South, and West continental US location were independently associated with ARDS/ALI.
Mortality after TBI decreased from 13% (95% CI, 12%-14%) in 1988 to 9% (95% CI, 9%-10%) in 2008 (P < .001). Admissions of patients with ARDS/ALI had a 3-fold increase in the odds of in-hospital mortality. ARDS/ALI-related mortality after TBI decreased from 33% (95% CI, 33%–34%) in 1988 to 28% (95% CI, 28%-29%) in 2008 (P < .001). Independent predictors of in-hospital mortality after TBI are shown in Table 3. The age-adjusted in-hospital mortality for TBI admissions was higher in older patients with ARDS/ALI and in younger patients without ARDS/ALI (Figure 2). The multivariate analyses and GEE models that accounted for patient clustering within hospitals did not substantively change the point estimates or CIs, suggesting that a significant hospital effect was not present in the model (Table 3).
The combination of codes used8,9 carried a PPV of 100%, sensitivity of 71%, and specificity of 100% (area under the curve 0.86) for the identification of cases with ARDS/ALI when the AECC criteria of PaO2/FiO2 plus chest x-ray (CXR) was applied; and PPV of 90%, sensitivity of 53%, and specificity of 73% (area under the curve 0.77) when AECC criteria of PaO2/FiO2, CXR, and exclusion of high left atrial pressures (normal echocardiography) was applied (Figure 3). We did not encounter any cases with measurements of PAOP by pulmonary artery catheterization.
ARDS/ALI is a common in-hospital complication after admission for TBI. Our study demonstrates that the mortality after TBI is decreasing in the United States, whereas rates of ARDS/ALI have increased significantly and may be explained by better diagnostic definition,4 epidemiological variables, comorbid conditions, and changing economic incentives for better coding and billing.25
In the general population, the incidence of ARDS/ALI increases with age, because the population is at higher risk for one of the most prevalent risk factors: sepsis.5,22,26 However, the effect of age on the development of ARDS/ALI may be modified by disease-specific characteristics. In our TBI cohort, younger patients were more likely to develop ARDS/ALI, which has been seen in general trauma patients,27 a phenomenon that may be explained by an interaction between specific disease characteristics, a healthy aging effect, the fact that older TBI patients die in the hospital before the onset of ARDS/ALI, that critically ill older TBI patients may be withdrawn from life support before onset of ARDS/ALI, and the higher age-specific incidence of TBI in younger cohorts.17
In our cohort, males and the white race were at a higher risk of developing ARDS/ALI. A recent study in blunt trauma patients demonstrated that black race/ethnicity had a protective effect in relation to the development of ARDS/ALI, and Hispanic ethnicity was associated with a higher mortality and case fatality.28 This effect may be related to genetic or hormonal factors that predispose to ARDS/ALI, difference in susceptibility to risk factors for ARDS/ALI, or disparities in the access to specific therapies for ARDS/ALI.29 The discrepancy between prior epidemiological studies in ARDS cohorts30 and our results may also be based on the stratification of race/ethnicity, which is confounded by socioeconomic status and underreporting of race status in general, which accounts for to up to 20% in the NIS cohort.15
Our analysis shows an independent effect of male sex and onset of ARDS/ALI after TBI. Males are uniformly at higher risk of TBI than are females, with the higher male-to-female ratios in adolescence and young adulthood.17 TBI severity may be associated with the higher risk of ARDS/ALI among males. The high male-to-female ratio seen in TBI cohorts is explained by interpersonal violence and motor vehicle collisions,17 which are generally associated with higher severity injury scores and multiorgan trauma, a predisposing risk for the development of complications associated with ARDS/ALI. Traditional risk factors for ARDS/ALI that have been reported in the literature include drug reactions, burns, inhalation injuries, pancreatitis, amniotic fluid embolism, and blood transfusions.31 In most clinical trials and cohort studies, most of the cases of ARDS/ALI are related to pneumonia, sepsis, and trauma.29 Risk factors associated with ARDS/ALI in our TBI cohort were sepsis, hypertension, renal failure, and congestive heart failure, whereas diabetes and cancer were protective. Hypertension and congestive heart failure may have biased the ascertainment of ARDS/ALI. It is plausible that patients with higher left atrial pressures and pulmonary edema for cardiac reasons may have been misdiagnosed as ARDS/ALI, because excluding cardiogenic or hydrostatic pulmonary edema on clinical grounds, even in the presence of PAOP measurements, is problematic.32 The use of markers of myocardial wall stretch (brain natriuretic peptide) does not distinguish cardiogenic from noncardiogenic respiratory failure in relevant cases in the ICU,33,34 which is supported by the results of our validation scheme. When we used echocardiography as a surrogate for exclusion of elevated left atrial pressures, our sensitivity for diagnosis of ARDS/ALI decreased from 71% to 53%, and its PPV decreased from 100% to 90%, meaning that, in the absence of PAOP or echocardiography as surrogates of normal left atrial pressures, a false positive rate of up to 20% with subsequent overestimation in the prevalence of ARDS/ALI may be expected.
Organ dysfunctions and sepsis were associated with ARDS/ALI and higher mortality. Sepsis may be a cause of ARDS/ALI,5,22,26 but multiorgan failure may not be causal but a result of sepsis or ARDS/ALI. In septic shock cohorts, diabetic patients were less likely to acquire ARDS/ALI even after adjusting for other confounders.35 A plausible explanation for this observation may be related to a direct biological effect or it might because these patients are more likely to develop sepsis from genitourinary sources rather than respiratory sources.29 Cancer has been associated with increased mortality and prevalence after ARDS/ALI.7,36,37 Although we found a higher mortality for TBI admissions with cancer, this comorbidity conferred a negative or “protective” effect for the development of ARDS/ALI. This might be because TBI patients with cancer die or are withdrawn from life support in the hospital before the onset of ARDS/ALI.
Admissions of TBI patients to private as well as academic hospitals in urban settings were more likely to be associated with the onset of ARDS/ALI. Larger centers may serve as trauma centers with multidisciplinary teams and act as referral centers for such patients from other smaller and rural establishments, which in turn favors transfers of “sicker” patients that may require a higher level of care. The overrepresentation of trauma centers, particularly levels III to V in the Midwest, West, and South of the US, could have explained our observed association.38 However, mortality after TBI was not significantly different when accounting for hospital location in multivariate analysis or with GEE modeling.
Our study demonstrates that the mortality after TBI is decreasing in the United States. Overall, our observed in-hospital mortality fell from 13% in 1988 to 9% in 2008. ARDS/ALI-related mortality decreased from 33% in 1988 to 28% in 2008, supporting results from smaller cohorts,14 and seems to be lower than the early mortality reported by cohort studies of nontraumatic patients, which was as high as 39%.5 The combination of declining in-hospital mortality and increasing incidence29 as the population ages suggests that caring for the long- term sequelae of survivors of ALI/ARDS after TBI will be an increasingly important public health problem in the future.1
The recognition of ARDS/ALI is important for both research and bedside management. However, the diagnosis of ARDS/ALI is challenging. There is no definite laboratory test, imaging study, or biological marker to diagnose ARDS/ALI; so patients would meet the syndromatic diagnosis of ARDS/ALI once they have met laboratory and radiographic criteria set by a group of experts.29 The extrapolation of this syndromatic diagnosis to epidemiological research offers an additional challenge, because the validity of the definition remains controversial.39 The oxygenation component depends on ventilator settings40; excluding cardiogenic or hydrostatic pulmonary edema on clinical grounds, even in the presence of PAOP measurements, is problematic32; and the CXR definition shows a high degree of observer bias even when assessed by experienced operators.41 The accuracy of ICD-9-CM coding for the identification of ARDS/ALI remains controversial, but its accuracy has been demonstrated previously.9,42 Thomsen and Morris9 assessed the accuracy of the ICD-9-CM documentation for ARDS in the primary hospital of the study; the sensitivity and specificity of the codes used were 88% and 99%, respectively. Several studies in different institutions have shown that clinical recognition, documentation, and accuracy in the ascertainment of ARDS/ALI in clinical practice may be variable, with sensitivities and specificities ranging between 20% to 49% and 97% to 100%, respectively.42–44 The modification of ascertainment of ARDS/ALI by Reynolds et al,8 used in our validation, includes descriptors ARDS/ALI and ventilator use and is superior because the combination of diagnostic criteria and ventilatory support may be mandatory when studying the epidemiology of ARDS/ALI in large administrative databases.8 Our validation step demonstrated that the combination of codes used8,9 carried a high PPV with higher sensitivities than those reported in the literature.42–44 Despite these shortcomings, the most important field test of the validity of the AECC criteria has been provided by the ARDS Network Clinical Trials, which have demonstrated its ability to identify patient populations that would respond to specific therapies in ARDS.45,46 The lack of PAOP data reflects current practices and trends in the use of pulmonary artery catheterization in general,47 arguing in favor of a revised definition of ARDS/ALI.39
Our study provides important information for clinical practice and public health. Accurate national estimates regarding the temporal trends in the epidemiology of TBI are important for the allocation of health care resources, for the evaluation of health care delivery related to this devastating disease, and for future research studies and budgets, so our data may have potential implications. Although we have plausible explanations for the disparities among races and between men and women, we must explain these differences in future studies. Similarly, the encouraging changes and certain observed disparities in mortality need to be investigated by future studies that will compare the effectiveness of care related to survival and disposition after TBI; and the efficacy and effect on outcomes, particularly cognitive outcomes,48 of future treatment strategies for ARDS/ALI and TBI may need to be evaluated against these historic trends.
Our study has limitations. First, our analysis was observational in nature, limiting the inferences that can be made about causal relationships, so our results should be interpreted with caution, particularly the changes in the prevalence of ARDS/ALI, which may have followed epidemiological variables such as better diagnostic definition and changing in coding and billing practices. Second, ICD-9-CM codes have historically been of questionable accuracy in general. However, our validation scheme using ICD-9-CM diagnostic and procedural codes achieved a high PPV with sensitivities superior to those reported in the literature, implying single-center bias in ascertainment. Extrapolating this to the NIS universe, a moderate sensitivity means an underestimation of the true prevalence of ARDS/ALI within TBI admissions, but, despite this limitation, the NIS database allows investigators to estimate national trends of several variables important for the understanding of the behavior of certain diseases and helping in the allocation of health care resources.22,49 Third, the NIS did not allow us to evaluate the effects of other variables that may have confounded our observed associations such as the Glasgow Coma Score, extent of the severity of trauma, trauma center designation, and effect of medications and interventions. Nevertheless, our study provides insight to the accuracy of the ICD-9-CM coding for identification of ARDS/ALI cases and on the burden of ARDS/ALI on TBI patients, particularly on mortality.
The prevalence of ARDS/ALI is significant after TBI. Critical care specialists caring for these patients should be aware of risk factors and the impact of ARDS/ALI in patient's outcomes.
This is an excellent summary of data in the NIS Database at AHRQ. The changes in the incidence of ARDS/ALI associated with TBI from 1988 to 2008 are dramatic: 2% to 22%. This may in part represent improved detection and diagnosis, but because the definition of ARDS/ALI is clearly stated in terms of tests that were routinely obtained in 1988 I am inclined to think that the increase is real. The independent effect of hospital size, rural vs nonrural hospitals, and geographic location in the continental US are all concerning and emphasize the inconsistencies in quality of health care provided throughout the US. This is most especially illustrated by the finding that Black race is protective, and mortality rates are higher in Hispanics, strongly suggesting an important effect of access to care. The authors of this study have carefully considered all of the most important variables and presented important new insight into the contemporary care of patients with TBI associated with chest injury.
In this interesting epidemiological study the authors present a retrospective cohort analysis of patients registered in the National Inpatient Sample from the years 1998 to 2008. The authors identified patients over the age of 18 years who were registered with the diagnosis of traumatic brain injury (TBI) and acute lung injury (ALI) or acute respiratory distress syndrome (ARDS). They found a remarkable ten-fold increase in the incidence of ALI/ARDS in TBI patients between 1998 (2%) and 2008 (22%). However, as the authors point out improved coding and reporting in the database over time may have influenced this finding, especially since changing economic incentives over time may have favored more accurate coding. Given this limitation, this particular finding needs to be interpreted with caution. The authors stress that mortality related to ALI/ARDS decreased from 33% to 28% in the cohort studied. This is certainly an important finding, but in my view an even more interesting finding is the decrease in overall mortality in TBI patients over the 20-year study period from 13% to 9%. I am not sure that this substantial decrease in mortality can be explained by the decrease in ALI/ARDS related deaths. Although it will be difficult to sort out the primary determinates of the decrease in mortality due to limitations in the database, this is certainly a topic that deserves further study.
San Francisco, California
American-European Consensus Conference
acute lung injury
acute respiratory distress syndrome
generalized estimated equation
International Classification of Diseases, Ninth Revision, Clinical Modification
Nationwide Inpatient Sample
pulmonary artery occlusion pressures
positive predictive value
traumatic brain injury