Adult respiratory distress syndrome: a complication of induced hypertension after severe head injury

Transcription

1 J Neurosurg 95: , 2001 Adult respiratory distress syndrome: a complication of induced hypertension after severe head injury CHARLES F. CONTANT, PH.D., ALEX B. VALADKA, M.D., SHANKAR P. GOPINATH, M.D., H. JULIA HANNAY, PH.D., AND CLAUDIA S. ROBERTSON, M.D. Department of Neurosurgery, Baylor College of Medicine, Houston, and Department of Psychology, University of Houston, Texas A Object. The factors involved in the development of adult respiratory distress syndrome (ARDS) after severe head injury were studied. The presence of ARDS complicates the treatment of patients with severe head injury, both because hypoxia causes additional injury to the brain and because therapies that are used to protect the lungs and improve oxygenation in patients with ARDS can reduce cerebral blood flow (CBF) and increase intracranial pressure (ICP). In a recent randomized trial of two head-injury management strategies (ICP-targeted and CBF-targeted), a fivefold increase in the incidence of ARDS was observed in the CBF-targeted group. Methods. Injury severity, physiological data, and treatment data in 18 patients in whom ARDS had developed were compared with the remaining 171 patients in the randomized trial in whom it had not developed. Logistic regression analysis was used to study the interaction of the factors that were related to the development of ARDS. In the final exact logistic regression model, several factors were found to be significantly associated with an increased risk of ARDS: administration of epinephrine (5.7-fold increased risk), administration of dopamine in a larger than median dose (10.8-fold increased risk), and a history of drug abuse (3.1-fold increased risk). Conclusions. Although this clinical trial was not designed to study the association of management strategy and the occurrence of ARDS, the data strongly indicated that induced hypertension in this high-risk group of patients is associated with the development of symptomatic ARDS. KEY WORDS cerebral perfusion pressure traumatic brain injury severe head injury secondary ischemic insult adult respiratory distress syndrome intracranial hypertension DULT respiratory distress syndrome has been reported in approximately 20% of patients with severe head injury. 4 An even greater percentage of patients with severe head injury, perhaps as high as 50%, have increased extravascular fluid accumulation in the lungs. 12 The development of ARDS complicates the treatment of head-injured patients (Fig. 1), because many therapies that are protective for lungs afflicted with ARDS can raise ICP or decrease CPP. Fluid restriction, diuretic medications, and pulmonary vasodilating agents, as well as PEEP and permissive hypercapnia are commonly used in the treatment of patients with ARDS. In addition, hypoxia from ARDS can be a secondary insult that further injures the brain. Based on the TCDB studies, it has been estimated that the development of acute lung injury is Abbreviations used in this paper: AIS = Abbreviated Injury Scale; ARDS = adult respiratory distress syndrome; BP = blood pressure; CBF = cerebral blood flow; CPP = cerebral perfusion pressure; CT = computerized tomography; CVP = central venous pressure; DRS = Disability Rating Scale; ER = emergency room; FiO 2 = fraction of inspired oxygen; GCS = Glasgow Coma Scale; GOS = Glasgow Outcome Scale; ICP = intracranial pressure; IR = interquartile range; MABP = mean arterial blood pressure; PEEP = positive end-expiratory pressure; PT = prothrombin time; PWP = pulmonary wedge pressure; TCDB = Traumatic Coma Data Bank. associated with a threefold increased risk of dying or remaining in a vegetative state after a severe head injury. 4 Fluid administration and the management of BP and CPP can be particularly difficult issues in these patients. In patients with severe head injury, the occurrence of hypotension has been associated in several studies with a poor neurological outcome. Induced hypertension has been advocated to raise CPP to at least 70 mm Hg. Increased pulmonary hydrostatic pressures, however, can increase the amount of water accumulating within the lungs. 23 For a high-risk group such as head-injured patients, the practice of induced hypertension could increase the incidence and severity of symptomatic ARDS. The effect of induced hypertension on the incidence and the severity of ARDS in the head-injured population has not been closely studied; respiratory failure has not been a particular concern in recent clinical series in which induced hypertension was used. In a clinical trial in which the effects of two management strategies on the incidence of secondary ischemic insults and on long-term neurological outcome were compared, a significantly higher incidence of ARDS was observed in the group of patients treated with induced hypertension. 21 The purpose of this follow-up study was, first, to examine risk factors for development of ARDS in the head-injured patients enrolled in this trial; second, to ex- 560

2 Adult respiratory distress syndrome after head injury amine the role of induced hypertension in the development of ARDS; and third, to study the consequences of development of ARDS for long-term outcome. Clinical Material and Methods Design of the Study The design of this study is described in detail in our previous publication, 21 and was a randomized clinical trial of two management strategies. The eligibility criteria for the trial included the following: 1) that the patient be comatose due to head injury (motor GCS score 5 on admission or deterioration to motor GCS score 5 within 48 hours postinjury); 2) age 15 years or older; and 3) admission within 12 hours of injury. The exclusion criteria included the following: 1) GCS Score 3 with fixed and dilated pupils after resuscitation; 2) contraindication to placement of jugular bulb catheter; and 3) severe associated systemic injury. This study was approved by the institutional review board for human subjects, and informed consent was obtained from the nearest relative of the patients. Patients were randomly assigned to either the ICP-targeted or the CBF-targeted treatment groups. Randomization was performed as follows: the year was divided into three 4-month periods based on the rotation of neurosurgery residents in the intensive care unit. Both treatment protocols were used for 2 months during each 4-month period, but the order of the protocols was randomly assigned for each physician group rotation. All patients admitted to the intensive care unit during each 2-month time period received the same assigned treatment. The major differences between the ICP-targeted protocol and the CBF-targeted protocol were the treatment goals for MABP, CPP, and PaCO 2. In the ICP-targeted protocol, MABP of at least 70 mm Hg was maintained, and systolic BPs greater than 160 mm Hg were treated with antihypertension agents. The CPP was maintained at a minimum of 50 mm Hg. Although hyperventilation was not used routinely in either protocol, hyperventilation to a PaCO 2 of 25 to 30 mm Hg was used as a treatment for increased ICP in the ICP-targeted protocol. In the CBFtargeted protocol, MABP was kept at 90 mm Hg or more and CPP was maintained at a minimum of 70 mm Hg. Hyperventilation was not used to treat intracranial hypertension. Treatment of intracranial hypertension in both groups included cerebrospinal fluid drainage, sedation, and administration of mannitol, a neuromuscular blocking agent, followed by a barbiturate-induced coma if the ICP was not controlled with mannitol. To evaluate the incidence of ARDS, chest x-ray films and arterial blood gas data were reviewed; ARDS was identified according to the definition recommended by the American-European Consensus Report. 2 The criteria included the following factors: 1) acute onset; 2) PaO 2 / FiO or less; 3) bilateral pulmonary infiltrates; and 4) PWP 18 mm Hg or less, or no evidence of left-sided heart failure. The neurological outcome was determined at 3 months and 6 months postinjury by personnel who were unaware of the patient s treatment, based on the GOS 10 score and the DRS 19 score. FIG. 1. Chart demonstrating that the development of ARDS complicates the management of severe head injury. Hypoxia is an important cause of secondary injury to the brain. The management strategies intended to improve oxygenation and to protect the lung from barotrauma, however, can reduce BP, increase ICP, and result in hypercarbia. I:E = inspiration/expiration; NO = nitric oxide; pco 2 = PaCO 2. Statistical Methods Categorical data are presented as proportions. Comparisons of categorical and nominal data were performed using the chi-square test for contingency tables or, when the expected data set sizes were small, the Fisher exact test. Continuous data are presented as the means standard deviation or the medians and IRs. Comparisons of continuous data were performed using the Mann Whitney or Kruskal Wallis tests for comparing medians. To evaluate the effects of more than one predictor on the occurrence of ARDS, logistic regression models were fit. The number of patients with ARDS (18 cases) was too small to allow screening of a large number of variables in a single logistic model. To address this statistical problem, the following approach was adopted: logistic regression was used to screen candidate variables as described later, then exact logistic regression models were built from the variables found to be significant in the screening models. Initially, the independent variables were grouped into those representing the baseline characteristics of the patient and those representing treatment. Based on the relationship between the individual variables and ARDS, a set of candidate regressors was selected for each regression model. To screen these variables, a series of logistic regression models were estimated using commercially available software (S-Plus 2000, release 3; Insightful Corp., Seattle, WA). The robust variance estimate was used. An indicator variable for the protocol was included in each model (1 = ICP-targeted, 0 = CBF-targeted). A model was then fit with the protocol and one of the candidate regressors for each regressor. The model that was most significant, as measured by the change in log likelihood from the model with just protocol as a regressor, was then used in the next phase of modeling as the base model. As before, each of the remaining candidate variables was added singly, and the most significant model was used as the base model in the following step. When no other candidate variables were found to contribute significantly to the model, the final model was examined. 561

3 C. F. Contant, et al. TABLE 1 Comparison of demographic characteristics in patients with and without ARDS* No. of Patients (%) p Characteristic W/O ARDS W/ ARDS Value age (yrs) 30 (IR 23 43) 29.5 (IR 21 39) male sex 143 (83.6) 18 (100) ethnicity Asian 7 (4.1) 2 (11.1) African American 40 (23.4) 2 (11.1) hispanic 70 (40.9) 4 (22.2) caucasian 54 (31.6) 10 (55.6) history of drug abuse present 60 (35.1) 11 (61.1) absent 45 (26.3) 0 (0.0) unknown 66 (38.6) 7 (38.9) GCS score in ER postresus (29.8) 6 (33.3) (40.9) 6 (33.3) 8 50 (29.2) 6 (33.3) best GCS score on Day (16.4) 3 (16.7) (56.7) 12 (66.7) 8 46 (26.9) 3 (16.7) 1 or both pupils abnormal ER 93 (54.4) 10 (55.6) Day 1 98 (57.3) 13 (72.2) presence of systemic injuries ISS (64.3) 13 (72.2) ISS (8.2) 2 (11.1) ISS (18.1) 3 (16.7) ISS (9.4) 0 (0.0) prehospital hypoxia 60 (35.1) 6 (33.3) prehospital hypotension 17 (10.0) 4 (22.2) * ISS = Injury Severity Scale; postresus = postresuscitation. From this model, each variable was excluded, one at a time, to determine the variable that contributed the least to the overall model. This variable was then removed, and the process was repeated until all the remaining variables were significant. The variables in this final logistic regression model were used to fit a model by using the exact logistic regression methods implemented in version 2.1 of the LogXact program (Cytel Software Corp., Cambridge, MA). Exact logistic regression is a method for estimating the coefficients of a model when the number of patients with the outcome is small. It uses sophisticated methods to provide coefficient estimates that are appropriate for small samples. In fitting the exact models, continuous variables could not be used as such. These variables were transformed so that a two-category variable was produced, indicating whether the measurement was above or below the median for that variable. This process was repeated for both the baseline and treatment variables. In both types of models, the variable indicating which treatment group was being assessed (ICP-targeted or CBF-targeted) was included. The variables found to be significant in each of the two exact logistic regression models (baseline and treatment variables) were then combined into the final model. Finally, because all the cases of ARDS occurred in male patients, the models were refit to include only the male patients. Results Demographic and Injury Characteristics in Patients With ARDS One hundred eighty-nine patients were entered into the study: 100 in the CBF-targeted group and 89 in the ICP-targeted group. The two treatment groups were well balanced for demographic characteristics and for injury severity parameters, as discussed in detail in a previous publication. 21 Overall, 18 (10%) of the 189 patients enrolled in the trial experienced ARDS. The median PaO 2 / FiO 2 ratio was 99 (IR ). All of the patients required PEEP at a median level of 10 cm (IR cm), and a high FiO 2 to achieve adequate oxygenation. In five (2.6%) of the 189 patients, ARDS either directly caused or contributed to the patient s death. The demographic characteristics of the patients with ARDS are compared with those without ARDS in Table 1. The initial diagnosis was isolated head injury in 13 of the 18 patients with ARDS. Only five patients had other systemic injuries that might have predisposed them to pulmonary complications: one had rib fractures, three had lower extremity fractures, and one had an abdominal injury, but the highest AIS score for any of these associated injuries was 3. The severity of the brain injury, judged by the ER personnel, and taking into account the best Day 1 GCS score and pupillary reactivity, or by the occurrence of preadmission hypoxia, was not significantly different in the patients with ARDS compared with those without ARDS. The only demographic characteristics that distinguished the ARDS group were that all of the patients were male (p = 0.048) and that they were more likely to have had a history of drug abuse (p = 0.011). There was a suggestion that a higher proportion of the patients without ARDS were noncaucasian (p = 0.063) and were more likely to have suffered hypotension previously (p = 0.105). In Table 2, the injury characteristics of the patients with ARDS are contrasted with the characteristics of those without ARDS. There were no significant differences in the types of injury between the two groups of patients, either on the admission CT scan or on the worst CT scan obtained during the hospital stay. The patients with ARDS tended to have a higher incidence of signs of brain swelling on admission CT scans; such signs included absent basal cisterns (33.3% compared with 17.5%), compressed ventricles (50% compared with 39.2%), and midline shifts greater than 15 mm (16.7% compared with 8.2%), but none of these differences were significant (p = 0.159, p = 0.651, and p = 0.109, respectively). In a greater percentage of the patients with ARDS, however, the appearance of their injury worsened on CT scans obtained during the hospitalization. Among the patients who did not have ARDS, 18 (10.5%) of 171 developed a worsening appearance of their injury on CT scans compared with five (27.8%) of the 18 patients with ARDS (exact test, p = 0.049). Among the 18 patients in the group without ARDS, 11 changed from having a diffuse injury to having a mass lesion, whereas changes in all five of the patients with ARDS were of this type (exact test, p = 0.022). There were no statistically significant differences between patients with and those without ARDS with respect to variables measured in the ER, with the exception of the 562

4 Adult respiratory distress syndrome after head injury TABLE 2 Comparison of admission CT scan characteristics in patients with and without ARDS* Admission CT Scan Worst CT Scan No. W/O No. W/ No. W/O No. W/ Characteristic ARDS (%) ARDS (%) p Value ARDS (%) ARDS (%) p Value type of injury closed head injuries 158 (92.4) 17 (94.4) (92.4) 17 (94.4) diffuse injury 1 & 2 65 (41.1) 6 (35.3) (36.1) 4 (23.5) diffuse injury 3 & 4 36 (22.8) 5 (29.4) 36 (22.8) 2 (11.8) evacuated mass 53 (33.5) 6 (35.3) 61 (38.6) 11 (64.7) unevacuated mass 4 (2.5) 0 (0.0) 4 (2.5) 0 (0.0) gunshot wound 13 (7.6) 1 (5.6) 13 (7.6) 1 (5.6) SAH 87 (50.9) 10 (55.6) (43.3) 8 (44.4) IVH 23 (13.4) 1 (5.6) (15.2) 3 (16.7) basal cisterns absent 30 (17.5) 6 (33.3) 27 (16.1) 6 (33.3) compressed 40 (23.4) 5 (27.8) 88 (52.4) 6 (33.3) normal 101 (59.1) 7 (38.9) 53 (31.5) 6 (33.3) ventricular system compressed 67 (39.2) 9 (50.0) (49.4) 12 (66.7) midline shift (mm) (59.1) 6 (33.3) 84 (50.0) 6 (33.3) 5 32 (18.7) 6 (33.3) 37 (22.0) 4 (22.2) (14.0) 3 (16.7) 34 (20.2) 5 (27.8) (8.2) 3 (16.7) 13 (7.7) 3 (16.7) * IVH = intraventricular hemorrhage; SAH = subarachnoid hemorrhage. PT (p = 0.026). Prothrombin time and partial thromboplastin time were not measured, however, in a large number of patients. Treatment Variables for Patients With ARDS The incidence of development of ARDS was nearly five times higher in the CBF-targeted group, 15% compared with 3.3% (p = 0.007). The resulting odds ratio was To study the contribution of the differences in management to the development of ARDS in the two treatment groups, all of the physiological and management data collected during the trial were analyzed. The patients in the CBF-targeted group received a higher fluid intake and larger doses of pressor agents to maintain the higher targeted BP and CPP. Hyperventilation was not used to treat intracranial hypertension in the CBF-targeted group. There were no other differences in the management protocols between the two treatment groups, including the treatment required for management of intracranial hypertension and the use of barbiturate coma. 21 Table 3 summarizes the management-related variables that were found to be different in patients with and without ARDS. For the fluid intake and output data, the volumes recorded hourly were summed for each 24-hour period. The average of the daily values was then calculated for each patient, and the means of these data were compared in the two groups. Drug administration data includes the number of patients in each group who were ever given each specific treatment during the course of the study, the average daily dose of each drug, and the average rate for the drugs given as a continuous infusion. Patients who suffered from ARDS more commonly received mannitol, barbiturate drugs for coma induction, dopamine, epinephrine, and phenylephrine; they also received these drugs in higher daily doses than did patients without ARDS. Patients who experienced ARDS also received higher daily doses of morphine and vecuronium. In addition, patients in whom ARDS developed received higher fluid intakes and had a more positive daily fluid balance. Finally, Fig. 2 shows the occurrence of ARDS throughout the study in relation to the rotation of neurosurgery resident groups through the intensive care unit. Patients treated by the third group of residents had the overall highest incidence of ARDS. Table 4 summarizes the physiological variables that were different in the patients with and without ARDS. The values given were obtained by averaging the hourly values recorded during the course of the study for each patient, and comparing the mean of the averaged values for patients in the two outcome groups. The average arterial O 2 saturation value was significantly lower in the patients who experienced ARDS. The average CVPs and PWPs were higher in the patients who suffered from ARDS, and these patients also had more severe intracranial hypertension, with a higher average ICP and lower average CPP. In addition, a higher serum sodium concentration and osmolarity probably reflected the higher mannitol requirements for treating elevated ICP in the group with ARDS. Outcome Measures The clinical trial had three outcome measures: incidence of jugular venous desaturation, incidence of refractory intracranial hypertension, and GOS score at 3 months. These measures were compared in the patients with and without ARDS. Incidence of Secondary Insults The primary goal of the CBF-targeted management protocol was to reduce the incidence of secondary insults to the injured brain, and the results indicated that this ap- 563

5 C. F. Contant, et al. TABLE 3 Comparison of treatment variables that differed in patients with and without ARDS* Variable ARDS Absent ARDS Present p Value fluid intake & balance (ml/day) fluid intake 2874 ( ) 4059 ( ) fluid output 2718 ( ) 3611 ( ) fluid balance 278 ( 6 to 587) 478 (88 922) pressor agents dopamine dose, mg/kg/day 35.1 ( ) ( ) infusion rate, mg/kg/min 7.53 ( ) ( ) no. (%) receiving 101 (59.1) 18 (100.0) phenylephrine dose, mg/day 0 (0 0) 0 (0 6.4) infusion rate, mg/min 0 (0 0) 0 (0 1.53) no. (%) receiving 28 (16.4) 8 (44.4) epinephrine dose, mg/day 0 (0 0) 0 (0 4.6) infusion rate, mg/min 0 (0 0) 0 (0 0.43) no. (%) receiving 7 (4.1) 5 (27.8) intracranial hypertension agents morphine dose, mg/day 51.3 ( ) 81.5 ( ) no. (%) receiving 168 (98.3) 18 (100.0) vecuronium dose, g/day 41.2 ( ) 67.5 ( ) no. (%) receiving 163 (95.3) 18 (100.0) mannitol dose, g/day 12.5 (0 44.4) 33.0 ( ) no. (%) receiving 97 (56.7) 18 (100.0) barbiturate coma dose, mg/day 0 (0 0) ( ) no. (%) receiving 32 (18.7) 9 (50.0) * Unless specified in Column 1 as number of patients and percentages, the values are medians, with IRs in parentheses. proach was highly successful in preventing hypotensionand hypocapnia-related insults. In the full trial, the incidence of jugular venous desaturation was reduced from 50.6% in the ICP-targeted group to 30% in the CBF-targeted group (p = 0.006). The development of ARDS did not significantly alter the incidence of jugular venous desaturation. Incidence of Refractory Intracranial Hypertension The CBF-targeted management was expected to reduce the incidence of refractory intracranial hypertension if it successfully decreased the incidence of secondary ischemia. In the full trial, the mean values for ICP, the duration of time in which ICP was greater than 25 mm Hg, and the incidence of refractory intracranial hypertension were not significantly different in the two treatment groups. In addition, treatment of intracranial hypertension followed a standardized protocol during the trial. There were no differences between the two treatment groups in either the duration of treatment period with or the dose of any agent used to treat intracranial hypertension. Only the use of hyperventilation, which was not permitted in the CBF-targeted management strategy, was significantly different between the two groups. These parameters, however, were significantly different in the patients with ARDS, compared with those without ARDS. As described earlier, the average ICP was higher, and the amount of treatment required to manage intracranial hypertension was greater in the patients who experienced ARDS. In addition, the incidence of refractory intracranial hypertension was 55.6% in the patients with ARDS, compared with 22.2% in the patients without ARDS (p = 0.004). Neurological Outcome The CBF-targeted management was expected to improve long-term neurological recovery, if it successfully decreased the incidence of secondary ischemia. In the full trial, neurological recovery was assessed at 3 months and 6 months postinjury by using GOS and DRS scores. The proportion of patients with a favorable recovery (good recovery or moderate disability) at 3 months was 37% in the ICP-targeted group and 31.9% in the CBF-targeted group (p = 0.554). The median DRS was 6 (IR 4 23) in the ICPtargeted group and 12 (IR 5 30) in the CBF-targeted group. At 6 months postinjury, 49.3% of the patients in the ICP-targeted group and 39.8% of those in the CBF-targeted group had a favorable outcome. None of these differences was statistically significant. The 3-month and 6-month GOS scores are contrasted for the patients with and without ARDS in Table 5. The percentage of patients who were in a vegetative state or dead at 6 months postinjury was twice as high among the patients with ARDS: 68.8% compared with 29.7% (p = 0.010), and there were few favorable recoveries among the patients with ARDS. 564

6 Adult respiratory distress syndrome after head injury TABLE 4 Physiological parameters that differed in patients with and without ARDS* Parameter ARDS Absent ARDS Present p Value ICP (mm Hg) 16.3 ( ) 21.5 ( ) CPP (mm Hg) 76.1 ( ) 69.9 ( ) SaO 2 (%) 99.2 ( ) 98.3 ( ) CVP (mm Hg) 6.8 ( ) 8.9 ( ) PWP (mm Hg) 11.2 ( ) 12.7 ( ) Na concentration ( ) ( ) (meq/l) osmolarity ( ) 302 ( ) (mosm/l) * Values are given as medians, with IRs in parentheses. Abbreviation: SaO 2 = arterial saturation of oxygen. FIG. 2. Bar graph showing that ARDS occurred in patients admitted throughout the 32-month clinical trial, but primarily in the CBF-targeted treatment groups. Each pair of bars represents a different resident physician group. Results of Logistic Regression These initial analyses indicated that the development of ARDS could not be clearly predicted based on a more severe initial neurological injury. The only measures that even suggested a more severe injury were the more common presence of preadmission hypotension and the more common finding of midline shift on the initial CT scan, but these differences were not significant. The patients who experienced ARDS had more severe intracranial hypertension, by measures of ICP, by measures of the intensity of treatment directed at lowering ICP, and by the proportion of patients in whom refractory intracranial hypertension developed. This finding, however, could represent either a more severe neurological injury, or it could have been caused by the development of ARDS and its subsequent treatment. Otherwise, the primary differences between the two groups were related to treatments intended to maintain a higher BP and CPP. For these analyses, the independent variables were grouped into those representing baseline demographic and injury severity characteristics of the patients, and those representing treatment. To screen for these variables, a series of logistic regression models was created. The results of the logistic model screening for the baseline variables and the treatment variables are shown in Table 6, including all of the variables screened, the probability value for inclusion in the initial model (with protocol [1 = ICP targeted, 0 = CBF targeted] as the only other variable), and the probability value for all variables in the final models. The final model for the baseline demographic and injury severity variables included history of drug abuse and presence of midline shift on admission CT scan. The final model for the treatment variables included ever receiving epinephrine, cumulative dose of dopamine, serum sodium concentration, and treatment by the third group of residents. The variables in the final logistic model for baseline variables and for treatment variables were used to fit a model using exact logistic regression methods. Table 7 presents the odds ratios, 95% confidence limits, and probability values of the exact logistic models for the variables in the final models, which were derived from the screening process for the baseline and treatment models. The variables found to be significant in each of the two exact logistic regression models were then combined into the final model. Table 7 also contains the final exact logistic result obtained when combining the variables from the baseline and treatment exact analyses. The exact model for the baseline variables indicates significant effects for the presence of midline shift, a history of drug abuse, and the treatment protocol. For the treatment variables, the final model included whether the patient had ever received epinephrine, the total dose of dopamine over the first 10 days of the study, and the mean sodium concentration. Whether the patient was treated by the third group of residents and the treatment protocol were included with marginal effects. The final model, which was obtained by combining the baseline and treatment models, included whether the patient had ever received epinephrine, dopamine dose, history of drug abuse, and a marginally significant effect of treatment protocol. The models including only the male patients resulted in selection of the same set of variables for inclusion in the exact logistic regression modeling; the results of fitting these models is not shown. TABLE 5 Comparison of 3-month and 6-month GOS and DRS scores in patients with and without ARDS No. of Patients (%)* Outcome W/O ARDS W/ ARDS p Value GOS score at 3 mos good recovery/moderate disability 55 (35.5) 4 (23.5) severe disability 54 (34.8) 1 (5.9) vegetative/dead 46 (29.7) 12 (70.6) GOS score at 6 mos good recovery/moderate disability 65 (47.1) 3 (18.8) severe disability 32 (23.2) 2 (12.5) vegetative/dead 41 (29.7) 11 (68.8) DRS score at 3 mos 8 (4 23) 30 (6 30) at 6 mos 6 (3 30) 30 ( ) * The last two values are given as the median, with the IR in parentheses. 565

7 C. F. Contant, et al. TABLE 6 Results of the logistic regression model screening of baseline and treatment variables* p Value Initial Model Variables Included Variable (protocol + variable) in Final Model logistic regression model screening of baseline variables prehospital hypotension worst CT category or both pupils reactive in ER GCS motor component in ER history of drug abuse caucasian age presence of midline shift in ER presence of gunshot wound AIS score 3 in lower extremity logistic regression model screening of treatment variables ever received barbiturate drugs ever received epinephrine cumulative dopamine daily dose cumulative mannitol daily dose cumulative fluid balance cumulative morphine dose cumulative fluid intake cumulative fluid output mean daily Na concentration ever had PT measurement ever had PTT measurement ever had platelet measurement treated in 1st or 2nd resident rotation treated by 3rd group of residents * PTT = partial thromboplastin time. Discussion Pulmonary complications, including pneumonia, atelectasis, and ARDS, are important complications of trauma, and contribute significantly to the morbidity and mortality rates. Head injury is a major risk factor for the development of pulmonary complications after trauma. In one large series of 3406 cases of severe trauma, the incidence of ARDS was 12%, and ARDS was responsible for 16% of the deaths. 20 In another large series of 3280 patients with multiple trauma, pulmonary complications occurred in 368 (11.2%) of the cases, accounting for one third of all disease-related complications. 9 Head injury (AIS score 3) was a risk factor for the pulmonary complications of pneumonia, atelectasis, and respiratory failure. Surgery for head injury was a risk factor for ARDS (odds ratio 4.5). Other risk factors for the development of ARDS after multiple trauma are pelvic fracture, 18 blunt trauma, a low trauma score, 9 and treatment of femur fracture. 17 Noncardiogenic pulmonary edema, or ARDS, is a common finding in patients who die of head injury. In an autopsy study of 21 patients who died within a few minutes of a penetrating head injury, 17 (81%) had associated pulmonary edema. 13,22 In those who survive long enough to be admitted to the hospital, approximately 15 to 20% of severely head injured patients develop ARDS. 6,7,9,11 The overall 10% incidence of ARDS in the current study is slightly lower than in these published studies, but this may TABLE 7 Results of the exact logistic regression models for baseline and treatment variables 95% Confi- Variable Odds Ratio dence Limits p Value exact logistic regression model for baseline variables treatment protocol CBF-targeted ICP-targeted , presence of midline shift yes no , history of drug abuse no or possibly yes , exact logistic regression model for treatment variables treatment protocol CBF-targeted ICP-targeted , ever received epinephrine no yes , dopamine dose below median above median , mean daily Na level below median above median , neurosurgery resident group not 3rd resident group rd resident group , exact logistic regression model for combined baseline and treatment variables treatment protocol CBF-targeted ICP-targeted , ever received epinephrine no yes , dopamine dose below median above median , history of drug abuse no or possibly yes , * The comparison value for each variable has an odds ratio of The odds ratios for the other levels are protective for ARDS if the values are less than , and increase risk of ARDS if they are greater than be partially due to the differences in diagnostic criteria for ARDS among the different studies. The important difference in this study is the close association of the incidence of ARDS to the treatment differences in a randomized management trial. Although the trial was not specifically designed to examine the effect of the treatment on the incidence of ARDS, the evidence seems convincing that the management strategy of maintaining an elevated BP causes an increased incidence of symptomatic ARDS. Some limitations of the methods should be acknowledged. First, logistic regression modeling when there are few outcome events and a large number of possible explanatory variables is very difficult, and the usual variable selection methods are inappropriate. Using the guideline of 10 outcomes for each explanatory variable in the model would allow screening of, at most, two variables at a time 566

8 Adult respiratory distress syndrome after head injury by using the standard methods. Exact logistic regression offers a statistically valid alternative; however, it would have been computationally impossible to attempt to screen variables by using the exact model methods. The procedure adopted for this study was a practical solution to this limitation of logistic regression modeling. Second, many of the variables were closely related. Table 3 lists several variables that were significantly associated with ARDS when used individually in the regression model, but which were not significant when other variables were added to the model. These changes are largely due to the correlation among variables that are in the model, and the variables that were not selected for inclusion. The exact logistic regression model for the baseline variables (Table 7) provided probability values that were somewhat less impressive than the screening analysis results. In this model, the presence of midline shift and a history of drug abuse were both risk factors for the development of ARDS. The apparent protective effect of protocol membership remains in this model. The exact model for the treatment effects would indicate that use of epinephrine, the dose of dopamine, and, to a lesser extent, the resident group treating the patients and the mean sodium concentration were related to the development of ARDS. In this model, the ICP-targeted management protocol is still protective, although the probability value is greater than Again, large differences in the probability value were seen in the screening models and in the exact model. With these limitations of the analyses in mind, the following conclusions are justified. 1) The major risk factors for developing ARDS in the clinical trial were treatment related, and not due to injury severity. In fact, relatively few demographic and injury severity characteristics were even related to the development of ARDS. All of the patients in whom ARDS developed were male, but sex is not generally thought to be a risk factor for the development of ARDS, and sex was not an important factor in any of the logistic regression models. It is likely that this finding was an artifact caused by the small percentage of women in the patients studied. A history of drug abuse was significantly related to the development of ARDS, and continued to be a significant risk factor, even in the final logistic regression model. Drug abuse, especially of cocaine and heroin, has been associated with the development of ARDS 3,15,24 and would likely be an important complicating factor in patients with traumatic injuries. Finally, the presence of midline shift was associated with a significantly increased risk of developing ARDS in both the logistic regression screening and the exact logistic regression models for the baseline characteristics. Although this variable fell out of the final model when the baseline and treatment variables were combined, data from other studies indicate that this may be an important risk factor. In the analysis of the TCDB studies, ARDS was found to be associated with certain types of injury. 4 The patients in our study who had diffuse injury and midline shift greater than 5 mm had a 10-fold increased risk, and the patients with nonevacuated mass lesions had a fivefold increased risk of developing ARDS. All of these CT scan findings are likely to be identifying patients who develop severe intracranial hypertension. Intracranial hypertension has been one of the experimental models used to produce neurogenic pulmonary edema and increased pulmonary capillary permeability. 8,14 Therefore, the patients with severe traumatic brain injury who are at greatest risk of developing ARDS include those with a history of drug abuse and those with midline shift on their admission CT scan. Nevertheless, because the incidence of these findings was not significantly different between the two treatment groups, these risk factors probably do not explain the increased risk of ARDS in the CBF-targeted group. 2) The treatment-related variables that were associated with an increased risk of ARDS reflected the goals of the CBF-targeted management strategy (greater fluid intake and especially more frequent use of pressor agents, higher CVP and PWP). The analysis of individual treatment variables in Table 3 and the physiological variables in Table 4 revealed significant associations between the protocol-designed differences in the two treatment groups and the occurrence of ARDS. That is, the MABP and CPP were planned to be higher in the CBF-targeted group than in the ICP-targeted group. The management differences required to obtain these targeted pressure goals are the same treatment variables that were significantly associated with the development of ARDS. The final exact logistic regression model includes two of these treatment effects: ever receiving epinephrine, and dopamine dose, and one baseline effect: history of drug abuse. Membership in the ICP protocol is still protective, but the probability value is greater than Examination of the confidence limits around the dopamine effect is a reminder of the lack of stability of all of these estimates. It cannot be determined with a high degree of certainty that the cause of the ARDS in the CBF-targeted group is the use of dopamine and epinephrine, because all of the treatments are highly correlated with one another and with the treatment group. Instead, the inclusion of dopamine and epinephrine use in the final exact logistic regression model probably is best interpreted as an indication that the management strategy of maintaining MABP and CPP at artificially high values, which included use of pressor agents to maintain BP, causes an increased risk of ARDS. 3) The development of ARDS after severe head injury was associated with more severe intracranial hypertension. This complication of the CBF-targeted treatment may have offset the beneficial effects of reducing secondary ischemic insults on ICP. The patients in whom ARDS developed had higher ICP and lower CPP, and required more mannitol, more sedation and neuromuscular blocking agents, and more barbiturates to control ICP. It is impossible to attach a cause and effect relationship to these associations, and two explanations are possible. On one hand, the patients with more severe intracranial hypertension may have been at greatest risk for developing ARDS. As discussed earlier, the presence of midline shift, which indicates a greater tendency toward intracranial hypertension, was a risk factor for the development of ARDS in the present and in other studies. In addition, it is known from laboratory studies that intracranial hypertension can produce pulmonary edema. On the other hand, once ARDS develops in a patient with severe head injury, hypoxia may worsen the neurological injury, 5,16 and the ventilator and hemodynamic management 567

9 C. F. Contant, et al. required to obtain adequate oxygenation in a patient with ARDS may make it more difficult to control intracranial hypertension. 1 Because this study was a randomized trial, however, and because there were no significant differences in injury severity measures between the two treatment groups, 21 the second explanation is more likely. 4) The development of ARDS after severe head injury was associated with a significantly worse neurological outcome. This complication of the CBF-targeted treatment may have offset the beneficial effects on neurological outcome of reductions in secondary ischemic insults. Both GOS scores and DRS scores were significantly worse at 3 and 6 months postinjury in the patients in whom ARDS developed. At 6 months postinjury, the number of patients who were dead or in a vegetative state was more than doubled in the ARDS group, from 30 to 69%. It was thought that ARDS directly or indirectly resulted in the death of at least five patients in the study. This is similar to the findings in the TCDB series, in which the development of acute lung injury was associated with a 2.8-fold increased risk of death or vegetative status at 6 months postinjury. 4 As with intracranial hypertension, it is impossible to determine a cause and effect relationship in the association between neurological outcome and the occurrence of ARDS. Because the assignment to a treatment group was randomized and because there was no significant difference in the severity of injury between the two treatment groups, it is most likely that the morbidity and mortality associated with the development of ARDS resulted in a worse outcome in the CBF-targeted groups and at least partially offset any beneficial effects of the CBF-targeted protocol. References 1. Baigelman W, O Brien JC: Pulmonary effects of head trauma. Neurosurgery 9: , Bernard GR, Artigas A, Brigham KL, et al: Report of the American-European consensus conference on acute respiratory distress syndrome: definitions, mechanisms, relevant outcomes, and clinical trial coordination. Consensus Committee. J Crit Care 9:72 81, Bird DJ, Markey JR: Massive pulmonary edema in a habitual crack cocaine smoker not chemically positive for cocaine at the time of surgery. Anesth Analg 84: , Bratton SL, Davis RL: Acute lung injury in isolated traumatic brain injury. Neurosurgery 40: , Chesnut RM, Marshall LF, Klauber MR, et al: The role of secondary brain injury in determining outcome from severe head injury. J Trauma 34: , Demling R, Riessen K: Pulmonary dysfunction after cerebral injury. Crit Care Med 18: , Fulton RL, Jones CE: The cause of post traumatic pulmonary insufficiency in man. Surg Gynecol Obstet 140: , Garcia-Uria J, Hoff JT, Miranda S, et al: Experimental neurogenic pulmonary edema. Part 2: The role of cardiopulmonary pressure change. J Neurosurg 54: , Hoyt DB, Simons RK, Winchell RJ, et al: A risk analysis of pulmonary complications following major trauma. J Trauma 35: , Jennett B, Bond M: Assessment of outcome after severe brain damage. A practical scale. Lancet 1: , Katsurada K, Yamada K, Sugimoto T: Respiratory insufficiency in patients with severe head injury. Surgery 73: , Mackersie RC, Christensen JM, Pitts LH, et al: Pulmonary extravascular fluid accumulation following intracranial injury. J Trauma 23: , Martin AM Jr, Simmons RL, Heisterkamp CA III: Respiratory insufficiency in combat casualties. I. Pathologic changes in the lungs of patients dying of wounds. Ann Surg 170:30 38, McClellan MD, Dauber IM, Weil JV: Elevated intracranial pressure increased pulmonary vascular permeability to protein. J Appl Physiol 67: , Meisels IS, Loke J: The pulmonary effects of free base cocaine: a review. Cleve Clin J Med 4: , Miller JD, Sweet RC, Narayan R, et al: Early insults to the injured brain. JAMA 240: , Pape HC, Regel G, Dwenger A, et al: Influences of different methods of intramedullary femoral nailing on lung function in patients with multiple trauma. J Trauma 35: , Poole GV, Ward EF, Griswold JA, et al: Complications of pelvic fractures from blunt trauma. Am Surg 58: , Rappaport M, Hall KM, Hopkins K, et al: Disability Rating Scale for severe head trauma: coma to community. Arch Phys Med Rehabil 63: , Regel G, Lobenhoffer P, Grotz M, et al: Treatment results of patients with multiple trauma: an analysis of 3406 cases treated between 1972 and 1991 at a German Level I Trauma Center. J Trauma 38:70 78, Robertson CS, Valadka AB, Hannay HJ, et al: Prevention of secondary insults after severe head injury. Crit Care Med 27: , Simmons RL, Martin AM Jr, Heisterkamp CA III, et al: Respiratory insufficiency in combat casualties. II. Pulmonary edema following head injury. Ann Surg 170:39 44, Smith WS, Matthay MA: Evidence for a hydrostatic mechanism in human neurogenic pulmonary edema. Chest 5: , Sporer KA: Acute heroin overdose. Ann Intern Med 130: , 1999 Manuscript received June 26, Accepted in final form May 8, This work was supported by National Institutes of Health Grant No. PO1-NS Address reprint requests to: Claudia Robertson, M.D., Department of Neurosurgery, Baylor College of Medicine, 6560 Fannin, Suite 944, Houston, Texas claudiar@bcm.tmc.edu. 568