INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Acute respiratory distress syndrome (ARDS), initially described by Ashbaugh and colleagues (1), is a major contributor to mortality and morbidity of adult patients admitted to intensive care units (ICU). ARDS is characterized by arterial hypoxemia, bilateral infiltrates on chest radiographs, and a normal pulmonary artery occlusive pressure. It is the most severe form of a wide spectrum of pathologic processes designated as acute lung injury (ALI) (2).

Although the average mortality is 50%, reported mortality rates in individual series vary from 30 to 70% even with optimal conventional therapies (3). In a recent review of 101 clinical investigation papers reporting both the mortality and the oxygenation status, Krafft and colleagues (3) found that 80% of studies did not report any classification system for the severity of patients, and only three reported all necessary information pertaining to ventilatory support. Thus, variations in mortality among studies may be attributable to differences in the severity of illness of the patients. Reports claiming improvement (4) in survival are potentially limited by variations in patient selection (5). Therefore, it would be useful to have a system assessing the probability of mortality based on objective data. Most of the reported predictive factors for survival in patients with ARDS were analyzed using univariate tests (6). These analyses may confound interactions among various factors, and thus independent predictive factors for survival in ARDS should be analyzed using multivariate methods with models including a widely used general severity score such as SAPS or APACHE.

In an attempt to select early discriminant prognostic factors of survival in mechanically ventilated patients with ARDS, we evaluated mortality predictive values of clinical, respiratory, and hemodynamic parameters obtained at the onset of ARDS, using a multivariate analysis including several general severity scores.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Data Collection

We screened all mechanically ventilated patients admitted to one intensive care unit between January 1992 and December 1995. Chest radiographs and clinical data of these patients were reviewed for ARDS criteria by two independent physicians. Mechanically ventilated patients who met ARDS criteria during their ICU stay were selected for the study. Patients who survived more than 24 h after the onset of ARDS were included into a database.

According to the American-European consensus conference (2), a diagnosis of ARDS was assigned by the following criteria: acute onset; arterial hypoxemia with Pa_O2/FI_O2 lower than 200 (regardless of PEEP level); bilateral infiltrates seen on chest radiograph; pulmonary artery occlusive pressure lower than 18 mm Hg or no clinical evidence of left atrial hypertension; and compatible risk factors.

Data compiled for each patient included demographic information (age, sex) and risk factors for ARDS as defined by the American- European consensus conference. The triggering risk factor for ARDS was defined as the specific clinical disorder associated with ARDS during the first 24 h and classified as either direct or indirect lung injury. If a patient was admitted to our ICU after a previous hospital stay, the initial setting was categorized as medical, surgical, or ICU. Documented preexisting chronic diseases were recorded and severity of the underlying medical conditions were stratified according to the criteria of McCabe and Jackson (12) as nonfatal (score 1), ultimately fatal (score 2), or fatal (score 3). Immunodeficiency was categorized and noted as follows: hematologic malignancy, cancer, immunosuppressive therapy, and HIV infection with CD4⁺ cell count below 200 per cubic mm. Right ventricular dysfunction was defined by a right atrial pressure (PRA) higher than pulmonary artery occlusive pressure (PRA > Ppao) on right heart catheterization (13). Right heart catheter placement was decided by the patient's attending physician.

Severity scores calculated for the first day of ARDS were Simplified Acute Physiology Score (SAPS) (14), SAPS-II (15), Organ System Failure (OSF) (16), and Lung Injury Score (LIS) as defined by Murray and colleagues (17).

On the first day of ARDS, we recorded hemodynamic data, ventilatory parameters, and blood gases at the lowest Pa_O2/FI_O2 ratio level, in a steady state without evidence of untreated pneumothorax. The oxygenation index was defined as mean airway pressure × FI_O2 × 100/ Pa_O2. Mechanical ventilation duration before the onset of ARDS in days was also noted.

Statistical Analysis

For all variables, in the cases of discordance between the two observers, parameters were reviewed and a consensus value was noted for further statistical analysis.

To develop and validate a predictive model of survival, two-thirds of the patients in the database were randomly selected to constitute the developmental sample and the remaining third composed the validation sample.

Using the developmental sample, all recorded variables were screened for association with mortality using chi-square tests for categoric data and Mann-Whitney U tests for numerical data. A p value lower than 0.05 was accepted as indicating statistical significance. All statistical tests were two-tailed.

A stepwise multiple logistic-regression model was used to evaluate the independent role of each variable by including all significant or nearly significant parameters (p < 0.10) of univariate analysis. The multiple logistic regression model produced an equation in which each significant variable had an associated coefficient  that was a measure of the relative weight of the variable while controlling for all other variables.

To obtain a point value for each significant variable, the corresponding  coefficient was multiplied by a factor of 10 and rounded to a whole number. Thus, for each patient, an ARDS severity score was calculated by summing the points associated with each variable. The ARDS severity score was then used as the single variable of a logistic regression analysis, producing an equation of the form: logit = ₀ + ₁ (ARDS severity score) (15). The logit was then converted to a probability of hospital mortality as Pr = e^logit/(1 + e^logit), where Pr indicated the probability and e the mathematical constant 2.71882.

The assessment of model performance was the final stage of the analysis. To evaluate model calibration, Hosmer-Lemeshow goodness-of-fit tests, comparing observed with expected mortality, were performed on the developmental and the validation sample, and a p value above 0.1 was assumed to indicate a good agreement (18). To evaluate the model's discrimination, the area under the receiver operating characteristic (ROC) curve was determined, both for the developmental and the validation sample, and a value above 0.80 was accepted to indicate a good discrimination (19). Analyses were done on a personal computer using STATA software (Stata Corp., College Station, TX).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Population Characteristics

Between January 1992 and December 1995, 3,511 patients were admitted to our ICU. Of these patients, 259 met ARDS criteria and had an overall mortality rate of 65%. A discordance between observers for at least one variable was noted for 26 patients and then corrected.

One hundred seventy-seven patients were included into the developmental sample and 82 into the validation sample.

The characteristics of the study population are summarized in Table 1. As shown in Table 1, mortality reached 81% in immunocompromized patients and 96% in patients with cirrhosis. Respiratory parameters are summarized in Table 2.

Univariate Analysis

In univariate analysis, the following parameters were found to be significantly associated with mortality: age, COPD, chronic renal failure, cirrhosis, location prior to ICU admission, McCabe Score, immunodeficiency, the mechanism of lung injury (direct or indirect), SAPS, SAPS-II, OSF Score, epinephrine or norepinephrine prescription, mean arterial pressure, right ventricular dysfunction (PRA > Ppao), arterial bicarbonate, and arterial pH. The oxygenation index reached a nearly significant probability (p = 0.09). Other respiratory and hemodynamic variables were not statistically significant.

Multivariate Analysis

The results of multivariate logistic regression are shown in Table 3. The SAPS-II was the most relevant severity score used in univariate analysis, and it remained significant in multivariate analysis. The McCabe Score was highly discriminant to determine the outcome with an odds ratio (OR) of four per score rising. Among underlying diseases, cirrhosis was independently associated to mortality with a very high OR of 27. Other recorded underlying diseases became nonsignificant in the multivariate analysis. The length of mechanical ventilation prior to ARDS was associated with an elevated risk of death (OR = 1.1 per day).

A direct lung injury etiology remained significantly associated with a higher mortality in the multivariate model. The elevation of the oxygenation index was also associated with mortality (OR = 1.05 per point rising). Among the hemodynamic parameters, only the right ventricular dysfunction retained a significant relationship. Patients who had a PRA higher than the Ppao had an OR of 5.1, as compared with those who had a PRA lower or equal to Ppao.

The ARDS severity score points, defined using the  coefficient of the multivariate model, are shown in Table 4. The ARDS score was then calculated for each patient and used as the only term in a new logistic regression analysis. This analysis provided the following equation for the logit: logit = 7.6697 + 0.09385 (ARDS Score).The logit was then converted to a probability of hospital mortality for each patient: Pr = e^logit/(1 + e^logit). A graphic representation of the hospital probability of death using the model calculations for the logit and probability is shown in Figure 1. For this model, the area under the ROC curve was 0.95 for the developmental sample and 0.92 for the validation sample. The p value of Hosmer-Lemeshow goodness-of-fit test was 0.84 for the developmental sample and 0.72 for the validation sample, with 10% intervals of mortality.

View larger version (11K): [in this window] [in a new window]   Figure 1.   Curve showing the relationship between the defined score and the probability of hospital mortality.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

This study has demonstrated that on the first day of mechanically ventilated ARDS: (1) among several severity scores (LIS, OSF, SAPS, and SAPS-II), SAPS-II was the best; (2) the McCabe Score adds to the predictive ability of SAPS-II for the prediction of outcome; (3) the triggering risk factor affects the probability of survival; (4) patients who had a long period of mechanical ventilation prior to the development of ARDS had a worse prognosis; (5) right ventricular dysfunction is associated with a higher risk of death; (6) the oxygenation index, unlike other respiratory parameters, has a prognostic value.

This study is the first descriptive study on ARDS combining a data collection by two independent observers, a report of all respiratory parameters, the determination of general severity scores for all patients, and statistical analysis using multivariate methods. We chose to collect data by two independent observers to optimize the quality of recorded data. The importance of interobserver variation in clinical medicine is widely underestimated (20) and, according to our knowledge, this method of data collection has never been used in descriptive studies of ARDS. We believe that this method is essential for the reliability of collected data.

We chose the American-European consensus conference definition of ARDS to select our study population. It is more sensitive and more specific than other definitions (e.g., LIS) (21). In fact, the use of ventilator parameters in the criteria defining ARDS can produce both false positive and false negative results (21).

SAPS-II is a third generation multipurpose ICU scoring system (22). Its performance as predicting hospital mortality was superior to that of SAPS and OSF, which were nonsignificant in multivariate models. We chose SAPS-II in this study because SAPS-II is extensively used in European ICUs.

The McCabe Score had a high prognosis value in our study and other variables of underlying states or immunodeficiency, with the exception of liver cirrhosis, became nonsignificant in the multivariate model. Using univariate analysis, Cox and colleagues (23) reported a significant association between the severity of underlying disease and mortality of patients with acute respiratory failure; however, in their study, respiratory and hemodynamic parameters were not analyzed. We found that patients who had a severe underlying disease with a short life expectancy had a very bad prognosis, which might have been related to their underlying disease instead of their ventilatory parameters.

We found liver cirrhosis to be independently associated with a very poor outcome in patients with ARDS (OR = 27). Although poorly understood, there is a growing recognition that organ interactions must be taken account to understand the determinants of ARDS. Among extrapulmonary organs, the liver plays a central role in regulating cytokine kinetics relevant to acute lung injury (ALI). After endotoxemia, lung microvascular permeability, neutrophilic alveolitis, and mortality were augmented in animals with acute liver injury, portocaval shunt, or cirrhosis (24, 25). Matuschak and colleagues (26) observed a high incidence of sepsis-induced acute lung injury in patients with end-stage liver failure. Liver cirrhosis is associated with a very high mortality in patients with septic shock (27) and Doyle and colleagues (11) stressed the poor outcome of patients with liver cirrhosis developing ALI. Our study suggests that liver cirrhosis, after adjustment for the McCabe Score, oxygenation index, and the severity of the acute disease remains highly predictive of hospital mortality.

In our analysis, mortality was related to the triggering risk factor of ARDS. Patients with a direct lung-injury-induced ARDS had a lower survival rate than did others. This finding agrees with the report of Seidenfeld and colleagues (28) who analyzed the survival in a population of patients with ARDS and observed a survival rate of 13% in lung infections and 59% in abdominal infections. In their study of patients with acute respiratory failure, Knaus and colleagues (29) also found that surgical patients, i.e., patients with an indirect lung injury, had a higher survival rate than did patients with a medical diagnosis. Thus, there is some agreement in the published data stressing the disparity of hospital mortality according to triggering risk factor. However, we must emphasize that our study population had a few number of trauma patients (nine). Therefore, the prognostic value of a traumatic direct lung injury (i.e., lung contusion) remains unclear.

We also found that the length of mechanical ventilation before the onset of ARDS is an independent risk factor for mortality. Patients who developed ARDS after several days of mechanical ventilation had an OR of 1.1 per day of prior ventilation. It has previously been reported that the length of respiratory failure and the improvement of oxygenation over a 7-d course had a prognosis value (8). In the study of Knaus and colleagues (29), previous hospital length of stay before ALI criteria was also associated with mortality (OR = 1.6 per 10 d). The association of prior hospital stay or prior mechanical ventilation duration with mortality might reflect the severity of initial illness as well as the occurrence of a nosocomial infection such as nosocomial pneumonia. In a cohort study of ventilated ICU patients, Fagon and colleagues (30) reported that nosocomial pneumonia is an independent predictor of dying. The development of ARDS after several days of mechanical ventilation might be associated with the occurrence of nosocomial pneumonia. Patients who meet ARDS criteria after several days of mechanical ventilation may constitute a high-risk group and should be distinguished from patients who require mechanical ventilation for ARDS.

After adjustment for the presence of a pulmonary artery catheter, the relationship between PRA and Ppao was the only hemodynamic information strongly related to mortality in the multivariate analysis. In patients monitored by a pulmonary artery catheter, a PRA higher than Ppao increases the probability of dying (OR = 5.1). The relationship between PRA and Ppao is determined by right cardiac function. Right ventricular (RV) dilation is commonly present in ARDS (31). Right ventricular dysfunction may also be present in the most severe cases. In a study of severe acute respiratory failure in trauma patients, Laghi and colleagues (32) analyzed the right (RVSWI) and left (LVSWI) ventricular stroke work index and found a significantly higher RVSWI/LVSWI ratio in nonsurvivors. In 15 patients with ARDS, Steltzer and colleagues (33) investigated the right ventricular performance using a pulmonary artery catheter with rapid response thermistor and described a significantly depressed right ventricular ejection fraction in nonsurvivors. However, these studies performed only a univariate analysis. We found that at the onset of ARDS, right ventricular performances, assessed by the PRA/Ppao ratio, is an independent predictive factor of survival. Interestingly, as can be seen from our multivariate regression model, we were unable to demonstrate that the presence of the pulmonary artery catheter independently altered outcome in the patient data set. However, the sample size is likely too small for further conclusions.

Several investigators have emphasized that statistical comparisons of the Pa_O2/FI_O2 ratio of survivors and nonsurvivors were not significant on the first day of ARDS (3, 11) and these reports are consistent with our univariate analysis. Bone and colleagues (8) found that the Pa_O2/FI_O2 ratio became significantly higher in survivors only after more than 24 h of conventional mechanical ventilation. However, we found in the multivariate analysis that a high oxygenation index at the onset of patients with ARDS is an independent risk factor for mortality. The oxygenation index has been used primarily in neonatal studies as the mean airway pressure cost of oxygenation. In a study of high-frequency oscillatory ventilation for adult patients with ARDS, Fort and colleagues (34) found that nonsurvivors had a higher oxygenation index than did survivors. However, to our knowledge, the predictive value of the oxygenation index at the onset of ARDS has never been studied in an adult population. We found that the oxygenation index has some value to predict outcome of patients with ARDS; however, its value may be analyzed only after adjustment for underlying diseases, the mechanism of lung injury, and other organ dysfunctions.

The mortality rate (65%) in our study is similar to that in some other studies. Knaus and colleagues (29) have reported a much lower mortality rate in their study (37%). However, their population was more heterogeneous: mechanical ventilation was not required and a more liberal level of oxygenation (Pa_O2/FI_O2 < 300) without radiographic selection criteria was used to select patients. As recommended by the American-European consensus report (2), only mechanically ventilated patients were selected for our study. In addition, our study included a higher ratio of medical patients when compared with other studies. It has been suggested that mortality of medical patients is much higher than that of surgical or trauma patients with ARDS (4, 11, 29).

In conclusion, we found that prediction of hospital survival after the onset of ARDS can be made by taking into account a general severity score, underlying diseases, the mechanism of the lung injury, the pressure cost of oxygenation (i.e., oxygenation index) and the presence of right ventricular dysfunction. Our analysis, by refining the association of these factors with mortality, might constitute a useful tool for the assessment of the real severity of medical and surgical patients with ARDS and the potential impact of novel therapeutic approaches.