INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The increasing importance of patient-centered outcomes and fiscal constraints are leading critical-care researchers to recognize the importance of long-term mortality and quality of life after critical illness in addition to considering physiologic and intensive care unit (ICU) outcomes when evaluating the utility of treatments (1). Without understanding these effects of critical illness, it is impossible to construct reference-case cost-effectiveness analyses (2, 3). Recent data suggest that mortality associated with sepsis continues for years after hospital discharge as compared with the mortality of matched controls (4). If a substantial proportion of mortality after critical illness occurs in the months following hospital discharge, hospital and ICU mortality may be inadequate endpoints for clinical trials of critical care. Moreover, long-term outcomes are essential to patients and families making decisions about continued life support and aggressive medical care.

During the 30 yr since the original description of ARDS, a great deal of information has been collected on the risk factors, comorbidities, and hospital mortality associated with this syndrome. Recent single-center studies suggest that hospital mortality in patients with ARDS has been declining (8, 9). In the acute setting, ARDS doubles hospital mortality over that of patients with similar risk factors who do not develop ARDS (10). There are data on the return of lung function and quality of life in survivors of ARDS (11). However, despite this body of clinical research, there have been no published studies of the long-term mortality of patients with ARDS.

The purpose of this study was to determine the long-term mortality of patients who survived an episode of ARDS and to assess the impact of other clinical factors on long-term mortality after this syndrome. In an effort to discriminate the specific effect of lung injury on long-term mortality from the effects of ARDS-associated risk factors and critical illness in general, we evaluated survival in ARDS patients and in a comparable control group.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Study Design and Setting

We previously reported on the assembly, collection of data from, and follow-up of a subset of the cohort in the present study in an earlier study on quality of life in ARDS survivors (11). The present study was a prospective, matched, parallel cohort study. Patients with ARDS associated with trauma or sepsis who survived to hospital discharge constituted the subject cohort. A group of critically ill patients who also survived to hospital discharge but did not have ARDS were the potential controls. The study site was Harborview Medical Center (HMC), a 411-hospital-bed, 48 ICU-bed municipal medical center in Seattle, Washington. HMC is also the only Level 1 Trauma Center in a four-state area. The University of Washington Institutional Review Board approved the research protocol for the study.

Subjects

As part of an ongoing ARDS Specialized Center of Research (SCOR) program, all ICU patients at HMC are screened daily to identify those with ARDS or acute lung injury (ALI). In addition to severity of illness and demographic and health-care utilization data, the clinical risk factor for developing ARDS is assessed according to previously reported criteria (15). For the purpose of this study, all patients with ARDS with either trauma or sepsis as their ARDS clinical risk factor, who were admitted to an ICU at HMC between January 1, 1994 and July 31, 1996 and who survived to hospital discharge, were identified from the ARDS SCOR database (Figure 1). ARDS was defined as the acute onset of diffuse, bilateral pulmonary infiltrates involving  50% of at least three lung quadrants, a Pa_O2/FI_O2 < 150 or Pa_O2/FI_O2 < 200 with a positive end-expiratory pressure of  5 cm H₂O, and a pulmonary capillary wedge pressure of  18 mm Hg or, if no pulmonary capillary wedge pressure was available, no clinical evidence of congestive heart failure (9, 12, 15). The presence of comorbidity was determined according to the definitions in the Acute Physiology and Chronic Health Evaluation (APACHE) III (16).

View larger version (28K): [in this window] [in a new window]   Figure 1.   Identification and construction of the study cohorts.

Sepsis as a risk factor for ARDS was defined as any two of the following: temperature > 39° C, white blood cell count  12,000/mm³,  20% band forms, or a known or strongly suspected source of infection. In addition, one of the following findings had to be present with no alternative explanation: anion gap > 20 mEq/L or base deficit > 5 mEq/L, systemic vascular resistance < 800 dynes/s/cm², or systolic blood pressure (BP) < 90 mm Hg for > 2 h (15). Trauma as a risk factor for ARDS was defined as any one of the following: (1) fracture of two or more major long bones; (2) unstable pelvic fracture; (3) fracture of one major long bone and a major pelvic fracture; (4) thoracic trauma including pulmonary contusion; or (5) massive blood transfusion ( 15 units/24 h) associated with trauma (15).

Controls

The control group was selected to reflect a population of patients who were similar to the ARDS subjects in important respects, with the exception of developing ARDS. Because one goal of this study was to examine the effect of ARDS on mortality after hospital discharge, only controls who survived to hospital discharge were selected. Therefore, control patients were selected from patients admitted to the ICU with a diagnosis of trauma or sepsis, or who developed sepsis during their ICU stay, who did not develop ARDS during their hospitalization, and who survived to hospital discharge.

Potential trauma controls were identified through the HMC Trauma Registry, and were included as potential controls if they were admitted to the ICU between January 1994 and December 1996 and survived to hospital discharge (17). Sepsis controls were identified from the screening logs of three sepsis studies (from December 1989 to December 1997) at HMC, and were verified through chart review. ARDS patients with trauma as their clinical risk factor (trauma-ARDS) were matched to trauma controls by having an Injury Severity Score (ISS) within five points of the latter and a date of hospital admission within 6 mo of the controls (18).

The potential sepsis controls met a common definition of sepsis that included having all four of the following criteria: temperature  39° C or  35.5° C, heart rate  90 beats/min, respiratory rate > 20 breaths/min or minute ventilation > 10 L/min, and a known or strongly suspected source of infection. In addition, sepsis controls met one of the following indicators of organ failure within 48 h of meeting the foregoing criteria: ongoing metabolic acidosis with an anion gap > 20 mEq/L or base deficit > 5 mEq/L, systemic vascular resistance < 800 dynes/s/cm², unexplained hypotension with a systolic BP  90 mm Hg for > 2 h or vasopressor use, platelet count < 81,000/mm³, urine output < 0.5 ml/kg/h or  30 ml/h, Pa_O2 < 70 mm Hg on room air or Pa_O2/ FI_O2 < 333, or an alteration in mental status of  2 points on the Glasgow coma scale (GCS  13). ARDS patients with sepsis as their clinical risk factor (sepsis-ARDS) were matched to controls through having an APACHE III score within 15 points and date of hospital admission within 6 mo of the controls (16).

Data Collection

Demographic information, including residential addresses and telephone numbers of study patients and their next-of-kin, was obtained from medical records. Patients were initially contacted in writing by their in-hospital attending physician. Subjects were then contacted in writing by the investigators and after that by telephone about participation. Initial attempts to determine the survival status of study subjects were made by telephone contact. Attempts to contact next-of-kin were made only if a subject was not contacted by telephone. If telephone contact with the next-of-kin revealed that the patient was dead, no further verification was performed. Death records from Washington State were queried for all study patients reported by next- of-kin as being alive or who could not be located despite an exhaustive search of available contact information. If direct contact could not confirm the patient's death, and if the Washington State Center of Health Statistics database did not contain a record of the patient's death, the patient was presumed to be alive for the purposes of this study.

Statistics

Data are presented as median and range unless otherwise indicated. Since ARDS survivors were matched to comparably ill or injured controls, the ARDS-control pair was the unit of analysis for statistical testing. Descriptive and severity-of-illness measures of ARDS survivors and matched controls were compared through Wilcoxon's signed rank test for matched pairs for continuous variables, and through chi-square tests for categorical variables. Survival time was compared with the Kaplan-Meier method and Cox proportional hazards model. All tests were two-sided and results were considered significant at p < 0.05. SAS PC statistical software (SAS Institute, Cary, NC) was used for all analyses.

We used data from a study on the long-term effects of sepsis on mortality, and reasoned that ARDS would have a similar effect during our observation period (relative risk [RR] for death = 3.0) (7). Using a two-sided value of  = 0.05 and assuming a mortality in the control patients of 10%, we calculated that 72 patients per group would be necessary to detect this effect with a value of (1-) = 0.80.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Between January 1, 1994 and July 31, 1996, 328 patients were identified with ARDS at HMC (Figure 1). Of the ARDS cases, 207 (63%) were determined to have had sepsis (n = 119) or trauma (n = 88) as their ARDS clinical risk factor. Hospital mortality was 14% (n = 12) among patients with trauma-ARDS and 43% (n = 51) among patients with sepsis-ARDS. Of the 79 hospital and posthospital deaths during the follow-up period among the 207 patients with trauma- or sepsis-ARDS, 63 (80% of all deaths) occurred in the hospital, 61 (77% of all deaths) occurred by Day 30 after the onset of ARDS, and 70 (89% of all deaths) occurred by Day 100 after the onset of ARDS. There were 76 survivors of trauma-ARDS and 68 survivors of sepsis-ARDS. The median follow-up time for subjects and controls was 753 d (range 8 to 1,503 d). When both hospital and posthospital deaths were considered, the median time until death after onset of ARDS was 10 d. Sixteen ARDS patients (two with trauma-ARDS, three with sepsis-ARDS who did not have controls, and 11 with sepsis-ARDS) died after hospital discharge during the follow-up period.

All trauma-ARDS patients were matched to a control. Matching was not possible for 17 (25%) of the sepsis-ARDS patients. The primary reason for failure to find a matching control for these patients was the lack of sepsis registry data for a period in 1995. The 17 unmatched sepsis-ARDS survivors were similar to the 51 matched sepsis-ARDS study patients with respect to age, sex, APACHE III score, comorbidities, length of stay, late mortality, and state of residence. The remaining 127 ARDS patients, 76 with trauma-ARDS and 51 with sepsis-ARDS, were matched 1:1 with controls who were similar with respect to age, sex, APACHE III score, and ISS, and who had similarly low rates of most comorbidities (Table 1). ARDS patients had a longer hospital length of stay, a higher proportion of non-Washington State residence, and a higher proportion of cirrhosis. We were unable to confirm the status of 55 patients by direct contact or through the Washington State Center of Health Statistics database. Seven of these patients were sepsis-ARDS patients, seven were sepsis controls, 19 were trauma-ARDS patients, and 22 were trauma controls. Subsequent analyses focused on the 254 ARDS patients and matched controls (127 matched pairs) to evaluate the effect of ARDS and risk factor on late mortality. Figure 2 depicts the survival of the study patients after hospital discharge. There were 31 (12%) deaths in this group during the follow-up period. Thirteen of the deaths involved ARDS patients (10% late mortality) and 18 deaths involved controls (14% late mortality) (p = 0.4 for comparison of late mortality in ARDS patients and controls). No significant difference in mortality was noted between subjects with sepsis-induced ARDS and matched sepsis controls (p = 0.4), or between subjects with trauma-induced ARDS and matched trauma controls (p = 0.65).

View larger version (13K): [in this window] [in a new window]   Figure 2.   Survival plot of ARDS patients and controls.

To evaluate the independent effects of ARDS and risk group on survival, and to generate confidence intervals (CIs) around our survival estimates, we performed two multivariate Cox survival analyses. The hazard ratio for ARDS after controlling for age, risk group, and comorbidity was 1.00 (95% CI: 0.47 to 2.09). To see whether the effect of ARDS on survival was similar in sepsis and trauma patients, we added an interaction term to the model. This term was not statistically significant (p = 0.78), and the effect of ARDS on late mortality was therefore similar in sepsis and trauma. To evaluate the effect of sepsis as compared with trauma on survival independent of the effect of ARDS, we examined the coefficient for risk factor. The hazard ratio for sepsis as compared with trauma was 5.45 (95% CI: 2.05 to 14.45). After controlling for risk and age, we found that comorbidities, interpreted as the presence of any of the APACHE III comorbid diagnoses, had a significant effect on long-term survival (hazard ratio: 2.93; 95% CI: 1.17 to 7.33). After controlling for risk and comorbidities, we found that age, interpreted as the increase in mortality rate per year of age, also had a significant effect on long-term survival (hazard ratio: 1.05; 95% CI: 1.03 to 1.07).

To test the sensitivity of our analyses to the assumption that patients not known to be dead were alive, we recalculated the survival model and mortality figures under the worst-case scenario that all patients presumed to be alive for the primary analysis died on the day after hospital discharge (Tables 2 and 3). The multivariate estimate of the effect of ARDS on survival did not change, but its CI narrowed. Because the assumption preferentially increased the number of deaths in the trauma group (a younger group with fewer comorbidities), the effect of sepsis, age, and comorbidity on late survival was eliminated. Estimates of late mortality were higher for all groups under the assumptions for the worst-case scenario.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

To our knowledge, this is the first study to describe long-term mortality following ARDS. ARDS has no effect on survival after hospital discharge as compared with that of equally ill patients with trauma or sepsis. Trauma patients who survived to hospital discharge had a very low long-term mortality rate regardless of whether or not they had ARDS. Patients with ARDS related to sepsis died at a much higher rate after discharge from the hospital, but at no greater rate than sepsis patients without ARDS.

Although ARDS has been observed to increase in-hospital mortality, it had no effect on posthospital survival in our study. After matching for ISS (in trauma patients) and APACHE III score (in sepsis patients), and with control for age, risk factor, and comorbidity in a multivariate model, ARDS had no effect on survival in patients who were discharged alive from the hospital. Late mortality was strongly influenced by the risk factor for ARDS; sepsis patients had a 6-fold greater mortality rate than did trauma patients. The presence of a serious comorbid diagnosis nearly tripled the mortality rate after hospital discharge. Furthermore, older patients had a higher mortality rate after hospital discharge. Sixty-year-old patients discharged from the hospital after critical care for sepsis or trauma had a rate of death 2.8 times that of 40-yr-old patients when risk factor, comorbidities, and the presence of ARDS were controlled.

No prior study has described the long-term mortality following ARDS, and we therefore cannot compare our findings with those of other studies. However, the importance of risk factor and comorbidity in determining hospital morbidity and mortality after ALI has been reported by others (19, 20). It is not surprising that risk factor is a major determinant of long-term survival after ARDS. The observation that sepsis reduces long-term survival when compared with trauma is consistent with the observations of reduced long-term survival in other sepsis cohort studies. The cumulative 1-yr mortality following sepsis has been reported to be 43 to 48% (4). Perl and colleagues followed patients for 2 yr after an episode of sepsis and found their mortality to be 59% (6). Our results indicate that the cumulative 1- and 2-yr mortalities of sepsis-ARDS patients from 1994 to 1996 at HMC were 47% and 51%, respectively. Therefore, sepsis-ARDS patients appear to have a long-term mortality similar to the reported long-term mortality of patients with sepsis regardless of the presence of lung injury.

Although it has been extensively studied, there are conflicting data on the effect of age on critical-care outcome (21). The interpretation of the studies producing these data, including our own results, is complicated by several factors. Age is highly correlated with comorbidity, and it may be difficult to separate their independent effects because measures of comorbidity are relatively crude. Critical-care outcome can only be studied in patients who are selected for admission to the ICU, and therefore only reflects the outcome of a subset of elderly patients. Additionally, some elderly patients, in a factor that may vary by geography or disease, choose less aggressive care in the ICU and will have higher mortality rates without an underlying biologic explanation.

Because we observed no statistically significant independent increase in long-term mortality associated with ARDS, an important potential limitation of our study was its power, with 254 patients and 31 deaths reflected in the CI around the estimated Cox hazard ratio of 1.00 (95% CI: 0.47 to 2.09) for ARDS. However, there was not even a trend toward increased late mortality associated with ARDS.

Another potential limitation to our study was incomplete or biased ascertainment of mortality. To the extent that this mortality ascertainment problem occurred, it should have affected the assessment of ARDS subjects and controls equally and should therefore not have introduced a bias. In addition, a sensitivity analysis of a worst-case scenario in which all of the patients presumed alive were assumed to die immediately after hospital discharge did not materially affect our results. The definition of sepsis was slightly different for the sepsis-ARDS and sepsis control groups. Because all of the organ failure variables that differed in the two definitions of sepsis are contained in the APACHE III score, the severity-of-illness matching should have balanced any residual small variations. It is possible that failure to include the unmatchable ARDS patients biased the results. This seems unlikely, because there were no statistically significant differences between the ARDS patients in the cohort and those excluded because of lack of a matched control. We compared only the long-term mortality of sepsis patients with that of trauma patients, and therefore cannot comment on the comparison of these groups' mortality with the long-term mortality of our ICU population as a whole. Because of the younger age and lack of comorbid illness of trauma patients, their long-term mortality is likely to be somewhat better than that of the general ICU population. Therefore, the difference between long-term mortality after sepsis and trauma as found in our study may overestimate the difference between survival in sepsis and survival of a general ICU population. We also cannot comment on the long-term effects of ARDS stemming from other risk factors, but in most series, sepsis and trauma account for the majority of cases of ARDS. It is important to realize that the effect of ARDS on mortality following hospital discharge depends to some degree on the timing of hospital discharge. In our center, 9% of ARDS patients were discharged to chronic-care facilities while still on ventilators during this period. Similar studies, performed in a setting in which patients with ARDS are discharged earlier to other settings while still on ventilators, may count as posthospital deaths some deaths that were captured as hospital deaths in our study. Although we are confident that none of the control patients met our criteria for ARDS, it is possible that some met the less severe hypoxemia criterion for ALI (25). If ALI and ARDS have similar effects on long-term survival, and if a significant number of patients with ALI were included among our controls, this would reduce the effect of ARDS on late survival observed in our study.

A unique feature of this study was our ability to compare the mortality of ARDS patients with that of a group of matched controls. We could therefore evaluate the independent contributions of risk factor, age, comorbidity, and ARDS to long-term mortality. We felt that it was important not to control for duration of mechanical ventilation or length of hospital stay, since these factors might have been part of the causal pathway between ARDS and an effect on long-term mortality. We restricted our analysis to hospital survivors because the effect of ARDS on hospital mortality has been described, and we were specifically interested in the independent effect of ARDS on late mortality. An additional strength of our study was the rigor with which each patient's vital status was ascertained. It is unlikely that deaths went undetected. The prospective identification of ARDS subjects and controls makes it unlikely that there was a bias in the assembly of either group.

This study has important implications for physicians caring for patients with ARDS. Patients with ARDS who survive to hospital discharge do not incur an additional risk of mortality later in life, other than the risk imposed by their risk factor for developing ARDS and their comorbidities. The study also has important findings for investigators. Clinical trials that use hospital mortality will capture 80% and those that use 100-d mortality as an endpoint will capture 89% of of all deaths that occur within 2 yr after ARDS associated with sepsis or trauma. Data on long-term outcomes of critical illness are essential for modeling the cost-effectiveness of therapies in critical care. For example, the data on long-term outcome from the present study can be combined with the data on quality of life in ARDS survivors and data on the cost and efficacy of an ARDS treatment to generate accurate cost-utility models (2, 11). Cost-utility ratios can be compared with other economic data to determine the relative value of treatments for critically ill patients. In conclusion, our study suggests that survival after hospital discharge in patients with ARDS associated with trauma or sepsis can be modeled on survival data based on underlying risk factor, age, and comorbidities.