INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Acute renal failure (ARF), which is a syndrome rather than a disease, may have numerous causes and can be observed in a wide range of clinical settings. The mortality associated with ARF is high (about 50 to 60%) and stable over time (1). This high mortality rate is still observed despite marked advances in hemodialysis techniques. Controlled studies have shown that the nature of hemodialysis membranes can affect the mortality rate in ARF (2).

More accurate identification of the factors associated with outcome in patients with ARF is, therefore, a crucial issue. A review of the literature shows that there are some methodological problems with published epidemiological studies. For example, the definition of ARF differs, from different levels of increase in serum creatinine to dialysis requirement (1). The causes of ARF are important predictors of survival. A single nephrotoxic insult usually has a good prognosis, whereas for ARF in multiple organ dysfunction syndrome (MODF) the prognosis is poor. The causes of ARF currently observed in intensive care units (ICUs) are generally combined insults, such as hypotension, sepsis, and toxic exposure. The manner in which renal replacement therapy is used in ICUs has also changed. At present, we do not know the extent to which these changes have affected patient outcome. Some information is available from a few prospective multicenter epidemiological studies that applied multivariate analysis to data from a reasonable number of patients (5).

However, the effect of ARF that is present at ICU admission or that develops during the ICU stay has been only scantly investigated. In a one prospective multicenter French study the delayed occurrence of ARF was identified by multivariate analysis as an independent predictive factor of in-hospital death (7). However, ARF was considered as being delayed when it occurred after the first day of the ICU stay. Some patients are admitted to the ICU with an ARF episode, but because not all the definition criteria are present initially, the patients are classified as having delayed ARF when they fulfill the criteria later. They should be classified as having ARF on admission. In the present study we aimed to improve the description of the time course of occurrence of ARF in a large number of patients and to test the hypothesis that acquired ARF is associated with a poorer prognosis than ARF present on ICU admission.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

This prospective, uncontrolled observational study was conducted from March 1, 1996 to the end of May 1997. The study involved a 12-mo inclusion period and an additional 3-mo follow-up period.

Population

All adult patients (age, > 18 yr) admitted to 1 of the 28 medical or medical-surgical ICUs in the Rhône-Alpes area of France between March 1, 1996 and February 28, 1997 with ARF present at the time of admission or occurring during the stay in the unit were included. These 28 ICUs represent all the multidisciplinary ICUs of the Rhône-Alpes area (southeastern France), which is the second-largest area of France in terms of population. These ICUs had a total capacity of 313 active beds. During the study period there were 4,485,147 inhabitants over 18 yr of age in the area. Two specialized renal ICUs were excluded from the survey, because it was thought that the population would be different in these ICUs compared with the others. The 28 ICUs that participated in the study were located in 20 hospitals (12 general hospitals, 4 private hospitals, 3 university-teaching hospitals, and 1 military hospital).

ARF was defined by the criteria of Fagon and coworkers (8), as a serum creatinine concentration > 300 µmol/L, or a urine output < 500 ml/24 h (or < 180 ml/8 h), or the need for hemodialysis. Patients with chronic renal failure were included if their serum creatinine concentrations increased by more than 100% from their baseline values. Patients undergoing chronic dialysis were excluded, as were those transferred from another ICU.

Data Collection

At time of ICU admission, the following variables were recorded: age, sex, identification of center, date of admission, Simplified Acute Physiology Score II (SAPS II) (9), MacCabe Scale of previous chronic health status (10) (level 0, no fatal disease; level 1, underlying disease expected to be fatal between 1 and 5 yr; and level 2, underlying disease expected to be fatal in less than 1 yr), main diagnosis, number and type of organ dysfunction (ODF) (8), infection (8), need for mechanical ventilation, origin (home, hospital, emergency room), and indication (medical, scheduled surgery, or emergency surgery) for ICU admission. According to Fagon and coworkers (8), infection was defined as the presence of one or more of the following criteria associated with clinical evidence of infection: two or more positive blood cultures or presence of gross pus in closed space or source of infection determined during hospitalization, or at autopsy in case of death within the 24 h.

At time of ARF the following variables were recorded: date, SAPS II, number and type of ODF, including infection (8), need for mechanical ventilation, oliguria (daily diuresis < 500 ml) or nonoliguria (daily diuresis  500 ml). The following variables, which are either possible mechanisms or etiologies of ARF, were also recorded: prerenal ARF (increase in urine output after volume expansion and/or diuretic therapy), shock (8), sepsis (8), toxic exposition, disseminated intravascular coagulation (DIC), hemolysis, acute pancreatitis, rhabdomyolysis, postrenal ARF (acute obstruction of the urinary tract), and vascular obstruction (thrombosis or emboli of the large renal artery). For each episode of ARF, the beginning and end dates, the renal function outcome, and the use of hemodialysis were recorded. The criteria for renal replacement therapy were not defined by a uniform protocol because the various centers had different practices. However, when hemodialysis was not performed, we determined whether this was due to the absence of indication, a rapid fatal issue for the patient, or a do-not-resuscitate order. The end of each ARF episode was defined as the disappearance of the three diagnostic criteria for ARF (8).

Patient Follow-up

After inclusion, the patients were prospectively monitored up to hospital discharge on a regular basis. At discharge they were classified as dead or alive, and the data were used to calculate the in-hospital mortality rate.

Statistical Analysis

Definition of groups. Patients were grouped on the basis of the time delay to ARF occurrence:

Group A. Criteria for ARF present on ICU admission or during the first two d of ICU stay

Group B. Criteria for ARF present from Day 3 to 6 (inclusive) of ICU stay, with the suspected cause of the ARF episode not present on ICU admission

Group C. Criteria for ARF present from Day 7 of ICU stay or after

Patients were classified into the following five etiological categories on the basis of the suspected mechanisms of ARF:

Prerenal ARF. Increase in urine output after volume expansion and/ or diuretic therapy without any other recognized cause

Vascular obstruction. Thrombosis or emboli of a/the large renal artery(ies) without any other recognized cause

Postrenal ARF. Acute obstruction of the urinary tract without any other recognized cause

Acute tubular necrosis (ATN). ATN was subdivided into toxic exogenous ATN (exposure to antibiotic, contrast medium, or other nephrotoxic injury); toxic endogenous ATN (hemolysis, rhabdomyolysis); ischemic ATN (resulting from shock, DIC, or sepsis); and mixed ATN (resulting from both toxic and ischemic renal insults).

Undertermined. The remaining cases were considered as having an undetermined etiology

Analysis. Univariate comparisons were performed to compare patients in Groups A, B, and C, to compare survivors and nonsurvivors at hospital discharge. In the latter analysis, the time course of ARF occurrence was classified as early (definition for Group A, above) or late (definition for Group B or C, above). Because ischemic ARF was prevalent in our population, we defined two groups depending on the absence or presence of ischemic ATN. Univariate statistical analyses were performed by ² tests for comparing proportions and by analysis of variance (ANOVA) for comparing means. We used a ² test for linear trend analyses (extension of Mantel) to take into account the inherent order among the categories of ARF.

Multivariate logistic regression analysis was performed, taking survival at hospital discharge as a dependent variable, and the variables found to be significant in the univariate analyses (p up to 0.10) for survivors and nonsurvivors as independent variables. In an initial phase, we applied the logistic regression model to a sample of 510 patients obtained computer-derived randomization (11). The ability of the subset of variables obtained from the logistic regression model to predict in-hospital survival was then tested in the 576 remaining patients. The power of the equation to predict in-hospital mortality was studied in this second set of patients by a receiver operating characteristic (ROC) curve. The odds ratios were used to estimate the association between the covariates and the dependent variable. The values are expressed as means ± SD unless otherwise stated. A p value < 0.05 was considered significant (two-tailed). The data were analyzed with EPI-Info (version 5; Centers for Disease Control, Atlanta, GA) and SPSS software (version 7.5 for Microsoft Windows 95; SPSS, Chicago, IL).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Whole Population at ICU Entry

During the study period, 14,116 patients entered the 28 ICUs, and the overall mortality rate in the study period was 15% (95% confidence interval [CI], 8-22%). Among these patients, 1,086 (736 men) presented the criteria of ARF, either on admission or during the ICU stay, giving an overall incidence rate for ARF of 7.7% during the study period. The number of patients with ARF included by each center ranged from 2 to 102. Sixty-two patients had more than one ARF episode during the same ICU stay: 57 patients had 2 episodes, 3 patients had 3 episodes, and the remaining 2 had 4 episodes. The mean number of ARF episodes per patient was not different between Groups A, B, and C. Mechanical ventilation, either invasive or noninvasive, was necessary for 806 (74%) patients.

Time of ARF Occurrence

On the basis of the time of ARF occurrence in relation to ICU entry, 736 (68%) patients were classified as Group A, 202 (19%) patients as Group B, and 148 (13%) patients as Group C. Table 1 shows the distribution of age, sex, MacCabe Scale score, clinical indication for ICU admission, and origin of admission. Patients in Group A were less often male and more frequently admitted for a medical indication than those of the two other groups.

The severity of patient illness at time of both ICU admission and ARF is shown in Table 2. At the time of ICU admission, the patients in Group A were more severely ill than those in the other two groups. The mean SAPS II score and the number of ODFs (including renal dysfunction) were significantly greater in Group A than in either of the other groups. The incidence of cardiovascular dysfunction was higher in Group A than in the other groups, and this significantly decreased across the groups. Similar findings were observed for hepatic, hematological, and neurological dysfunctions, although the difference was significant only between Groups A and C. Opposite results were found for respiratory dysfunction and the need for mechanical ventilation, which were less frequent in Group A, with the incidence significantly increasing from Group A to Group C. Infection was present in 36% of the entire population and its incidence was lower in Group A than in Group C. Except for the incidence of infection, Groups B and C were quite similar (Table 2). At time of ARF, the mean values of SAPS II were not different between the three groups. Between ICU admission and ARF occurrence, the mean values of SAPS II were unchanged in Group A but increased significantly in Groups B and C (p < 0.01). At the time of ARF, there were significantly more ODFs in Groups B and C than in Group A. Between ICU admission and ARF, the mean number of ODFs was unchanged in Group A but increased significantly in Groups B and C (p < 0.01). Cardiovascular dysfunction was significantly more frequent in Groups B and C than in Group A. 

The criteria used for diagnosing the first episode of ARF are given in Table 3. Oliguria was the most frequent criterion used to define ARF, followed by serum creatinine and hemodialysis requirement. Only hemodialysis requirement was significantly different between Groups A, B, and C; the later the ARF, the more frequently hemodialysis requirement used as the diagnostic criterion for ARF. The mean values of serum creatinine at the time of ARF were 350 ± 272 µmol/L in Group A, 277 ± 105 µmol/L in Group B, and 238 ± 107 µmol/L in Group C (Group A versus B and Group A versus C, p < 0.0001; Group B versus C, p < 0.001). The occurrence of oliguria was similar in the three groups. Hemodialysis was more frequently required the later the ARF occurred after ICU admission. Hemodialysis was not undertaken in 517 patients for the following reasons: judged to be unnecessary by the attending physician in 316 patients (61%); do-not-resuscitate order in 113 patients (22%); and rapid death of the remaining 88 patients (17%).

The etiological factors implicated in the first episode of ARF are listed in Table 4. The majority (83%) were due to ATN, with this being ischemic ATN most frequently. The second, but far less frequent, factor involved was prerenal mechanism. Ischemic ATN was more frequent in delayed ARF, whereas prerenal ARF was more frequent in earlier ARF.

The average length of ICU stay was 16 ± 21 d for the whole population, and 11 ± 15, 19 ± 20, and 37 ± 29 d for Groups A, B, and C, respectively (p < 0.0001 between Groups A, B, and C). The mean duration of hospital stay was 22 ± 25 d for the whole population, and 18 ± 22, 24 ± 26, and 39 ± 31 d for Groups A, B, and C, respectively (p < 0.0001 between Groups A, B, and C).

In-hospital Mortality

For all patients the in-hospital mortality rate was 66% (716/ 1,086); it was 61% (450 of 736) for Group A, 71% (144 of 202) for Group B, and 82% (122 of 148) for Group C (p < 0.0001 in a ² test for linear trend between the three groups). Among the 517 patients who did not receive hemodialysis, the mortality rate was 61% (314 of 517), which is not significantly different from that for the overall population. However, all patients who did not receive hemodialysis because of a do-not-resuscitate order died.

ARF was more frequently oliguric (82 versus 53%, p < 0.001), more frequently treated with hemodialysis (58 versus 46%, p < 0.001), more frequently delayed (37 versus 23%, p < 0.001), and more frequently due to ischemic ATN (66 versus 31%, p < 0.001) in nonsurvivors than in survivors (Table 5). Nonsurviving patients were older and had more severe illness than did survivors. Nonsurvivors were more frequently stage 1 or 2 on the MacCabe Scale and had transferred to the ICU from the hospital (Table 5).

We randomly selected a sample of 510 patients from the total population. This sample did not differ from the overall population for the variables listed in Tables 1234. The in-hospital mortality rate was 67% in the sample versus 66% in the overall population. Between sample and population, the mortality rates were 65 versus 68%, 22 versus 19% and 13 versus 13% (p = NS) in Groups A, B, and C, respectively. The differences between nonsurvivors and survivors were similar in the sample and the overall population, except for the origin of admission (Tables 5 and 6), thus confirming that the random sample was representative of the entire population. The covariates associated with in-hospital survival of the 510 patients are shown in Table 7. The probability of patient survival was reduced with increasing age, increasing values of SAPS II on ICU admission, increasing number of ARF episodes, and underlying rapid fatal illness (stage 2 of the MacCabe Scale). In contrast, the absence of oliguria, the absence of hemodialysis, and the absence of ischemic ATN were predictive of an increased survival probability. The model was tested and validated in the remaining 576 patients (Figure 1). The area under the ROC curve was 0.769 (95% CI = 0.732-0.803, SEM = 0.02).

View larger version (14K): [in this window] [in a new window]   Figure 1.   ROC curve of equation obtained by logistic regression of data collected from 510 randomly selected patients and tested in the remaining 576 patients.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The results of our study, which is to date the largest multicenter prospective cohort of ICU ARF cases (12), showed the following in our sample: (1) the incidence of ARF in ICUs was 8% per year; (2) the in-hospital mortality rate was high (66%) and increased as ARF occurred later relative to ICU admission; and (3) after controlling for confounding factors, however, the time of occurrence of ARF was not predictive of patient outcome.

Incidence of ARF

We used the definition of ARF introduced by Fagon and colleagues to build a new severity-of-disease classification model to predict ICU outcome (8). With the same definition, we found that the prevalence of ARF during the first 24 h after ICU admission was 4.5% (638 patients of 14,116 admissions) as compared with 26.4% of 1,070 consecutive admissions studied by Fagon and coworkers (8) between October 1, 1987 and March 1, 1990. This marked difference may be explained by differences in case mix, since our study was performed in 28 ICUs, and the Fagon and coworkers study was performed in 1 ICU (8). The overall incidence of ARF in ICUs was 242 cases per million in our study, a value that is relatively close to that of 209 cases per million reported in the Madrid area in 13 tertiary-care hospitals using, however, a different definition of ARF (13).

Time of Occurrence of ARF

In our ARF population, failure was present in two-thirds of the patients at the time of ICU admission and occurred during the ICU stay in the remaining third. Acquired ARF represented 2.5% of the 14,116 patients admitted to ICUs during the study period, which is somewhat lower that the previously published prevalence rates for acquired ARF: 14.9% (14) and 6.6% (15) in ICU; and 4.9% (16) in hospital. Differences in the definition of ARF, case mix, or sample size may explain these differences (14).

Hypotension, sepsis, and toxic renal injury are well-recognized risk factors for hospital- or ICU-acquired ARF (14). Ischemic ATN was the major cause of ARF in our study, involving more than half of the patients. ATN was not biopsy-proven in our study, but was, rather, a differential diagnosis; ischemic ATN was, nevertheless, significantly more frequent in late ARF than in early ARF. The data in Table 2 show clearly that this stems from the combined effect of shock and infection, since cardiovascular dysfunction and infection are more frequent in late ARF than in early ARF. Thus, it would seem that late ARF is related to nosocomial infection and septic shock.

The ARF severity was different between early and late ARF. The serum creatinine level was significantly higher in early ARF than in late ARF, a finding already observed by Brivet and coworkers (7). This may be explained by a greater muscle mass in patients at the time of ICU admission, or conversely, by denutrition, or sepsis-induced protein breakdown, or dilution of extracellular fluid in patients with late ARF. Oliguria was as frequent in early as in late ARF in our study, whereas Brivet and coworkers (7) observed that the incidence of oliguria was 88 versus 52% in early and late ARF, respectively, an important difference that was reported not to be significant. However, by using the data provided in Table 2 in the report by Brivet and colleagues (7), we found that this difference was highly significant (² = 41.68, p < 0.0000001). In our study, the need for hemodialysis was significantly higher in late ARF than in early ARF, although in late ARF the serum creatinine value was lower and the occurrence of oliguria was similar in both. Brivet and coworkers (7) observed opposite findings, with 86% of patients requiring hemodialysis in early ARF versus 45% in late ARF. Although the difference was reported as being not significant, when we recalculated using their data, we found a significant result (² = 47.85, p < 0.0000001).

In our study, the patients in Groups B and C had a poorer prognosis than those in Group A. This is due to two phenomena. First, by definition, the patients in Groups B and C were developing ARF during their ICU stay, and so their condition worsened. Second, they presented cardiovascular dysfunction and infection more frequently than did the patients in Group A. There was fewer differences in terms of other organ dysfunctions, and hence the total number of organ dysfunctions increased by twofold in patients in Groups B and C and was higher than that for patients in Group A, an observation that is in complete opposition to what was observed at the time of ICU admission. Therefore, the clinical condition of patients with late ARF declined after ICU admission, probably as a result of nosocomial infection in the majority of cases. It is, therefore, understandable that patients with late ARF died more often than those with early ARF. Similarly, Brivet and colleagues (7) showed that SAPS (19), APACHE II (20), and number of organ dysfunctions (21) increased from the time of ICU admission to ARF occurrence in patients with late ARF. However, at the time of ARF, the number of organ dysfunctions was not different, on average, between early and late ARF, suggesting that the increase in organ dysfunctions was due only to the occurrence of ARF. Consistently, in late ARF, the mean number of organ dysfunctions was lower than in our study (1.8 versus more than 3).

Patient Outcome

The mortality rate was high in our study (66%) compared with that of 15% for the total admissions during the study period. However, this level is in the range of that reported in the most recent studies conducted in ICUs (56.7 to 79%; Table 8).

There are many mortality risk factors with interaction, so that multivariate analysis is required to take into account any confounding variables. In our study seven significant predictors of in-hospital survival were selected by the logistic regression analysis in the representative sample of 510 patients. Stage 2 of the MacCabe Scale and oliguria did not reach a significant difference. These covariates have previously been reported several times as risk factors (22).

Many general scores assessing patient severity have been used to predict patient outcome in the event of ARF, such as SAPS, APACHE II, or APACHE III (28). To our knowledge, SAPS II has not been previously studied in this situation. We found that a high value for SAPS II at the time of ICU admission, but not at the time of ARF, predicted a low in-hospital survival. In the study by Brivet and colleagues (7), SAPS at the time of ARF, but not at the time of ICU admission, predicted a high in-hospital mortality. In their study, APACHE II was not predictive of patient outcome either at the time of ICU admission or at the time of ARF occurrence. We have suggested an explanation for these discrepant results for SAPS and SAPS II between the two studies. The power of APACHE II to predict mortality in ARF is controversial (22, 24, 26). Nevertheless, in light of the higher value of serum creatinine in the nonsurvivors than in survivors observed in our study, it may not be surprising that APACHE II does not predict outcome in ARF. Unlike SAPS and SAPS II, APACHE II includes serum creatinine, which has an increasing contribution to patient death as its value increases.

Previous health status was found to be a significant contributing factor to patient outcome in our study. This was mainly due to the Level 0 score of the MacCabe scale, which significantly predicted patient survival. Previous health status, as assessed by another index, was also predictive of patient outcome in the study by Brivet and coworkers (7).

We found that no hemodialysis was a significant predictor of patient survival, as observed in the univariate analysis (i.e., mortality of 72% in patients who were dialyzed versus 62% in those who were not [² = 6.76, p = 0.009, odds ratio 1.64, 95% CI 1.11-2.42]). This result was obtained despite the fact that hemodialysis was not offered to 64 patients owing to a do-not-resuscitate order, all of whom died, and that 44 patients died rapidly, before hemodialysis could be performed. When these patients were excluded from the analyses, similar results were obtained. This observation raises the question of how to use renal replacement therapy in ICU. Does this finding suggest that hemodialysis performed in ICUs may be detrimental? Or does it mean that hemodialysis is not offered to the patients with less severe illness? This latter possibility is unlikely because the variable absence of hemodialysis is thought to predict patient outcome independently of the other covariates, particularly SAPS II. Brivet and coworkers (7) reported that the need for hemodialysis had an effect on mortality but only in their univariate analysis. None of the more recent studies on ARF has reported an effect of renal replacement therapy on patient outcome in multivariate analysis. In all of the centers involved in our study, biocompatible membranes were used, essentially polyacrylonitrile membranes. We are currently analyzing our data for the dialyzed patients.

Delayed occurrence of ARF was not a risk factor for patient outcome after adjusting for confounding variables. Thus, although the mortality rate was higher when the onset of ARF was most delayed, late-onset ARF was not the sole factor explaining the outcome. This is in opposition to the results in the study by Brivet and colleagues (7), who found by logistic regression analysis that delayed ARF was a risk factor. Differences in definitions for ARF itself, or for late and early ARF, and in patient severity may explain this discrepancy.

Finally, in our study, ischemic ATN was the most powerful prognosis factor with the highest odds ratio. This is not surprising since our etiological classification resulted in ischemic ATN being the most frequent cause of ARF. Moreover, ischemic ATN, as defined in our study, included shock and sepsis, which are well-recognized risk factors for mortality in ARF (17). Since ischemic ATN was highly correlated with cardiovascular dysfunction and infection, it is not surprising that these two other variables were not in our final model.

One strength of our study is that our model has been validated internally. As shown in Figure 1, the ROC curve obtained in the test group was excellent. Douma and coworkers (29) retrospectively compared different models of outcome prediction in 238 patients with ARF treated by hemodialysis in a single ICU. Seven general prediction models, i.e., APACHE II, APACHE III, SAPS, SAPS extended (30), SAPS reduced (30), a sickness scoring model (31), and a mortality prediction model (32), and four specific prediction models for ARF (5, 6, 22, 33) were tested. APACHE III and the specific index of Liano and colleagues (6) were the most likely to identify the subgroup of patients with a quasi-100% in-hospital mortality despite dialysis. The area under the curve in our study was close to those obtained with APACHE III and the Liano equation (6) in the study of Douma and coworkers (29).

In conclusion, our prospective multicenter study confirmed the consistently high mortality rate associated with ARF in ICUs in the late 1990s. Delayed occurrence of ARF during the ICU stay affected patient outcome only in univariate analysis. A model was developed on the basis of a random representative sample from our population, by logistic regression analysis, and this model was validated internally on the remaining patients. This model included the following factors: age, SAPS II on ICU admission, number of ARF episodes, MacCabe Scale, oliguria, hemodialysis, and ischemic ATN.