This one-month (November 2004) prospective multicenter observational study was conducted in 41 adults surgical (n = 22) or medical/medico-surgical ICUs (n = 19) in 29 teaching university and 12 non-teaching hospitals. Participating ICUs, volunteers participating in the study, were widely distributed throughout France. These were closed units of more than six beds, non-specialized units (avoiding cardiac and neurosurgical ICUs), with a critical care specialist and microbiology laboratory on hand 24 hours a day.

Legal organization of day shifts and "out-of-hours" hours in French ICUs has been previously described 16 . Briefly, day shifts as defined by law run from Monday to Friday, 8:30 am to 6:29 pm, and Saturday from 8:30 am to 12:59 pm; the remaining period corresponds to off hours. Overall during the study period, day shifts accounted for 218 hours (30.2%) in a total of 720 hours of work.

In these units, day-shift medical teams consisted of a median of three (range, 1 to 6) senior physicians board certified in critical care medicine, a median of one (range, 0 to 3) critical care specialist in training (certified medical specialist in anesthesiology, or medical specialty), and a median of two (range, 0 to 5) residents. During out-of-hours, one critical care specialist (board certified or in training) was on call on site, either alone (in 14 ICUs) or with a medical resident.

In each center, the principal investigator was the senior critical care specialist leading the team and fully responsible for the ICU. All consecutive adult patients admitted to the ICU during the study period were eligible for enrollment. Criteria used for diagnosis, microbiologic techniques and the decision to prescribe AT were left to the physician's discretion. Ethics Committee approval for the protocol was obtained. In accordance with French law, as the study protocol was strictly observational and did not modify clinical practice, information was given to the patients and their familly but no written informed consent was obtained from our patients. Approval of the CNIL (Commission Nationale de l'Informatique et des Libertés) was obtained, ensuring that patient data were kept confidential according to French regulations. A Scientific Committee independently designed the study and reviewed all data collected.

For each ICU admission, demographic characteristics, underlying diseases, severity of illness, and type of admission were recorded on a standardized report form. Severity of illness on admission was assessed using the simplified acute physiology score II (SAPS II score) 17 . Underlying diseases were classified as not ultimately fatal, ultimately fatal (death expected in <5 years) or rapidly fatal (in <1 year) according to the McCabe score 18 .

To assess the incidence of AT during the ICU stay, the patients were classified into four categories: (I) patients not receiving AT either at the time of admission, or during their ICU stay; (II) patients suspected of having bacterial infection and already receiving AT at the time of admission; (III) patients with known infection with identification and susceptibility testing of the pathogen at the time of admission on which AT was based; (IV) patients receiving new AT for a new suspicion of infection during their ICU stay (Figure 1). This last subgroup was analyzed specifically. In patients who developed several infections during their ICU stay, only the first episode of new AT was considered. A preceding seven-day course free of antibiotics was required before considering a new course of AT. Antibiotic prophylaxis was not analyzed in the current study.

In each center, the presence and number of empiric AT protocols were assessed. The period of initiation of AT was defined by categorizing the week into day shifts and out-of-hours. The type of prescriber was assessed: fellow or senior physician (assistant professor, senior critical care specialist). The individual or team decision (>2 physicians) for initiation of AT was assessed. When infectious disease specialists were involved in the decision-making progress, they were considered as a part of the team. Patients with one of the following diagnoses were classified as being immunosuppressed: febrile neutropenia, splenectomized patients, cirrhosis, solid organ transplantation, steroid therapy, and HIV infection 19 . Therapeutic emergencies were defined as septic shock, hypoxemic pneumonia or multiple organ failure (MOF) 19 . The sequential organ failure assessment (SOFA) score was calculated at the time of initiation of AT 20 . The supposed source of infection was recorded.

Applied microbiologic techniques were based on the recommendations of the French Society for Microbiology 21 . Microbiologic results were recorded as part of the decision-making process for initiation or changes of AT. The definitions used for the site of infection, true pathogens, contaminants and commensals were those recommended by the French Society of Anæsthesiology and Critical Care Medicine 22 . The following timing of AT prescription was analyzed: in the absence or before microbiologic sampling; after microbiologic sampling; on the results of Gram-stained direct examination, on the results of microbiologic cultures (24 to 48 hours); on the results of susceptibility testing (Figure 1). In patients with negative cultures, the decisions were assessed 48 hours after collection of the samples when the cultures demonstrated no growth. Apart from adaptation to microbiologic results, the other reasons for antibiotic changes were recorded: clinical worsening, new site of infection, antibiotic side effect, de-escalation (withdrawing the non-pivotal antibiotic or switching to a narrow-spectrum antibiotic) and discontinuation of aminoglycosides. The quality of antibiotic prescription (dose, intervals, and so on) according to pharmacokinetic/pharmacodynamic criteria was not analyzed.

Patients treated without any microbiologic sampling of their suspected infection or having their treatment based only on microbiologic identification without susceptibility testing were considered to have a low level of microbiologic confirmation of infection. In patients undergoing susceptibility testing of their microbiologic samples, appropriateness of AT was assessed by the principal investigator at the end of the therapeutic course. In order to replicate real life conditions as much as possible, all positive microbiologic cultures were analyzed 22 but appropriateness of AT was only considered for true pathogens. Therapy was judged appropriate if, according to the susceptibility testing 21 , all bacteria considered true pathogens were targeted by at least one of the drugs administered. The other cases were classified as inappropriate AT. The antibiotic selection was judged appropriate or inappropriate on the basis of the culture results obtained. Considering that severe infections encountered in ICU cases require emergency AT, the scientific committee classified the delayed introduction of AT at the time of susceptibility testing as arbitrary and inadequate AT. Fungi were excluded from the analysis of appropriateness and antifungal therapy was not considered.

All patients were followed from the day of admission until ICU discharge. Death during ICU stay was recorded. Links between ICU mortality and clinical features of new AT were assessed.

Patient characteristics according to AT during their ICU stay were analyzed. Characteristics of AT were assessed and their relationships with death were determined.

Data were analyzed using Stata 9.2™ (Stata Corporation, College Station, TX, USA). We assessed that the continuous variables were normaIly distributed using the Shapiro-Wilk test. Variables were expressed as mean with standard deviation and range or numbers with proportions. Groups were compared using the Chi-square test with Yates' correction if necessary for qualitative parameters and ANOVA for quantitative data. Bonferroni correction was used for multiple comparisons. To identify factors independently associated with death, a multivariate stepwise logistic regression analysis was performed among the factors found to be significant at the 15% level in univariate analysis 23 . A backward Wald model was used. The probability to enter in the model was 0.05 and to remove 0.1. Hosmer-Lemshow goodness of fit Chi-square was assessed. The median value of the population was used as a cut-off for quantitative data. Odds-ratio (OR) and their 95% confidence intervals (95% CI) were calculated. Statistical significance was accepted at the 5% level.

A total of 1,702 patients (Figure 1) was studied. The mean number of admissions in each unit was 42 ± 21 pts. Overall, 54 ± 30% of patients were admitted for a medical reason, 9 ± 12% following scheduled surgery, and 37 ± 25% following emergency surgery.

Overall, 34 ± 21% of patients did not receive any AT during their ICU stay, 29 ± 21% were already treated at the time of admission, 4 ± 7% received an AT with identification and susceptibility testing available at admission, and 34 ± 16% received new AT (Table 1). The large variation in the amount of antibiotics used by the different ICUs is illustrated by Figure 2.

Written protocols for empiric AT were available in 25 (61%) ICUs in accordance with national guidelines and adapted to local epidemiology, including antibiotic resistance frequencies. These protocols were defined for community-acquired infections (mainly pneumonia n = 19, intra-abdominal infections n = 19, meningitis n = 18) and nosocomial infections (mainly ventilator-associated pneumonia (VAP) n = 21, postoperative intra-abdominal infections n = 16, septic shock n = 16) with a mean of 6 ± 3 protocols per ICU. No difference was observed between teaching and non-teaching hospitals in terms of the availability (63% vs 57%, P = 0.72) and mean number of protocols (3 ± 3 vs 4 ± 3, P = 0.96). The number and availability of protocols were similar in surgical, medical and medico-surgical units.

Among the 509 patients receiving new AT during their ICU stay, the main underlying diseases were immunosuppression (n = 61; 12%), respiratory and cardiovascular comorbidities (n = 62; 12%), cirrhosis (n = 31; 6%) and scored as ultimately (24%) or rapidly (11%) fatal. The mean SOFA score at the time of AT prescription was 6 ± 5. Therapeutic emergencies were reported in 42% (n = 215) of cases, including septic shock (n = 122; 24%), MOF (n = 47; 9%) and hypoxemic pneumonia (n = 1 01; 20%) with high SOFA score (11 ± 6; 13 ± 6; 9 ± 6, respectively). The most frequently suspected sites of infection were lung (n = 241; 47%), bacteremia (n = 121; 24%), and intra-abdominal (n = 105; 21%).

AT was initiated at the time of suspicion of infection in 363 cases (71%), based on the results of direct examination by Gram-stain in 105 cases (21%), on microbiologic cultures (n = 25; 5%) or susceptibility testing (n = 16; 3%) (Figure 1). New AT was decided on day shifts in 227 cases (45%) and out-of-hours in 282 cases (55%). New empiric AT was initiated in 213 (76%) patients out-of hours and in 150 (66%) patients on day shifts (P = 0.03). Treatment was based on the results of Gram-stain direct examination in 49 (17%) patients out-of-hours and in 56 (25%) cases on day shifts (P = 0.055), on microbiologic cultures in 14 (5%) and 11 (5%) patients, and on susceptibility testing in 6 (2%) and 10 (4%) patients, respectively. In most cases, the decision to prescribe AT was made by a single senior physician (n = 397, 78%, involving a senior critical care specialist (n = 340; 67%) or an assistant professor (n = 57; 11%)), and more rarely by the team (n = 87; 17%), or a fellow (n = 25; 5%).

Among the 215 patients with therapeutic emergencies, AT was initiated empirically on suspicion of infection in 152 cases (71%), in 195 (91%) at the time of the Gram-stain, on the results of microbiologic cultures in 206 cases (96%) or susceptibility tests in 214 (99.5%). Among the 121 patients suspected of bacteremia, 86 (71%) of them were treated before Gram-stain examination, 34 (28%) at the time of pathogen identification and 1 (1%) at the time of susceptibility testing. The AT decision-making process is shown in Table 2.

No difference in the severity of the cases (assessed by SAPS II and SOFA scores) was observed according to the timing of prescription, the type of prescriber, or the time to initiation of AT.

When comparing ICUs with written empiric AT protocols and those without protocols, the proportion of empiric AT among all antibiotic prescriptions was similar (33% (305 patients) of the cases per center versus 32% (204 patients), respectively) and severity scores were similar. The number of patients receiving antibiotics in units with written protocols and those without protocols was similar whenever the number of patients (12 ± 6 vs 13 ± 5 patients, P = 0.75) or their proportions (35 ± 19 vs 32 ± 10%, P = 0.56) were considered.

When compared to ICUs without protocols, a higher proportion of prescriptions was made by fellows in ICUs with written protocols (48 (14.7%) vs 12 (6.6%) in other ICUs, respectively, P = 0.01), AT prescriptions were more frequent at the time of suspicion of infection in ICUs with protocols (251 (76.7%) vs 112 (61.5%), respectively; P = 0.003) and prescription was more frequent out-of-hours in the units with a written protocol (192 (59%) vs 90 (49.5%), respectively; P = 0.04).

Overall, empiric ATs were interrupted in 14 patients and modified in 163 patients following Gram-stained direct examination, microbiologic examination and susceptibility testing. Time of stopping and changes in empiric AT is summarized in Table 2.

Overall, in 346 (68%) patients no change of the new AT was made, while 191 changes were observed in 163 (31%) patients: 137 patients (27%) had one AT change, 24 (5%) two changes, and 2 (0.2%) three changes. The timing of these AT changes is presented in Table 2. Among these patients with modified AT, changes were unrelated to microbiologic reasons in 98 (19%) patients but were linked to clinical deterioration n = 21 (4%), to new site(s) of infection n = 14 (3%), to interruption of aminoglycosides n = 36 (7%), to adverse effects n = 6 (1%), or to de-escalation therapy n = 40 (8%).

Among the 215 patients with therapeutic emergencies, changes of AT were reported for the following reasons: 21 (10%) de-escalation, 18 (8%) interruption of aminoglycosides, 14 (6%) clinical deterioration, 4 (2%) new site(s) of infection and 2 (1%) adverse events.

Overall 248 (49%) patients had a low level of microbiologic assessment of infection. Eighty (16%) patients (mean age 55 ± 21) received new AT without any microbiologic sampling of their suspected infection. Among these patients with a mean SAPS II score of 33 ± 15 on ICU admission, 49 (61%) were admitted for a medical diagnosis, 26 (33%) for emergency surgery and 5 (6%) for scheduled surgery. Eight (10%) were immunosuppressed, 6 (7.5%) had comorbidities and 19 (24%) had an ultimately or rapidly fatal underlying disease. Their mean SOFA score was 5 ± 5 and 10 (12.5%) had signs of therapeutic emergencies. Most of these patients were suspected of having pulmonary infection (n = 35, 44%) or intra-abdominal infection (n = 14, 18%).

In the remaining 168 cases, AT was continued with only limited microbiologic confirmation. In 59 (12%) cases, AT was prolonged and based on microbiologic identification without susceptibility testing, while 109 (21%) patients had negative cultures. Among these 59 cases with only organisms identification (SAPS II score on admission of 44 ± 17 and SOFA score of 9 ± 6 at the time of initiation of therapy), therapeutic emergencies were observed in 25 (42%) cases while therapeutic emergencies were reported in 39 (36%) of the 109 cases with negative samples (SAPS II score on admission of 38 ± 17 and SOFA score of 7 ± 6 at the time of initiation of therapy).

Overall, 51 AT changes were made among these 248 patients without susceptibility testing (including clinical deterioration in 16 cases and new site(s) of infection in 6 patients). Among the 80 patients who received a new AT without microbiologic sampling, only 11 (2%) changes were made (clinical deterioration in 4 patients, new site of infection in 2, interruption of aminoglycosides in 3, adverse effects in 2), 17 (29%) changes were made among the 59 cases who had only identification of causative organisms and 23 (21%) among the 109 patients with negative cultures.

Susceptibility testing and assessment of appropriateness of a new AT were obtained in 261 (51%) patients homogenously distributed throughout the centers. Antibiotic therapy was judged inappropriate in 58 patients (22%), involving mainly pneumonia (n = 26; 37.7%), bacteremia (n = 13; 18.8%), urinary tract (n = 14; 20.3%), and intra-abdominal infections (n = 13; 18.8%). Among the 215 cases with therapeutic emergencies, susceptibility testing and assessment of appropriateness was obtained in 126 cases (59%). Antibiotic therapy was considered appropriate in 100 cases (80%).

Patients with appropriate and inappropriate AT had similar SAPS II scores (43 ± 13 vs 42 ± 19) on admission to ICU and SOFA scores (7 ± 6 vs 7 ± 5) on initiation of AT. The clinical features at the time of initiation of AT were assessed in these 261 patients (Table 3). Some organisms initially considered as contaminants (coagulase negative staphylococci) or commensals (enterococci) turned out to be true pathogens. Consequently, the cases were classified at the end the clinical course as inappropriately treated. The reasons for additional antibiotic changes not related to susceptibility testing are shown in Table 3.

The mean duration of ICU stay for the whole cohort was 10.8 ± 9.6 days. A 20% mortality rate (n = 101) was observed among the 509 patients receiving new AT with no significant differences according to gender, type of admission or type of infection (Table 4). No significant link was evidenced between mortality rate and type of institution (18% of death in university teaching hospitals compared to 23% in non-university hospitals (P = 0.17)) or type of ICU (17% of death in surgical ICUs, 19% in medical ICUs and 23% in medico-surgical ICUs (P = 0.35)). No significant link was evidenced between mortality rate and time of prescription, type of prescriber, appropriateness of AT or subsequent changes of treatment. Six the 80 patients (7.5%) who received a new AT without any microbiologic investigation finally died (including 2 of those who had changes in AT), while death was reported in 33 (30%) of the 109 cases with negative samples and 11 (19%) of the 59 patients where only the organism(s) was identified.

Among the 509 cases, only the progress of 27 (5.3%) patients was tracked in the ICU for more than 30 days (6 deaths and 21 survivors). In the three most frequent sites of infection, mortality rates between patients receiving appropriate and inappropriate AT were not significantly different: 24/96 (25%) vs 3/26 (12%), 14/65 (22%) vs 5/15 (33%), 8/46 (17%) vs 3/13 (23%), in pneumonia, bacteremia and intra-abdominal infections, respectively. In contrast, underlying diseases and severity at the time of initiation of AT were associated with a higher mortality rate (Table 4).

Among the 98 patients who had AT changed for non-microbiologic reasons, death was observed in 8/21 (38%) patients who deteriorated clinically, in 2/14 (14%) patients who developed a new site(s) of infection, in 3/36 (8%) of those whose aminoglycosides were stopped and in 3/40 (7.5%) of those who had de-escalation therapy.

Among the 126 patients with therapeutic emergencies in whom appropriateness of AT was assessed, death was reported in 4 (15%) of the 26 patients who had inappropriate AT and 31 (31%) of the 100 patients where AT was appropriate.

Univariate and multivariate analysis assessed predictive factors of mortality in the population of patients receiving new AT (Tables 4 and 5). Hosmer-Lemshow goodness of fit Chi square was 5.06, P = 0.75. Among the identified risks of mortality, the absence of AT protocols was the only criterion not related to underlying disease or severity at the time of initiation of AT.