Empiric broad-spectrum antibiotic therapy of nosocomial pneumonia in the intensive care unit: a prospective observational study

Introduction Antibiotic de-escalation, which consists of the initial institution of empiric broad-spectrum antibiotics followed by antibiotic streamlining driven by microbiological documentation, is thought to provide maximum benefit for the individual patient, while reducing the selection pressure for resistance. Methods To assess a carbapenem-based de-escalating strategy in nosocomial pneumonia (NP), a prospective observational study was conducted in critically ill patients with NP treated empirically with imipenem ± aminoglycoside/glycopeptide in 24 intensive care units of Spanish general hospitals. Overall, 244 patients were assessable (91% with late-onset NP). The primary outcome was therapeutic success 7–9 days post therapy. Results Microbial identification – based on cultures of tracheal aspirates in 82% of patients, cultures of protected specimen brush in 33%, and cultures of bronchoalveolar lavage in 4% – was only available for 131 (54%) patients. Initial antibiotics were inadequate for 23 (9%) patients. Of the remaining patients, antibiotics were streamlined in 56 (23%) patients and remained unchanged in 14 (6%) patients based on microbiology data, in 38 (16%) patients despite microbiology data favouring de-escalation, and in 113 (46%) patients due to unknown aetiology. Overall, de-escalation was implemented in only 23% of patients with potentially multiresistant pathogens, compared with 68% of patients with the remaining pathogens (P < 0.001). Response rates were 53% for patients continuously treated with imipenem-based regimens and 50% for the de-escalated patients. Higher Acute Physiology, Age, and Chronic Health Evaluation II scores were associated with greater mortality, whereas adequate empiric antibiotic therapy protected against fatal outcomes. No increase of superinfection rates caused by emerging pathogens was observed. The costs associated with de-escalation were mainly dependent on the duration of hospitalization. Conclusion This study mainly highlights the current practice of a specific algorithm of de-escalation solely based on the available microbiological data, and highlights the barriers to using it more widely. In this setting, de-escalation was less likely to occur in the presence of potentially multiresistant pathogens. Prior antibiotic administration and the low use of bronchoscopic techniques may have influenced negatively the implementation of de-escalation. Optimization of de-escalation strategies for NP should rely on a correct choice of empiric antibiotics, on appropriate microbiological investigations, and on a balanced interpretation of microbiological and clinical data.

De-escalation, which consists of the initial institution in severely ill patients of broad-spectrum antibiotics covering the most probable causative pathogens followed by antibiotic streamlining driven by microbiological documentation, is thought to provide maximum benefit for the individual patient, while reducing the selection pressure for resistance [11,12]. There are concerns, however, about the emergence of multidrug-resistant pathogens prompted by delayed streamlining of unnecessary antibiotics and the potential negative influence on diagnostic accuracy and economic impact of this strategy. De-escalation was the main cause of antibiotic modification in a series of patients with ventilator-associated pneumonia (VAP) from one institution [13], being more feasible in earlyonset pneumonia and less frequent in the presence of nonfermenting Gram-negative bacilli (NFGNB) [13].
We designed a prospective multicentre observational study -Analysis of Antibiotic De-escalation for Nosocomial Pneumonia (ADANN) -to assess the clinical, microbiological, and pharmacoeconomic features of a carbapenem-based deescalating strategy in intensive care unit (ICU) patients with NP. Our hypothesis was that de-escalation was feasible, yet highly dependent on the quality of microbiological investigations and the patient's clinical characteristics.

Study design
This prospective, observational study was conducted in 24 ICUs of Spanish general hospitals in accordance with local regulations. The protocol was approved by Hospital del Mar's Ethical Committee.
Eligible patients aged ≥18 years had NP, were to receive empirically imipenem-based antibiotic regimens, and were required to have blood cultures and respiratory sampling performed before starting the study therapy. Written informed consent was obtained from the patients. Other diagnostic procedures were performed as needed.

Clinical management and study drug dosing regimen
The only therapeutic, diagnostic, or monitoring interventions allowed were those considered standard clinical practice. Visit 1 corresponded to the onset of study therapy. On visit 2 (3-5 days later), investigators streamlined antibiotics if favourable microbiological data were available. Patients were re-evaluated after discontinuation of therapy (visit 3) and again 7-9 days later (visit 4).
Antibiotics were generally dosed as recommended in the corresponding package inserts. The most common dose of imipenem was 1 g/8 hours daily. Patients with suspected Pseudomonas infection also received amikacin (15 mg/kg/ day) or tobramycin (5 mg/kg/day). If MRSA infection was suspected, vancomycin (1 g/12 hours) or teicoplanin (400 mg/ 12-hour load, 400 mg/day thereafter) were added.

Definitions
NP was defined per the Center for Disease Control and Prevention clinical criteria [14]. The bacteriologic diagnosis required one or more of the following criteria: tracheal aspirate (TA) cultures yielding ≥10 5 colony-forming units/ml; protected specimen brush cultures yielding ≥10 3 colony-forming units/ ml; bronchoalveolar lavage (BAL) cultures yielding ≥10 4 colony-forming units/ml; blood or pleural fluid cultures yielding the same pathogen as respiratory samples; histopathologic evidence of pneumonia; or positive serology/identification of Legionella pneumophila.
The proposed de-escalation approach encompassed two stages. The first stage consisted of administering broad-spectrum empiric antibiotics (imipenem ± aminoglycoside and/or glycopeptide). Empiric therapy was considered adequate if ≥1 antibiotic to which all isolates were susceptible was used after the initial respiratory sampling. Carbapenem-resistant, aminoglycoside-susceptible Pseudomonas infections were always considered inadequately treated, even if they were treated with imipenem in combination with an active aminoglycoside. The second stage involved antibiotic streamlining based on available microbiological data. The investigators were provided with no rules and with no guidance for antibiotic de-escalation. Streamlining was considered adequate if the empiric regimen was changed to piperacillin/tazobactam, or to an anti-pseudomonal cephalosporin if susceptible Pseudomonas aeruginosa was isolated, or to a non-antipseudomonal β-lactam (witholding aminoglycosides and/or glycopeptides if possible) if NFGNB were absent. Three independent reviewers judged the adequacy of the empiric therapy and the correctness of de-escalation.
Clinical success required complete resolution of all attributable symptoms, signs, and radiographic and laboratory abnormalities. Failures included unsatisfactory clinical response, any antibiotic addition or substitution, or death of any cause. Deaths were attributed to NP if they occurred before any objective response to antimicrobials or if NP contributed to death in patients with comorbidities. The judgement of attributable mortality was made by the investigator who followed up the patient at each study site.

Statistics
The primary outcome was therapeutic success 7-9 days post study therapy, which was estimated at ~70% for sample size calculations. Using a significance level of 0.05, a power of 94%, and 10% of patients lost to follow-up, 247 patients were required for the 95% confidence interval of the response rate to fall between 63% and 76%. Preplanned secondary objec-tives included descriptive and comparative effectiveness, toxicity, mortality, superinfections, and cost analyses.
Effectiveness analyses were performed on modified intentionto-treat (MITT) and patient-evaluable (PE) populations. The MITT population only excluded patients misdiagnosed or dying before treatment began. PE patients met MITT criteria, underwent follow-up until visit 4 or death and had sufficient information to determine outcomes, and received study therapy for ≥7 days for cured cases and for ≥3 days for failures.
Chi-square tests were used for effectiveness and mortality contingency analyses. Forward stepwise logistic regression analysis with a cutoff P value of 0.05 was used to determine the relationship between outcome variables and independent variables previously identified in univariate analyses (P < 0.05). For the costs, chi-square and Kruskal-Wallis tests were Study overview: flow chart representing the patient flow throughout the different study visits Study overview: flow chart representing the patient flow throughout the different study visits. Visit 1, onset of study therapy; visit 2 (3-5 days later), investigators streamlined antibiotics if favourable microbiological data were available; visit 3, patients were re-evaluated after discontinuation of therapy, visit 4, patients were re-evaluated 7-9 days later. MITT, modified intention-to-treat; D, dead; LFU, lost to follow-up. used for categorical variables and continuous variables, respectively.

Results
Two hundred and fifty-eight patients with a first episode of NP were enrolled from April 2000 to June 2001. Fourteen patients did not meet the MITT criteria and 31 patients were lost to follow-up ( Figure 1). No selection bias was identified in a com- Group I, patients with an unknown aetiology and unmodified therapy; Group II, patients with resistant organisms, who had unmodified therapy; Group III, patients with susceptible organisms, who had unmodified therapy; Group IV, patients who had susceptible organisms and whose therapy was modified accordingly; and Group V, patients who initially received inadequate antibiotic therapy, which was later modified on the basis of cultures. APACHE, Acute Physiology, Age, and Chronic Health Evaluation. a Chi-square test, P < 0.01. b Chi-square test, P = 0.052. Favourable response rates among MITT patients were 75.4% upon completion of therapy and were 50.4% 7-9 days later, with no significant differences across treatment groups ( Table  3). The PE analysis was supportive of the primary MITT analysis. Increasing baseline Acute Physiology, Age, and Chronic Health Evaluation II scores were associated with a lower likelihood of favourable response (P < 0.01 . Both the overall mortality and the NP-attributed mortality were higher in inadequately treated patients despite modification of initial antibiotics within 72 hours. A logistic regression analysis confirmed the association between inadequate therapy and overall mortality (Table 3).
Susceptible strains were identified in 108 (44.3%) patients, yet antibiotics were only streamlined in 56 patients (Group IV). Group I, patients with an unknown aetiology and unmodified therapy; Group II, patients with resistant organisms, who had unmodified therapy; Group III, patients with susceptible organisms, who had unmodified therapy; Group IV, patients who had susceptible organisms and whose therapy was modified accordingly; and, Group V, patients who initially received inadequate antibiotic therapy, which was later modified on the basis of cultures. The duration of antibiotic therapy was similar in patients who were de-escalated (median, 18 days; range, 4-55 days) versus that in patients who were not de-escalated (median, 16 days; range, 3-65 days; P > 0.05), yet was longer than for patients with unknown aetiology (median, 13 days; range, 4-72 days; P = 0.02). Not surprisingly, imipenem was less used in the former (median, 5 days versus 14 days versus 11 days; P < 0.001). Superinfections were diagnosed in 16 (6.6%) patients, most of whom had prior antibiotic exposure. Bacteriologic testing of TA (12 patients), protected specimen brush (one patient), and BAL (one patient) yielded P. aeruginosa (six Table 3 Effectiveness

Effectiveness response rates
Modified Forward stepwise logistic regression analysis (cut-off P value of 0.05) was used to determine the relationship between mortality and independent baseline variables previously identified in univariate analyses (P < 0.05), including: age, mechanical ventilation, Acute Physiology, Age, and Chronic Health Evaluation (APACHE) II score, treatment group, and adequacy of initial empiric therapy. Group I, patients with an unknown aetiology and unmodified therapy; Group II, patients with resistant organisms, who had unmodified therapy; Group III, patients with susceptible organisms, who had unmodified therapy; Group IV, patients who had susceptible organisms and whose therapy was modified accordingly; and Group V, patients who initially received inadequate antibiotic therapy, which was later modified on the basis of cultures. Table 4 Microbiologic

documentation available in visit 2 in modified intention-to-treat patients with nosocomial pneumonia
Prior antibiotic therapy O v e r a l l P e r c e n t a g e performed Antibiotic-related adverse reactions included seizures (three episodes probably or possibly related to imipenem), thrombocytosis (one episode probably related to cefotaxime), rash (one episode probably related to vancomycin), acute tubular necrosis (one episode possibly related to amikacin), and elevated serum aminotransferase levels (one episode possibly related to imipenem). Adverse reactions were generally mild to moderate, only causing therapy discontinuation in three patients (two seizures, one rash).
A cost analysis focusing on the duration and costs of hospitalization was conducted from Spanish National Health Insurance perspective ( Table 5). The duration and costs of the ICU stay were higher for patients with microbiological diagnosis who were not de-escalated (P < 0.001 and P = 0.001, respectively), while the costs of diagnostics and drugs other than antibiotics were higher for inadequately treated patients and for patients not de-escalated despite favourable microbi-ology (P = 0.04). Although this resulted in significant differences in total cost per NP episode (P = 0.01), there were no differences in the total daily costs between treatment groups, which averaged €326.07 (P = 0.35).

Discussion
In this study, initial empiric antibiotics were streamlined in 51.9% episodes with susceptible microbial isolates in patients who were given appropriate initial therapy. This proportion fell to 42.7% when episodes of unknown aetiology were excluded, to 25.3% when inadequately treated patients were excluded, and to 23.0% when all 244 evaluable patients were considered. These figures are in general agreement with an earlier report [13], where de-escalation was possible in 38% of VAP episodes with susceptible pathogens (34.2% and 31% when resistant strains and unknown aetiology were included, respectively). These data are, however, much higher than the 6.1% found in a previous cohort of patients with VAP [1]. The low de-escalation rate in this latter experience was probably due to the high prevalence of Pseudomonas (48.3%), much greater than that found in our study (30.6% (33/108 patients)) and than that recently reported by Rello and colleagues (17.4%) [13]. The impact of local susceptibility patterns on the de-escalation strategy is further evidenced by our observation that patients with potentially multiresistant pathogens (NFGNB and MRSA) were de-escalated less frequently than patients with other pathogens (23.1% versus 68.1%; P ≤ 0.001). with an unknown aetiology and unmodified therapy; Group II, patients with resistant organisms, who had unmodified therapy; Group III, patients with susceptible organisms, who had unmodified therapy; Group IV, patients who had susceptible organisms and whose therapy was modified accordingly; and Group V, patients who initially received inadequate antibiotic therapy, which was later modified on the basis of cultures.
(page number not for citation purposes) The feasibility of de-escalation may also depend on the patients' clinical characteristics. The lower de-escalation rate in our study compared with that previously reported in VAP patients [13] could be partly due to the intensive prior use of antibiotics (79.1%) in our series and the larger proportion of late-onset episodes in our study (90.6% versus 62.8%). Also, in our study de-escalation was not implemented in 15.6% of all patients with known aetiology. Similarly, Rello and colleagues [13] reported that empirical therapy remained unchanged in 10% of VAP patients due to delayed clinical resolution.
De-escalation was not carried out in 113 patients with negative cultures, which resulted in a prolonged administration of imipenem (median, 11 days). In our view, a wider use of the bronchoscopic techniques, with higher specificity than TA cultures, may have facilitated the proposed de-escalation strategy in our study. Although we believe the high proportion of patients with negative cultures who were not de-escalated was probably influenced by the lack of specific recommendations for de-escalation in this patient group, whether or not deescalation was still possible remains an issue to be defined. In Rello and colleagues' study [13], patients with unknown aetiology were re-evaluated 2-3 days after the onset of therapy; those clinically improved continued to receive short-course antibiotics, while nonresponders were switched to different antibiotics. The authors concluded that de-escalation was not possible in patients with negative cultures since antibiotics were only changed in 30% of the episodes due to clinical deterioration. The results of a more recent prospective study, however, suggest that patients with a clinical suspicion of VAP and culture-negative BAL can have empiric antimicrobial therapy safely discontinued within 72 hours or in some cases withheld altogether [15].
De-escalation is expected to minimize inappropriate initial antibiotic therapies, which result in higher mortality [1][2][3][4][5][6][7][8]. In our study, inadequate initial empiric therapy occurred in 9.4% of MITT patients and was associated with increased mortality (Table 3). This rate was similar to that recently reported in VAP patients [13]. Pathogens identified in this patient group included carbapenem-resistant Acinetobacter spp., MRSA (only in patients not given empiric glycopeptides), M. tuberculosis, and C. albicans. The question remains of how many of these patients could have benefited from an appropriate empiric antibiotic regimen, had they been previously identified as being at high risk for antibacterial-resistant infections.
While it has been shown that extensive carbapenem use may lead to increases in resistance [16,17], in some studies imipenem-based empiric strategies did not lead to higher rates of bacterial and fungal superinfections [18,19]. Yet, it is generally agreed that in order to avoid unnecessary selection pressure on microorganisms resulting from the use of broad-spectrum antibiotics [20][21][22][23], carbapenems should always be used for empiric therapy based on local epidemiology. In order to control potential confusing factors related to the different trends of antibiotic use across participating institutions, the study design only allowed the enrolment of patients empirically treated with imipenem-based regimens, which are common in Spanish institutions based on the local pathogens and antibiotic susceptibilities [24,25]. Interestingly, the incidence of superinfections was low despite the large proportion of patients (67.6%) continuously treated with imipenem until clinical resolution. Nonetheless, no general conclusion can be drawn regarding the impact of a carbapenem-based de-escalation strategy on the emergence of resistant pathogens based on this study, since the proportion of NP episodes empirically treated with imipenem varied significantly between institutions.
The cost analysis suggested that resource use associated with de-escalation for NP was mainly dependent on the costs of diagnostic and therapeutic procedures other than antibiotics and, probably, the duration of hospitalization (probably driven by the severity of disease), but not heavily influenced by antibiotics acquisition costs. These data challenge the assumption that the widespread use of broad-spectrum antibiotics in patients lacking a definite microbiologic diagnosis may escalate costs.
This observational study has several limitations; most notably, the lack of a control group and pre-established de-escalation recommendations, and the study focus being on patients empirically treated with imipenem-based regimens only. The study design was therefore not appropriate to draw any conclusions regarding the benefits of carbapenem-based deescalation strategies. Another major limitation was the large proportion of patients lacking microbiological documentation in the absence of guidance on the duration of empiric antibiotic therapy. Changes of therapy in this patient group were most probably prompted by nonresponse to empiric antibiotics. Finally, there were no standardized criteria for admission, diagnosis, and the initial choice of empiric antibiotics across participating ICUs. The relatively low cure rates observed could be partly due to suboptimal dosing of the different antibiotics used for empiric therapy resulting from the wide pharmacokinetic variations typically occurring in critically ill patients. Further studies are needed to evaluate the magnitude of changes in antibiotic use and clinical outcomes resulting from the implementation of de-escalation strategies.

Conclusion
The findings herein reported were based on a diagnostic strategy mainly supported by TA cultures obtained before the initiation of empiric imipenem-based therapy in a cohort of patients with 90% of late-onset NP episodes. De-escalation was implemented in 51.9% of episodes with susceptible microbial isolates but was not performed if the pathogen remained unknown. The intensive prior use of antibiotics and the infrequent use of invasive bronchoscopic techniques may have contributed to the large number of episodes with unknown aetiology. De-escalation was more likely to occur in the presence of potentially multiresistant pathogens, mainly NFGNB.
The results of this study suggest that optimization of de-escalation strategies for the treatment of critically ill patients with NP should rely on the following principles. First, the choice of empiric antibiotics should be based both on local pathogen prevalence and antibacterial susceptibility and on the identification of patients with selected clinical parameters at high risk of developing infections caused by multi-resistant microorganisms. Good-quality culture sampling is also needed before starting antibacterial therapy. Finally, specific recommendations need be developed for patients with unknown aetiology in order to reduce the duration of empiric therapy.

Key messages
• Inadequate empirical antibiotic therapy contributes substantially to attributed mortality in critically ill patients with NP. Appropriate antibiotic therapy for NP continues to be a major challenge in the ICU setting.
• The antibiotic de-escalation strategy has recently been introduced to increase coverage against the most frequently causative pathogens, including multiresistant strains, and to reduce the selection pressure for antimicrobial resistance.
• In a prospective observational multicentre study, the causative organism was only identified in 54% of patients with NP. De-escalation was used in 23% of cases. In 16% of cases, despite microbiology data favouring de-escalation, it was not applied. Empirical antibiotic treatment was inadequate in 9% of patients.
• Response rates were 53% for patients continuously treated with imipenem-based regimens versus 50% for those patients who were de-escalated. Patients with inappropriate empirical treatment had significantly higher mortality.