Effectiveness of tigecycline-based versus colistin- based therapy for treatment of pneumonia caused by multidrug-resistant Acinetobacter baumanniiin a critical setting: a matched cohort analysis

Patients with MDRAB pneumonia and treated with colistin or tigecycline were enrolled in this study from January 2009 through December 2010. MDRAB was defined as A. baumannii, which showed non-susceptibility to ≥ 1 agent in ≥ 3 antimicrobial categories [16]. Pneumonia was defined based on a prospective surveillance by the microbiology lab and infection controls at NTUH [17]. A patient with a pneumonia must have either a new onset of purulent sputum, a change in the character of sputum, rales or dullness to percussion from a physical examination of the chest, chest radiographic examinations indicating new or progressive infiltrates, consolidation, cavitation, or pleural effusion and MDRAB isolated from the blood, bronchoscopic bronchial brushing, lavage or suitable sputum (defined as > 25 polymorphonuclear neutrophil and < 10 squamous epithelial cells per low power field). We used qualitative culture through the study. Ventilator-associated pneumonia (VAP) is defined as pneumonia developed 48 h or longer after mechanical ventilation is given by means of endotracheal intubation or tracheostomy. Patients are further classified by ventilator status: early-onset VAP (< 5 d of ventilation) or late-onset VAP (≥ 5 d of ventilation).

NTUH restricts the prescription of colistin or tigecycline to infectious disease physicians. In the circumstance of MDRAB infection, the prescriptions are limited to culture proven MDRAB, which showed non-susceptibility to ampicillin/sulbactam and the following anti-pseudomonas antibiotics, included cephalosporins, extended-spectrum penicillins, carbapenems, aminoglycosides, and fluoroquinolones. Since there were no MIC interpretation criteria of tigecycline against A. baumannii, MIC of tigecycline was not routinely tested at our hospital. During the study periods, MIC of colistin was not routinely tested also. Therefore, we consider both colistin and tigecycline as definite treatments of MDRAB pneumonia [2]. The dose of colistin (Colimycin injection, TTY Biopharm Co., Ltd., Taipei, Taiwan) was colistin base activity (CBA) 2.5–5 mg/kg/day in 2–3 divided doses (2 million units or 160 mg colistin methanesulfonate equivalent to 66.8 mg CBA). The dose of tigecycline was a 100 mg loading, followed by 50 mg every 12 hours.

This study excluded patients under the age of 18 years or with incomplete medical records. We also excluded patients with concomitant infection, such as methicillin resistant Staphylococcus aureus or Pseudomonas aeruginosa. When patients had multiple episodes of MDRAB pneumonia and were treated with colistin or tigecycline, only the first episode was included.

This study identified A. baumannii by biochemical methods [18] and determined antimicrobial susceptibility by the disk diffusion method for gentamicin, amikacin, ciprofloxacin, levofloxacin, cefepime, ceftazidime, ticarcillin/clavulanate, meropenem, and ampicillin/sulbactam according to CLSI criteria [19]. We performed MICs of colistin by agar dilution methods according to CLSI criteria [19]. Lastly, MICs of tigecycline were determined by agar dilution methods (BBL, BD Diagnostic Systems, Sparks, MD), with isolates of MIC ≦ 2 μg/mL were considered as being susceptible [20].

We collected demographic data, underlying diseases, and the concomitant infections by a retrospective chart review, recorded regimens of antimicrobial combination, and evaluated the severity of pneumonia by the Acute Physiology and Chronic Health Evaluation (APACHE) II score [21]. The outcomes for evaluation are in-hospital mortality and nephrotoxicity. In patients with normal renal function, nephrotoxicity was defined as a doubling of baseline serum creatinine compared with treatment initiation, or a reduction in the calculated creatinine clearance (CLCr) of ≥ 50%. In patients with a pre-existing renal dysfunction, nephrotoxicity was defined as an increase of > 50% of the baseline creatinine level compared with the value at treatment initiation, a decrease of ≥ 20% of calculated serum CLCr from the baseline, or a decline in renal function that required renal replacement therapy [7, 22]. Combination therapy was defined as at least 3 days of concurrent use of the antimicrobial agent other than colistin or tigecycline.

We calculated the mean and standard deviation (SD) for continuous variables and percentages for categorical variables. We then compared the associations between the clinical presentations of colistin and tigecycline treatments using Student’s t test or Fisher’s exact test. A two-sided p-value ≤ .05 was considered significant.

Since there may be significant baseline characteristic differences between the colistin and tigecycline groups, we used propensity score matching to balance the baseline predictors for colistin or tigecycline use and mortality [23]. Propensity score matched analysis provides quasi-experimental design, which compared the outcome between two similar groups [24]. We estimated the probability of patients receiving colistin rather than tigecycline treatment with a logistic regression model, using their baseline characteristics before enrollment as variables. The fitted model was established by a stepwise variable selection procedure, with the significance levels for stay set to 0.2. Age and gender were forced to be considered in the final model. We conducted all matching on a one-to-one basis without replacement, set the caliper at 0.25, and chose the size of the caliper to be 0.0603.

For the matched cohorts, this study examined post-match balancing by using the paired t test or McNemar’s chi-square test [24]. We then calculated the risk difference (RD) in outcome, which included mortality, and nephrotoxicity, between the colistin group and the tigecycline group, or the average treatment effect of the treated (ATT). The data were analyzed with Stata software, version 12 (StataCorp, College Station, Texas). We used R 2.14.2 software (R Foundation for Statistical Computing, Vienna, Austria) with the Package Matching version 4.8-0 to conduct propensity score matching and ATT estimation.

A total of 294 ICU-admitted patients, who were found to have MDRAB pneumonia and were treated with colistin or tigecycline, were enrolled during the study period. Among the patients, 119 received colistin and 175 received tigecycline. Table 1 presents the demographic characteristics and underlying diseases of the 294 patients with MDRAB pneumonia. There were no significant differences among the different treatment groups for age, APACHE II score, and duration of treatment. However, patients treated with tigecycline had significantly lesser underlying malignancies (1.1% vs. 28.6%, P < .001). The most common combination regimen used was carbapenem combination (34/294, 11.6%).

There were no significant differences of underlying chronic kidney diseases (23.5% vs. 17.7%, P = .24) and baseline creatinine (1.66 vs. 1.81, P = .40) before treatment between the colistin and tigecycline groups. However, patients treated with colistin suffered from a significantly higher rate of nephrotoxicity (10.1% vs. 3.4%, P = .03). There was no significant mortality rate difference between the colistin and tigecycline groups (49.6% vs. 58.9%, P = .12). There was no significant mortality difference between carbapenem with the colistin combination (n = 15) and colistin without carbapenem combination (n = 104) (46.7% vs. 50.0%, P = .81). There was also no significant mortality difference between carbapenem with the tigecycline combination (n = 19) and tigecycline without carbapenem combination (n = 156) (42.1% vs. 60.9%, P = .12).

We performed a multiple variables logistic analysis to determine the significant independent variables associated with the choice of colistin rather than tigecycline (Table 2). The factors associated with using colistin were male, better baseline creatinine, insulin use, and underlying malignancy. The large number of differences among the independent variables in the colistin and tigecycline groups did not permit a definitive assessment of the effect on the outcome. Accordingly, we employed a propensity matching analysis to control for the effect of the numerous confounders.

We matched 84 pairs of patients. There were no suitable tigecycline treatment controls for 35 unmatched colistin treated cases. After matching, the two groups of patients showed no significant differences of potential prognostic factors and severity of illness between them (Table 3). The propensity scores were colistin: 0.370 vs. tigecycline: 0.370, (P = .97); underlying malignancy (2.4% vs. 2.4%, P = 1.00); and insulin use (23.8% vs. 25%, P = .85). 163 cases had positive sputum cultures of A. baumannii, among them 118 sputum samples were from tracheobronchial aspirate. 36 cases had positive bronchoscopic bronchial lavage culture. The mean time (± SD) elapsed between Acinetobacter isolation and the beginning of antibiotics of tigecycline or colistin group is 1.9 (± 1.8) and 2 (± 1.5) days, respectively (P = .72). There were three (one Chryseobacterium indologenes, one Escherichia coli, and one Enterobacter aerogenes) polymicrobial infections in tigecycline and four (one C. indologenes, one E. coli, one Enterobacter cloacae, and one Klebesiella pneumoniae) in colistin group (P = .71). In patient without renal dysfunction, the mean dose of colistin used was CBA 3.0 (±0.8) mg/Kg/day. All the distributions of the estimated propensity scores were well-balanced after matching (Figure 1). We saw a significant mortality difference between the colistin and tigecycline groups (44.1% vs. 60.7%), and the tigecycline group had a significant excess mortality (RD 16.7%, 95% confidence interval (C.I.) 0.9% – 32.4%, P = .04). The excess mortality remained significant even after being adjusted by propensity score, age, gender, and a combination of carbapenem use (adjusted RD 16.4%, 95% C.I. 0.9% – 31.8%, P = .04) (Table 4). The colistin group had a higher rate of nephrotoxicity (9.5% vs. 2.4%, P = .05). The excess nephrotoxicity remained significant even after being adjusted by propensity score (adjusted RD 7.4%, 95% C.I. 1.2% – 13.6%, P = .02).

The MIC₅₀ and MIC₉₀ of colistin were 1 and 2 μg/mL, respectively. All tested isolates were susceptible to colistin. Among patients who were treated with tigecycline, only 22 have isolates available for tigecycline MIC testing. The MIC₅₀ and MIC₉₀ of tigecycline were 2 and 8 μg/mL, respectively. The post-hoc analysis analyzing the mortality showed that the excess mortality of tigecycline is significant among those with MIC >2 mg/mL (10/12 vs. 37/84, P = .01), but not for those with MIC ≦2 mg/mL (4/10 vs. 37/84, P = .81).

We rechecked the result of the propensity score matching by a different matching scenario - the overfitted model - which means all baseline factors before colistin or tigecycline use were considered in the logistic regression model without stepwise selection. The results were similar. There was excess mortality of tigecycline (RD 19.7%, 95% C.I. 2.8% – 36.6%, P = .02), and the colistin group had a significantly higher nephrotoxicity (RD 9.8%, 95% C.I. 1.1% – 18.6%, P = .03).