A total of 143 patients were enrolled in the four clinical trials. Study inclusion criteria were met by 107 patients, with 128 organisms being evaluable. The mean age of the patients ± standard deviation was 68.6 ± 11.7 years. There were 64 males (60%) and 43 females (40%), with a mean weight of 69.4 ± 16.4 kilograms. The mean duration of therapy was 10.7 ± 3.5 days, with a range of 3 to 31 days. Patient baseline underlying disease states and case characteristics are listed in Table . Thirty-six patients were excluded from the evaluation; details concerning them are listed in Table . Serum concentrations were obtained for only ciprofloxacin in the third trial (ciprofloxacin versus imipenem); therefore, for trial 3 only the patients receiving ciprofloxacin were included in this analysis. Of the 128 organisms obtained from 107 patients, a single pathogen was evaluated for 90 patients, two pathogens were evaluated for 13 patients, and three pathogens were evaluated for 4 patients. Table lists all organisms. Baseline MICs for selected pathogens are shown. Resistance developed in 32 (25%) of the evaluable cases. Three additional organisms, *K. pneumonia* and two strains of *E. cloacae*, were isolated during therapy or during the follow-up period, and were found to be resistant. These organisms were not isolated at baseline; therefore, they were not included in the analysis.

| **TABLE 1**Baseline characteristics by clinicaltrial |

| **TABLE 2**Patients excluded from theanalysis |

| **TABLE 3**Median MICs for study organisms at baseline and at the end point(resistance) |

The rates of resistance development as determined for organism groupings are listed in Table . The greatest frequency of selected resistance was observed for *Pseudomonas* (group 1 organisms) (46.1%), followed by group 2 organisms (27%) and group 3 organisms (10%). Resistance was not observed in the diverse group of remaining organisms (group 4). When resistance was evaluated by treatment, the greatest rate of resistance was observed for cefmenoxime (42.9%), followed by ciprofloxacin (27.6%), ceftazidime (20%), and the ceftazidime-tobramycin combination (9.1%). Resistance was not observed in the ciprofloxacin-piperacillin combination treatment arm. For monotherapy the rate of selected resistance development was 30.7% (31 of 101 cases), while resistance developed in only 3.7% (1 of 27) of the cases of combination therapy. The rates of resistance development by treatment groups, monotherapy and combination therapy, and by organisms are listed in Table .

| **TABLE 4**Selected resistance rates by organism grouping andtreatment |

An initial univariate screen of patient, organism, and antimicrobial factors and their relationship to the development of resistance revealed that age, sex, weight, ventilator status, surgery, chronic obstructive pulmonary disease (COPD), diabetes mellitus, steroid use, malignancy, chemotherapy and/or radiation therapy, ciprofloxacin therapy, ceftazidime therapy, and the presence of group 2 gram-negative rods were not significant as predictors of resistance. However, several factors, including the AUC_{0–24}/MIC ratio, the presence of group 1 or group 3 gram-negative rods, cefmenoxime therapy, group 2 organisms treated with β-lactam monotherapy, and previous antimicrobial therapy, were determined to be significant by univariate analysis (Table ).

| **TABLE 5**Univariate analysis results of resistance development by patient risk factors andcharacteristics |

The median AUC_{0–24}/MIC values by organism and treatment groupings are presented in Table . The designation of susceptible or resistant indicates the antimicrobial exposure and applies to those organisms that either remained susceptible or became resistant, respectively, during treatment. Although the parameter AUC_{0–24}/MIC has great variability, there is an apparent trend of emergent resistance at lower levels of antimicrobial exposure.

| **TABLE 6**Summary statistics for AUC_{0–24}/MIC ratios by organism groups and by treatmentgroups |

CART analysis identified four factors as significant: AUC_{0–24}/MIC ratio, cefmenoxime treatment, and organisms of group 1 and group 2. Treatment and organism interactions, such as *Pseudomonas* treated with cefmenoxime and β-lactamase-producing organisms treated with cefmenoxime and/or ceftazidime, were scrutinized for possible significance. Further inspection of these findings and analysis with pharmacodynamic models revealed that *Pseudomonas* spp. and all other organisms, with the exception of β-lactamase-producing organisms (group 2 organisms) treated with β-lactam monotherapy (cefmenoxime or ceftazidime), exhibited an inverse relationship between the probability of developing resistance and the AUC_{0–24}/MIC ratio.

The final pharmacodynamic model describing this relationship is represented by the equation %*P* = [*P*_{0} − (*P*_{0} − *P*_{∞}) · AUIC^{H}/(AUIC_{m}^{H} + AUIC^{H})] · (1 − *R*_{2}) + *P*_{2} · *R*_{2}, where *P*_{0} is the asymptotic maximum percent probability of resistance as the AUC_{0–24}/MIC ratio goes to 0; *P*_{∞} is the asymptotic minimum percent probability of resistance as the AUC_{0–24}/MIC ratio goes to infinity; *R*_{2} is an indicator, either 0 or 1, of group 2 organisms (β-lactamase-producing organisms treated with β-lactam monotherapy); *P*_{2} is the percent resistance for cases when *R*_{2} is 1; AUIC_{m} is the AUC_{0–24}/MIC ratio at which %*P* = 0.5 · (*P*_{0} + *P*_{∞}); and *H* is Hill’s constant, which reflects the degree of sigmoidicity. Parameters fitted by this model were %*P*_{0} = 82.6%, %*P*_{min} = 9.2%, %*P*_{2} = 64.3%, AUC_{0–24}/MIC (AUIC_{m}) = 100, and *H* = 40 (fixed). A log-linear regression approach with weighting was utilized to determine the line of best fit for the group 2 organisms treated with β-lactam therapy. The equation describing the line is *y* = 1.409 − 0.2548 · log_{10} AUC_{0–24}/MIC. The final model fitted the data extremely well. The results of the model goodness-of-fit analysis are presented in Table . As described by this model, an inverse-effect relationship, which applies to all organisms and treatments with the exception of β-lactam monotherapy for group 2 organisms (consistent with type I β-lactamase-producing gram-negative organisms), exists. The observed and modelled data are graphically presented in Fig. . As shown in the figure the solid line represents the modelled response surface for all cases, classified as either susceptible or resistant, within the data set for which the AUC_{0–24}/MIC inverse-effect relationship applied. The dashed line represents the line of best fit for the group 2 organisms treated with β-lactam monotherapy. Observed cases identified by symbols and numbers are plotted as single points within an AUC_{0–24}/MIC ratio category at the median value for that category.

| **TABLE 7**Pharmacodynamic model goodness offit |

For those cases in which the model fits applied, *Pseudomonas aeruginosa* treated with ciprofloxacin represented a large number of the fitted cases. The observed percent resistance for ciprofloxacin monotherapy for pseudomonas organisms was 66.7%. This extremely high incidence of resistance was associated with the level of antimicrobial exposure. For example, the observed percent resistance for pseudomonas organisms treated with ciprofloxacin monotherapy, determined by utilizing the fitted AUC_{0–24}/MIC breakpoint, was 100% (10 of 10 cases) when the AUC_{0–24}/MIC ratio was <100 and 25% (2 of 8 cases when it was ≥100). In the case of group 3 organisms (other gram-negative rods), including all treatments, the observed percent resistance was 100% (2 of 2 cases) below the breakpoint, and 5.3% (2 of 38 cases) above the breakpoint. The observed data points of β-lactam monotherapy for group 2 organisms are clearly not reflective of the modelled response surface. The high percentage of resistance (>60%) within this group could not be explained by the pharmacodynamic measure of antimicrobial exposure, AUC_{0–24}/MIC, as resistance occurred throughout a large range of AUC_{0–24}/MIC ratios (217 to 14,190).

The median time to the observation of selected bacterial resistance was 6 days. The median time to resistance for all cases below the AUC_{0–24}/MIC ratio breakpoint was 7 days versus 6 days for those cases above the breakpoint. This difference was not significant, irrespective of treatment or organism. Figure is a Kaplan-Meier plot of the probability of remaining susceptible over time, from the initiation of therapy. The three curves represent three distinct groups: (i) the cases fit by the model below the AUC_{0–24}/MIC breakpoint of 100 (*n* = 17); (ii) the cases consistent with type I β-lactamase-producing organisms treated with β-lactam monotherapy, which did not exhibit an AUC_{0–24}/MIC relationship (*n* = 14); and (iii) the cases fit by the model above the AUC_{0–24}/MIC breakpoint of 100 (*n* = 97). For the organisms represented by curve i (*n* = 17), the times to selection of 25, 50, and 75% resistance occurred by days 5, 10, and 14, respectively. For the gram-negative rods represented by curve ii (*n* = 14), the time to selection of 25% resistance occurred by day 5, and the time to selection of 50% or greater resistance occurred by day 16. For all other organisms, the cumulative rate of resistance development was approximately 9% and remained relatively consistent over time, with all cases of resistance occurring by day 13. A statistically significant difference was noted between groups 1 and 3 (*P* < 0.001) and between groups 1 and 2 (*P* < 0.001). Groups 2 and 3 did not differ (*P* = 0.322). However, the numbers of cases in these two groups are small and a type II error may exist. For these two groups, which exhibit different relationships of antimicrobial exposure to response, the rates of selection of bacterial resistance are similar.