Suppl A 08 Forrest

Journal of Antimicrobial Chemotherapy (1997) 40, Suppl. A, 45–57

JAC

Pharmacokinetics and pharmacodynamics of oral grepafloxacin in patients with acute bacterial exacerbations of chronic bronchitis Alan Forrest, Sanford Chodosh, Michael A. Amantea, David A. Collins and Jerome J. Schentag* SUNY at Buffalo Clinical Pharmacokinetics Laboratory, Millard Fillmore Health System, Buffalo, NY; Veterans Administration Outpatient Clinic, Boston, MA, USA This analysis was designed to characterize the population pharmacokinetics and pharmacodynamics of oral grepafloxacin (OPC-17,116) in patients with acute bacterial exacerbations of chronic bronchitis (ABECB). The study group included 76 patients (43 male, 33 female) between 23 and 81 years of age, who were part of a multicentre, randomized, double-blind, dose–response study. Patients were randomly assigned to receive oral regimens of grepafloxacin, 200, 400 or 600 mg, each administered once daily for 14 days. Plasma samples for drug assay (typically eight per subject; four samples on either day 3, 4 or 5, plus troughs on other clinic visit days), were obtained during treatment. Population pharmacokinetic analysis was accomplished using iterative two-stage analysis. Cultures and quantitative Gram stains from serial 24 h collections of sputum were used to determine the time (in days) taken to eradicate each bacterial strain. Population pharmacodynamic analysis was performed for three measures of antibacterial response: probability of bacteriological cure, probability of clinical cure, and time to eradication. Grepafloxacin plasma concentration profiles were best fitted by a pharmacokinetic model with first-order absorption following a lag time between administration of the dose and onset of systemic absorption. All three measures of response were strongly related to the 24 h AUIC (AUC/MIC). At an AUIC of 175, it was 98% (P < 0.01). In conclusion, antibacterial response for grepafloxacin in ABECB patients was highly related to AUIC; values of 175 were optimal.

killing in these patients. Bacterial eradication was significantly faster when the AUIC was 250 than when it was between 125 and 250.1,2 For patients with nosocomial LRTI having an AUIC below the threshold of 125, the microbiological failure rate approached 70%.1,6 Factors that alter the AUIC in individual patients can explain different outcomes in clinical trial patients. In the early phases of clinical antibiotic development, studies of the AUIC of patient populations given a range of doses can assist in the determination of an optimal dosage for patients with known infecting pathogens. Studies in LRTI patients demonstrate that variability in the clinical and microbiological response can be the consequence of pronounced inter-patient differences in pharmacokinetics. The primary determinant of variability in pharmaco-

Introduction Achievement of an optimal relationship between the pharmacokinetics of fluoroquinolone antibiotics and the MIC of the infecting organism in the same patient, is predictive of eradication of the invading organism.1–6 Studies have advanced these principles in patients with nosocomial lower respiratory tract infections (LRTI). In a study of 73 LRTI patients given intravenous ciprofloxacin, the optimal area under the curve (AUC) to MIC ratio (AUC/MIC, or AUIC) was 125, which approximates 80% of the AUC above the MIC at dosing intervals of every 8–12 h.1,7 When the AUIC was 125, microbiological and clinical cure was achieved in significantly more of these patients. Ciprofloxacin displayed concentration-dependent bacterial

*Corresponding author. The Clinical Pharmacokinetics Laboratory, Millard Fillmore Health System, 3 Gates Circle, Buffalo, NY 14209, USA.

45 © 1997 The British Society for Antimicrobial Chemotherapy

A. Forrest et al. kinetics is typically renal function as a determinant of AUC and MIC, as it varies between organisms, even within the same bacterial species. The AUIC-versus-outcome principles developed for ciprofloxacin should apply to the new fluoroquinolone antibiotic grepafloxacin (OPC-17,116). However, there are some factors unique to this compound that should be studied. Unlike most other antibiotics of this class, grepafloxacin is not excreted renally to any significant degree. Thus inter- and intra-patient variability in AUC may be less than or more than that for antibiotics like ofloxacin, which is excreted almost entirely by the renal route,8 or ciprofloxacin, which is excreted by both renal and hepatic pathways.8 Grepafloxacin is not being developed as a parenteral formulation, thus the antibiotic will have limited use in the nosocomial LRTI population. Its lack of activity against Pseudomonas spp. makes it more suited to the treatment of out-patient respiratory tract infections. We tested the AUIC principles of this new fluoroquinolone in patients with acute bacterial exacerbations of chronic bronchitis (ABECB), a condition previously shown to be a highly suitable model for studies of AUIC and bacterial eradication.9,10 In these patients, serial sputum specimens are readily obtainable, bacterial pathogens are present in most patients and the patients are sufficiently ambulatory to allow frequent blood sampling to determine their pharmacokinetic parameters. There has been little study of serial cultures and frequent blood samples in this population, so the impact of the disease on either pharmacokinetics or pharmacodynamics was unknown. The opinion that cultures are of little value in ABECB9,11 did not discourage this exercise; rather, this study seemed a good opportunity to determine if there was a relationship between organism eradication and the outcome of ABECB.

Patients were excluded if there was a history of allergy to either -lactams or fluoroquinolones, if they required hospitalization or parenteral therapy, or if the chest radiograph indicated the presence of a pneumonia. Patients were also excluded if they had evidence of gastrointestinal disease or were taking either oral antacids or sucralfate, which could interfere with the absorption of oral medications8 or if they had evidence of hepatic disease, renal insufficiency (serum creatinine 132.6 mol/L (1.5 mg/dL)), or they were receiving warfarin or fenbufen.

Study design and sampling procedures In all three protocols, patients were randomized to receive once-daily oral doses of either 200, 400 or 600 mg grepafloxacin for 14 days. Study drug formulations were grepafloxacin 200 mg or placebo tablets. Patients were treated on an out-patient basis for 14 days with scheduled clinic visits on day 1; day 3, 4, or 5; day 8; day 10, 11 or 12; and day 14. On the days on which a clinic visit was scheduled, the patient was instructed to withhold the dose for that day and to bring all remaining study drug to clinic. The dose for that day was taken, under observation, at the prescribed time. On clinic days there was an assessment of the nature and quality of bronchopulmonary signs and symptoms, and vital signs. Blood was collected for haematology and serum chemistries and determination of trough grepafloxacin concentrations. In any patients taking concomitant theophylline, a theophylline plasma concentration was also determined. To describe the pharmacokinetics of grepafloxacin, it was necessary to obtain plasma samples at appropriate time points. In addition to the plasma trough samples obtained at clinic visits, four serum samples were obtained on either day 3, 4 or 5. These were obtained at pre-dose, 1 and 2 h after grepafloxacin administration, and a sample at least 1 h after the third sample but within 12 h after the dose. In total, each patient typically had eight samples obtained at 0, 72, 73, 74, 77, 168, 240 and 336 h. The times for these samples were based on optimal sampling theory12–14 applied to previous data obtained in normal volunteers.15,16 Sputum was collected by the patients over the 24 h before their clinic visits. Volume was measured, and an aliquot was analysed for cell numbers, cell types, and culture to identify organisms.17,18 If a respiratory pathogen or predominant organism was isolated from a sample, bacterial susceptibility to grepafloxacin, as an MIC, was determined. From daily measurements of these specimens, we determined the time taken to eradicate bacteria from pulmonary excretions, bacterial response, inflammatory response and clinical response. Additional measures of disease resolution were made using the sputum neutrophil levels, and clinical data including frequent evaluations of bronchopulmonary symptoms.

Patients and methods Patient selection and randomization Data were obtained during three prospective clinical trials of grepafloxacin for ABECB. Frequent sputum and blood samples were taken from all of these patients, to provide the data required for analysing the relationship between grepafloxacin pharmacokinetics and pharmacodynamics. Patients were older than 18 years, with a diagnosis of chronic bronchial disease, including but not limited to chronic bronchitis, chronic bronchial asthma, asthmatic bronchitis or bronchiectasis. Superimposed on this underlying disease, each enrolled patient had acute bacterial bronchial infection characterized by all of the following: (i) an increase in bronchopulmonary symptoms, (ii) an increase in inflammatory neutrophils in sputum and (iii) a sputum Gram stain analysis indicating the absence of contaminating oropharyngeal squamous cells and significant numbers of morphologically distinct bacteria. 46

Grepafloxacin pharmacokinetics intrinsic clearance (Cli) and the Michaelis constant (KM). Because the doses were oral, the fitted volumes and clearances are dependent on F, which could not be estimated in a study of this design. Several other pharmacokinetic parameter values can be derived from the fitted parameters. For example, the oral volume of distribution at steady-state, Vss/F Vc/F Vp/F; Cld Vc kcp Vp kpc (kcp and kpc are inter-compartmental transfer rate constants); and Vmax Cli KM. This model was based on earlier single- and multiple-dose pharmacokinetic studies, in normal young adults.15,16 However, other models (e.g., one with linear clearance), were also considered.

Drug assay Plasma samples were analysed for grepafloxacin by HPLC. In this method, grepafloxacin and the added internal standard, OPC-17,203, were extracted from plasma using a liquid–liquid extraction. This extract was then subjected to reversed phase HPLC, on a 5 m ODS column. Grepafloxacin and OPC-17,203 in the effluent were quantified using fluorescence detection. System calibration was accomplished for each batch of samples, by linear regression of the ratio of the peak height of grepafloxacin, to that of the added internal standard. Using 0.5 mL of plasma, the lower limit of quantification for grepafloxacin was 4.66 ng/mL (on the low curve) and 93.2 ng/mL (on the high curve). The method was linear up to at least 93.2 ng/mL and 2796 ng/mL on the low and high curves, respectively. As an independent estimate of accuracy and precision, blinded quality control samples were assayed in each batch, together with calibration and study samples.

Identification of factors predictive of grepafloxacin pharmacokinetics An initial statistical screen of the effects of age, race, gender, weight, smoking history, alcohol usage, laboratory values, pulmonary diagnosis, dose, clinical study site, concomitant medications (e.g., theophylline) and active medical problems, upon the pharmacokinetics of grepafloxacin, was performed by classification and regression tree (CART) analysis, a procedure which uses recursive partitioning, multiple stepwise linear regression20,21 and the Kruskal–Wallis non-parametric one-way analysis of variance.20 Those factors significantly associated with pharmacokinetic parameter values were incorporated into the population pharmacokinetic model and were tested by Akaike’s information criterion. Differences in variance were evaluated using the variance ratio (the F-test). Graphical depictions of differences between groups were accomplished using notch plots.22 Referring to Figure 5 as an example of a notch plot, the horizontal line, at the narrowest portion of the plot, is the median of the data (the 50th percentile); the notch (the indented region) indicates the 95% confidence interval about the ‘true’ value of the median; the vertical dimensions of the box delimit the inter-quartile range (the 25th–75th percentiles), i.e., the upper and lower horizontal lines are the 25th and 75th percentiles, respectively; the bars extending out from the box show the approximate 99% range for the observed data; outliers (determined by the graphing module) are indicated as an asterisk; distant outliers are indicated as an open circle. The outliers were not excluded from any of the analyses.

Pharmacometric modelling methods A population pharmacokinetic model, which consists of a vector of mean parameter values, a p p lower triangular covariance matrix (p is the number of structural parameter values), and a model for the residual variance of the observations (i.e. plasma concentrations), was fitted to the plasma concentration–time data. All grepafloxacin data in each subject were co-modelled. This included the trough plasma samples obtained from all clinic visits and the four plasma samples obtained on clinic days 3, 4 or 5. The individual and population parameter values were determined by the iterative two-stage analysis (ref. 14 and G. Prévost, personal communication). Modules developed within the ADAPT II12,13 package were used to implement the iterative two-stage program. In addition, individual estimates for each subject (pharmacokinetic point estimates and asymptotic covariance matrices), were also generated during this process. Model discrimination was accomplished using the rule of parsimony and Akaike’s information criterion19.

Final structural model The structural model employed is depicted in Figure 1. Drug is administered into an absorptive compartment, X(3) is the amount of drug in this compartment; F is the systemic bioavailability and, after a lag time (Tlag), grepafloxacin is absorbed (according to a first order rate constant, ka) into a central compartment of apparent volume Vc; in this model, X(1)/Vc is the plasma concentration. Drug in the central compartment equilibrates (Cld is the distributional clearance) with drug in the peripheral compartment of apparent volume Vp; in the model, X(2) is the amount of drug in this compartment, and it is eliminated from the central compartment by a saturable process, defined as the

Figure 1. Pharmacokinetic model for grepafloxacin.

47

A. Forrest et al. used as potential covariates of response, were computed as:

Modelling the pharmacodynamics of response Three measures of antimicrobial response were modelled: probability of a bacteriological cure, probability of a clinical cure, and time (in days) taken to eradicate bacteria from pulmonary secretions. Probability of cure was initially modelled using both multifactorial and stepwise logistic regression23 and CART. Then certain relationships were modelled using a Hill-type pharmacodynamic model.1,3 The time taken to eradicate bacteria was modelled using proportional hazards regression.24 The independent variables evaluated were the same for all three measures of antibacterial response. These included the bacterial species and MIC, and the patients’ age, pulmonary diagnosis (e.g., chronic bronchitis, chronic bronchial asthma, bronchiectasis, etc.), other active medical problems, concomitant drugs, laboratory values and the clinical site at which they were treated. The conventionally reported MIC is biased upwards; for example, a reported MIC of 1.0 mg/L usually means that there was growth at 0.5 mg/L and none at 1.0 mg/L; thus, the ‘true’ MIC is between these two values. For this reason, the MIC used in the statistical modelling was transformed to the ‘MIC midpoint’ (MICmp). This was computed as the midpoint between the reported MIC and the next lower value in the dilution series. Thus, for an MIC of 1.0, the MICmp was 0.75, the midpoint between 0.5 and 1.0 mg/L. It is easy to inter-convert MICmp and MIC values: MICmp MIC 3/4. The independent variables also included the following measures of drug exposure: dose (as mg/day), the 24 h area under the plasma concentration–time curve (AUC), the AUICmp (AUC/MICmp, AUIC 0.75 AUICmp), the peak plasma concentration achieved on day 5 (peak), the ratio of peak/MICmp, the trough following the fifth dose (trough), the trough/MICmp and the percent of time that plasma concentrations were above the MICmp (during the fifth dose interval). The plasma concentration profile for the fifth dose interval (approximately 96–120 h) was intensively reconstructed through simulation as demonstrated in Figure 2. The ‘near steady-state’ peak, trough and percent of time above the MIC were based on these data. The 24 h AUC, for each bacterial isolate, was determined by numerically integrating the plasma concentration profiles, from the beginning of treatment until DTE (the number of days of treatment until eradication of that strain from pulmonary secretions, or until treatment was ended (whichever occurred first), and then dividing that value by the number of days elapsed. Thus, the 24 h AUC, averaged over that period (0 to DTE), was obtained. This computation of AUC and AUIC is depicted graphically in Figure 2. A pharmacokinetic model (smooth curve) was fitted to the observations (closed circles) and the AUC0 DTE (shaded area) was obtained by numeric integration. The 24 h AUC and the AUICmp, which were

AUCR

AUC0 DTE DTE

and AUICmp

AUC0 DTE DTE MICmp

To perform logistic regression for a continuous potential covariate (such as AUIC), the range of values was divided into cells (using a geometric progression for endpoints), with reasonable numbers of cases in each. Then, in the course of analysis, adjacent cells that did not differ significantly (in probability of cure, for example), were sequentially collapsed, until only groups that were different remained. To perform CART, the continuous independent variables did not need to be empirically divided into cells. The CART procedure determined the optimal breakpoints for the continuous independent variables. The minimum number of cases for each side of a split (breakpoint that defines the cells), was constrained to four. Although the two methods are largely redundant, to determine the parameters which were significantly associated with likelihood of cure and the optimal breakpoints, CART and logistic regression were performed in parallel. Performing both, and investigating reasons for any differences, improved the probability of deriving the optimal model. Both logistic regression and CART are relatively nonparametric methods, which we use mainly as screening techniques; during the screening process, care is taken to minimize the imposition of a rigid model on the relationship between covariates and responses. After the significant covariates had been identified, a parametric model, a Hill-type equation, was considered. For a single

Figure 2. Determination of AUIC. An iterative two-stage program was used to fit a pharmacokinetic model (solid curve) to the observed plasma concentrations (closed circles); AUC0 DTE (shaded region) was determined by numeric integration; in this example, DTE is 3.5 days.

48

Grepafloxacin pharmacokinetics continuous, independent variable, the Hill-type model was of the following form: (Pmax P0) XH percent probability of cure P 0, XH XH 50%

Bias and precision of the fitted model The pharmacokinetic model and final parameter values fi t the data very well. This can be seen in Figure 3, which depicts the observed plasma concentrations (y-axis) plotted against fitted values (x-axis) for all of the observations in the 76 patients. By weighted, linear regression (a maximum likelihood objective function), the line of best fit through these data did not differ from the line of identity (diagonal). The fitted parameter values, for the residual variance model, were 0.11 and 5.0 mg/mL, for u1 and u2, respectively. This residual variance model suggests that, at a plasma concentration of 20 ng/mL, the standard deviation was approximately 7.3 ng/mL, and at 1000 ng/mL, was 117 ng/mL.

in which Pmax is the asymptotic maximum response rate, P0 is the asymptotic baseline response rate (the probability of cure as X approaches 0; a value that is not necessarily 0, because a proportion of patients diagnosed as ABECB would recover without drug treatment). X is the measure of drug exposure (AUIC, for example) and X50% is the drug exposure at which percent probability of cure is (P0 Pmax)/2; H is Hill’s constant (a variable that reflects degree of sigmoidicity). This model was implemented by coding each cure as response 1, and each failure as response 0 and then fitting the above model, using a maximum likelihood objective function.12,13

Fitted pharmacokinetic parameter values Table I summarizes the fitted pharmacokinetic parameter values for all 76 patients. The AUCavg was computed by simulating the plasma concentration profiles that would be achieved by giving 600 mg/65 kg/day for 7 days. The AUCavg was obtained by numerically integrating the AUC(0 7 day) and dividing by 7, thus yielding the 24 h AUC, averaged over days 1–7. With increases in assigned dose, the increase in actual (fitted) AUC values was disproportionately high (data not shown). This phenomenon was probably the consequence of saturable elimination, not differences between groups, in clearance and/or bioavailability. The notch plot in Figure 4 illustrates the lack of difference in pharmacokinetics between the three dosing groups. For patients that had been randomized to receive 200, 400 and 600 mg per day, this notch plot shows the equivalence of distributions of AUCavg. The figure demonstrates that, if all subjects were

Integration of pharmacokinetics and pharmacodynamics to determine optimal dosage regimens For all 76 patients, the fitted individual pharmacokinetic models were used to predict the plasma concentration profiles (including 24 h AUC values) that would have been predicted, for oral grepafloxacin regimens of 300, 400 and 600 mg/65 kg per day. The AUCs were also converted into the resulting 24 h AUIC values, that would have been achieved if the MIC were 0.125, 0.25 and 0.5 mg/mL. Thus each of the 76 patients had nine AUIC values predicted (three MICs and three regimens). The pharmacodynamic analyses provided models relating drug exposure and other covariates to probability of clinical and bacteriological cure and to time to eradication of bacteria from sputum. These models were used to predict the number of patients that would have been cured on each regimen and at each MIC value and the number of days of therapy that sputum would have remained culturepositive. These data enabled a recommendation of the optimal dose and duration of therapy, for treatment of ABECB, in this target population.

Results Demographics and distribution of assigned doses Seventy-six patients (43 males, 33 females) were randomized to once-daily, oral grepafloxacin dosage regimens (27 patients were given 200 mg, 24 patients given 400 mg and 25 patients given 600 mg). The median (range) weight was 74 (41–129) kg and age was 63 (23–81) years old. All patients except one were Caucasian. A total of 617 grepafloxacin plasma concentrations were obtained in these 76 patients; the median (and mean) was eight, with a range of two to 12 samples per subject.

Figure 3. An assessment of the precision of the fitted model by plotting observed and fitted grepafloxacin plasma concentrations for 76 subjects (586 data points). The line of identity (diagonal) did not differ from the line of best fit (y 0.997x, r2 0.986; computed by weighted, linear, least-squares regression).

49

A. Forrest et al. Table I. Pharmacokinetic parameters of grepafloxacin in patients with ABECB Population ABECB all (CV, %), 23–81 years ABECB 49 years ABECB 50–59 years ABECB 60–69 years ABECB 70 years

n

Ka (h–1)

Vss/F (L/65 kg)

Cl/F (L/h/65 kg)

76 19 12 24 21

0.53 (44) 0.55 (39) 0.56 (36) 0.52 (45) 0.52 (55)

496 (62) 495 (50) 510 (48) 483 (71) 504 (71)

36.2 (37) 39 (34) 40 (44) 34 (35) 35 (37)

AUC (mg.h/L)a 39 (68) 35 (75) 40 (74) 41 (67) 40 (63)

a

24 h AUC adjusted for 600 mg per 24 h dosing.

Figure 4. Notch plot of AUCavg (the computed 7 day average, on 600 mg/65 kg/day) for patients randomized to receive 200, 400 or 600 mg/day; median values did not differ significantly. Asterisks indicate ‘outliers’.

given the same weight-normalized dose, the median AUC values would have been similar between groups. Table I also summarizes the pharmacokinetic parameter values by age group: it includes a summary of pharmacokinetic parameter values in the 19 patients below 50 years of age, the 12 patients 50–59 years old, the 24 patients 60–69 years old and the 21 patients at least 70 years of age. None of these groups differed in concentrations achieved (or in AUCavg). In fact, neither age, gender, dose, nor study site (race was not assessed because 75/76 patients were Caucasian) were significant predictors of the pharmacokinetics of grepafloxacin. These trends are illustrated in Figure 5, which is a notch plot that shows the dispersion of computed values for AUCavg, for six study groups, given 600 mg/65 kg/day. Study group 1 comprised 20 young adults,15,16 ; group 2, 17 elderly volunteers;23 group 3, 19 ABECB patients less than 50 years old; group 4, 12 ABECB patients between 50 and 59 years old; group 5, 24 ABECB patients between 60 and 69 years old; group 6, 21 ABECB patients at least 70 years of age. There were no differences, between groups, in

Figure 5. Notch plot of AUCavg in six groups of subjects. Study group 1 is 20 normal, young adults; group 2 is 17 normal, elderly adults; group 3 is 19 ABECB patients younger than 50 years; group 4 is 12 ABECB patients 50–59 years; group 5 is 24 ABECB patients 60–69 years; group 6 is 21 ABECB patients at least 70 years of age. Median values for AUCavg did not differ; ABECB patients were more variable. Outliers and distant outliers are indicated by an asterisk and open circle, respectively.

median values of AUCavg; the inter-subject variance was greater in the ABECB patients.

Pharmacodynamic response Table II presents a summary of the exposure covariates developed out of the integration of the pharmacokinetics with the MIC. For peak and trough, the number of cases (n) is equal to the number of patients (76). These values were obtained from the ‘intensively reconstructed’ (simulated) fifth dose interval. For AUC (AUC0 DTE/DTE) and the measures related to MICmp, n is equal to the number of bacterial strains isolated (119). As can be appreciated from the data in Tables I and II, there was substantive inter-patient variance in AUC (e.g., coefficient of variation in percent (CV%) 92), peak (CV% 65) and trough (CV% 86) concentrations and 50

Grepafloxacin pharmacokinetics Table II. Summary of selected exposure covariates

AUCR (mg.h/L) AUICmp Peak (mg/L) Peak/MICmp Trough (mg/L) Trough/MICmp MICmp (mg/L)

n

Mean

CV%

119 119 76 119 76 119 119

15.4 762 1.60 96.5 0.357 20.7 0.211

92 233 65 291 86 258 226

Median

Range

10.0 242 1.36 24.2 0.255 5.10 0.0450

2.59–69.8 6.46–14765 0.398–4.54 0.705–2589 0.003–1.27 0.0447–430 0.001–3.0

Table III. Results of univariate analyses: relationship of % bacteriological and clinical cure (n in parentheses) to study variables Parameter Dose (mean, CV%) 200 mg once daily 400 mg once daily 600 mg once daily MIC 0.0–0.1 0.11–0.2 0.21–0.4 0.41–0.8 1.4–3.0 % Time above MIC 10% 10 100% 100% Peak/MICmp 0–2 2.1–4 4.1–8 8.1–16 16.1–32 32.1–2589

n

% Bacteriological cure

27 24 25

even larger variance in MICmp (CV% 226) and the measures that are related to MICmp (AUICmp, peak/MICmp and trough/MICmp). In addition, for six of 119 bacterial isolates, the plasma concentrations were above the MICmp for less than 10% of the day; for 17 of 119 isolates, the time above the MICmp was 55–60%; and, for 96 of 119 isolates, the trough was above the MICmp, so the time above the MICmp was 100%.

% Clinical cure

73 85 85

65 87 96

92 (77) 65 (23) 56 (5) 100 (3) 17 (6)

97 (58) 53 (15) 75 (8) 100 (3) 17 (6)

40 (5) 53 (17) 88 (96)

75 (4) 64 (14) 92 (72)

50 (10) 70 (10) 46 (13) 92 (12) 84 (19) 93 (54)

78 (9) 56 (9) 78 (9) 78 (9) 92 (12) 98 (42)

their apparent relationship to probability of bacteriological and clinical cure. The covariates tabulated include dose, bacterial species, MICmp, percentage of time above the MICmp, peak/MICmp and AUICmp,. Table V shows the probabilities of bacteriological and clinical cure compared with AUICmp (divided into three ranges) and bacterial species. For bacteriological cure there were 119 evaluable strains in 69 patients, while for clinical cure there were 90 evaluable strains in 59 patients. Probabilities of bacteriological and clinical cure are shown in Table III against dose, in Table V against bacterial species, and in Table III against MICmp. Univariate analyses (Fisher’s exact test) suggest

Pharmacodynamics of antibacterial response Univariate analyses. Tables III and IV present univariate statistical summaries of selected potential covariates and 51

A. Forrest et al. that the probability of clinical cure was higher with a 600 mg dose than with a 200 mg dose (P 0.005). When response was partitioned by bacterial species and AUICmp (Table IV), cell sizes were small, but it could be appreciated that Pseudomonas aeruginosa and Staphylo coccus aureus had lower probabilities of both bacteriological and clinical cure. Most of the species differences in AUICmp were a consequence of species differences in MICmp (Table V). For both microbiological and clinical cure, there was a decreasing probability of cure with increasing MICmp (Table III). Tables III and IV present activity measures that integrate the pharmacokinetics with the potency (e.g., MIC) of the drug. All of the factors tested (percent of time above the MICmp, AUICmp, peak/MICmp and trough/ MICmp) covaried strongly and all exhibited a trend for increasing likelihood of cure with increasing activity. AUICmp, as presented in Table III, was the ‘best’ (most predictive of outcome) of these variables. Figure 6 is a histogram showing the percent probability of clinical cure and failure as a function of the AUIC (not AUICmp; to convert AUIC to AUICmp, multiply by 4/3.)

Figure 6. Histogram of percent probability of clinical cure plotted against AUIC. The AUIC values on the x-axis are the medians of each cell; the values above each bar are the number of subjects in that cell.

Table IV. Results of univariate analyses: relationship of % bacteriological and % clinical cure (n in parentheses) to study variables Parameter

% Bacteriological cure

AUICmp 0–50 51–100 101–200 201–400 401–800 801–14,765

Multivariate analyses. Probability of cure and time to eradication are complex functions, usually dependent on a number of factors. To obtain an optimal assessment of the factors that affect antibacterial response, requires integration of the antibiotic dosage regimen and pharmacokinetics (to determine exposure profile), with the susceptibility of the bacteria to the drug (as reflected by MIC, for example). These and other factors should then be considered simultaneously, using the appropriate multivariate methods of analysis.

% Clinical cure

61 (28) 50 (8) 95 (19) 84 (19) 89 (19) 92 (25)

70 (23) 86 (7) 85 (13) 93 (14) 100 (14) 94 (19)

Table V. Probability of bacteriological cure and clinical cure versus bacterial species and AUICmp

Organism Moraxella catarrhalis Haemophilus spp. Enterobacter and Serratia spp. Escherichia coli & Klebsiella spp. Acinetobacter spp. Pseudomonas aeruginosa Staphylococcus aureus Streptococcus pneumoniae

Cure

0–92

AUICmp 92.1–230

230.1

bacteriological clinical bacteriological clinical bacteriological clinical bacteriological clinical bacteriological clinical bacteriological clinical bacteriological clinical bacteriological clinical

1/1 – 1/1 – 1/2 1/1 – – 1/1 – 4/13 9/13 1/2 0/2 7/8 7/8

1/1 1/1 – – 3/3 0/1 1/2 2/2 1/1 1/1 1/1 1/1 5/5 4/4 7/7 3/4

7/7 7/7 20/20 17/17 9/10 8/8 3/3 2/2 3/3 2/2 2/4 1/2 2/3 2/2 3/4 2/2

52

Grepafloxacin pharmacokinetics AUICmp greater than 230, 45/46 strains (98%) were associated with a clinical cure. With AUICmp in the model, none of the other drug exposure measures (e.g., dose, peak, peak/MIC, AUC, time above the MIC, etc.) nor MIC, infection site, etc., provided any additional information. One possible exception is bacterial species. As can be seen in Table V, there is a suggestion of some species-specific sensitivity, not ‘explained’ by AUIC. However, the cells in this table each contain relatively small numbers of bacteria. Examination of Table V reveals the possibility that P. aeruginosa might be significantly less sensitive and that Streptococcus pneu moniae might be more sensitive than might be predicted by the respective AUICmp. In our previous work, in patients with nosocomial lower respiratory tract infections receiving a different fluoroquinolone, no species-specific differences in response versus AUIC, were noted.1 It is possible that this could be an effect of ABECB, or Pseudomonas could be a marker of more severe pulmonary involvement, and thus an indirect measure of disease response. Given the continuous relationship between probability of cure and AUIC (see Tables IV and V and Figure 6), Hill-type models, relating percent probability of bacteriological (Figure 7) and clinical (Figure 8) cure, were developed. Figures 7 and 8 are ‘jitter plots’ (AUIC was divided into cells; the observed percent cure was weighted by the number of cases in that cell and plotted at the cell’s median AUIC value; each closed circle in the ‘jitter plot’ is one case, offset by the others in the same cell by a small amount of random noise). For bacteriological cure, the final model was:

Figure 7. Jitter plot of the Hill-type pharmacodynamic models for the percent probability of bacteriological cure versus AUICmp. The equation y (91 – 55)x8.9/888.9 x8.9) 55 depicts the fitted model. The x-axis is log-transformed for graphical clarity. No transformations were employed in the analyses.

percent probability of cure =

(91 55) AUICmp8.9 888.9 AUICmp8.9

55,

in which 91 (Pmax) is the asymptotic maximum response rate, 55 (P0) is the asymptotic baseline response rate (the probability of cure without drug treatment), 88 (AUIC50%) is the AUICmp at which percent probability of cure is (P0 Pmax)/2, and 8.9 is Hill’s constant (H). For clinical cure, the final model was:

Figure 8. Jitter plot of the Hill-type pharmacodynamic models for the percent probability of clinical cure versus AUICmp. The equation y (98 – 69)x2.0/882.0 x2.0) 69 depicts the fitted model. The x-axis is log-transformed for graphical clarity. No transformations were employed in the analyses.

percent probability of cure =

When multivariate, stepwise logistic regression and a recursive partitioning procedure (CART) were used to determine covariates to the probability of bacteriological and of clinical cure, both methods found that AUICmp was the most important determinant of response. For probability of bacteriological cure, there was a single breakpoint at an AUICmp of 92 (an AUIC of 69). Below this limit, 20/35 (57%) strains were eradicated, and above an AUICmp of 92, 75/83 strains (90%) were eradicated. For probability of clinical cure, two breakpoints, at an AUICmp of 92 and at a value of 230 (an AUIC of 173), were identified. Below an AUICmp of 92, 20/28 evaluable strains (71%) were associated with clinical cure; at an AUICmp between 92 and 230, 12/15 strains (80%) were cured; and at an

(91 69) AUICmp2.0 882.0 AUICmp2.0

69,

in which 98 is Pmax, 69 is P0, 128 is AUIC50%, and 2.0 is Hill’s constant. The very large baseline response (P0) is notable, and is probably a function of the disease. The last pharmacodynamic measure that we modelled is the time (in days) required to eradicate bacteria from the sputum. Proportional hazards analysis was performed on 117 evaluable strains in 76 patients. Again, of all of the potential covariates, AUIC was the most informative and, once AUIC was in the model, no other factor was significant. Figure 9 shows the Kaplan–Meier plots for the percent of strains remaining culture positive (vertical axis) plotted against days of therapy (horizontal axis), with each 53

A. Forrest et al. predicted to achieve an AUIC of 75. For an MIC of 0.25 mg/mL, 300 mg/65 kg/day would yield 62 patients (82%) with an AUIC of 75; 400 mg/65 kg/day would yield 44 patients (58%) with an AUIC of 75; and 600 mg/65 kg/day would be predicted to yield 10 patients (13%) with an AUIC of 75. For an MIC of 0.5 mg/mL, even 600 mg/65 kg/day would be predicted to yield 50 patients (66%) with an AUIC of 75. Figure 10 depicts the predicted rates of eradication for infections by bacteria with MICs of 0.5, 0.25 and 0.125 mg/L, for ABECB patients given 300, 400 and 600 mg/65 kg/day. Figure 11 shows the same information for doses of 600, 400 and 300 mg/65 kg/day, for the three MICs (0.125, 0.25 and 0.5 mg/L). A subset of the results contained in Figures 10 and 11 is summarized in Table VI. For all nine MIC–dose combinations, the predicted percentage of patients predicted to remain culture positive after 3, 5, 7 and 10 days of therapy is tabulated.

Figure 9. Kaplan–Meier plots of rates of eradication (117 bacterial strains in 76 ABECB patients). The three AUIC ranges (AUIC 75 (n 36); 75 AUIC 190 (n 23); 190 AUIC 11,000 (n 58)) for which rate of eradication differed are depicted.

Discussion

symbol marking a different AUIC range. After consolidating the AUIC regions that did not differ, three curves remained (Figure 9). At an AUIC of 75 (n 36), the median time (in days) to eradication (DTE50%) was 2.5 days and the time to eradication of 75% of strains (DTE75%), was infinity (at an AUIC of 75, the predicted proportion of strains eradicated was 55%; even extrapolating beyond the study period, more than 25% of strains would be predicted to remain culture-positive). At an AUIC of 75–190 and at an AUIC of 190, the DTE50% was 0.5 days and the DTE75% was 1.5 days. The first curve differed from the second two overall (P 0.001). The second two curves did not differ overall (0.05 P 0.1), but did differ over the period of 3.5–8.5 days (P 0.05).

Overall, grepafloxacin performed well in these ABECB patients, from both a pharmacokinetic and a clinical perspective. Considering that this agent is primarily excreted as metabolites, the pharmacokinetics were reasonable for a diverse group of patients with a wide range of age and underlying diseases. Although the patient population contained both smokers and non-smokers, the drug displayed reasonably predictable AUCs. Grepafloxacin bioavailability is not known, yet the serum profiles varied little between patients. On average the serum concentrations exceeded the MIC for at least a part of the dosing interval in virtually all ABECB patients. There were problem MIC values for this agent, as is typical of previously studied fluoroquinolones, and indeed, other classes of antibiotic as well.

Determination of optimal dose and duration Population pharmacokinetics

Table VI shows the results of the simulations that provided AUC and AUIC values. For an MIC of 0.125 mg/mL, 300 mg/65 kg/day would be predicted to yield 25 patients (33%) with an AUIC between 37.5 and 75; 400 mg/65 kg/day would yield six patients (8%) with an AUIC between 37.5 and 75; and, at 600 mg/65 kg/day, all patients would be

The population pharmacokinetic model fit the data very well (overall r2 0.986). Inter-patient variance in pharmacokinetic parameters and AUC values were substantive. AUCs were not adjusted for body weight, which may remove some of the variability.

Table VI. Percentage of patients predicted to be culture positive, based on MIC, dose and days of therapy

Days of therapy

MIC 0.125 mg/L grepafloxacin dose 300 400 600

MIC 0.25 mg/mL grepafloxacin dose 300 400 600

MIC 0.5 mg/L grepafloxacin dose 300 400 600

3 5 7 10

31.8 30.8 26.8 22.9

45.1 44.7 39.6 36.3

48.7 48.7 45.6 39.3

23.7 21.4 19.5 15.2

17.4 13.5 13.1 10.9

54

38.8 38.2 33.1 30.0

25.5 24.0 21.6 17.5

47.4 47.4 43.2 37.5

40.9 40.3 35.0 31.2

Grepafloxacin pharmacokinetics

Figure 10. Kaplan–Meier plots of predicted rates of eradication for bacteria with MICs of (a) 0.5 mg/L, (b) 0.25 mg/L and (c) 0.125 mg/L, in ABECB patients given oral grepafloxacin 300 ( ), 400 ( ) and 600 ( ) mg/65 kg/day.

floxacin.1 These differences are probably a consequence of differences in the immunocompetence of the patient, rather than of differences in pharmacodynamics between these two quinolones. Grepafloxacin 400–600 mg daily, as monotherapy, was able to eradicate organisms with MICs of 0.25 mg/L. For organisms with an MIC of 0.5 mg/L, only 33% could be expected to have organism eradication with a grepafloxacin regimen of 600 mg daily. Longer courses of therapy could also be considered. For an MIC of 0.5 mg/L, the predicted microbial failure rate (40.3%), for 600 mg/65 kg/day when given for 5 days, is almost 33% higher than after 10 days of therapy (31.2%). At an MIC of 0.25 mg/L or less, a regimen of 600 mg/65 kg/day provides significantly faster eradication and a better probability of cure. In patients who cannot tolerate this regimen, 400 mg/65 kg/day for longer courses can provide acceptable cure rates. Unless the infecting organism is very susceptible or the patient cannot tolerate larger doses for shorter courses, we would not recommend 300 mg/65 kg/day. All of these conclusions are based on a therapeutic goal of initial bacterial eradication, a situation that has been very well predicted by models of the relationships between pharmacokinetics as AUC and the organism MIC. In noso-

Population pharmacodynamics Probability of cure and days of treatment needed to eradicate bacteria from the sputum, were highly dependent on the AUIC. In univariate analyses, other measures of exposure were also significantly related to outcome, but none more strongly than AUIC. In multivariate analyses only AUIC was significant but it appeared that there might be some bacterial species-specific differences in outcome, not ‘explained’ by AUIC. Specific, continuous equations were developed that can predict outcome as a function of AUIC. Considering all three models of pharmacodynamics, it would appear that organisms will not be eradicated with an AUIC of 75 (an AUICmp of 100) during treatment of ABECB patients. An AUIC of 75–175 (AUICmp of 100–230) will be effective, but an AUIC of 175 is optimal, because it produces more rapid killing of the organism. No concentration–effect relationships have been identified for grepafloxacin toxicity, so the upper end of the optimal exposure is, in part, an empirical value. Also, risk of toxicity to the patient is probably associated with AUC, not AUIC. These AUIC breakpoints are lower than the value of 125 we had recommended for patients with lower respiratory tract infections, who were being treated with cipro55

A. Forrest et al.

Figure 11. Kaplan–Meier plots of predicted rates of eradication for daily doses of (a) 300, (b) 400 and (c) 600 mg/65 kg/day, used to treat ABECB caused by bacteria with MICs of 0.125 ( ), 0.25 ( ) and 0.5 ( ) mg/L.

comial pneumonia, failure to eradicate the organism leads to continued infection symptoms, organism resistance and clinical failure.1,2,4,25–27 In nosocomial pneumonia, organism eradication is a prerequisite to cure. This may or may not apply to patients with ABECB, where it appears that disease resolution is more common even when the antibiotic would not be active against the isolated organism.10,28 Based on studies in nosocomial pneumonia,25 resistance selection is also predicted in situations where ABECB patients do not have bacterial eradication. Even selected resistance in this patient population may not be associated with clinical failure, but the selection of resistance could result in a slower resolution of disease, and potentially cross-transmission of resistant organisms to other patients.

examination of dual individualization principles. (II) The rate of bacterial eradication at the same area under the inhibitory curve (AUIC) is more rapid for ciprofloxacin than for cefmenoxime. Annals of Pharmacotherapy 28, 863–8.

References

5. Schentag, J. J., Highet, V. S., Forrest, A. & Ballow, C. H. (1997). Antibiotic dosing issues in LRTI: population-derived AUIC is predictive of efficacy. Chest, in press

3. Forrest, A., Amantea, M., Collins, D., Chodosh, S. & Schentag, J. J. (1993). Pharmacodynamics of oral OPC17116 in patients with acute bacterial exacerbations of chronic bronchitis. In Programs and Abstracts of the Thirty-Third Interscience Conference on Antimicrobial Agents and Chemotherapy, New Orleans, 1993. Abstract 81, p. 134. American Society for Microbiology, Washington, DC. 4. Ballow, C. H., Amsden, G. W., Forrest, A., Highet, V. S. & Schentag, J. J. (1993). Control of antimicrobial activity exposure in the treatment of Gram-negative infections. In Programs and Abstracts of the Thirty-Third Interscience Conference on Antimicrobial Agents and Chemotherapy, New Orleans, 1993. Abstract 32, p. 134. American Society for Microbiology, Washington, DC.

1. Forrest, A., Nix, D. E., Ballow, C. H., Goss, T. F., Birmingham, M. C. & Schentag, J. J. (1993). The pharmacodynamics of intravenous ciprofloxacin in seriously ill patients. Antimicrobial Agents and Chemotherapy 37, 1073–81.

6. Peloquin, C. A., Cumbo, T. J., Nix, D. E., Sands, M. F. & Schentag, J. J. (1989). Intravenous ciprofloxacin in patients with nosocomial lower respiratory tract infections: impact of plasma concentrations, organism MIC, and clinical condition on bacterial eradication. Archives of Internal Medicine 149, 2269–73.

2. Goss, T. F., Forrest, A., Nix, D. E., Ballow, C. H., Birmingham, M. C., Cumbo, T. J. & Schentag, J. J. (1994). Mathematical

56

Grepafloxacin pharmacokinetics 7. Schentag, J. J., Nix, D. E. & Adelman, M. H. (1991). Mathematical examination of dual individualization principles. (I) Relationships between AUC above MIC and area under the inhibitory curve for cefmenoxime, ciprofloxacin, and tobramycin. DICP, Annals of Pharmacotherapy 25, 1050–7.

17. Baigelman, W., Chodosh, S., Pizzuto, D. & Sadow, T. (1979). Quantitative sputum Gram stains in chronic bronchial disease. Lung 156, 265–70. 18. Chodosh, S., Eichel, B., Ellis, C. & Medici, T. C. (1982). Comparison of trimethoprim–sulfamethoxazole with ampicillin in acute infectious exacerbations of chronic bronchitis: a double-blind, crossover study. Reviews of Infectious Disease 4, 517–27

8. Nix, D. E. & Schentag, J. J. (1988). The quinolones: an overview and comparative appraisal of their pharmacokinetics and pharmacodynamics. Journal of Clinical Pharmacology 28, 169–78.

19. Akaike, H. (1979). A Bayesian extension of the minimum AIC procedure of autoregressive model fitting. Biometrika 66, 237–42.

9. Chodosh, S. (1991). Treatment of acute exacerbations of chronic bronchitis: state of the art. American Journal of Medicine 91, Suppl. 6A, 87S–92S.

20. Wilkinson, L. (1990). SYSTAT: the system for statistics. SYSTAT Inc., Evanston, IL.

10. Ball, P. (1995). Epidemiology and treatment of chronic bronchitis and its exacerbations. Chest 108, 43S–52S.

21. Steinberg, D. & Colla, P. (1992). CART: a supplementary module for SYSTAT. SYSTAT Inc., Evanston, IL.

11. Lillington, G. A. (1993). Bronchitis, a primary care update. Consultant 33, Suppl. 9, 34–8.

22. McGill, R., Tukey, J. W. & Larson, W. A. (1978). Variations of box plots. American Statistician 32, 12–6.

12. D’Argenio, D. Z. & Schumitzky, A. (1979). A program package for simulation and parameter estimation in pharmacokinetic systems. Comp. Prog. Biomed. 9, 115–34.

23. Steinberg, D. & Colla, P. (1991). LOGIT: a supplementary module for SYSTAT. SYSTAT Inc., Evanston, IL.

13. D’Argenio, D. Z. & Schumitzky, A. (1992). Adapt II users’ guide. Biomedical simulations resource. University of Southern California, Los Angeles, CA.

24. Steinberg, D. & Colla, P. (1988). SURVIVAL: a supplementary module for SYSTAT. SYSTAT Inc., Evanston, IL.

14. Forrest, A., Hawtoff, J. & and Egorin, M. J. (1991). Evaluation of a new program for population PK/PD analysis: applied to simulated phase I data. Clinical Pharmacology and Therapeutics 49, 153.

25. Thomas, J. K., Forrest, A., Bhavnani, S. M., Hyatt, J. M., Cheng, A., Ballow, C. H. et al. (1997). Pharmacodynamic evaluation of factors contributing to the emergence of bacterial resistance in acutely ill patients. Antimicrobial Agents and Chemotherapy, in press.

15. Forrest, A., Collins, D. A., Pitlick, W. H. & Schentag, J. J. (1992). Pharmacokinetics of PO grepafloxacin (OPC 17,116) in normal subjects: single, dose-ranging (200–1200 mg) cross-over design. Program and Abstracts of the Thirty-Second Interscience Conference on Antimicrobial Agents and Chemotherapy, Anaheim, 1992. Abstract 429. American Society of Microbiology, Washington, DC.

26. Hyatt, J. M., McKinnon, P. S., Zimmer, G. S. & Schentag, J. J. (1995). The importance of pharmacokinetic/pharmacodynamic surrogate markers to outcome. Focus on antibacterial agents. Clinical Pharmacokinetics 28, 143–60. 27. Hyatt, J. M. & Schentag, J. J. (1996). Risk factors for selection of Pseudomonas aeruginosa resistance to ciprofloxacin in patients with previously susceptible strains. In Abstracts of the Sixth Annual Meeting of the Society for Healthcare Epidemiology of America, Washington, DC. Abstract 57 Society for Healthcare Epidemiology of America, Washington, DC.

16. Forrest, A., Killian, A. D., Collins, D. A., Pitlick, W. H. & Schentag, J. J. (1992). Multiple-dose pharmacokinetics of PO grepafloxacin (OPC 17,116) in normal subjects: 400 versus 800 mg, parallel design. In Program and Abstracts of the ThirtySecond Interscience Conference on Antimicrobial Agents and Chemotherapy, 1992. Abstract 430. American Society of Microbiology, Washington, DC.

28. Trigg C. J. & Davies R. J. (1994). Antibiotic prescribing for bronchitis: time for a change? Respiratory Medicine 88, 161–3.

57