Pharmacokinetic-Pharmacodynamic Determinants of Oseltamivir Efficacy Using Data from Phase 2 Inoculation Studies

Abstract

Given the limited understanding about pharmacokinetic-pharmacodynamic (PK-PD) determinants of oseltamivir efficacy, data from two phase 2 influenza virus inoculation studies were evaluated. Healthy volunteers in studies 1 and 2 were experimentally infected with influenza A/Texas (the concentration of neuraminidase inhibitor which reduced neuraminidase activity by 50% [IC₅₀] = 0.18 nM) or B/Yamagata (IC₅₀ = 16.76 nM), respectively. In study 1, 80 subjects received 20, 100, or 200 mg of oral oseltamivir twice daily (BID), 200 mg oseltamivir once daily, or placebo for 5 days. In study 2, 60 subjects received 75 or 150 mg of oral oseltamivir BID or placebo for 5 days. Oseltamivir carboxylate (OC) (active metabolite) PK was evaluated using individual PK data and a population PK model to derive individual values for area under the concentration-time curve from 0 to 24 h (AUC_0–24), minimum concentration of OC in plasma (C_min), and maximum concentration of OC in plasma (C_max). Exposure-response relationships were evaluated for continuous (area under composite symptom score curve [AUCSC], area under the viral titer curve, and peak viral titer) and time-to-event (alleviation of composite symptom scores and cessation of viral shedding) efficacy endpoints. Univariable analyses suggested the existence of intuitive and highly statistically significant relationships between OC AUC_0–24 evaluated as a 3-group variable and AUCSC, time to alleviation of composite symptom scores, and time to cessation of viral shedding. The upper OC AUC_0–24 threshold (∼14,000 ng · h/ml) was similar among these endpoints. Multivariable analyses failed to demonstrate the influence of study/strain on efficacy endpoints. These results provide the first demonstration of exposure-response relationships for efficacy for oseltamivir against influenza and suggest that OC exposures beyond those achieved with the approved oseltamivir dosing regimen will provide enhanced efficacy. The clinical applicability of these observations requires further investigation.

INTRODUCTION

Oseltamivir is an orally available prodrug of the influenza virus neuraminidase inhibitor (NAI), oseltamivir carboxylate (OC). While oseltamivir has been studied extensively, its efficacy and safety profile is known, and it has been licensed for more than a decade for the treatment and prophylaxis of influenza in persons aged ≥1 year (1, 2), fundamental information describing relationships between OC exposure and clinical and virologic response in humans is lacking. Such information is important, as it forms part of the scientific basis for optimizing treatment regimens for influenza, especially for viruses possessing greater pathogenicity than regular seasonal strains.

Indirect information from preclinical systems provides some important insights about pharmacokinetic-pharmacodynamic (PK-PD) relationships for OC and other NAIs, including zanamivir and peramivir. McSharry et al. and Brown et al. utilized an in vitro pharmacodynamic hollow-fiber infection model (HFIM) system to explore PK-PD determinants of oral oseltamivr (3) and intravenous (i.v.) zanamivir (4, 5) antiviral activity. In dose fractionation study designs for OC, the treatment schedule did not appear to alter suppression of viral replication, thus indicating that the PK-PD index associated with efficacy was the AUC_0–24/EC₅₀ ratio, the ratio of the area under the concentration-time curve from 0 to 24 h (AUC_0–24) to the drug concentration that reduces the number of plaque forming units (PFU) by 50% (EC₅₀). In contrast, it was demonstrated that time above EC₅₀ was the PK-PD index predictive of efficacy for the i.v. administration of zanamivir. The differences in these results were attributed to differences in the half-lives of the respective NAIs. Interestingly, when the i.v. zanamivir half-life was increased from 2.5 to 8 h (i.e., the same as the OC half-life), the PK-PD-linked index was the AUC_0–24/EC₅₀ ratio. The authors speculated that for oseltamivir, it may be possible to effectively treat influenza with a once-a-day schedule. Interestingly, the authors also noted an exposure-response relationship for OC that was less steep than that for other non-NAI antiviral agents, suggesting there may be opportunity to identify higher doses that may provide additional virologic benefit.

In vivo investigations in mice and ferrets have also provided some supportive information on exposure-response relationships for OC, with much of the published information reported from investigations in highly pathogenic virus subtypes. In ferrets lethally challenged with highly pathogenic influenza A/Vietnam/1203/04 (H5N1) given oseltamivir treatment initiated 24 h postinfection, dosing regimens providing similar exposures to those achieved by giving 75 mg twice daily (BID) (approved dosing regimen) in humans were insufficient to prevent death; doses 2.5-fold higher were necessary to prevent death in ferrets in this model (6). Similar observations were noted in mice challenged with differing H5N1 clades, experiments for which oseltamivir treatment was initiated 4 h prior to inoculation (7). Logistical constraints make it difficult to obtain PK for PK-PD examinations directly from animals infected with such highly virulent viruses. Thus, a limitation for such studies is that the inferences are based primarily on dose-response data and PK is inferred. Recently, a PK-PD evaluation was performed in ferrets inoculated for influenza B/Yamagata/1988 in which both OC PK and PD were determined. In contrast to more virulent strains, only mild disease was induced following inoculation, thereby limiting the ability to detect PK-PD relationships. Despite this limitation, the authors noted a PK-PD association between increasing OC AUC and positive impact on the weight of ferrets in this study (8).

In summary, the available preclinical findings from HFIM and in vivo infection models suggest the existence of exposure-response relationships for efficacy for OC. These data also suggest that AUC is the more important exposure measure associated with OC efficacy. However, there is no clear evidence that the maximal effect (E_max) has been attained for all influenza viruses at OC exposures comparable to those achieved with the labeled oseltamivir dosing regimen of 75 mg BID.

In an effort to further characterize the exposure-response relationships between OC exposure and virologic and clinical efficacy endpoints, analyses were undertaken using data from two well-controlled phase 2 influenza virus inoculation studies performed in healthy volunteers (9, 10).

(Results of this investigation were presented at the XIV International Symposium on Respiratory Viral Infections, 23 to 26 March 2012, Istanbul, Turkey.)

MATERIALS AND METHODS

Study population, drug dosage, and administration.

The study population included data from two phase 2 studies, study PV15616 (study 1) and study NP15717 (study 2), evaluating the safety, tolerability, and antiviral activity of various doses of oseltamivir in healthy adults experimentally inoculated with either influenza virus A/Texas/36/91 (H1N1) or influenza virus B/Yamagata/16/88, respectively (9, 10). Study 1 was a single-center, multiple-dose, double-blind, randomized, placebo-controlled study which evaluated four oral oseltamivir dosing regimens, 20 mg BID for 5 days, 100 mg BID for 5 days, 200 mg BID for 5 days, 200 mg once daily (QD), and matching placebo administered BID for 5 days (9). Study 2 was a single-center, multiple-dose, double-blind, randomized, placebo-controlled study evaluating two oral dosing regimens of oseltamivir, 75 mg BID and 150 mg BID, and placebo given orally BID for 5 days (10).

The inclusion criteria for both studies were similar. Healthy adult subjects enrolled in these studies were ≥18 years of age, had an influenza virus antibody level of ≤1:8 (study 1) or <1:10 (study 2), had no significant health abnormalities as determined by evaluation of medical history, general physical exam, vital signs, laboratory tests, and electrocardiogram, were nonsmokers or consumed an average of less than 10 cigarettes per day (study 1), and were able to give informed consent.

Healthy subjects were excluded if they met any of the following criteria: had hepatitis B infection; were transplant recipients; were taking steroids or receiving immunosuppressant therapy; had a known HIV infection; had a hypersensitivity to oseltamivir or structurally similar compounds; had a known allergy to components of the virus suspension; were asthmatic and were receiving chronic therapy for asthma; had experienced a previous episode of acute respiratory tract infection, otitis, bronchitis, or sinusitis or had received antibiotics for any of these conditions within 2 weeks prior to the start of the study; had pyrexia within 3 days of study start; had a clinically relevant history of alcohol or drug abuse; or had been given an influenza vaccine less than 6 months prior to study start (9, 10).

Concomitant medications that might interfere with oseltamivir metabolism, gut mobility, or renal excretion were excluded. Healthy subjects were allowed to receive paracetamol for the relief of fever and discomfort, and continued use of the oral contraceptive pill was permitted during the study. Other medications required for disease symptoms or other medical conditions arising during the study were permitted at the discretion of the investigators.

The two studies were carried out in accordance with the principles of the Declaration of Helsinki and applicable local laws. Full ethical committee approval was obtained.

Schedule of assessments.

In study 1, healthy subjects underwent screening within 4 weeks prior to the first dose during which a complete medical history, physical examination, vital signs, safety laboratory samples (hematology, biochemistry, and hepatitis B and HIV antibodies), urine drug screen, and urinalysis were conducted. A sample size of 80 subjects was chosen, and subjects were randomized to one of five treatment groups. Subjects were admitted to the isolation unit on day 1 and remained in the unit until discharge on the morning of day 9 (9). Vital signs, blood samples for safety laboratory samples and hemagglutination inhibition antibody titers, urine samples for urinalysis, nasal wash samples for virus culture and titer, and nasal discharge weights were taken prior to inoculation. A physical exam was performed prior to inoculation, and symptom assessment scores (self-rated by the subject) were assessed prior to inoculation. Inoculation with the human influenza virus A/Texas/36/91 (H1N1) was performed on the evening of day 1 via nasal drops containing a median tissue culture infective dose (TCID₅₀) of virus of 10⁶, following which repeat PD assessments (as described below), vital signs, and temperature were recorded (9).

Treatment with study drug was initiated on the evening of day 2, 28 h after inoculation. On days 2 to 8, the following assessments were performed: adverse event history; modified physical examination if indicated by the adverse event history or medical evaluation; twice daily vital signs measurements; temperature recordings 4 times daily; twice daily symptom score self-assessment (nasal stuffiness, ear ache, runny nose, sore throat, cough, breathing difficulty, myalgia, fatigue, headache, feeling feverish, hoarseness, sneezing, chest discomfort, and overall discomfort); nasal wash samples for viral culture and viral titer taken twice daily on days 2 and 3 and once daily subsequently. On day 9, after completion of the assessments as taken on days 2 to 8 in addition to blood for safety samples, subjects were allowed to leave the isolation unit at the discretion of the supervising physician and if the symptoms of influenza had resolved. If symptoms persisted, subjects remained in the isolation unit and were monitored until the resolution of all symptoms. Three to 4 weeks after virus inoculation, subjects returned to the clinic for a final assessment which included an adverse event history and blood sampling to determine virus antibody titer (9).

The study schedule for study 2 was similar to that of study 1 with the following exceptions: subjects were screened within 2 weeks prior to the start of the study, and a sample size of 60 subjects with baseline antibody titers of <1:10 was chosen with 20 subjects randomized to each of the three treatment groups; subjects were admitted to the isolation unit 24 h (day −1) prior to inoculation with influenza B/Yamagata/16/88 virus and remained in the isolation unit until discharge on day 8; subjects underwent twice daily nasal washes for virus culture on days 1 to 3 and then once daily from days 4 to 8; and blood samples for the assessment of laboratory endpoints for safety were collected on days 3 and 5 (10).

NA inhibition assay was performed with viruses standardized to equivalent NA activity and incubated with NAIs at concentrations of 0.00005 to 100 μM with 2′-(4-methylumbelliferyl)-α-d-N-acetylneuraminic acid as a substrate (final concentration of 100 μM). The IC₅₀ was determined by plotting the dose-response curve of inhibition of NA activity at 60 min as a function of the compound concentration. Each experiment was performed in triplicate, and the final mean ± standard deviation (SD) IC₅₀ was 16.76 ± 4.10 nM for influenza B/Yamagata and 0.18 ± 0.11 nM for influenza A/Texas.

Efficacy endpoints.

PK-PD analyses described below involved the evaluation of five efficacy endpoints. These endpoints included three continuous and two time-to-event variables. The continuous efficacy endpoints included composite symptom score AUC, viral titer AUC, and peak viral titer. The composite symptom score was calculated using seven individual symptom scores, including feeling feverish, headache, muscle ache, sore throat, cough, overall discomfort, and nasal symptoms (defined as the maximum of “nasal stuffiness” or “runny nose”) (10). During the first 9 days of the study, each subject self-ranked the severity of each symptom twice daily using a scale for which 0 represents the absence of symptoms and 3 represents the most severe symptoms. The seven scores were added to form the composite symptom score. The AUC of this composite symptom score-versus-time curve over 9 days was derived using the linear trapezoidal rule. The AUC of the viral titer-versus-time curve over 9 days was likewise calculated using the linear trapezoidal rule. Peak viral titer was defined as the maximum viral titer value for each subject observed during the study period.

The time-to-event efficacy endpoints included time to alleviation of composite symptom score and time to cessation of viral shedding. The time to alleviation of the composite symptom score was calculated as the time at which any of the seven individual symptom scores, as described above, was rated as >1 (time = 0) until the time at which all applicable individual symptom scores were ≤1. The time to cessation of viral shedding was calculated using the viral culture data acquired using nasal lavage. The time to cessation of viral shedding was defined as the time from the first positive viral culture result (time = 0) until the time of the first negative viral culture result.

Pharmacokinetic sampling.

In study 1, plasma samples for PK assessment of oseltamivir and OC were taken prior to the morning dose of medication on days 3, 4, and 7. In study 2, samples for PK assessment were taken each morning just prior to drug administration on days 2 to 4 and complete 12-hour plasma drug PK profiles were obtained following the first dose on days 1 and 5. Additional details describing the plasma drug sample assay are provided elsewhere (11).

Determination of plasma drug exposures.

The steady-state AUC_0–24 and the maximum and minimum plasma OC concentrations (C_max and C_min, respectively) represented the exposure measures evaluated for the PK-PD analyses described herein. In the accompanying article by Kamal et al. (11), a population PK model for oseltamivir and OC, which was based on data from 13 clinical studies, including the two studies described herein, was used to obtain post hoc PK parameter estimates for the subjects in the current analysis. Using these post hoc PK parameter estimates, exposure measures were computed for each subject (11). In brief, the model simultaneously described the plasma PK data for both oseltamivir and OC using two compartments for oseltamivir with first-order absorption and direct conversion of oseltamivir to OC and one compartment for OC with first-order elimination. A covariate analysis demonstrated that weight and creatinine clearance, and to a lesser degree age, were statistically significant predictors of the PK of oseltamivir and OC. As evidenced by the agreement between both the population mean predicted (r² = 0.741) and individual predicted (r² = 0.969) and observed plasma OC concentrations, the model fit the data well (see Fig. S1 in the supplemental material).

Post hoc PK parameter values were used to generate individual predicted steady-state concentrations every 0.1 h during day 5 of therapy for all subjects receiving oseltamivir. The OC C_max and C_min were determined by direct observation. Steady-state AUC_0–24 was calculated via the linear trapezoidal rule using the individual predicted concentration-time data.

Pharmacokinetic-pharmacodynamic analyses.

Univariable and multivariable PK-PD analyses were conducted as described below using R 2.11.1 (12), and data from all evaluable subjects in the two studies, including those who received placebo. Evaluable subjects were those for whom OC AUC_0–24 values could be computed and for whom adequate data for at least one efficacy endpoint were available. Subjects who received placebo and for whom efficacy data were available were also considered; an OC AUC_0–24 of zero was assumed for these subjects.

Univariable analyses.

Univariable relationships for each continuous efficacy endpoint were evaluated using the F-test from linear regression or analysis of variance. Univariable relationships for each time-to-event efficacy endpoint were examined using log rank tests for categorical independent variables and the likelihood ratio test from Cox proportional hazard regression for continuous independent variables.

The independent variables evaluated in these analyses included the OC exposure variables, AUC_0–24, C_max, C_min, and IC₅₀. Each continuous OC exposure variable was evaluated in its original form and as 2- and 3-group categorical variables to account for potential nonlinearity and/or nonmonotonicity. The categorical forms of these independent variables were constructed using thresholds that were optimally determined for the given efficacy endpoint. Two-group independent variables were constructed by using the resulting split of a regression tree for a continuous efficacy endpoint and by using a cutoff maximizing the log rank test derived from a univariable Cox proportional hazard regression model for a given time-to-event efficacy endpoint. Three-group independent variables were constructed by determining a pair of cutoff values that minimized the likelihood ratio P value using linear regression for a continuous efficacy endpoint. For a time-to-event efficacy endpoint, minimization of the log rank P value derived from Cox proportional hazard regression was used to determine the pair of cutoff values to define the 3-group independent variable. For both 2- and 3-group independent variables, a minimum subgroup size of 10 subjects was imposed to construct such categorical variables. OC exposure measures were also evaluated as an empirically divided categorical variable; each measure was divided into quartiles. IC₅₀, which included two values, 0.18 and 16.76 nM, was evaluated as a categorical variable.

Multivariable analyses.

Multivariable analyses were carried out for each efficacy endpoint, with consideration of separate models that included each of the above-described exposure measures and IC₅₀. Each continuous efficacy endpoint was analyzed using linear regression, while each time-to-event efficacy endpoint was analyzed using Cox proportional hazard regression. Multivariable models considering the interaction between individual forms of OC exposure and IC₅₀ were also assessed. The statistical significance of the model parameters was tested using Wald P values for single parameters (e.g., 2-group and continuous forms of an independent variable) and likelihood ratio P values for multiple parameters (e.g., 3-group and quartile forms of an independent variable). Model discrimination was accomplished using the corrected Akaike's information criterion (AICc) (13), an assessment that balances improvement in goodness of fit with model complexity (e.g., degrees of freedom, number of fitted parameters). If the AICc value was closely similar among more than one model, the final multivariable model was chosen based on clinical judgment, including the biological plausibility of the nature of the relationships for OC exposure or IC₅₀ retained in the model.

RESULTS

Subject population.

A total of 115 subjects were evaluated for inclusion in these analyses. In study 1, 69 (45 female and 24 male) subjects received either active treatment (n = 56) or placebo (n = 13) and were inoculated with influenza A/Texas/36/91 (H1N1) virus. For study 2, 46 (14 female and 32 male) subjects received either active treatment (n = 30) or placebo (n = 16) and were inoculated with influenza B/Yamagata/16/88 virus. All 86 subjects who received study drug had PK data available. Summary statistics of baseline characteristics for all evaluable subjects, stratified by study, are presented in Table 1.

Table 1.

Summary statistics of baseline demographic characteristics for all evaluable subjects

Characteristic	Mean (% CV) or % (n/N)
	Study 1	Study 2	Both studies
Age (yr)	22.3 (20.3)	25.0 (29.3)	23.4 (25.4)
Ht (cm)	170 (5.72)	176 (4.44)	172 (5.45)
Wt (kg)	68.6 (20.5)	73.7 (13.4)	70.6 (18.0)
Creatinine clearance (ml/min/1.73 m2)	108 (23.8)	124 (17.8)	114 (22.2)
Sex
Male	35.0 (24/69)	70.0 (32/46)	49.0 (56/115)
Female	65.0 (45/69)	30.0 (14/46)	51.0 (59/115)
Race
Black	11.6 (8/69)	0 (0/46)	6.96 (8/115)
White	78.3 (54/69)	93.5 (43/46)	84.4 (97/115)
Other	10.4 (7/69)	6.50 (3/46)	8.70 (10/115)
Total daily dose (mg)
40	21.7 (15/69)	0 (0/46)	13.0 (15/115)
150	0 (0/69)	32.6 (15/46)	13.0 (15/115)
200	39.1 (27/69)	0 (0/46)	23.5 (27/115)
300	0 (0/69)	32.6 (15/46)	13.0 (15/115)
400	20.3 (14/69)	0 (0/46)	12.2 (14/115)
None (placebo)	18.8 (13/69)	34.8 (16/46)	25.2 (29/115)
Treatment regimen (mg)
20 BID	21.7 (15/69)	0 (0/46)	13.0 (15/115)
75 BID	0 (0/69)	32.6 (15/46)	13.0 (15/115)
100 BID	20.3 (14/69)	0 (0/46)	12.2 (14/115)
150 BID	0 (0/69)	32.6 (15/46)	13.0 (15/115)
200 QDa	18.8 (13/69)	0 (0/46)	11.3 (13/115)
200 BID	20.3 (14/69)	0 (0/46)	12.2 (14/115)
None (placebo)	18.8 (13/69)	34.8 (16/46)	25.2 (29/115)

^aQD, once daily.

Summary of exposure measures.

Figure 1 shows the comparison of distribution of the OC AUC_0–24 values for all evaluable subjects by study as represented by box plots. While the median AUC_0–24 values were closely similar, there was a wider range of exposures for subjects in study 1 than for those in study 2 as would be expected given the dose ranges for each study. When examining the Spearman rank correlation coefficient among OC AUC_0–24, C_max, and C_min, high correlations among all three exposure measures were evident (>0.96). Given these findings and the fact that the PK-PD index reported to be the most associated with the efficacy of NAIs in preclinical models was the AUC_0–24/EC₅₀ ratio (3, 4), univariable and multivariable analyses described herein were carried out using OC AUC_0–24.

Fig 1.

Comparison of OC AUC_0–24 values for all evaluable subjects administered active treatment stratified by study. The OC AUC_0–24 values are shown in nanograms · hour/milliliter. The box shows the 25th to 75th percentile values and the line in the box shows the median value, while the whiskers extend from the minimum and maximum values.

Pharmacokinetic-pharmacodynamic analyses. (i) Univariable analyses.

A summary of the P values and directional assessments for the univariable relationships between the efficacy endpoints and OC AUC_0–24, evaluated as continuous and 2-group, 3-group, and quartile categorical variables or IC₅₀ (categorical variable), is shown in Table 2. Scatterplots showing the relationship between each continuous efficacy endpoint and OC AUC_0–24 evaluated as a continuous variable, with the OC AUC_0–24 ranges encompassing the 2- and 3-group variables indicated by a triangle symbol and dashed vertical lines, respectively, are presented in Fig. S2 in the supplemental material.

Table 2.

Summary of P values and directional assessments for univariable relationships between efficacy endpoints and OC AUC_0–24 or IC₅₀^a

Efficacy endpoint	Directionality and P value for efficacy endpointb
	OC AUC0–24				IC50c
	2-group	3-group	Quartile	Continuous
Composite symptom score AUC	/, <0.001	/, <0.001	/, 0.036	/, 0.010	\, 0.06
Viral titer AUC	/, 0.004	/, 0.010	N, 0.042	/, 0.06	\, 0.035
Peak viral titer	/, 0.049	∩, 0.004	0.26	0.55	/, 0.011
Time to alleviation of composite symptom score	/, <0.001	/, <0.001	/, 0.006	/, <0.001	\, 0.06d
Time to cessation of viral shedding	/, 0.012	/, 0.007	/, 0.09	/, 0.037	\, 0.026d

^aSummary of P values and directional assessments for univariable relationships between efficacy endpoints and OC AUC_0–24 evaluated as a continuous, 2-group, 3-group, and quartile categorical variables or IC₅₀.

^bAs described in Materials and Methods for the univariable analyses, P values for univariable relationships for continuous efficacy endpoints were based on the F-test, while those for univariable relationships for time-to-event efficacy endpoints were based on the log rank test for categorical independent variables and the likelihood ratio tests from Cox proportional hazard regression for continuous independent variables. Directionality is reported only for P values of ≤0.10. Directionality for the OC exposure variables is reported as follows: ∩, the drug effect was highest for the middle AUC_0–24 group compared to the low and high OC AUC_0–24 groups; U, the drug effect was lowest for the middle AUC_0–24 group compared to the low and high OC AUC_0–24 groups; /, the drug effect increased as OC AUC_0–24 increased; \, the drug effect decreased as OC AUC_0–24 increased; N, the drug effect was lowest with the first and third OC AUC_0–24 quartiles and highest with the second and fourth OC AUC_0–24 quartiles.

^cIC₅₀ was evaluated as a categorical variable, since it only had values of 0.18 and 16.76 nM for studies 1 and 2, respectively. Directionality for IC₅₀ is reported as follows: /, the drug effect was larger as the IC₅₀ increased; \, the drug effect was larger as the IC₅₀ decreased.

^dRelationship directionality is based on time to 50% of the population achieving the endpoint.

Significant relationships with at least one form of OC AUC_0–24 were evident for all five endpoints. For four of the endpoints, composite symptom score AUC, viral titer AUC, time to alleviation of composite symptom score, and time to cessation of viral shedding, the relationships with OC AUC_0–24 evaluated as a 2- and/or 3-group variable incorporated comparisons with the lowest OC AUC _0–24 group, which was mostly or entirely comprised of subjects who received placebo. However, in the case of composite symptom score AUC and time to cessation of viral shedding and alleviation of composite symptom score, statistically significant 3-group categorical assessments demonstrated increased response between the middle and higher OC AUC_0–24 groups. Biologically plausible univariable relationships between viral titer AUC or peak viral titer and OC AUC_0–24 were not apparent.

Univariable relationships for IC₅₀ are also presented in Table 2. However, given that OC AUC_0–24 were not balanced across studies (as shown in Fig. 1) and that each study evaluated a specific virus with a defined IC₅₀ value colinearity between OC AUC_0–24 and IC₅₀ was expected. Thus, the evaluation of the impact of OC AUC_0–24 and IC₅₀ were assessed as part of the multivariable analyses.

Parameter or hazard ratio estimates for univariable models describing the relationship between composite symptom score AUC, time to alleviation of composite symptom score, or time to cessation of viral shedding and OC AUC_0–24 evaluated as a 3-group variable are provided in Table 3. As shown in Table 4, the influence of the middle and higher OC AUC_0–24 groups relative to the lower OC AUC_0–24 group on the efficacy endpoints is demonstrated by comparing the time (in days) to 25, 50, and 75% of the population achieving time to alleviation of composite symptom score or time to cessation of viral shedding or the mean value for composite symptom score AUC among OC AUC_0–24 groups. The above-described univariable relationships are also shown graphically by stratified Kaplan-Meier curves for time to alleviation of composite symptom score and time to cessation of viral shedding and by box plots for composite symptom score AUC by OC AUC_0–24 group in Fig. 2.

Table 3.

Parameter or hazard ratio estimates for selected univariable models describing the relationship between continuous or time-to-event efficacy endpoints and OC AUC_0–24^a

Efficacy endpoint	Reference group for OC AUC0–24 (ng · h/ml)	Comparison group for OC AUC0–24 (ng · h/ml)	Univariable modelb
			Parameter estimate for continuous efficacy endpoint or hazard ratio estimate for time-to-event efficacy endpoint (95% CIc)	Pairwise P value	Overall P value
Composite symptom score AUC	≤1,495	>1,495 to ≤14,497	−5.50 (−8.87, −2.31)	0.002	0.001
	≤1,495	>14,497	−7.88 (−12.42, −3.34)	<0.001
	>1,495 to 14,497	>14,497	−2.38 (−6.44, 1.68)	0.25
Time to alleviation of composite symptom score	≤1,568	>1,568 to ≤13,638	1.94 (1.08, 3.50)	0.028	<0.001
	≤1,568	>13,638	5.87 (2.71, 12.73)	<0.001
	>1,568 to 13,638	>13,638	3.03 (1.51, 6.06)	0.002
Time to cessation of viral shedding	0	>0 to ≤14,180	1.77 (1.01, 3.10)	0.048	0.007
	0	>14,180	2.85 (1.46, 5.57)	0.002
	>0 to 14,180	>14,180	1.62 (0.941, 2.77)	0.082

^aParameter or hazard ratio estimates for selected univariable models describing the relationship between continuous or time-to-event efficacy endpoints and OC AUC_0–24 evaluated as a 3-group variable.

^bAll possible pairwise comparisons were evaluated.

^c95% CI, 95% confidence interval.

Table 4.

Influence of OC AUC_0–24 subgroups on the time to 25, 50, and 75% of the population achieving the continuous or time-to-event efficacy endpoint

Efficacy endpoint	Subgroup threshold for OC AUC0–24 evaluated as a 3-group variable (ng · h/ml)	No. of subjects	Time to 25, 50, and 75% of the population achieving time-to-event dependent variable (days) or mean value for continuous variablea
Composite symptom score AUC	≤1,495	30 (29 placebo)	14.6
	>1,495 to ≤14,497	64	9.1
	>14,497	18	6.7
Time to alleviation of composite symptom score	≤1,568	20 (18 placebo)	2, 3.5, 4.5
	>1,568 to ≤13,638	30	1, 1.5, 3.5
	>13,638	14	0.5, 0.75, 1.5
Time to cessation of viral shedding	0	22 (22 placebo)	2.5, 4.75, 7
	>0 to ≤14,180	49	1, 2.5, 5.5
	>14,180	21	1, 2, 3

^aAll possible pairwise comparisons were evaluated.

Fig 2.

Univariable relationships between composite symptom score (A), time to alleviation of composite symptom score (B), and time to cessation of viral shedding (C) and OC AUC_0–24 (ng · h/ml) evaluated as a 3-group variable. ANOVA, analysis of variance.

Highly statistically significant relationships between each of the three efficacy endpoints and OC AUC_0–24 evaluated as a 3-group variable were apparent (P ≤ 0.007). Not surprisingly, the OC AUC_0–24 thresholds for composite symptom score AUC and time to alleviation of composite symptom score when OC AUC_0–24 was evaluated as a 3-group categorical variable were similar (≤1,495, >1,495 to ≤14,497, >14,497 and ≤1,568, >1,568 to ≤13,638, >13,638 ng · h/ml, respectively). OC AUC_0–24 thresholds for time to cessation of viral shedding were 0, >0 to ≤14,180, and >14,180 ng · h/ml. The highest group thresholds for the OC AUC_0–24 were similar for time to cessation of viral shedding, composite symptom score AUC, and time to alleviation of composite symptom score (>14,180, >13,638, and >14,497 ng · h/ml, respectively).

The influence of higher OC AUC_0–24 on lower symptom score AUC was evident as assessed by the mean values for composite symptom score AUC, which were 14.6, 9.1, and 6.7 for OC AUC_0–24 groups with threshold values of ≤1,495, >1,495 to ≤ 14,497, and >14,497 ng · h/ml, respectively. The influence of higher AUC_0–24 on earlier time-to-event variables was evident as assessed by the lengths of time for 50% of the population to achieve alleviation of their composite symptoms, which were 0.75, 1.5, and 3.5 days for OC AUC_0–24 groups with threshold values of >13,368, >1,568 to ≤13,368, and ≤1,568 ng · h/ml, respectively. The influence of higher OC AUC_0–24 on earlier time to cessation of viral shedding was evident as assessed by the lengths of time for 50% of the population to achieve this endpoint, which were 2, 2.5, and 4.5 days for OC AUC_0–24 groups with threshold values of >14,180, >0 to ≤14,180, and 0 ng · h/ml, respectively.

(ii) Multivariable analyses.

A summary of the P values and directional assessments for the relationships between the efficacy endpoints and OC AUC_0–24, adjusted for IC₅₀ based on multivariable linear regression or Cox regression models, is provided in Table 5. For each efficacy endpoint, AICc values and the nature of each exposure-response relationship for OC AUC_0–24 evaluated in a given form were assessed to discriminate among candidate models. For models for which AICc values were statistically indistinguishable (e.g., 458.6 versus 459.8 for composite symptom score and 603.8 versus 604.3 for time to cessation of viral shedding), a final multivariable model was selected based upon consideration of which model was more informative of the nature of the exposure-response relationship (e.g., the model evaluating AUC_0–24 as a 3-group variable would take precedence over that evaluating AUC_0–24 as a 2-group variable). The multivariable models that were considered to be the final and hence most informative are represented by the bold P and AICc values.

Table 5.

Summary of P values and directional assessments for multivariable relationships between efficacy endpoints and OC AUC_0–24^a

Efficacy endpoint	Directionality and P value or AICc for the form of OC AUC0–24 evaluatedb
	2-group		3-groupc		Quartile		Continuous
	Directionality and P value	AICc	Directionality and P value	AICc	Directionality and P value	AICc	Directionality and P value	AICc
Composite symptom score AUC	/, <0.001	458.6	/, 0.003	459.8	/, 0.08	466.8	/, 0.029	465.1
Viral titer AUC	/, 0.004	342.0	/, 0.009	342.5	N, 0.042	345.5	/, 0.06	346.9
Peak viral titer	/, 0.033	99.1	∩, 0.022	97.6	0.12	101.2	0.15	101.6
Time to alleviation of composite symptom score	/, <0.001	381.9	/, <0.001	379.6	/, 0.036	388.4	/, 0.002	384.1
Time to cessation of viral shedding	/, 0.026	603.8	/, 0.033	604.3	0.13	607.4	0.11	606.8

^aSummary of P values and directional assessments for multivariable relationships between efficacy endpoints and OC AUC_0–24 evaluated as a continuous, 2-group, 3-group, and quartile categorical variables, adjusted for IC₅₀.

^bAs described in Materials and Methods for multivariable analyses, P values for multivariable relationships for continuous efficacy endpoints were based on the Wald test for single parameters (i.e., 2-group and continuous forms of an independent variable) and likelihood ratio test for multiple parameters (3-group and quartile forms of an independent variable). Directionality is reported only for P values of ≤0.10. Directionality for the OC AUC_0–24 variables is reported as follows: ∩, the drug effect was highest for the middle OC AUC_0–24 group compared to the low and high OC AUC_0–24 groups; /, the drug effect increased as OC AUC_0–24 increased; N, the drug effect was lowest with the first and third OC AUC_0–24 quartiles and highest with the second and fourth OC AUC_0–24 quartiles. The relationships between the efficacy endpoints and OC AUC_0–24 have been adjusted for IC₅₀.

^cThe multivariable models that were considered to be the most informative and thus, final, are represented by the bold P and AICc values.

While exposure-response relationships were apparent for composite symptom score AUC, time to alleviation of composite symptom score, and time to cessation of viral shedding, such relationships were not apparent for viral titer AUC and weak for peak viral titer endpoints. Specifically, relationships between viral titer AUC and OC AUC_0–24 evaluated as a 2- or 3-group categorical variable were due to the contrasts between placebo- versus drug-receiving subjects, while the relationship with OC AUC_0–24 as a quartile was N shaped. The relationship between peak viral titer and OC AUC_0–24 evaluated as a 2-group categorical variable was only marginally significant (P = 0.049), while that for OC AUC_0–24 evaluated as a 3-group categorical variable was ∩ shaped; no discernible association between peak viral titer and OC AUC_0–24 evaluated as a continuous or quartile variable was evident. Thus, further consideration was not given to the multivariable models for viral titer AUC and peak viral titer.

A summary of those models for composite symptom score AUC, time to alleviation of composite symptom score, and time to cessation of viral shedding that were considered to be most informative and thus, final, is provided in Table 6. For each model shown, OC AUC_0–24 was evaluated as a 3-group variable. While IC₅₀ was not significant in any of these models, the previously described relationship between IC₅₀ and OC AUC_0–24 warranted its retention as a model covariate.

Table 6.

Summary of the final multivariable models for continuous or time-to-event efficacy endpoints containing OC AUC_0–24 evaluated as a 3-group variable^a

Efficacy endpoint	Independent variableb	Reference group	Comparison group	Parameter estimate for continuous efficacy endpoint or hazard ratio estimate for time-to-event efficacy endpoint (95% CI)	P value
Composite symptom score AUC	IC50 (nM)	0.18	16.76	1.77 (−1.31, 4.85)	0.257
	AUC0–24 (ng · h/ml)	≤1,495	>1,495 to 14,497	−5.36 (−8.73, −1.99)	0.0021
		≤1,495	>14,497	−7.03 (−11.8, −2.27)	0.0042
		>1,495 to 14,497	>14,497	−1.68 (−5.91, 2.56)	0.434
Time to alleviation of composite symptom score	IC50 (nM)	0.18	16.76	0.82 (0.46, 1.45)	0.494
	AUC0–24 (ng · h/ml)	≤1,568	>1,568 to 13,638	1.85 (1.01, 3.39)	0.045
		≤1,568	>13,638	5.34 (2.37, 12.07)	<0.001
		>1,568 to 13,638	>13,638	2.89 (1.43, 5.85)	0.003
Time to cessation of viral shedding	IC50 (nM)	0.18	16.76	0.66 (0.38, 1.14)	0.136
	AUC0–24 (ng · h/ml)	0	>0 to 14,180	1.72 (0.98, 3.03)	0.059
		0	>14,180	2.42 (1.20, 4.84)	0.013
		>0 to 14,180	>14,180	1.40 (0.80, 2.46)	0.240

^aSummary of the final multivariable linear regression or Cox regression for continuous or time-to-event efficacy endpoints containing OC AUC_0–24 evaluated as a 3-group variable.

^bPairwise comparisons were made between the comparison and reference groups.

The comparison of box plots and stratified Kaplan-Meier curves for subject cohorts defined by independent variables based on the above-described final multivariable models is shown in Fig. 3. The graphic illustrations provided in Fig. 3 demonstrate the influence of OC AUC_0–24 group and IC₅₀ on composite symptom score AUC (Fig. 3A) and time to alleviation of composite symptom score (Fig. 3B) and time to cessation of viral shedding (Fig. 3C).

Fig 3.

Composite symptom score AUC (A), time to alleviation of composite symptom score (B), and time to cessation of viral shedding (C) by subject cohorts defined by independent variables based on the final multivariable models containing OC AUC_0–24 (ng · h/ml) evaluated as a 3-group variable.

Evaluation of the interactions between OC AUC_0–24 group and IC₅₀ failed to show any significance for each of the three models described above. As shown by the comparison of composite symptom score AUC among the six cohorts of subjects defined by two IC₅₀ and three OC AUC_0–24 groups in Fig. 3A, the lack of influence of IC₅₀ was evident by the same pattern of improved efficacy among subjects with higher OC AUC_0–24 values. Similarly, as shown by the Kaplan-Meier curves for six cohorts of subjects defined by the two IC₅₀ and three OC AUC_0–24 groups in Fig. 3B, earlier times to alleviation of composite symptom score were evident among subjects in higher OC AUC_0–24 groups, irrespective of the IC₅₀. Last, as shown by the Kaplan-Meier curves for the six cohorts of subjects defined by the two IC₅₀ and three OC AUC_0–24 groups in Fig. 3C, earlier times to cessation of viral shedding were also evident among subjects in higher OC AUC_0–24 groups, irrespective of the IC₅₀ value.

DISCUSSION

The results of the analyses described herein represent the first robust exploration of exposure-response relationships for efficacy for an anti-influenza NAI using clinical data. Prior development of a population PK model enabled OC exposures to be determined for all subjects across two phase 2 studies, regardless of whether sparse or intensive PK samples were collected.

Despite the broad dose range in the two phase 2 studies, the limited number of schedules (i.e., dosing intervals) studied led to highly correlated exposure measures. Thus, it was not possible to evaluate which exposure, AUC_0–24, C_min, or C_max, was most associated with efficacy. Given the preclinical findings derived from HFIM investigations of NAIs (3, 4), AUC_0–24 was chosen for evaluation in the PK-PD analyses carried out.

A two-stage evaluation for exposure-response relationships was then performed. The first stage involved the conduct of univariable analyses, with the primary objective to examine the relationship between each efficacy endpoint and OC AUC_0–24. The second stage of evaluation included multivariable analyses, for which the impact of IC₅₀ (which also represented “study” or “influenza strain”) on each efficacy endpoint was also considered.

The key findings of the univariable analyses suggested the existence of biologically plausible and highly statistically significant relationships between OC AUC_0–24 evaluated as a 3-group variable with three efficacy endpoints. The optimally determined thresholds for the analyses based on evaluation of the 3-group OC AUC_0–24 were, as might be expected, closely aligned for the clinical endpoints, composite symptom score AUC, and time to alleviation of composite symptom score (∼1,500 and ∼14,000 ng · h/ml). The upper OC AUC_0–24 threshold of 14,000 ng · h/ml was also closely similar to that for time to cessation of viral shedding, thus suggesting that the exposures required for maximal antiviral effect correlate with those required for maximal clinical outcome. Such findings are expected, given that NAI like OC, have highly specific antiviral activity.

Exposure-response relationships were not readily apparent for either peak viral titer or viral titer AUC endpoints. In retrospect, this was not unexpected based on the inoculation study design, as maximal virus titers following inoculation occurred early in therapy and are not primarily a function of drug exposure. Unlike time to cessation of viral shedding, the viral titer AUC is heavily influenced by peak virus titer. Thus, both peak virus titer and viral titer AUC are suboptimal efficacy endpoints for PK-PD evaluations when using data from an influenza inoculation study.

The observation that AUC_0–24 beyond ∼14,000 ng · h/ml is associated with greater efficacy supports the rationale for investigating whether doses higher than the approved oseltamivir 75-mg BID dosing regimen, the average AUC_0–24 for which is ∼6,000 ng · h/ml, will provide added benefit (2). This finding was consistent with observations from a small study comparing two different oseltamivir dosing regimens, 75 mg BID and 225 mg BID, in patients infected with pandemic H1N1 (pH1N1), from which provisional results suggest cessation of viral shedding at day 5 occurred in 75% of subjects for the higher dose versus 12% for the standard dose (P = 0.012) (M. A. Kumar, personal communication).

The findings described above are in contrast to the outcomes of two other studies in which 75 mg oseltamivir BID and 150 mg oseltamivir BID were compared in patients with seasonal influenza (15, 16) and findings from another study evaluating approved versus double dosing of oseltamivir in avian and severe influenza (17). These three studies failed to demonstrate superiority of the higher dose versus the approved regimens. However, these studies were not designed to test whether exposure-response relationships existed beyond any apparent lack of a dose-response relationship. We also suspect, that by virtue of the study designs, the resultant PK exposures in patients in each dose group may not have been sufficiently different given only a 2-fold difference in doses investigated. Thus, this may have effectively reduced the “power” of a study to detect a difference in efficacy by dose and may be the basis for why Kumar et al. appear to have demonstrated a difference between doses with fewer patients but a greater dose (and thus exposure) separation (225 mg versus 75 mg BID) (M. A. Kumar, personal communication).

The second stage of evaluation included the conduct of multivariable analyses for efficacy endpoints. The primary purpose of these analyses was to explore relationships between efficacy endpoints and both OC AUC_0–24 and IC₅₀. The final multivariable models considered to be most informative for the three efficacy endpoints, composite symptom score AUC, time to alleviation of composite symptom score, and time to cessation of viral shedding, were those for which OC AUC_0–24 was evaluated as a 3-group variable. It was noted that some of the pairwise comparisons between OC AUC_0–24 groups were nonsignificant. Nonetheless, given the arrays of the significant comparisons across all the endpoints, the consistency of model estimates with increasing efficacy, and especially, the significant differences across each pairwise comparison for time to alleviation of composite symptom score, the available evidence is supportive of the existence of an exposure-response relationship for efficacy of oseltamivir.

Among the above-described final multivariable models, no strong evidence was seen that would suggest IC₅₀ (i.e., a surrogate for study name or virus type) had influence on efficacy endpoints. This was an unexpected finding even in the inoculation study setting, as it suggests that the AUC_0–24 thresholds were not impacted by influenza A/Texas versus influenza B/Yamagata or the ∼100-fold range difference between the IC₅₀ values (0.18 to 16.76 μM). Further exploration in natural influenza infection is needed to explore the impact of IC₅₀ and influenza virus strain and subtype on the OC exposure-response relationships for efficacy.

There are several limitations to the study data utilized for the analyses described herein. These analyses represent a retrospective pharmacometric assessment of pooled data from phase 2 inoculation studies. Data were from experimental pharmacology studies and may not represent efficacy in natural influenza infection. Nevertheless, the highly controlled nature of these types of studies enhances the ability to explore exposure-response relationships and generate hypotheses to optimize dosing regimens for further study in the setting of natural influenza.

The results of these analyses, which utilized data from healthy subjects in two phase 2 inoculation studies, provided the first demonstration of the existence of exposure-response relationships for the efficacy of oseltamivir against influenza. These findings also suggest that OC exposures beyond those that are achieved with the approved oseltamivir dosing regimen will provide enhanced efficacy. The confounded factors of this study, influenza strain and IC₅₀ did not influence efficacy endpoints. The clinical applicability of these observations requires further investigation but may have important implications for the future management of influenza.