PHARMACOKINETICS Population pharmacokinetics and exposureresponse of osimertinib in patients with nonsmall cell lung cancer

Transcription

1 British Journal of Clinical Pharmacology Br J Clin Pharmacol (2017) PHARMACOKINETICS Population pharmacokinetics and exposureresponse of osimertinib in patients with nonsmall cell lung cancer Correspondence Dr Karthick Vishwanathan, 35 Gatehouse drive, Waltham, MA 02451, USA. Tel.: ; Fax: ; karthick.vishwanathan@astrazeneca.com Received 7 September 2016; Revised 8 December 2016; Accepted 18 December 2016 Kathryn Brown 1, *, Craig Comisar 2, Han Witjes 2,JohnMaringwa 2,RikdeGreef 2, Karthick Vishwanathan 3, Mireille Cantarini 4 and Eugène Cox 2 1 UCB, Slough, UK, 2 Quantitative Solutions, a Certara company, Breda, The Netherlands, 3 AstraZeneca, Waltham, USA, and 4 AstraZeneca, Alderley Park, UK *Current affiliation, at time of study was AstraZeneca, Alderley Park, UK. Keywords Drug safety, Patient safety, Pharmacokinetics, Pharmacodynamics, Modelling and Simulation AIMS To develop a population (pop) pharmacokinetic (PK) model for osimertinib (AZD9291) and its metabolite (AZ5104) and investigate the exposure response relationships for selected efficacy and safety parameters. METHODS PK, safety and efficacy data were collected from two non-small cell lung cancer (NSCLC) patient studies (n=748) and one healthy volunteer study (n=32), after single or multiple once-daily dosing of mg osimertinib. Nonlinear mixed effects modelling was used to characterise the poppk. Individual exposure values were used to investigate the relationship with response evaluation criteria in solid tumours (RECIST 1.1) efficacy parameters and key safety parameters (rash, diarrhoea, QTcF). RESULTS A poppk model that adequately described osimertinib and its metabolite AZ5104 in a joint manner was developed. Body weight, serum albumin and ethnicity were identified as significant covariates on PK in the analysis, but were not found to have a clinically relevant impact on osimertinib exposure. No relationship was identified between exposure and efficacy over the dose range studied. A linear relationship was observed between exposure and the occurrence of rash or diarrhoea, and between concentration and QTcF, with a predicted mean (upper 90% confidence interval) increase of 14.2 (15.8) ms at the maximum concentration for an 80 mg once-daily dose at steady state. CONCLUSIONS PopPK and exposure response models were developed for osimertinib and AZ5104. There was no relationship between exposure and efficacy but a linear relationship between exposure and safety endpoints (rash, diarrhoea and QTcF) was observed. DOI: /bcp The British Pharmacological Society

2 Population PK and exposure-response of osimertinib in advanced NSCLC WHAT IS ALREADY KNOWN ABOUT THIS SUBJECT Osimertinib has efficacy in patients with EGFR T790 M mutation positive lung cancer. Osimertinib treatment is associated with adverse events typical of wild-type EGFR inhibition (rash, diarrhoea). Osimertinib plasma concentrations are sustained throughout the dosing interval, which is considered to be optimal for efficacy. WHAT THIS STUDY ADDS Covariates that affect osimertinib PK have been identified (body weight, serum albumin, ethnicity); none have a clinically relevant impact. There was no evidence of an osimertinib PK-efficacy relationship at the dose range studied, with all doses being efficacious. Linear osimertinib PK-safety relationships have been characterised for rash, diarrhoea and QTcF. Tables of Links TARGETS Enzymes [2] EGFR CYP3A4 LIGANDS Osimertinib EGFR Inhibitor These Tables list key protein targets and ligands in this article that are hyperlinked to corresponding entries in the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY [1], and are permanently archived in the Concise Guide to PHARMACOLOGY 2015/16 [2]. Introduction Osimertinib (AZD9291) is a potent irreversible inhibitor of the epidermal growth factor receptor mutation positive (EGFRm; tyrosine kinase inhibitor [TKI] sensitivity-conferring mutations) and dual EGFRm/T790 M mutation positive (T790 M; TKI resistance-conferring mutation) receptor forms of EGFR but designed to have limited activity against wild-type EGFR [3]. Osimertinib is effective in EGFR T790 M mutation positive non-small cell lung cancer (NSCLC) patients who are treatment naïve or have had disease progression during prior therapy with EGFR TKIs [4, 5]. Osimertinib treatment is associated with manageable adverse events (AEs), primarily rash and diarrhoea, typical of wild-type EGFR inhibition [4]. Two metabolites of osimertinib (AZ5104 and AZ7550) were measured in plasma in all clinical studies. Both are formed directly from osimertinib, predominantly via CYP3A4/5. At steady state, both have pharmacokinetic (PK) profiles similar to osimertinib and circulate at ~10% each of osimertinib exposure [6]. The in vitro cell potency in EGFRm and EGFR-wild type [3] suggests AZ5104 may contribute to safety and efficacy, whilst AZ7550 is unlikely to; therefore, AZ7550 was not included in the modelling. A population PK (poppk) analysis was conducted to characterise osimertinib and AZ5104 PK and to assess the impact of patient covariates on exposure. Individual exposure estimates of osimertinib and AZ5104 were then used in exposure response (ER) evaluations. Best overall confirmed objective response, duration of response (DoR), and best percentage change in target lesion size from baseline were assessed in efficacy ER analyses. Occurrence of rash and diarrhoea were assessed in safety ER analyses. The relationship between observed concentrations and QTcF was also assessed. These analyses were performed to support dose selection and to assess whether dose adjustment was required in any subpopulation. The analyses included data from two patient studies: D5160C00001 (AURA, phase I/II, NCT ) and D5160C00002 (AURA2, phase II, NCT ). The poppk also included healthy volunteer (HV) data from D5160C00005 (Study 5, NCT ; Table 1). Subjects and methods All studies were conducted in accordance with the principles of the Declaration of Helsinki and the Good Clinical Practice guidelines of the International Conference on Harmonisation. The protocols were approved by local institutional review boards at each participating site. All subjects provided written informed consent prior to screening. Patients AURA was a phase I/II study in advanced EGFRm positive NSCLC patients; the phase II component (AURA extension) assessed patients who had progressed following treatment with at least one EGFR TKI, with additional regimens permitted. AURA2 was a phase II study in a similar patient population as the AURA extension. The poppk and safety ER analyses included 337 patients from AURA; 143 patients were included in the efficacy ER analyses (i.e. those patients who were confirmed T790 M mutation positive). Patients received osimertinib doses from 20 to 240 mg, primarily as a capsule formulation with one cohort receiving a phase 1 tablet. In the majority of patients, PK samples were collected predose, 1, 1.5, 2, 4, 6, 8, 10, 12, 24, Br J Clin Pharmacol (2017)

3 K. Brown et al. Table 1 Baseline characteristics and doses administered Number of patients by study (AURA, AURA2 and Study 5) and dose Study Dose (mg), number of subjects (%) 20 (n = 53) 40 (n = 58) 80 (n = 552) 160 (n = 96) 240 (n = 21) Total number of subjects (n = 780) AURA escalation 6 (11.3%) 6 (10.3%) 18 (3.3%) 6 (6.3%) 7 (33.3%) 43 (5.5%) AURA expansion 15 (28.3%) 52 (89.7%) 123 (22.3%) 90 (93.8%) 14 (66.7%) 294 (37.7%) AURA extension 0 (0.0%) 0 (0.0%) 201 (36.4%) 0 (0.0%) 0 (0.0%) 201 (25.8%) AURA2 0 (0.0%) 0 (0.0%) 210 (38.0%) 0 (0.0%) 0 (0.0%) 210 (26.9%) Study 5 32 (60.4%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 32 (4.1%) Baseline demographic data (AURA, AURA2 and Study 5) Continuous variables Covariate (units) Mean SD Median Range n Missing Body weight (kg) (0.1%) Age (years) (0.0%) Body mass index (kg m 2 ) (1.2%) Body surface area (m 2 ) (1.2%) Estimated creatinine a 1(0.1%) clearance (ml min 1 ) a Albumin(gl 1 ) (0.6%) ALT (U l 1 ) (0.0%) AST (U l 1 ) (0.1%) Alkaline phosphatase (U l 1 ) (0.1%) Bilirubin (μmol l 1 ) b b 2(0.3%) Categorical variables Covariate Category n % Sex Female Male Study AURA1 c AURA Study Smoking status Never Current Former Ethnicity Caucasian Asian (not Japanese or Chinese) Japanese Chinese Other Missing Population Healthy subjects Non-small cell lung cancer patients Dose level (mg) d,e (continues) 1218 Br J Clin Pharmacol (2017)

4 Population PK and exposure-response of osimertinib in advanced NSCLC Table 1 (Continued) Categorical variables Covariate Category n % Formulation e Capsule Phase 1 tablet Film-coated tablet Solution Food status e Fasted Fed ALT, alanine transaminase; AST, aspartate aminotransferase; n, number of patients; SD, standard deviation. a <30, n =3;30 59, n =149;60 89, n =330; 90, n =295. b <19, n =751;19 28, n =22;29 56, n =3; 57, n =0. c From the 536 AURA1 patients with measurable concentrations there were 337 in the dose escalation and expansion (phase I component) and 199 in the extension (phase II component). d Dose level modelled as a continuous covariate. Distinct dosing levels shown for reference. e Non-patient-specific variable, will not sum to 100%. Sources: m5/datasets/er pooled-1-2/analysis/legacy/datasets/nmer.xpt (data), m5/datasets/er pooled-1-2/analysis/legacy/programs/stud0dose.sas (analysis) and m5/datasets/er pooled-1-2/analysis/legacy/ programs /tab_study_dose.rtf (output) \AZ\9291\Analysis\ 9291ReportScriptfinalv1.RMD and dcov5.mod 48, 72 and 120 h postsingle dose, and predose, 1, 1.5, 2, 4, 6, 8, 10, 12 and 24 h postmultiple once-daily dosing. The analyses included 411 phase II patients, from AURA phase II (AURA extension) and AURA2. In AURA extension and AURA2, all patients received 80 mg osimertinib as a film-coated tablet once daily. PK samples were collected following a single dose (predose, 1, 2, 4, 6 and 8 h postdose) andaftermultipledosing(predose,1,2,4,6,8,10,12and 24 h postdose). AURA2 data were used for concentration-qtc (c-qtc) analyses. Baseline electrocardiogram (ECG) measurements were collected over a 24-h period, to enable time matched baseline calculations, as well as after single and multiple dosing. The following ECG study entry exclusion criteria were applied in the AURA2 study: resting QTcF >470 ms obtained from three ECGs; any clinically important abnormalities in rhythm, conduction or morphology of resting ECG (e.g., complete left bundle branch block, second- or third-degree heart block, or PR interval > 250 ms); any factors that increase risk of QTcF prolongation or risk of arrhythmic events such as heart failure, hypokalaemia, congenital long QT syndrome, family history of long QT syndrome or unexplained sudden death in first-degree relatives aged <40 years, or use of any concomitant medication known to prolong QT interval. The c-qtc analysis included 3187 QTcF values with time-matched PK observations from 210 patients. Healthy volunteers The poppk included one phase I HV study (D5160C00005, 32 subjects), which assessed the relative bioavailability of different oral formulations (solution, capsule and phase 1 tablet) and an exploratory food effect assessment (high fat breakfast). A single 20 mg osimertinib dose was administered. PK samples were collected predose, 0.5, 1, 1.5, 2, 3, 4, 6, 8, 10, 12, 24, 48, 72, 120, 168, 216, 336 and 504 h postdose. Sample collection and quantification Plasma samples were analyzed by Covance UK Ltd (Harrogate, UK) for osimertinib, AZ5104 and AZ7550 simultaneously, after protein precipitation, using a validated high-performance liquid chromatography tandem mass spectrometry method. The lower limit of quantification was 0.05, and nmol l 1 for osimertinib, AZ5104 and AZ7550, respectively. For AURA, accuracy for osimertinib was % andprecision(relativestandarddeviation;rsd) %. For AZ5104 and AZ7550, accuracy was % and RSD %. For Study 5, accuracy for osimertinib was % and RSD %. For AZ5104 and AZ7550, accuracy was % and RSD %. Concentrations below the assay limit of quantification were excluded. A total of plasma concentrations from 780 individuals were included in the final poppk. Model development and evaluation Exploratory graphical analysis, model development, analysis and simulation were performed using R (version 2.15 or above), SAS (version 9.0 or above, SAS Institute Inc., Cary, NC, USA), or NONMEM (version 7.0 or above, ICON, Hanover, MD, USA). PopPK model evaluation was based on successful minimisation and completion of covariance steps in NONMEM, assessment of standard goodness of fit (GOF) plots and prediction-corrected visual prediction checks (pcvpcs), reductions in NONMEM objective function value Br J Clin Pharmacol (2017)

5 K. Brown et al. for hierarchical models and reductions in interindividual (IIV) and residual variability. All models were estimated using the first order conditional estimation method with betweensubject variability and residual variability interaction (FOCE-I) in NONMEM. Base poppk model. Multicompartmental PK models were developed to jointly characterise the PK profile of osimertinib and AZ5104. Based on preliminary clinical PK data, a twocompartment linear structural model was investigated. The IIV model was described as a lognormal distribution. Covariance between IIV parameters was also assessed. A combined proportional and additive residual error model was used. The fraction of osimertinib metabolised to AZ5104 was fixed to an arbitrary value of 25%, to ensure mass balance between parent and metabolite. Since patient-specific random-effects were estimated for clearance of AZ5104, this also accounted for IIV in the rate of metabolite formation. Covariate poppk model. After identification of the base model, covariates were evaluated using a stepwise covariate model process. Potential covariate-parameter relationships were selected based on known or hypothetical factors that could affect PK. Body weight, age, sex, race, alanine transaminase (ALT), creatinine clearance (CrCl) and disease state (NSCLC vs. HV) on parent and metabolite clearance were evaluated. In addition, dose, food state, baseline bilirubin level and smoking status on parent clearance were also evaluated. The impact of body weight and baseline albumin level on parent and metabolite volume of distribution were evaluated, along with dose, formulation, food state and disease state on relative bioavailability. Dose was defined as most recent dose received. Relationships between covariates and PK parameters were examined graphically then tested by forward inclusion (P < 0.01) followed by backward elimination (P < 0.001), evaluated using the likelihood ratio test (LRT). The following models were used for continuous and discrete covariates respectively: Xij θi P j ¼ θ k and P j ¼ θ k 1þ ð θ i MXj ð Þ where Pj is the jth population estimate of parameters, Xij the covariate of patient i for parameter P j, θ k the typical value of parameter P j, θ i a coefficient that reflects the covariate s effect on the parameter, and M(Xj) the median of covariate X for the patient population. The significance of a covariate effect was evaluated with respect to clinical and/or physiological relevance, with any covariate not meeting a 20% relevance threshold being removed from the model (for continuous covariates, 20% was tested against the 5 th and 95 th percentiles, with the exception of dose which was tested at the highest and lowest doses). For race and ethnicity, any subcategories with 30 individuals were included as separate categories. Final poppk model. The final poppk model was used to obtain individual exposure estimates of osimertinib and Þ X ij AZ5104, and to simulate an 80-mg steady-state exposure. In the efficacy analysis, calculations of area under the plasma concentration time curve during any dosing interval at steady state (AUC ss ) were based on most prevalent dose received, since this is likely to be the most accurate reflection of AUC ss experienced during the study. In the safety response analyses, AUC ss was calculated using the first dose, thereby assuming that dose reductions were most likelytobedrivenbyaes. ER models. Separate models were developed for osimertinib and AZ5104 and their GOF with respect to probability of response or AE was compared based on likelihood principles. Since osimertinib and AZ5104 exposures are highly correlated, they were not included simultaneously in one model to avoid problems of multicollinearity. Since the majority of rash and diarrhoea events were grade 1, analyses stratified by grade were not warranted due to insufficient data across different grades. Overall response rate and probability of having rash or diarrhoea were assessed using logistic regression. Linear AUC ss effect, piece-wise linear, maximum drug-induced effect (E max ), and sigmoid E max models were investigated. For a linear effect of AUC ss, the probability of being a responderortheoccurrenceofanae,p i,forpatienti, was related to the individual s exposuremetricaucss i by: p log i ¼ α 0 þ α 1 AUCss i 1 p i where α 0 and α 1 are coefficients to be estimated representing the intercept and the effect (slope) respectively of AUC ss on the log odds of the probability of being a responder or the occurrence of an AE. Nested models were compared using LRT and non-nested models using the Akaike information criteria. GOF based on statistical significance(inthelrt)wasusedtotestformodel effects. Possible confounding risk factors were evaluated by plotting the distribution of prespecified factors stratified by AUC ss quartiles. DoR was explored using Kaplan Meier (KM) curves, stratified by AUC ss quartiles and categories of time to onset of response. The relationship between best percentage change in target lesion size from baseline and AUC ss was assessed graphically. A nonparametric regression approach using a locally weighted scatterplot smoothing (LOESS) function was overlaid on the observed data to evaluate trends. c-qtc model The time-matched change in the QT interval from baseline, corrected for heart rate using Fridericia s formula(δqtcf) was used in the characterisation of the relationship between QTcF interval and observed plasma concentrations: QTcF ¼ QT=RR 1=3 ; with QT expressed in ms, RR in s. At each time point, three digital ECGs were taken. To obtain a single value of QT and RR at each specified time point, the mean of the triplicate values was calculated, which was then used in the c-qtc analyses Br J Clin Pharmacol (2017)

6 Population PK and exposure-response of osimertinib in advanced NSCLC Linear and nonlinear mixed effects models were applied, with separate models developed for osimertinib and AZ5104. Effect of sex and nominal time point was tested. Boxplots of single dose concentrations and ΔQTcF against time were constructed to evaluate potential presence and magnitude of a delayed effect (hysteresis). Relative performance of the models was evaluated using Akaike information criteria and parameter estimate precision. The following linear model was applied: ΔQTcF ij ¼ θ 1 þ η 1;i þ θ2 þ η 2;i Cij þ ε ij where ΔQTcF ij is the time-matched baseline-adjusted QTcF interval for patient i at time j with plasma concentration C ij, θ 1 and θ 2 are the population mean intercept and slope, respectively; η 1,i and η 2,i are the intercept and slope for patient i, respectively; ε is the residual error. The final model was used to predict the expected ΔQTcF at the maximum steady-state concentration (C ss,max )for80mg osimertinib once-daily, together with its associated two-sided 90% confidence interval (CI). Results Model development The final base model comprised two compartments (one for osimertinib followed by a second for AZ5104). Oral absorption of osimertinib from the depot site into the central compartment was modelled as a first-order process. Based on simulations, the typical NSCLC population values of osimertinib AUC ss,c ss,max and the minimum steady-state concentration (C ss,min )foran80-mgdosewere11258nmoll 1.h, 501 nmol l 1 and 417 nmol l 1 respectively. For AZ5104 AUC ss, C ss,max,andc ss,min were 1271 nmol l 1.h,56nmoll 1 and 52 nmol l 1. PK was dose proportional and time independent across the mg dose range. In the final poppk model, the typical (standard error) value of clearance, volume of distribution and t 1/2 for osimertinib were 14.2 (0.3) l h 1,986(28) l, and 48 (2) h, respectively. For AZ5104 the typical values of clearance and volume of distribution were 31.5 (0.9) l h 1 and 207 (9) l. These typical values were affected by specific covariates in the analysis population. Covariate analysis The final poppk model included the parameter covariate relationships presented in Table 2. The largest effect was related to body weight. To investigate this effect further, the relative change of AUCss was calculated for the 5 95% range of body weight in the analysis data set (43 90 kg, Table 1). Within this body weight range, osimertinib AUC ss ranged from 20 to +30% compared to the median body weight of 62 kg, and from 40 to +50% for AZ5104 AUCss. Stepwise covariate model identified several effects of race/ethnicity on the apparent clearance of AZ5104. For all ethnic classes tested (Chinese, Japanese, non-chinese/non- Japanese Asian, non-asian non-caucasian), a decrease of AZ5104 AUCss of ~10 23% compared to Caucasian patients was predicted. There was no effect of race/ethnicity on the apparent clearance of osimertinib. There was no evidence of a difference in PK between the four formulations (capsule, solution, Phase 1 tablet and film-coated tablet) used across the studies. Similarly, age, sex and smoking status had no impact on the PK. The analysis indicated only minimal effects of surrogate markers of hepatic function on PK of osimertinib. Although an effect of ALT on apparent clearance of AZ5104 was identified, this effect was not considered clinically relevant (<20%) and was thus not included in the final model. Other markers of liver function (aspartate aminotransferase) or renal function (CrCl) were not significant. Serum albumin, an indicator of hepatic synthetic function, impacted osimertinib volume of distribution and was therefore maintained in the final model, however there was no clinically relevant impact on AUCss due to serum albumin. A difference in apparent clearance between NSCLC patients and HV was identified, which resulted in ~35% and ~55% lower AUCss values in HV for osimertinib and AZ5104, respectively. GOF plots for the final population PK model indicated the individual predictions fitted well along the identity line relative to observed concentrations, and the conditionally weighted residuals were well distributed along the zero line relative to population predictions (Figure 1 presents osimertinib; AZ5104 is presented in Supplementary Figure S1). There was some misspecification of concentrations related to times >24 h but this was not considered critical as the main objective was to estimate steady-state exposures for the 24-h dosing period. The pcvpcs of the final poppk model indicated the model describes the observed data well, and model predictions were generally within 90% prediction intervals (Supplementary Figure S2). The parameters of the final model (Table 2) were estimated with sufficient precision, with relative standard error typically <10% for structural model parameters and <20% for covariate effects. Shrinkage was small ( 5%) for clearance and higher (20 40%) for volume of distribution and absorption rate constant, which is not unexpected given thesparsedata. ER-efficacy There was no evidence of a relationship between exposure and probability of objective response, DoR or best percentage change in target lesion size (Figure 2 presents osimertinib; AZ5104 is presented in Supplementary Figure S3). Logistic regression indicated a linear effect of AUC ss and did not give a statistically significant improvement in model fit when compared to constant probability (intercept only; P=0.29, P=0.15, respectively). As the KM analysis indicated no clear trend in the relationship between DoR and AUC ss, this relationship was not pursued further. Similarly, as the graphical exploration of the best percentage change in target lesion size with AUC ss suggested no relationship, further evaluation was not pursued. ER-safety The probability of a patient experiencing rash or diarrhoea increased with exposure (Figure 3). A linear effect of log AUC ss Br J Clin Pharmacol (2017)

7 K. Brown et al. Table 2 Final model parameter estimates Parameter (units) Estimate %RSE IIV% (%RSE) NONMEM model estimates CLparent/F (l h 1 ) (3.2) Vparent/F (l) (3.3) ka (1 h 1 ) (6.1) CLmetabolite/F (l h 1 ) (5.3) Vmetabolite/F (l) (3.3) Healthy subject population effect (vs. NSCLC patients) on CLparent/F Effect of body weight on CLparent/F Ethnic Asian other population effect on CLmetabolite/F Ethnic Asian Chinese population effect on CLmetabolite/F Ethnic Asian Japanese population effect on CLmetabolite/F Ethnic non-asian non-caucasian population effect on CLmetabolite/F Healthy subject population effect (vs. NSCLC patients) on CLmetabolite/F Effect of body weight on CLmetabolite/F Effect of albumin on Vparent/F Effect of body weight on Vparent/F ηclparent ηclparent-ηclmetabolite covariance a η ηclmetabolite ηka ηvparent ηvmetabolite Additive error (g l 1 ) Proportional error (%) Shrinkage (%) CLmetabolite, apparent clearance of AZ5104; CLparent, apparent clearance of osimertinib; F, relative bioavailability; IIV, interindividual variability, ka absorption rate constant; NONMEM, nonlinear mixed effects modelling software; NSCLC, non-small cell lung cancer; RSE, relative standard error; Vmetabolite, apparent AZ5104 volume of distribution, Vparent apparent osimertinib volume of distribution. a Correlation coefficient (r) between random effects of CLparent and CLmetabolite =0.90. Sources: \AZ\9291\Analysis\dcov5.lst Source: \AZ\9291\Analysis\ dcov5.mod best described the data. At the mean osimertinib AUC ss, predicted probabilities of having rash (95% CI) for patients who received 20, 40, 80, 160 and 240 mg osimertinib were 0.28 ( ), 0.36 ( ), 0.46 ( ), 0.55 ( ) and 0.60 ( ), respectively. For diarrhoea, the predicted probabilities (95% CI) were 0.29 ( ), 0.37 ( ), 0.45 ( ), 0.54 ( ) and 0.58 ( ), respectively. A comparison of the model-predicted probabilities to observed probabilities indicated that the model captured the trend in the data well. c-qtc relationship A linear relationship between ΔQTcF and osimertinib concentrations (P < ) was identified(figure4).themean increase in ΔQTcF was ms per 10 nmol l 1 increase in osimertinib concentration with a 2-sided 90% CI of ms. Nominal time after dose had a significant effect on ΔQTcF and was included in the model (P < ). Sex did not have a significant effect (P = 0.97). No hysteresis was observed. The observed osimertinib C ss,max at the 80 mg approved dose was 525 nmol l 1.Osimertnibshowsaflat PK profile at 1222 Br J Clin Pharmacol (2017)

8 Population PK and exposure-response of osimertinib in advanced NSCLC Figure 1 General goodness-of-fit for osimertinib from the final population pharmacokinetics model (LOESS smoothing line as solid line) Figure 2 Observed response probabilities (with 95% confidence interval as vertical bars) and model prediction based on osimertinib AUC ss. AUC ss = area under the concentration time curve at steady-state conditions for a 24-h dosing interval. Note: The points are observed response probabilities in categories of osimertinib AUC ss, created from 8 quantiles of AUC ss (with probability = 0.125); the continuous line is the model prediction. The right panel shows the model-prediction including 95% confidence interval on the prediction as dashed lines. The vertical lines in both panels show the mean (5 th 95 th percentiles) for osimertinib AUC ss in patients receiving 80-mg osimertinib; ( ) nmol l 1.h Br J Clin Pharmacol (2017)

9 K. Brown et al. Figure 3 Observed probabilities and model prediction of rash (top panel) and diarrhoea (bottom panel). AUC ss = area under the plasma concentration time curve during any dosing interval at steady state; CI = confidence interval. Note: Categories of osimertinib AUC ss were created from 8 quantiles of AUC ss (with equal probability = 0.125); the continuous line is the model prediction. The right panel shows the model prediction including 90% CI (dashed lines) on the prediction. The vertical lines show the mean (5 th 95 th percentiles) of osimertinib AUC ss in patients receiving 80 mg osimertinib: ( ) nmol l 1.h steady state with a C ss,max to C ss,min ratio of 1.6 [6]. Applying the linear mixed effects model, the predicted mean ΔQTcF at this C ss,max concentrationis14.2mswithanupperboundof 15.8 ms (two-sided 90% CI). Discussion Figure 4 Scatterplot of ΔQTcF vs. plasma concentration of osimertinib with the fitted regression line obtained with the linear mixed effects model This study is the first population-based approach to evaluate osimertinib PK and has enabled assessment of ER relationships for key efficacy and safety endpoints. The positive clinical activity observed with osimertinib from the first cohort of patients meant the first time in a patient study (AURA) recruited patients rapidly, enabling relatively rich data collection in a substantial number of patients across the mg dose range. Two phase II studies (AURA extension and AURA2) provided further data at 80 mg and an investigation into potential effects on QTcF. Challenges associated with model development included presence of an active metabolite, addressed by developing a joint poppk model for parent and metabolite. The conversion to metabolite was fixed at 25%, since it was not possible to determine the actual value from the data collected, probably 1224 Br J Clin Pharmacol (2017)

10 Population PK and exposure-response of osimertinib in advanced NSCLC as a result of high correlation between individual plasma concentration time profiles of osimertinib and AZ5104. This may be an overestimation of the fraction of osimertinib metabolised to AZ5104, since plasma exposure of AZ5104 is ~10% of osimertinib plasma exposure [6]; however, since the fraction metabolised is a scaling factor estimated jointly with all other parameters in the model, the value selected does not impact on the goodness of fit of the data, nor model-based simulations of osimertinib and AZ5104 concentrations. The finalpoppkmodelwascomprisedoffirst order oral absorption of osimertinib followed by two compartments in series: one for osimertinib followed by a compartment for AZ5104. The primary predictors of PK variability were body weight, baseline serum albumin level and ethnicity. A 20% to +30% change in osimertinib AUC ss (compared to the AUC ss for the 62 kg median body weight) would be expected across a body weight range of kg, and a 40% to +50% change for AZ5104 AUC ss.adecreasein AZ5104 AUC ss of 10 23% versus Caucasian patients may be expected in Chinese, Japanese, non-chinese/non- Japanese Asian and non-asian non-caucasian patients; the reason for this is unclear since osimertinib and AZ5104 are predominantly metabolized by CYP3A [7]; therefore, there is minimal likelihood that metabolism would be impacted by genetic polymorphisms. Age, sex, formulation and smoking status had no impact on PK of osimertinib or AZ5104. Renal and hepatic function would not be expected to impact exposure, based on assessment of surrogate markers of liver (aspartate aminotransferase, ALT or bilirubin) and renal (CrCL) function. A lower clearance was observed in NSCLC patients compared to HV, leading to lower exposure in HV compared to patients at the same dose; the reason for this was not further evaluated; however, this is not uncommon in oncology [8]. The probability of response did not change with osimertinib or AZ5104 exposure over the mg dose range. Similarly, there was no evidence of a relationship between exposure and DoR or best percentage change in target lesion size from baseline, although DoR data were immature for patients participating in the phase II studies at the time of this analysis. Similarly, the data were too immature to assess progression free survival or overall survival. Since all doses studied were efficacious, the lack of an ER relationship could be because the maximum response has been achieved at this exposure range, with evidence of a lower magnitude of drug effect at lower exposures essentially absent. Whilst there were no obvious efficacy ER relationships, safety ER relationships were apparent. The probability of developing rash or diarrhoea, AEs typical of wild-type EGFR inhibition [4], increased with exposure and a linear relationship between ΔQTcF and osimertinib concentrations was identified. The rash and diarrhoea effects at the approved 80 mg dose were considered limited compared to alternative treatments [4] and quantification of the predicted ΔQTcF effect provides valuable information for prescribers. The integrated pharmacometric approach presented in this paper has enabled an optimally informed benefit risk assessment for the use of osimertinib in the treatment of patients with EGFR T790 M mutation positive NSCLC. A large therapeutic window is observed with osimertinib, with clinical activity evident at all doses studied and no maximum tolerated dose defined. Whilst covariates that affect osimertinib PK were identified, none would be expected to increase exposure >2-fold (greater than that of a 160 mg dose) or decrease below 50% (lower than that of a 40 mg dose), and therefore there were no covariates identified that would necessitate dose adjustments. These analyses support the conclusion that a once-daily dose of 80 mg, the dose approved in the USA, EU and Japan [9 11], provides a positive benefit risk profile that maximises clinical activity whilst minimizing the probability of adverse reactions across different population based covariates. Competing Interests All authors have completed the Unified Competing Interest form at (available on request from the corresponding author). K.B., K.V. and M.C. were AstraZeneca employees at the time the study was conducted. K.V. and M.C. reported shares in AstraZeneca. H.W., R.d.G. and E.C. received a grant from AstraZeneca as employees of Quantitative Solutions. This research was sponsored by AstraZeneca. C.C. and J.M. declared no competing interests. We thank the investigators and patients from the studies that provided data that informed the development of this model. Contributors Conception and design of the work: All authors. Collection of clinical data: M.C. Analysis and interpretation of the data: K.B., C.C., H.W., J.M., R.d.G., E.C. Drafting or revising the manuscript: K.B. Final approval of the manuscript: All authors. References 1 Southan C, Sharman JL, Benson HE, Faccenda E, Pawson AJ, Alexander SP, et al. The IUPHAR/BPS Guide to PHARMACOLOGY in 2016: towards curated quantitative interactions between 1300 protein targets and 6000 ligands. Nucl Acids Res 2016; 44: D Alexander SPH, Kelly E, Marrion N, Peters JA, Benson HE, Faccenda E, et al. TheConciseGuidetoPHARMACOLOGY 2015/16: Enzymes. Br J Pharmacol 2015; 172: Cross DA, Ashton SE, Ghiorghiu S, Eberlein C, Nebhan CA, Spitzler PJ, et al. AZD9291,anirreversibleEGFRTKI,overcomes T790 M-mediated resistance to EGFR inhibitors in lung cancer. Cancer Discov 2014; 4: Jänne PA, Yang JC, Kim DW, Planchard D, Ohe Y, Ramalingam SS, et al. AZD9291 in EGFR inhibitor-resistant non-small-cell lung cancer. N Engl J Med 2015; 372: Ramalingam S, Yang JCH, Lee C, Kurata T, Kim DW, John T, et al. AZD9291 in treatment-naive EGFRm advanced NSCLC: AURA first-line cohort. Presented in mini oral session, WCLC Br J Clin Pharmacol (2017)

11 K. Brown et al. Denver 6 9 September J Thorac Oncol 2015; 10 (9,Suppl 2): S319. MINI PlanchardD,BrownKH,KimDW,KimSW,OheY,FelipE,et al. Osimertinib Western and Asian clinical pharmacokinetics in patients and healthy volunteers: implications for formulation, dose, and dosing frequency in pivotal clinical studies. Cancer Chemother Pharmacol 2016; 77: Dickinson PA, Cantarini MV, Collier J, Frewer P, Martin S, Pickup K, et al. Metabolic disposition of osimertinib in rats, dogs and humans: insights into a drug designed to bind covalently to a cysteine residue of epidermal growth factor receptor. Drug Metab Dispos 2016; 44: Coutant DE, Kulanthaivel P, Turner PK, Bell RL, Baldwin J, Wijayawardana SR, et al. Understanding disease-drug interactions in cancer patients: implications for dosing within the therapeutic window. Clin Pharmacol Ther 2015; 98: Japan MHLW TAGRISSO (osimertinib) prescribing information. TAGRISSO 40 mg and 80 mg tablets. In, FDA TAGRISSO (osimertinib) prescribing information. Highlights of prescribing information: TAGRISSO (osimertinib) tablet, for oral use. Available at drugsatfda_docs/label/2015/208065s000lbl.pdf. (last accessed 20 January 2016). _Product_Information/human/004124/WC pdf. (last accessed 20 January 2016). Supporting Information Additional Supporting Information may be found in the online version of this article at the publisher s web-site: Figure S1 General goodness-of-fit for AZ5104 from the final population pharmacokinetic model (LOESS smoothing line as solid line) Figure S2 Prediction-corrected visual predictive checks (pcvpc) for the final a population pharmacokinetic model; (A) following single dose and (B) within a steady-state interval (24 h since last dose) Figure S3 Observed response probabilities (with 95% CI as vertical bars) and model prediction based on AZ5104 area under the plasma concentration time curve during any dosing interval at steady state 11 EMA TAGRISSO (osimertinib) Summary of Product Characteristics. European Medicines Agency. Available at Br J Clin Pharmacol (2017)