Development of a Population-Based Pharmacokinetic

Development of a population-based pharmacokinetic- pharmacodynamic model to simulate treatment efficacy in metastatic non-small-cell lung cancer patients on erlotinib

By Michelle Lui

A thesis submitted in conformity with the requirements for the degree of Master of Sciences

Pharmaceutical Sciences University of Toronto © Copyright by Michelle Lui (2015)

Development of a population-based pharmacokinetic- pharmacodynamic model to simulate treatment efficacy in metastatic non-small-cell lung cancer patients on erlotinib

Michelle Lui

Master of Sciences

Pharmaceutical Sciences University of Toronto

2015 Abstract

Erlotinib improves progression-free survival (PFS) in metastatic non-small-cell lung cancer

(mNSCLC) patients, but cause toxicities leading to dose reductions. The survival impact of dose reductions is unknown. We constructed a population-based pharmacokinetic-pharmacodynamic

(PK-PD) erlotinib model predicting PFS of mNSCLC patients, consisting of an erlotinib population PK component, a PD component describing the concentration-kill-constant relationship, and a tumour growth component tracking tumour size over time. This model was fit against clinical trial data to simulate population-based PFS estimates in mNSCLC patients and externally validated using other trial data. The model simulated population-based PFS estimates of placebo and erlotinib arms with 0.74-9.14% and 0.26-19.83% error, respectively. The model was externally validated with population-based PFS estimate errors of 11.3% and 16.7% in epidermal growth factor receptor (EGFR)-mutant and EGFR-wild-type mNSCLC, respectively.

It predicted significant PFS decreases with dose reductions. This is the first population-based

PK-PD erlotinib model that predicts PFS in mNSCLC patients.

Acknowledgments

This project would not have been possible without the help of many individuals. I would like to personally thank each and every patient for whom I have had the pleasure and privilege to provide care. Without their willingness to have me as their clinic pharmacist, I would not have developed my passion or knowledge in lung cancer management. As a tribute to them, this project is dedicated to all of my current and future lung cancer patients and their family members.

The true masterminds of this model are Dr. Carlo De Angelis, my supervisor, and Scott Walker, who have taken me on as their Master’s student and assumed supervisory roles in my journey at Sunnybrook. Their patience, time, knowledge, mentorship and foresight have allowed me to grow throughout these two years as a clinical researcher. Carlo has been most inspirational as my role model in my growth as an oncology pharmacist and researcher. It is difficult to put into words the enormity of time and mentorship he has given me throughout these last two years. He has taught me the true meaning of being a lifelong learner in this ever-growing field, has fueled my passions in constantly improving patient outcomes and patient-centered care and has provided me ample opportunities to improve my research, clinical, presentation and grant-writing skills. He’s been a thoughtful and caring teacher, mentor and savior throughout the years in my budding career and has been very influential in my growth as a person. Scott’s passion for pharmacokinetic-pharmacodynamic modeling was truly contagious and it provided a key opportunity for me to merge two areas that I’ve always been passionate in – lung cancer management and mathematics. I am extremely grateful for the time that he takes out of his busy schedule to be so involved in this project. I am deeply honoured to have had the opportunity to work alongside both individuals and will cherish any future opportunities to work with either individual.

I would like to give my thanks to Dr. Sunil Verma who took me into the thoracic clinic, which became my clinical home for two years and educational base where I learned about how to provide care for lung cancer patients. He allowed me to become part of his clinical team and eventually as part of the entire thoracic oncology team. I appreciated the time that he took to attend my advisory committee meetings and provide insight about the project. I would also like to thank the entire thoracic oncology team, particularly Dr. Jacques Raphael, Dr. Suneil Khanna, iii

Dr. Yee Ung, Dr. Parneet Cheema, Dr. Susanna Cheng, Magdalene Winterhoff and the OCC Lung Nurses, for allowing me to be part of the clinical team and provide care for these lovely but unfortunate patients, as well as teaching me anything that I wanted to learn to refine my clinical skills as a pharmacist, care provider for their patients and team player in the interprofessional group.

Last but most definitely not least, I would also like to thank the Pharmacy Department at the Odette Cancer Centre. They have allowed me to make my place as one of the oncology pharmacists in the team and provided every opportunity for me to babble on clinical pearls that were related to lung cancer management. It would be my great privilege and honour to continue working with such an esteemed group of clinicians beyond my journey as a Master’s student and Pharmacy Oncology Fellow.

Table of Contents

Abstract ------ii

Acknowledgements ------iii

Table of contents ------v

Glossary of Abbreviations ------viii

List of tables ------x

List of figures ------xii

Chapter 1 – Introduction ------1

Chapter 2 – Literature review ------5

2.1 Overview of non-small-cell lung cancer ------6

2.1.1 Epidemiology ------6

2.1.2 Risk factors and pathogenesis of non-small-cell lung cancer ------6

2.1.3 EGFR mutations ------8

2.1.4 Role of erlotinib in standard treatment of metastatic non-small-cell 8 lung cancer

2.1.5 Efficacy outcomes of metastatic non-small-cell lung cancer ------11

2.1.6 Prognosis ------13

2.1.7 Current dosing strategies of antineoplastic agents ------13

2.2 Pharmacokinetic-pharmacodynamic models ------14

2.2.1 Population pharmacokinetics of erlotinib ------15

2.2.1.1 Pharmacokinetic models ------15

2.2.1.2 Population pharmacokinetics of erlotinib ------16

2.2.2 Pharmacodynamics of erlotinib ------18

2.2.2.1 Mechanism of action of erlotinib ------18

2.2.2.2 Effect of erlotinib on tumour growth ------19 v

2.2.2.3 Pharmacodynamic modelling of erlotinib ------19

2.2.2.4 Modeling cancer cell growth ------21

2.2.3 Fitting of models ------24

2.2.4 Validation of models ------25

2.3 Factors influencing efficacy of erlotinib ------26

2.3.1 Patient-based factors ------26

2.3.2 Tumour-based factors ------28

2.3.3 Measurement-based factors ------29

2.3.4 Erlotinib-based factors ------30

2.3.4.1 Factors affecting population pharmacokinetics of erlotinib 30

2.3.4.1.1 Absorption ------30

2.3.4.1.2 Distribution ------31

2.3.4.1.3 Metabolism ------32

2.3.4.2 Factors affecting pharmacodynamics of erlotinib ------34

2.3.4.2.1 Methods of EGFR testing ------34

2.3.4.2.2 Intertumoral heterogeneity ------34

2.3.4.2.3 Intratumoral heterogeneity ------35

2.3.4.2.4 The tumour microenvironment ------36

2.4 Current pharmacokinetic-pharmacodynamic models ------37

2.5 Current literature on the exposure-efficacy relationship of erlotinib ------38

2.6 Summary ------40

Chapter 3 – Research objective and methodological steps ------41

3.1 Objective ------42

3.2 Methodological Steps ------42

Chapter 4 – Methods ------43

4.1 Overview of model ------43

4.2 The pharmacokinetic model component ------43

4.3 The pharmacodynamics model component ------48

4.4 The tumour growth (TG) model component ------50

4.5 The fitting process and determining internal validity ------53

4.6 The external validation and applications of the model ------57

Chapter 5 – Results ------59

5.1 The pharmacokinetic component ------60

5.2 The pharmacodynamic component ------65

5.3 The tumour growth component ------67

5.4 Simulation and prediction of PFS and response from erlotinib ------75

5.5 Applications of model ------86

Chapter 6 – Discussion------99

6.1 The pharmacokinetic component ------100

6.2 The pharmacodynamic component ------103

6.3 The tumour growth component ------105

6.4 The simulation and prediction of PFS data ------107

6.5 Applications of the model ------109

6.6 Limitations of the model ------110

6.7 Future research ------112

Chapter 7 – Conclusion ------115

Chapter 8 – References ------119

vii

Glossary of Abbreviations

General abbreviations Abbreviation Terminology ATP Adenosine triphosphate CACS Cancer anorexia-cachexia syndrome CAF Cancer-associated fibroblastic cells CT Computed tomography CR Complete response CYP450 Cytochrome P450 DCR Disease control rate DNA Deoxyribonucleic acid ECOG Eastern Cooperative Oncology Group EGF Epidermal growth factor EGFR Epidermal growth factor receptor EGFR-WT Epidermal growth factor receptor-wild type EMA European Medicines Agency FDA Food and Drug Administration FGF Fibroblast growth factors H2RA Histamine-2 receptor antagonist HGF Hepatocyte growth factor IGF-1 Insulin-like growth factor-1 IIC Infiltrating immune cells IL-2 Interleukin-2 IL-6 Interleukin-6 INF-γ Interferon-gamma KRAS Kirsten rat sarcoma viral oncogene MAP-K Mitogen-activated protein kinase MTD Maximum tolerated dose NNK Nicotine-derived nitrosaminoketone NSCLC Non-small-cell lung cancer OBD Optimum biologic dose ORR Objective response rate OS Overall survival PAH Polycyclic aromatic hydrocarbon PD Pharmacodynamic PD-1 Programmed cell death 1 PD-L1 Programmed cell death 1 ligand 1 PFS Progression-free survival pH Logarithm of the reciprocal of hydrogen ion activity PI3-K Phosphatidylinositol 3-kinase viii

PK Pharmacokinetic pKa Logarithm of the reciprocal of the acid dissociation constant PLC-γ Phospholipase C-γ PPI Proton pump inhibitor PR Partial response PrD Progressive disease RECIST Response Evaluation Criteria in Solid Tumours SCC Squamous cell carcinoma SCLC Small cell lung cancer SD Stable disease SLD Sum of longest diameter STAT Signal transducers and activators of transcription TKI Tyrosine kinase inhibitor TNF-α Tumour necrosis factor-α TTP Time to progression USA United States of America VEGF Vascular endothelial growth factor

Abbreviations specific to pharmacokinetics and pharmacodynamics Abbreviation Terminology AUC Area under the curve AUC CT Area under the curve of the Kaplan-Meier curve from the clinical trial AUC sim Area under the curve of the Kaplan-Meier curve from the model Cl Clearance Cmax Peak serum concentration Cmin Trough serum concentration Css Steady state concentration in serum Emax Maximal efficacy concentration F Bioavailability gr Growth rate constant of resistant cells gs Growth rate constant of sensitive cells IC50 Half maximal inhibitory concentration ka Absorption rate constant ke Elimination rate constant kr Kill rate constant of resistant cells ks Kill rate constant of sensitive cells t Time Tmax Time at peak concentration Vd Volume of distribution x Proportion of resistant to total number of tumour cells

List of Tables

Table 1: Tumour response criteria by RECIST version 1.1 (adapted from Eisenhauer et al.)

Table 2: Formulae of PK parameters

Table 3: Median Overall Survivals of Asian and Caucasian patients with metastatic non-small- cell lung cancer (adapted from Soo et al)

Table 4: ECOG Scale of Performance Status

Table 5: Effect of acid-suppressing therapies on erlotinib Cmax and AUC (adapted from Kletzl et al.)

Table 6: Parameters of pharmacokinetic component of model

Table 7: Search strategy of literature research of population pharmacokinetic studies of erlotinib

Table 8: Inclusion and exclusion criteria of literature search for population pharmacokinetic studies of erlotinib

Table 9: Formulae used to determine population-based estimates of pharmacokinetic parameters

Table 10: Parameters of pharmacodynamics component of model

Table 11: Parameters of the tumour growth equation

Table 12: Derivation method for the value of all model parameters

Table 13: Summary of values of erlotinib pharmacokinetic parameters reported in the included studies

Table 14: Summary of all other pharmacokinetic parameters for validation

Table 15: Summary of pooled values of all pharmacokinetic parameters

Table 16: T-test to test for heterogeneity between single-dose and multiple-dose erlotinib pharmacokinetic studies

Table 17: Comparison of values of PK parameters from model and other PK studies

Table 18: Steady state serum concentrations of erlotinib at different doses

Table 19: Summary of Emax values of sensitive and resistant cell groups for each patient subgroup population

Table 20: Summary of k s and k r values calculated from the PD equation

Table 21: Comparison of EC 50 and E max values of the model and Chmielecki et al study

Table 22: Breakdown of best response for each patient subgroup population

Table 23: Summary of mean and standard deviation values for g s gr and x for each patient subgroup

Table 24: Summary of simulated k r gr-kr, k s and g s-ks values for each patient subgroup population (n= 500)

Table 25: Summary of AUC sim /AUC CT ratios of placebo and erlotinib curves for each patient subgroup

Table 26: Summary of the effects of PK parameters on the shape of the PFS Kaplan-Meier curve

Table 27: Comparison of RECIST response rates of model simulation and clinical trial data

Table 29: Minimum serum erlotinib concentration at steady state (C ss ) required to gain objective response (at least 30% reduction in tumour size) in each patient subgroup

Table 30: Summary values of model parameters for all patient populations

List of Figures

Figure 1: Erlotinib penetration from the intravascular space to the intracellular space of NSCLC cells

Figure 2: Graphical representation of the Gompertz function

Figure 3: Derivation of the tumour growth equation

Figure 4: Flow diagram of the literature search: Screening and selection of erlotinib pharmacokinetic studies for the PK component

Figure 5: Concentration-kill constant relationships of sensitive and resistant cell groups

Figure 6: Tumour size changes over time based on best response simulated by model

Figure 7: Distributions of g-k values of resistant and sensitive cell groups

Figure 8: Comparison of model and clinical trial Kaplan-Meier PFS curves for all-comers and each patient subgroup

Figure 9: Breakdown of a PFS Kaplan-Meier curve

Figure 10: External validation of model – Comparison of Kaplan-Meier PFS curves from model and other clinical trials

Figure 11: Comparison of RECIST response rates between model and clinical trial data

Figure 12: Dose-PFS relationship of erlotinib in all patient subgroups

Figure 13: Extent of PK and PD variability on variability of erlotinib efficacy

Figure 14: Receiver-operating characteristic (ROC) curves of C ss predicting for at least a 30% reduction in tumour size (objective response)

xii

Chapter 1

Introduction

1 Introduction

Lung cancer is the second most common cancer in Canada, accounting for 14% of new cancer cases in 2014. The majority of lung cancer patients have non-small-cell lung cancer (NSCLC), which accounts for 85% of cases, and the remaining 15% of lung cancer patients have small-cell lung cancer. (1) It is also the leading cause of cancer-related mortality in Canada, and contributed to 26.8% of all cancer deaths in 2014. (1) Furthermore, the 5-year survival is only 4.7% for patients initially diagnosed with metastatic NSCLC. (2) The poor prognosis of metastatic NSCLC is quite concerning, as 57% of lung cancer patients are initially diagnosed with metastatic disease. (2) As a result, a substantial portion of research in lung cancer management is focused on uncovering new therapeutic agents and enhancing the efficacy of existing agents to improve survival rates in metastatic NSCLC patients. One of the most significant milestones in the management of NSCLC has been the emergence of targeted therapies such as the epidermal growth factor receptor tyrosine kinase inhibitors (EGFR TKIs).

One of the first EGFR TKIs to be approved is erlotinib. In 2002, a phase I study had identified 150mg/day to be the maximum tolerated dose of erlotinib and showed preliminary data suggesting potential reduction in tumour size in some patients. (3) This eventually paved the way to a series of phase III clinical trials, with BR.21 as the first clinical trial providing evidence for its ability to improve survival in metastatic NSCLC patients. (4)

The recommended starting erlotinib oral dose for all patients is 150mg daily. It is also available in two other strengths – 100mg and 25mg. (5) Dose reduction algorithms have been outlined by clinicians specializing in the care of lung cancer patients and are aimed at reducing the occurrence and severity of dose-limiting toxicities. (6,7) While such dose reductions frequently lead to the resolution of symptoms and prevention of future occurrences of severe side effects, clinicians do not generally take note of the potential consequences of these dose reductions on the efficacy of erlotinib, specifically on the response and survival of metastatic lung cancer patients while on the medication. Rather, these dose reductions are based on the convenience of dosing with commercially available strengths rather than an understanding of how different exposures of erlotinib may yield differences in efficacy.

While the dose reduction algorithm enables the treatment of dose-dependent toxicities of erlotinib, a poor delineation of the dose/efficacy relationship for erlotinib prevents clinicians from assessing the impact dose reductions on the treatment outcome. Although various in vitro and clinical studies have consistently demonstrated a relationship between exposure and cell kill, they do not provide good insight on the extent to which variations in exposure affect survival and response outcomes in patients and the incremental impact on survival and response with each dose reduction. (8- 11) Pharmacokinetic-pharmacodynamic (PK-PD) models have been constructed to explain the exposure-efficacy relationship of erlotinib; however, their application to further define dosing strategies to optimize efficacy in metastaticNSCLC patients are limited. (12)

The purpose of this thesis was to develop a population-based PK-PD model that simulates population estimates of progression-free survival (PFS) in metastatic NSCLC patients treated with erlotinib. Specifically, this model aimed to provide insight into the relationship between erlotinib serum concentration, tumour susceptibility and erlotinib efficacy and addressing some of the factors that seem to contribute variability in each of those elements. The development of this model provides insight into the ramifications of dose reductions on the expected PFS of metastatic NSCLC patients.

Chapter 2

Literature Review

2 Literature Review 2.1 Overview of Non-Small-Cell Lung Cancer

2.1.1 Epidemiology

Lung cancer is a highly heterogeneous group of cancers. It is broadly classified into non-small- cell lung cancer (NSCLC), which consists of 85% of all lung cancers, and small-cell lung cancer (SCLC), which makes up the remaining 15% of lung cancers and is not further discussed in this thesis. (13) NSCLC is similarly a very heterogeneous group of lung cancers; however, two histology types dominate – adenocarcinoma and squamous cell carcinoma (SCC). To simplify our discussion of NSCLC patients, NSCLC will henceforth refer to only adenocarcinoma and SCC histologies.

Adenocarcinoma represents 40-45% of NSCLC cases, while SCC consists of 25-30% of the NSCLC cases. Lung cancer has been traditionally portrayed as a disease of elderly males with a smoking history. 75-90% of patients with NSCLC were active or past smokers. (14, 15) The incidence of lung cancer in North America is higher in males compared to females by a ratio of 1.2-1.4 to 1. (1, 2) Furthermore, the median age of diagnosis in United States is 70 years old. (2)

2.1.2 Risk factors and pathogenesis of non-small-cell lung cancer

The most prominent risk factor of NSCLC is smoking. Smokers are 10 to 20 times more likely to develop lung cancer compared to non-smokers. While individuals exposed to second-hand smoke are exposed to lower levels of carcinogens compared to smokers, the levels are sufficiently high to increase the risk of lung cancer by 20-27% compared to individuals who were not exposed to second-hand smoke. (14, 15)

However, 15% of NSCLC in men and 53% of NSCLC in women are estimated to be non- smoking-related. (14, 15) Other risk factors can contribute towards the pathogenesis of NSCLC in these patients. Radon, which accumulates in mines and indoor home environments, has been

documented to be a contributor of non-smoking-related lung cancer. Household pollution due to cooking oil vapours and indoor coal burning contributes to the production of polycyclic aromatic hydrocarbons (PAHs) and benzo[a]pyrene, both of which are metabolized into reactive metabolites that lead to key oncogenic mutations when they bind covalently to DNA. Estrogen was found to be associated with increased risk of lung cancer in females due to its potential role in simulating NSCLC cell proliferation, altering the metabolic activation of carcinogen and regulating epidermal growth factor receptor (EGFR) expression. Viruses, such as human papilloma virus, have been linked to an increased incidence of lung cancer. (14, 15)

Despite being exposed to these environmental risk factors, genetic factors play a significant role in determining the predisposition to developing NSCLC. While the gene responsible for increased susceptibility to NSCLC has yet to be found, it is thought to be associated with chromosome 6q23-25. (14, 15) Differences in susceptibility to lung cancer may be due to differences in metabolism or DNA repair capacity.

The pathogenesis of NSCLC may differ depending on the smoking history of the patient. Current and past smoking patients developed NSCLC through exposure to numerous carcinogens in the smoke, including PAH, benzo[a]pyrene and nicotine-derived nitrosaminoketone (NNK). The metabolism of these carcinogens by cytochrome P450s (CYP450s) lead to the production of reactive metabolites that bind covalently onto DNA to form bulky adducts. While some adducts are typically rectified by the induction of apoptosis thus preventing these mutations from being passed on, some of these adducts create mutations in several pro-oncogenes, such as Kirsten rat sarcoma viral oncogene (KRAS) and tumour suppressor genes like TP53 contributing to the pathogenesis of lung cancer. (13, 14) While never-smokers may also be exposed to other currently unidentified carcinogens through non-smoking-related environmental risk factors, it is thought that these carcinogens may preferentially cause mutations in the EGFR gene and different kinds of TP53 mutations instead. Therefore, tumours from patients with a smoking history are generally different in mutational signature and frequency than tumours from patients who have never smoked. (14) As a result, patients with a smoking history are more likely to have tumours with KRAS mutations, while patients who have never smoked tend to have tumours with EGFR mutations.

2.1.3 EGFR mutations

Mutations of the EGFR gene, which is located on chromosome 7p12-13, were among the first that became pivotal in determining tumour sensitivity to different treatments. (13) While there are several mutations that lead to dysregulated activity of EGFR, contributing to uncontrolled proliferation, metastasis and survival (13, 16), three specific genetic mutations have become increasingly relevant in clinical practice: the deletion of 9 to 24 nucleotides around codons 746- 750 from exon 19 (henceforth called exon 19 del) (16-19), amino acid substitution from leucine to arginine at position 858 of exon 21 (henceforth called exon 21 L858R) (16, 20) and amino acid substitution from threonine to methionine at position 790 of exon 20 (henceforth called exon 20 T790M) (21). Exon 19 del and exon 21 L858R make up 48.2% and 42.7%, respectively of clinical relevant EGFR mutations and are mutations of the kinase domain of EGFR gene that lead to increased sensitivity to erlotinib. (16, 17, 19, 20) While exon 20 T790M is found in less than 5% of all untreated EGFR-mutant tumours (21, 22), it is present in 50% of EGFR-mutant resistant tumours previously treated with a first-generation TKI. (23, 24)

EGFR mutations are found in 10-15% of NSCLC patients in United States (25) and 35-51% of NSCLC patients in Asia. (24, 25) The vast majority of these EGFR mutations are found in adenocarcinoma tumours. Patients are more likely to have an EGFR-mutant NSCLC if they are female, East Asian and/or never-smokers. EGFR mutations are found in 28% of NSCLCs in female patients versus 18% in male patients. Furthermore, never-smokers are three times more likely to have an EGFR-mutant NSCLC compared to smokers. (26) In contrast, only 1.1-1.9% of SCC tumours are found to harbour an EGFR mutation. (27, 28) Therefore, SCC tumours are considered to be EGFR-wild-type in general.

2.1.4 Role of erlotinib in standard treatment of metastatic non-small-cell lung cancer

Standard treatment of metastatic NSCLC is determined by mutation status of the tumour. Since EGFR mutations almost always appear in adenocarcinoma tumours, EGFR mutation testing is only performed in patients who are thought to have metastatic adenocarcinoma of the lung. (26,

29) Nonetheless, EGFR mutation testing may be done in patients who have tumours that are predominantly non-adenocarcinoma; however, an adenocarcinoma component cannot be excluded, particularly if they possess demographic characteristics that increase the likelihood of their tumours harbouring an EGFR mutation. (26, 29, 30)

Patients with metastatic NSCLC that have an EGFR sensitizing mutation – exon 19 del or exon 21 L858R –would be treated with an EGFR TKI, such as gefitinib, afatinib or erlotinib, as first- line treatment. (31, 32) In patients who do not have an EGFR mutant NSCLC, their tumours are tested for other mutations (e.g. ALK) that can be treated with corresponding targeted agents (e.g. crizotinib, ceritinib, alectinib) or can be treated with doublet platinum chemotherapy if the tumours were wild-type to other oncogenes. (31, 32) Patients who have responded to doublet platinum chemotherapy may receive erlotinib as an option for maintenance therapy. If erlotinib was not used for maintenance therapy in the past, patients with EGFR-mutant metastatic NSCLC who progressed from chemotherapy may have the option of using erlotinib as 2 nd or 3 rd line. (31, 32)

The role of erlotinib in the management of metastatic NSCLC is based on multiple clinical trials evaluating erlotinib in different metastatic NSCLC patient populations. In 2004, Shepherd et al. published the results of the BR.21 phase III clinical trial, which reported an improvement in progression-free survival (PFS) from 1.8 months in the placebo group to 2.2 months in the erlotinib group in metastatic NSCLC patients of any EGFR mutation status who had failed at least one line of treatment. (4) The result of the BR.21 trial led to the approval from the Food and Drug Administration (FDA) in 2004 for the use of erlotinib as monotherapy of metastatic NSCLC after failure of at least one prior line of chemotherapy. (33) In 2010, Cappuzzo and company reported on the SATURN trial, which concluded that metastatic NSCLC patients on erlotinib 150mg daily had longer PFS (11.1 versus 12.3 weeks) compared to patients who were on placebo for maintenance therapy after responding to platinum doublet chemotherapy. (34) This study only recruited patients who had EGFR-wild-type (EGFR-WT) metastatic NSCLC. These results prompted the FDA to issue erlotinib a second indication for maintenance therapy in metastatic NSCLC patients who have received four cycles of platinum doublet chemotherapy and have not progress. (35) In 2012, the results of EURTAC were published. This phase III,

parallel-arm, randomized controlled trial reported a significant increase in median PFS from 5.2 (95% CI: 4.5-5.8) to 9.7 (95% 8.4-12.3) months compared to platinum doublet chemotherapy. (36) Furthermore, the response rates were significantly higher than that of conventional first-line chemotherapy; the objective response rate increased from 15% to 58%. (36) Subsequently, FDA approved erlotinib to be used as first-line therapy in metastatic NSCLC patients with either exon 19 del or exon 21 L858R mutations. (37) Globally, the vast majority of NSCLC patients receive erlotinib, either as first-line, second-/third-line or maintenance therapy, depending on the EGFR mutation status of their tumours. (31, 32)

Despite the fact that the literature from guidelines and clinical trials suggest that erlotinib can be used in any of these three indications (31, 32), funding criteria from the Ontario Drug Benefit program limit the use of erlotinib in Ontario for the management of metastatic NSCLC. The criteria for approved funding of erlotinib for the treatment of metastatic NSCLC in the Exceptional Access Program in Ontario are as follows: • Erlotinib is used as monotherapy for the 2nd- or 3rd-line treatment after failure of prior chemotherapy (any regimen) in patients 70 years of age or older. (38) • Erlotinib is used as monotherapy for the 2nd- or 3rd-line treatment of patients with clinically documented incurable progressive non-small cell lung cancer (NSCLC) despite prior chemotherapy including both docetaxel and a platinum-based treatment (i.e. cisplatin or carboplatin). (38) • Erlotinib is used as monotherapy for the 3rd-line treatment of patients with clinically documented incurable progressive non-small cell lung cancer (NSCLC) despite prior chemotherapy including both a platinum-based therapy (i.e. cisplatin or carboplatin) AND either pemetrexed or topotecan. (38) • Erlotinib is used as monotherapy for 2nd line treatment of NSCLC after 1st line platinum- based therapy, where no other chemotherapy will be given and erlotinib is used as the last treatment for the patient. (38)

Therefore, in Ontario, erlotinib is mostly used in the setting of 2 nd or 3 rd line treatment after failure of prior treatment.

2.1.5 Efficacy outcomes of metastatic non-small-cell lung cancer

There are multiple endpoints used to measure efficacy of therapeutic options for the management of metastatic NSCLC. Clinical trials frequently use overall survival (OS) and PFS as primary outcomes. (4, 34, 35, 39, 40) OS is defined as the time from when the patient is randomized into a treatment arm to when the patient passes away from any cause (36, 41), while PFS is defined as the time from study randomization to the time when the patient either progresses or dies. (41) Other endpoints include time to progression (TTP) or treatment response by Response Evaluation Criteria in Solid Tumours (RECIST) criteria. TTP is defined as the time from randomization into the clinical trial to the time at which the patient is found to have progressed. (41) There are four different treatment responses by RECIST criteria. (42) Table 1 provides the definition for each treatment RECIST response.

Table 1: Tumour response criteria by RECIST version 1.1 (adapted from Eisenhauer et al.(42)) Response Definition (Target lesions) Definition (Non-Target lesions) Complete Disappearance of all extranodal target Disappearance of all extranodal response (CR) lesions. All pathological lymph nodes non-target lesions. All lymph must have decreased to <10mm in short nodes must be non-pathological axis. in size (<10mm short axis). Normalization of tumour marker level. Partial response > 30% decrease in the sum of longest Persistence of one or more non- (PR) diameters (SLD) target lesion(s) and/or Stable disease Neither sufficient shrinkage to qualify maintenance of tumour marker (SD) for PR nor sufficient increase to qualify level above the normal limits. for PrD Progressive SLD increased by at least 20% from Unequivocal progression of disease (PrD) existing non-target lesions smallest value on study (including (subjective judgement by baseline, if that is the smallest). The experienced reader). SLD must also demonstrate an absolute increase of >5mm. Reappearance of a lesion in the same area as an observed CR. Unequivocal and confirmed new lesions in anatomic regions in which a tumour was previously not seen at baseline.

Best RECIST response may be reported in one of three ways. It may be separated based on CR, PR, SD, or PrD. It may be reported as an objective response rate (ORR), which is the proportion of patients who had either a PR or CR. Clinical trials may also report a disease control rate (DCR), which is the proportion of patients who have SD, PR or CR. (41)

Therapeutic efficacy is monitored through computer tomography (CT) scans of the chest, head and abdomen/pelvis every six to eight weeks. Each consecutive pair of current and past scans of each region are compared to determine whether progression has occurred. Progression is noted if the size of the measurable lesion has increased by 20% from its documented minimal size or when at least one new lesion is observed on the scan. (41)

While OS is the gold standard in evaluating efficacy of anticancer treatments and continues to be the basis for FDA approval of new anticancer drugs in USA (41, 43), clinical trials require increasingly long follow-up times and larger sample sizes in order for a statistically and clinically significant benefit to be seen. Since this limits patient access to medications and leads to greater costs for funding of these clinical trials, PFS is quickly gaining ground as a more feasible clinical endpoint for evaluating treatment efficacy. (43) Currently, the European Medicines Agency (EMA) also accepts prolonged PFS as another efficacy outcome considered to be beneficial for the well-being of patients. (44, 45)

PFS has multiple advantages as an efficacy endpoint for evaluating treatments for metastatic NSCLC. PFS only measures the efficacy of the single line of therapy evaluated in the clinical trial and is not skewed by potential survival benefit associated with subsequent lines of therapy, which is otherwise noted in the measurement of OS. Thus, the use of PFS as an outcome is advantageous in the setting of metastatic NSCLC, as there are multiple treatment options of chemotherapy and targeted agents that currently exist for its management. (43) Furthermore, PFS allows for shorter time of follow-up compared to OS. (42, 43) The achievement of this efficacy endpoint is more feasible with respect to patient sample size and follow-up duration, in clinical trials compared to OS. (43, 44) PFS is also unaffected by any crossover between treatment arms, which may occur after progression of the initial intervention in a randomized controlled trial. (36, 39)

2.1.6 Prognosis

Despite the recent advances in the management of metastatic NSCLC, lung cancer remains the leading cause of cancer-related death in the world. (46) Lung cancer causes nearly 30% of cancer-related mortality in Canada and USA. (1, 2) Furthermore, the 5-year survival rate for metastatic NSCLC is low at 4.7%. (2) The poor prognosis of metastatic NSCLC patients is concerning, as 57% of NSCLC patients are diagnosed with metastatic disease. (2) As a result, a substantial portion of research in lung cancer management is focused on identifying new therapeutic agents and enhancing the efficacy of existing agents to improve survival rates in metastatic NSCLC patients. In addition to the ongoing need for additional therapies to improve the prognosis of metastatic NSCLC patients, there is a need to evaluate approaches that optimize the use of current therapies to realize their fullest potential in improving patient outcomes.

2.1.7 Current dosing strategies of antineoplastic agents

Due to the existence of a narrow gap between doses that lead to toxicity and those that garner therapeutic benefit, clinicians use the maximum tolerated dose (MTD) approach to determine the dose used for subsequent human clinical trials. (47, 48) The maximum tolerated dose is defined by Phase I clinical trials as the highest dose level at which one third or less of patients experience a dose-limiting toxicity, where a dose-limiting toxicity is an adverse event thought to be caused by the medication that is considered unacceptable due to its severity, irreversibility or other negative consequence to the patient. (47) The principle of using a MTD as the therapeutic dose is thought to ensure that patients will get the highest degree of efficacy with toxicities that are still considered to be reasonable.

The main limitation of MTD is its assumption that all patients have similar responses in efficacy and toxicity to a single dose of medication. In reality, a one-size-fits-all approach does not guarantee that each patient will garner the same degree of efficacy and toxicity with the same dose. In a population treated based on an MTD approach, only 45% of patients have

concentrations thought to fall within the therapeutic window, 17% of patients have concentrations beyond the therapeutic window and may discontinue therapy due to toxicities, and 38% of patients are thought to have subtherapeutic concentrations and are unlikely to derive benefit. (49) Therefore, 55% of these patients would likely gain less therapeutic benefit than projected based on a one-size-fits-all approach.

Other researchers have suggested the use of optimum biologic dose (OBD), which is defined as the dose associated with achieving a certain level of effect on a targeted biomarker (e.g. EGFR inhibition on a tumour by erlotinib). (47, 48) While this is a more rational approach towards linking drug exposure to its effect, the biological effect of the drug may not correlate well with clinical outcomes of the patient. For instance, there is no evidence to suggest that complete EGFR inhibition will lead to prolonged survival for metastatic NSCLC patients.

Therefore, it is suggested that the effect of the drug should be translated into a desirable clinical outcome and clinicians should instead aim for the optimum clinical concentration to give this outcome. (48) To determine the optimal concentration of erlotinib, a comprehensive understanding of the pharmacokinetic-pharmacodynamic (PK-PD) relationship and its effect on clinically relevant measures of efficacy is required.

2.2 Pharmacokinetic-pharmacodynamic models

PK-PD models are designed to provide insight about the relationship between drug exposure and drug effect, or tumour inhibition in the oncology setting, and the mechanism by which the drug causes tumour inhibition. (50, 51) Ideally, the PK-PD model requires consideration of the pharmacokinetic (PK) profile of the drug, its pharmacodynamics (PD) properties affecting tumour growth inhibition and an understanding of how tumours grow over time.

2.2.1 Population pharmacokinetics of erlotinib 2.2.1.1 Pharmacokinetic models

The population PK profile of many orally administered drugs can be described using a one- compartment oral administration model with first-order absorption and elimination. The one- compartment oral model with first-order absorption and elimination has the following assumptions: - The drug instantly distributes into the body after its absorption, which is depicted as a single compartment (52) - The rates of elimination and absorption are proportional to concentration (52)

The equation of the PK model used to calculate drug concentration over time is

(1) (52)

Where - F is the oral bioavailability of the drug - Dose is the administered dose of the drug

- ka is the absorption rate constant

- ke is the elimination rate constant

- Vd is the apparent volume of distribution of the drug, and - t is the time from drug administration

The parameters of this model can be used to determine other PK parameters, including clearance

(Cl), time at peak concentration (T max, ), trough serum concentration (C min ), peak serum

concentration (C max ) and area under the curve (AUC). Table 2 provides a method of calculating each PK parameter. (53)

Table 2: Formulae of PK parameters (53) Parameter Equation Clearance (Cl)

Time at peak concentration (T max )

Trough serum concentration (C min )

Peak serum concentration (C max )

Area under the curve (AUC)

2.2.1.2 Population pharmacokinetics of erlotinib

The population PK of erlotinib has also been represented using a one-compartment model with first-order absorption and elimination. (54, 55) Its absorption is observed to be solubility- dependent; as a weak base with a pKa of 5.42, it is only slightly soluble in water, but demonstrates increased solubility at lower pHs. (56) This is because erlotinib requires protonation of its secondary amine to allow for it to have a positive charge and become more hydrophilic. The presence of food in the stomach may stimulate stomach acid secretion and

hence decrease stomach pH. Therefore, erlotinib, when taken concurrently with food, will exhibit increased dissolution and absorption, which is reflected in observing its bioavailability increase from 60% to 100%. (56) With respect to distribution, it has a moderately sized apparent volume of distribution of around 233 L, and is a highly protein bound drug that binds predominantly to albumin and α-1 acid glycoprotein. (56) Erlotinib is predominantly metabolized by CYP3A4/5, followed by CYP1A1 and to a lesser extent by CYP1A2 into a variety of inactive and active metabolites. (5) The most abundant active metabolite is OSI-420, which is found to be of a similar potency as erlotinib but is present at only 10% of erlotinib’s plasma concentrations. (56) Since erlotinib is primarily metabolized by cytochrome P450s, 83% of erlotinib is consequently eliminated in feces and only 8% of it eliminated in urine as unchanged drug.

The current population PK literature of erlotinib is based on the cancer patients who received the medication. Since this medication is not given to healthy patients, pediatric patients, patients with hepatic dysfunction or patients with hepatorenal syndrome in clinical practice, the population PK literature will not reflect the behaviour of erlotinib in these patients. Pediatric patients are documented to have significant differences in metabolism and volume of distribution compared to adult patients. (57-59) Medications, including erlotinib, also typically have altered pharmacokinetics in cancer patients compared to healthy patients (60, 61), likely due to the presence of chronic inflammation in the former group. As erlotinib is predominantly hepatically metabolized by CYP450s, impaired hepatic function caused by cirrhosis or liver disease can significantly decrease metabolism and elimination of erlotinib and therefore, the pharmacokinetics of erlotinib to be altered significantly from that of patients with normal hepatic function. (62) While only 8% of erlotinib is thought to be eliminated renally, patients with renal disease commonly present with hepatorenal syndrome due to the accumulation of renally cleared toxic byproducts into the liver that may cause hepatic impairment. (62)

2.2.2 Pharmacodynamics of erlotinib 2.2.2.1 Mechanism of action of erlotinib

Erlotinib is a first-generation, selective and reversible tyrosine kinase inhibitor of EGFR. It is a 4-anilinoquinazoline with a molecular weight of 429.9 g/mol, and can penetrate into the cell and prevent EGFR from autophosphorylating tyrosine residues on its tyrosine kinase domain by forming multiple interactions with the ATP binding pocket of the tyrosine kinase domain. (63) The lack of phosphorylation of tyrosine residues in this domain hinders the formation of docking sites for cytoplasmic proteins, which prevent the activation of several downstream intracellular signaling pathways. These include the Ras/Raf/mitogen-activated protein kinase (MAP-K), phosphatidylinositol 3-kinase (PI3-K)/Akt, phospholipase C-γ (PLC-γ), signal transducers and activators of transcription (STAT) and Src kinase pathways, which lead to increased cell proliferation, angiogenesis, apoptosis resistance, cell survival and cell migration necessary for metastasis formation. (64) By blocking autophosphorylation of tyrosine residues in the EGFR tyrosine kinase domain, erlotinib induces apoptosis and cell cycle arrest, thereby inhibiting growth of cancer cells. (65-67)

Erlotinib is effective as a treatment option for metastatic NSCLC patients because the EGFR gene is known to be amplified in 43-89% and mutated to be constitutively active in 25% of NSCLC. (68) Tumours that have an EGFR amplification have multiple copies of the EGFR gene, leading to an excess of EGFR expression and subsequently increased signaling favouring proliferation, angiogenesis, apoptosis resistance, cell migration and survival. (68) While erlotinib is not well studied in tumours that have EGFR amplification, it is expected that these tumours may have some sensitivity to erlotinib.

NSCLC tumours that have EGFR-sensitizing mutations, such as exon 19 del and exon 21 L858R, have increased susceptibility to erlotinib. Both mutations cause a change in the kinase domain that makes it constitutively active, promoting increased proliferation and survival of the tumour. These mutations cause the binding site to have a greater affinity to erlotinib compared to ATP, causing the tumour to have increased sensitivity to erlotinib and allowing erlotinib to be especially efficacious in treating tumours with these mutations. (8-10) In contrast, the exon 20 T790M mutation confers decreased sensitivity to erlotinib. (8) Therefore, precise identification 18

of the EGFR mutation through mutation testing is mandatory to select the most appropriate treatment option to manage metastatic NSCLC patients.

2.2.2.2 Effect of erlotinib on tumour growth

In vitro studies indicate that erlotinib can cause growth inhibition in a variety of cancer cells, but cancer cell lines possessing various mutations seem to have different sensitivities to erlotinib. Chmielecki et al. used PC-9 cells, a lung cancer cell line that harbours the exon 19 del mutation, to show that erlotinib was a potent growth inhibitor of sensitive PC-9 cells with a half maximal inhibitory concentration (IC50) of around 25 nM and reaching complete growth inhibition at around 50 nM, but was significantly less potent in inhibiting the growth of T790M mutated PC-9 cells, as its dose-response relationship shifted to the right with an IC50 of >4000 nM and never reaching complete growth inhibition in the study. (8) A similar dose-response relationship is observed in other cell lines harbouring the same or other clinically relevant mutations. Yuza et al. determined that erlotinib yielded a similar relationship in Ba/F3 and HCC827 cells, both lung cancer cell lines harbouring the exon 19 del mutation, with similar IC50 values ranging from 4.9 to 14nM. (9) A similar relationship was found with NCI-H3255 and 11-18 cell lines, both of which carry the exon 21 L858R mutation and have IC50 values of 6.9-10nM. (9, 10) It was observed that erlotinib was a slightly less potent growth inhibitor of cells with the L858R mutation than those with the exon 19 del. (69)

2.2.2.3 Pharmacodynamic modelling of erlotinib

Current understanding of erlotinib’s mechanism of action can be characterized using a PD model that demonstrates the ability of erlotinib to both kill and inhibit the growth of NSCLC cells through receptor-ligand binding at the EGFR tyrosine kinase domain. While there are multiple mathematical models used to depict the PD of systemic agents (50), the basis of most non- stochastic PD models is the commonly used E max sigmoidal equation (70-72), which has been

widely accepted as the equation describing a dose-response relationship involving receptor- ligand binding as

(2)

Where

- Emax is the concentration at which the drug has achieved its maximal effect - [C] is the concentration in which the effect takes place, and

- EC 50 is the concentration at which the drug has achieved 50% of its maximal effect

For our purposes the effect would be an observation such as tumour growth inhibition. Linking two models, one to describe the potency of erlotinib and another to relate the pharmacological effects to tumour growth, is needed to fully describe the pharmacological effects of erlotinib on tumour growth inhibition.

Drug concentration at the site of action can be directly measured and regulated in in vitro studies, but has not been measured in clinical trials. Phase I trials often use serum drug concentration as a surrogate of drug that can penetrate into the intracellular environment where the receptor- ligand binding occurs. There is little literature describing in vivo intracellular concentrations that result from regular doses of antineoplastic agents.

Several in vitro studies may provide some insight about the extent of erlotinib penetration in metastatic NSCLC patients. Figure 1 depicts the path by which erlotinib penetrates from serum to the intracellular space. Based on the fact that 5% of erlotinib is unbound, erlotinib’s pKa is 5.42 and the physiologic pH is 7.2, it can be determined that 3% of erlotinib is unbound and un- ionized using the Henderson-Hasselbach equation, and therefore, is free to defuse into the tumour microenvironment. Work by Tannock et al. indicates that there is poor penetration of chemotherapeutic agents into a given tumour ranging from 3-50% of the serum concentration. (73) With respect to tumour penetration data specific to erlotinib, around 10% of erlotinib in the plasma was found to have penetrated into the tumour microenvironment in 13 NSCLC patients who received neoadjuvant therapy of erlotinib and subsequently underwent resection. (74)

From the tumour microenvironment, an in vitro study using erlotinib-resistant A-431 cells reported that around 1% of erlotinib would penetrate into the intracellular environment. (75) Therefore, the fraction of drug in the serum that can penetrate into the intracellular environment of NSCLC cells is in the magnitude of 0.00005.

Figure 1: Erlotinib penetration from the intravascular space to the intracellular space of NSCLC cells

2.2.2.4 Modeling cancer cell growth

Multiple equations have been used to model cancer cell growth that can be extrapolated to describe the growth of NSCLC cells. The Gompertz function has been widely cited as a model describing tumour growth in multiple studies (71, 76) and is characterized by the equation

(3)

The equation models a sigmoidal growth pattern in which growth is initially slow, but accelerates to exponential growth and eventually plateaus to a higher asymptote that marks the maximal size of the tumour. The parameters, λ, µ and A, determine the initial lag phase, maximum growth rate and maximum tumour size, respectively, as observed on Figure 2. (77)

Figure 2: Graphical representation of the Gompertz function

Another function that has been used to describe the sigmoidal curve of tumour growth is the logistic function (71), which is described using the following equation

(4)

where - dV/dt is the relative growth rate (relative change in volume over time)

- a is a coefficient related to proliferation kinetics - V is the volume of the tumour, and - K is the maximal volume or carrying capacity

The equation depicts the rate of growth increasing proportionally to the tumour volume to a maximum rate, which is then followed by the rate decreasing proportionally to the tumour volume until it reaches the carrying capacity. (71)

More complex models depicting a sigmoidal pattern of tumour growth in in vitro settings exist. An example of this is the dynamic carrying capacity model (71), which can be represented with two equations:

(5) and

(6)

where - dV/dt is the relative change in volume over time - dK/dt is the relative change of the carrying capacity over time - a and b are coefficients related to the proliferation of the tumour - V is the volume of the tumour and - K is the carrying capacity

This model was intended to describe the effects of angiogenesis on tumour growth. Changes in blood supply in the tumour will affect the ultimate carrying capacity of the tumour and will therefore, affect how the tumour grows through time. (71)

A simpler model that describes tumour growth patterns without the plateauing of tumour growth at larger sizes and with the feature of potential tumour inhibition (negative tumour growth) is the exponential growth function, described as

(7)

This function is also widely used due to its simplicity and ability to portray the binary division of tumour cells over time. (78, 79) The equation features an exponential growth constant, which is expressed as the difference between the intrinsic proliferation rate and intrinsic death rate of the cells. In the case of tumour growth, the proliferation rate is greater than the death rate; therefore, the growth constant will be a positive number. In contrast, the growth constant is a negative number if the death rate is greater than the proliferation rate, which can occur with the presence of an agent that induces death at a higher rate and/or inhibits proliferation of tumour cells than the rate at which cells can proliferate. This function can also be additive to describe exponential growth of two or more groups of cells within a tumour that have different growth rates. This characteristic of the exponential function is convenient for describing intratumoral heterogeneity with respect to natural growth and death in the presence of an inhibitor. These features thus make the exponential model most suitable as the basis of a tumour growth equation as part of a PK-PD model.

2.2.3 Fitting of models

In order for the PK-PD model to simulate the relationship between population pharmacokinetics, mechanism of action, tumour growth inhibition and progression-free survival for a given agent, it must reflect the mechanics and the values that correspond with the agent in question. Therefore, the construction of PK-PD model requires an understanding of how the drug behaves at each component of the model. However, values from actual data representing each component must be inputted into the model based on current knowledge. This is done through a fitting processes.

The objective of model fitting is to determine a model that explains as much about the system in question and minimize the amount of data that cannot be explained by the model. Fitting methods to estimate parameter values of non-stochastic models from data are generally limited to visual inspection of closeness. (50, 79) Visual inspection involves how well the model and the trial data agree with each other, which is typically done by determining how well their resulting 24

curves superimpose on each other. While this is a popular method, it does not provide an objective means of evaluating goodness of fit of the model to the data and therefore, results on model fit from visual inspection should be supplemented with statistics. While the calculation of a least sum of squares may be used for this purpose, its use is generally limited to linear regression models. (72) Models that involve more complex functions may require other approaches, such as a comparison of areas under the curve (AUC) to allow analysis of the entire curve.

2.2.4 Validation of models

After the model has undergone a fitting process, its simulation data must be evaluated for its accuracy and generalizability to reality through validation. According to Rehman and Pedersen (80), models can be validated based on confirmative validation and subvalidation. Confirmative validation involves comparing the results of the model with observations from empirical knowledge. Subvalidation involves confirmative validation of all the components of the model, which requires comparing the results of the pharmacokinetics, pharmacodynamics and effects of tumour inhibition of erlotinib from the model to observations of each aspect as reported in the literature. (80)

Confirmative validation of the model and of its components can be used for internal validation and external validation. Internal validation requires the simulated data to be compared to the clinical trial data used to fit the model and is used to ensure that the model is generating correct data based on the information used to construct it. By providing a microscopic view of the model, it ensures the validity of the theory and equations that are set as the foundation of each component. External validation is required to ensure that the data from the simulations are an accurate portrayal of reality; this is generally performed by comparing simulation data with other clinical trial data independent of what is used to develop the model. By doing so, external validation provides a macroscopic assessment of the model’s ability to depict the properties of the clinical trial data representing the system. (80) Both validation processes are required at the

component and model levels to ensure that the model and components are accurately representing the system in question.

2.3 Factors influencing efficacy of erlotinib

Factors that affect treatment RECIST response and PFS of erlotinib can be patient-, tumour-, erlotinib-, or measurement-based.

2.3.1 Patient-based factors

Patient-based factors include ethnicity, gender, performance status, smoking status and the receipt of early palliative care. Patients with Asian ethnicity are found to have improved survival compared to Caucasian patients. (81) A meta-analysis by Soo et al (81) found this trend of improved OS in Asian compared to Caucasian patients regardless of the treatment that was given. Table 3 summarises the median OS of Asian and Caucasian patients for each type of treatment regimen.

Table 3: Median Overall Survivals of Asian and Caucasian patients with metastatic non-small- cell lung cancer (adapted from Soo et al (81)) Asian patients Caucasian p-value patients All chemotherapy Pre-EGFR TKI 9.1 months 7.6 months P < 0.001 regimens Pre- and post- 10.1 months 8.0 months P < 0.001 EGFR TKI Monotherapy Pre-EGFR TKI 8.9 months 6.5 months P < 0.001 chemotherapy Pre- and post- 9.9 months 6.8 months P < 0.001 EGFR TKI Platinum doublet Pre-EGFR TKI 9.1 months 7.5 months P < 0.001 chemotherapy Pre- and post- 10.4 months 8.6 months P < 0.001 EGFR TKI Three or more Pre-EGFR TKI 9.3 months 7.6 months P < 0.003 drugs in Pre- and post- 9.4 months 8.0 months P < 0.001 combination EGFR TKI EGFR TKI: Epidermal growth factor receptor tyrosine kinase inhibitor

Female metastatic NSCLC patients have also been found to have consistently better survival compared to their male counterparts. A retrospective cohort study of over 20,000 lung cancer patients showed that males had a 15% higher risk of death compared to females, after adjustment of age, lung cancer histology, performance status and stage of disease. (82)

Furthermore, higher performance status, which is evaluated using the Eastern Cooperative Oncology Group (ECOG) Scale of Performance Status (Table 4) (83), is an independent predictor of mortality in metastatic NSCLC patients. (84) Multiple studies have consistently found that superior performance status was associated with improved survival. (82, 84) In particular, it was found that median OS was improved from 15.4 months in ECOG 1 patients and 51.5 months in ECOG 0 patients. (84) Superior performance status is also associated with a greater improvement of survival with treatment. While Schiller et al. (85) found no significant difference between four doublet platinum chemotherapy regimens used in the treatment of metastatic NSCLC patients prior to the standard use of EGFR-TKIs, they observed that the median TTP differed significantly between patients with ECOG Grade 0, 1 and 2. Patients of ECOG 0 had a median TTP of 4.3 months, ECOG 1 patients with 3.5 months and ECOG 2 patients of 1.5 months. The observation of reduced survival in ECOG 2 patients thus discouraged the use of doublet platinum chemotherapy in most of these patients. (85)

Table 4: ECOG Scale of Performance Status Grade Description 0 Fully active, able to carry on all pre-disease performance without restriction 1 Restricted in physically strenuous activity but ambulatory and able to carry out work of a light or sedentary nature, e.g. light house work, office work 2 Ambulatory and capable of all self-care but unable to carry out any work activities. Up and about more than 50% of waking hours 3 Capable of only limited self-care, confined to bed or chair more than 50% of waking hours 4 Completely disabled. Cannot carry on any self-care. Totally confined to bed or chair 5 Dead

An active smoking history independently predicts higher risks of mortality compared to patients who are never-smokers. Kawaguchi et al found a large increase in median OS between active smokers (19 months) and never smokers (30 months). (84) Smoking cessation was also associated with decreased all-cause mortality and recurrence in NSCLC patients. (86)

Patients who receive palliative care upon their diagnosis of metastatic NSCLC benefit from having improved OS compared to patients who do not receive palliative care. (31, 87) In 2010, Temel et al. observed an improvement in survival in patients who received early palliative care compared to patients who received standard of care (11.6 months vs 8.9 months, p = 0.02). (87) Unfortunately, there are multiple barriers that may preclude patients from receiving timely palliative care (88), which may affect the survival of metastatic NSCLC patients.

2.3.2 Tumour-based factors

Tumour-based factors include the presence of EGFR-sensitizing mutations (particularly exon 19 del or exon 21 L858R) and tumours that are poorly differentiated.

Metastatic NSCLC patients with EGFR-sensitizing mutations have improved treatment efficacy when treated with an EGFR-TKI compared to their counterparts who are EGFR-wild-type. (25, 89) Patients with EGFR-sensitizing mutations were found to have superior OS compared to patients who were EGFR-wild-type while on gefitinib therapy (13.6 months vs. 10.4 months, p = 0.034). (90) Furthermore, Health Quality Ontario concluded that EGFR-mutant patients treated with an EGFR-TKI tend to have significantly higher RECIST responses compared to EGFR- wild-type patients. (91)

Patients with poorly-differentiated tumours tend to poorer OS when matched on NSCLC histology. Saijo et al. had initially observed that the median survival of patients with well- differentiated metastatic adenocarcinoma of the lung was higher than those with poorly- differentiated tumours (11.8 months vs. 5.6 months, p < 0.05). (92) This was confirmed by Barletta et al. who found that the median OS for patients with well-differentiated

adenocarcinoma of the lung was 72.4 months, 39.5 months for moderately differentiated adenocarcinoma, and 8.7 months for poorly differentiated adenocarcinoma. (93)

2.3.3 Measurement-based factors

Factors due to measurement error of PFS and treatment RECIST response can affect the documented efficacy of erlotinib on clinical trials. The measurement of PFS is highly dependent on the frequency of follow-up, the measurement of the tumour and the clinical acumen of progression. Because patients are generally followed up every six to eight weeks, it is possible that patients may progress during the gap in between CT scans. Therefore, PFS measurements may be falsely inflated by up to six to eight weeks. This is a clinically significant difference considering that many systemic therapies had been approved on the market with an improvement of PFS by only 1.3 weeks. (4, 34)

Furthermore, there is inherent error with the measurement of tumour size. Various radiologists may choose different dimensions as their obligatory longest and shortest dimensions, which may spark debates about the occurrence of progression and need to alter treatment. (42) Moreover, splice errors may occur in which the tumour’s size may vary by a few millimetres depending on how the images had cross-sectioned the scanned area; however, the difference of a few millimetres may not be critical in determining the occurrence of progression.

Finally, the interpretation of progression in the palliative setting carries an inherent degree of subjectivity when evaluating therapeutic response, which can deeply affect the measurement of PFS. Consider two scenarios of progression where one patient had several new pulmonary nodules appear in the scan and another patient whose main tumour had shrunk by 10%, but a new liver metastasis has appeared on the scan. Clinicians are more inclined to consider the first patient to have failed therapy and switch to another line of therapy, but the second patient would be continued on therapy and undergo radiation for the new metastasis that do not seem to respond to current therapy. Therefore, the two patients would have significantly different PFS values despite the fact that both patients have technically progressed from their therapy. These sources of error in accurately determining the PFS of an agent can contribute towards uncertainty 29

about the true efficacy of the drug. Models that predict exactly the moment when progression occurs in the target lesion circumvent issues in the measurement and interpretation of tumour progression in all cancer patients.

2.3.4 Erlotinib-based factors

Erlotinib-based factors are plentiful and those that affect drug concentration in the serum (PK of erlotinib), and drug penetration into the intracellular environment of the NSCLC cell (PD of erlotinib).

2.3.4.1 Factors affecting population pharmacokinetics of erlotinib

There are many factors that affect the PK profile of erlotinib. Variability in pharmacokinetics exists from absorption to elimination of erlotinib.

2.3.4.1.1 Absorption

Alterations in stomach pH can affect the absorption of erlotinib. Such alterations may be drug- induced or disease-induced. Interactions with medications (e.g. histamine-2 receptor antagonists (H2RAs), proton pump inhibitors (PPIs) and antacids) or conditions (e.g. achlorhydria, hypochlorhydria) that increase stomach pH, as well as the presence of food may affect the dissolution of erlotinib. As only molecules in solution can be absorbed, this will directly affect the amount of erlotinib absorbed. Indeed, concurrent use of acid-suppressing medications reduce the peak concentration (C max ) and AUC of erlotinib and OSI-420 by around 48-69% compared to

values in patients not on acid-suppressing medications (Table 1). The reduction in C max and AUC values may be less in H2RAs compared to PPIs because H2RAs are reversible antagonists that have incomplete stomach acid blockade, while PPIs are irreversible blockers that cause stomach acid blockade. In contrast, the reduction in erlotinib absorption from staggered H2RA use is much less than that seen with consistent use of PPIs and H2RAs, as their C max and AUC

values of erlotinib and OSI-420 are within 81% to 85.1% of the values without any acid- suppressive therapies (Table 5). (94) Therefore, the dosing and timing strategies, as well as the presence and type of acid-suppressive therapies may affect erlotinib exposure.

Table 5: Effect of acid-suppressing therapies on erlotinib Cmax and AUC (adapted from Kletzl et al.) Proportion with Proportion with Proportion with PPI/without PPI concurrent staggered H2RA/without H2RA H2RA/without H2RA Erlotinib OSI-420 Erlotinib OSI-420 Erlotinib OSI-420 Cmax 61% 69% 54% 62% 85.1% 82.6% AUC 48% 58% 33% 41% 83.0% 81.0%

Cmax : maximum serum concentration AUC: area under the curve OSI-420: main active metabolite of erlotinib PPI: proton pump inhibitor H2RA: histamine-2 receptor antagonist

The effect of conditions characterized by decreased or no secretion of stomach acid on erlotinib absorption (as measured by C max and AUC) has not been studied; however, it is theorized that patients with decreased or no secretion of stomach acid will have a profoundly reduced systemic exposure of erlotinib.

The effect of food taken concomitantly with erlotinib has been studied. Patients who were fed a

meal concurrently with erlotinib administration for a week had 33% higher C max and AUC values compared to patients who were administered erlotinib in a fasting state. (95) This is likely because the presence of food in the stomach stimulates the production of stomach acid, which in turn makes the stomach environment more acidic and more conducive to the dissolution of erlotinib. Since increased dissolution of erlotinib can increase erlotinib bioavailability, which may increase toxicity of the medication, patients are recommended to take erlotinib on an empty stomach.

2.3.4.1.2 Distribution

Since erlotinib is highly protein-bound, drug interactions and conditions that modulate protein binding of erlotinib can affect the free fraction of erlotinib, the portion of drug that is able to 31

penetrate into the tumour cells. Patients who take erlotinib concurrently with highly protein- bound medications (e.g. phenytoin, warfarin) may displace bound drug from the plasma protein. (96)

Furthermore, conditions, such as cancer anorexia-cachexia syndrome (CACS), chronic renal disease and increasing age that decrease protein production can reduce protein binding of erlotinib. (96) This is a significant issue in the metastatic NSCLC patient population, as CACS affects 61% of NSCLC patients and 80% of patients with advanced cancer (97), the median age upon lung cancer diagnosis is 70 years of age (2), and chronic renal disease affects around 25% of patients over 70 years of age. (98, 99) Conversely, cancer, as a chronic inflammatory disease, is a condition that can increase alpha-1-acid glycoprotein, also known to be an acute phase reactant and plasma protein that binds primarily to neutral or weakly basic molecules, which can also increase protein binding and reduce the free fraction of erlotinib. (96, 100) Effective control of metastatic NSCLC by erlotinib can contribute towards a reduction of inflammation within the patient, which in turn may lead changes in erlotinib alpha-1-acid glycoprotein binding. While it is hypothesized that CACS and proteinuria associated with chronic renal disease may lead to decreased protein binding, and inflammation may increase erlotinib binding to alpha-1- acid glycoprotein, the overall effect of protein displacement associated with the simultaneous presence of these factors on erlotinib efficacy is currently unknown.

2.3.4.1.3 Metabolism

Variation in CYP450 activity may further contribute to the pharmacokinetic variability of erlotinib. CYP3A4 is known to be polymorphic, with over 30 single nucleotide polymorphisms (SNPs) and differences in CYP3A4 gene transcription copies contributing towards a 40-fold variation in its activity among patients. (101, 102) Similarly, genetic variability is attributed to CYP3A5-mediated polymorphism, with significant contribution from SNP within intron 3 (A6986G transition) being the primary cause of CYP3A5 polymorphism, also known as CYP3A5*3. (101, 103) This mutation is observed to yield low hepatic CYP3A5 protein content and thus lead to low levels of CYP3A5-mediated metabolism.

In addition to genetic polymorphisms, drug interactions that induce or inhibit CYP450 function can significantly alter drug metabolism. CYP3A4 inhibitors, such as macrolides (azithromycin, clarithromycin), antifungals (ketoconazole, itraconazole), protease inhibitors (indinavir, ritonavir) and grapefruit juice, can lead to significant decreases in CYP3A4-mediated metabolism of erlotinib, and can increase AUC by up to two-fold. (104) Conversely, CYP3A4 inducers, such as rifampin, some antiepileptics (carbamazepine, phenytoin, oxcarbazepine) and St. John’s Wort, can increase CYP3A4-mediated metabolism, leading to a decrease of 34% in erlotinib AUC. (105) Despite the fact that CYP3A4 is the primary metabolic pathway of erlotinib (104), interactions that induce CYP1A1/1A2 function can lead to clinically significant changes in erlotinib exposure. Actively smoking metastatic NSCLC patients have erlotinib AUC half that of non-smoking NSCLC patients on the same dose. (106) In order to compensate for the smoking-related induction of CYP1A1/1A2, smokers would have to double the erlotinib dose typically used in non-smoking patients. (107)

Besides drug interactions, inflammation also plays an important role in the variability of CYP450-mediated metabolism of erlotinib. Many pro-inflammatory cytokines have been implicated to downregulate or decrease the activity of various CYP450s. Examples of these cytokines include INF-γ, IL-2, and IL-6, all of which downregulate both CYP1A2 and CYP3A4. (108, 109) Given the fact that each patient has a variation in type, proportion and amount of cytokines released over time, there is likely a significant variability in their effects on erlotinib metabolism and therefore, erlotinib exposure; however, the clinical significance of these effects remain unknown.

Changes in CYP450 metabolism may also lead to differences in the production of active metabolites, predominantly OSI-420. While this has been less studied, the observation that OSI- 420 is produced from the CYP450-mediated metabolism of erlotinib (110) suggests a decrease in CYP450 metabolism leads to a higher ratio of erlotinib/OSI-420, while an increase in CYP450 metabolism leads to a lower ratio of erlotinib/OSI-420.

2.3.4.2 Factors affecting pharmacodynamics of erlotinib

There is significant heterogeneity in the ability of erlotinib to inhibit growth and induce apoptosis in cancer cells within the tumour. This variation is likely due to the different EGFR testing methods, genetic variability in tumours between patients (intertumour heterogeneity), intrinsic genetic variation observed among the cells within the tumour (intratumoral heterogeneity), as well as the extrinsic conditions of the tumour microenvironment affecting the sensitivity of erlotinib on the tumour.

2.3.4.2.1 Methods of EGFR testing

Different methods have been developed to perform EGFR mutation testing in NSCLC, each with their own specificity and sensitivity. Lopez-Rios et al found that three different polymerase chain reactions (PCR) kits – The Therascreen EGFR29 Mutation Kit, Sanger Sequencing and 454 Sequencing) – agreed 98.8% of the time when the tumour was EGFR-mutant, but only agreed 79.3% of the time when the tumour was EGFR-wild-type (EGFR-WT) (111), implying that there may be patients who could be falsely classified as EGFR-mutant when they lack the mutation. As erlotinib is not as potent and efficacious against EGFR-WT NSCLC compared to EGFR-mutant NSCLC, this may lead to variability in therapeutic efficacy in patients classified as EGFR-mutant NSCLC.

2.3.4.2.2 Intertumoural heterogeneity

The adenocarcinomas of various patients may have different EGFR mutations that, in turn, lead to differences in susceptibility of erlotinib. Erlotinib’s affinity to the ATP binding pocket of EGFR and therefore, its therapeutic effect to induce cancer cell death and prevent proliferation may differ depending on mutations affecting the tyrosine kinase domain of the EGFR. The concentration at which 50% of tumour inhibition is observed (IC50), frequently measured in in vitro studies, is used as a surrogate for the affinity of erlotinib to the EGFR binding pocket and erlotinib’s potency as an EGFR TKI in inhibiting tumour growth. Erlotinib typically has an IC50

of 0.047 mg/L (110 nM) on EGFR-WT. (69) The IC50 of erlotinib on tumours containing EGFR with the exon 19 del mutation can range from 0.0021 to 0.0086 mg/L (5 to 20 nM), while the IC50 on tumours with EGFR with the exon 21 L858R mutation can range from 0.003 to 0.012 mg/L (6.9 to 30 nM). (9, 10) Thus, EGFR-mutant tumours are more susceptible to erlotinib compared to EGFR-WT tumours. However, there are some EGFR mutations that cause tumours to have decreased susceptibility to erlotinib. Tumour cells with an exon 20 T790M point substitution have a mean IC50 of over 1.72 mg/L (4000 nM), which means that the vast majority of cancer cells with this mutation would be resistant to erlotinib. (8, 69) Given the dependence of erlotinib efficacy on EGFR mutation status, molecular testing is required to identify patients with susceptible mutations.

2.3.4.2.3 Intratumoural heterogeneity

In addition to the intertumour heterogeneity among tumours of the same histology, genetic heterogeneity within lung cancer tumours is substantial. It has been observed that a median of 30% of mutations were only found in certain parts of the tumour (range = 4-63%). (112) As a result, the probability of a false negative is on average 42% (range 0-67%). This implies that tumours should be sampled multiple times during biopsies to gain an accurate representation of the mutations present within the tumour. Furthermore, many of these mutations led to cells within the same tumour of different subclonal origins. Taniguchi et al. found both EGFR-WT and EGFR-mutant cancer cells within the biopsied tissue in 6 of 21 (29%) patients and found that this heterogeneity influenced survival of these patients. Patients with EGFR genetic heterogeneity present in their tissue samples had significantly shorter time to disease progression and overall survival compared to patients of tumours with less EGFR genetic heterogeneity. (113) In two other studies that each looked at roughly 100 tumour-metastases pairs, discordance in EGFR mutation status was present in 7-16.8% of primary tumour-lymph node metastases pairs. (114, 115) This discordance is likely due to a combination of factors of locoregional differences in EGFR mutation status within the tumour, as well as the molecular profile of these cells changing during metastasis to lymph nodes due to selection of cells that have certain oncogenic mutations. (116)

2.3.4.2.4 The tumour microenvironment

In addition to the tumour itself, components of the tumour microenvironment, including cancer- associated fibroblastic cells (CAFs), vasculature and immune cells, affect the tumour’s susceptibility to various therapeutic agents, including erlotinib. CAFs promote tumorigenesis by enhancing extracellular matrix production for improved tumour support, releasing various cytokines (vascular endothelial growth factor (VEGF), hepatocyte growth factor (HGF), epidermal growth factor (EGF), insulin-like growth factor-1 (IGF-1), etc.) that promote proliferation, and repairing damage of the tumour microenvironment caused by treatment. (117, 118) Therefore, to overcome the oncogenic effects of CAFs in the lung tumour microenvironment, erlotinib must reverse the proliferative effects resulting from EGF-EGFR binding. Nonetheless, varying CAFs and their cytokines may exist among different regions within the tumour, which may account for intratumoral heterogeneity in susceptibility to and efficacy of erlotinib.

Angiogenesis is increased significantly in the tumour microenvironment to increase nutrient uptake by the cancer cells and thus promote proliferation. The abnormal vasculature within the tumour paradoxically serves as a barrier to drug delivery. The increased density of the vasculature causes vessels to be compressed and blood flow to be obstructed (118), which generates new hypoxic tumour regions in which cells grow and proliferate more slowly and thus become more resistant to therapy. This leads to variability in treatment susceptibility within the tumour, with hypoxic areas being more treatment-resistant and well-nourished cells being more treatment-sensitive. (118, 119)

In addition to CAFs and the tumour’s abnormal vasculature, the immune system influences tumour growth support and treatment response. The extracellular matrix houses infiltrating immune cells (IICs), which supply many growth mediators (EGF, transforming growth factor-β (TGF-β), tumour necrosis factor-α (TNF-α), fibroblast growth factors (FGFs), various interleukins, etc.) that promote proliferation of the cancer and stromal cells. They also help to repair any treatment-inflicted damages on the extracellular matrix by expressing various

proteolytic enzymes that cleave and modify the structure of the extracellular matrix. These changes in the extracellular matrix are thought to activate multiple downstream signaling pathways that sustain neoplastic proliferation. (117) In addition to inducing treatment resistance and promoting tumour proliferation, the tumour can downregulate T-lymphocytes by expressing various ligands, particularly the tumour-associated programmed cell death 1 ligand 1 (PD-L1), which binds to programmed cell death 1 (PD-1) receptor on the T-lymphocyte. This ligand- receptor interaction is attributed to the inhibition of CD8+ cytotoxic T-lymphocyte proliferation, survival and effector function, induction of apoptosis of T-cells and the differentiation of CD4+ T-cells into regulatory T cells, which further suppresses the immune system and reinforces immune tolerance of cancer cells. (120) Differences in the tumour’s ability to use the IICs for neoplastic proliferation and take advantage of the PD-L1-PD1 receptor interaction to evade the immune system would lead to differences in tumour sensitivity to erlotinib.

2.4 Current pharmacokinetic-pharmacodynamic models of erlotinib

Wu et al. constructed a pharmacokinetic-dynamic (PK-PD) model to describe the relationship between plasma concentration of erlotinib and tumour volume in mice with xenografted NSCLC tumours. (121) In this study, female BALB/c nude mice were subcutaneously injected with SPC-A-1 cells (a NSCLC EGFR-WT cell line) in their right flanks. Once the NSCLC cells had grown to a size of 0.15-0.25 cm 3, the mice were administered erlotinib at a daily dose of 0mg/kg (control), 4mg/kg, or 50mg/kg for three weeks or until the death of the mouse. The values of PK parameters of erlotinib were calculated based on a two-compartment PK model. The degree of EGFR inhibition was described by the proportion of inhibition of phosphorylated EGFR. Tumour growth in the presence of different doses of erlotinib was described using the Gompertz model. Phosphorylated EGFR inhibition was approximately 50% and 90% at doses of 4mg/kg and 50mg/kg, respectively and tumour growth slowed with increasing doses of erlotinib.

However, there are shortcomings to this model that limit its application in clinical practice. The in vivo model of a subcutaneously injected tumour into a specimen poorly simulates the event of

metastatic cancer as the xenografted tumour is quite isolated from the rest of the specimen’s vasculature. Consequently, it does not model the presence of metastases or the process of hematogeneous and lymphangitic spread of cancer cells, which is present in metastatic NSCLC patients. (122) The isolated tumour also does not replicate the tumour microenvironment in which erlotinib is present. In addition, the xenografted tumour was constructed based on a single NSCLC cell line. The homogeneity of the NSCLC cells within the xenografted tumour does not reproduce the intratumoral heterogeneity that is known to exist in human NSCLC. While SPC- A-1 cell line is thought to exhibit similar pathological features as lung adenocarcinoma (123), it is unknown whether this cell line is similar to NSCLC in clinical practice. An additional limitation is the population chosen to construct the model. While it is likely more feasible to use a rat instead of a human as the basis of the model, there is poor translation of clinical knowledge from animal to human model for the main reason that the mice population does not represent the human patients typically treated in practice. (124) Mice included in study are homogeneous in their diets and living conditions and are generally healthy and young. In contrast, human lung cancer patients are typically older, have additional comorbidities and are exposed to a multitude of potential carcinogens in their living conditions. (2, 124) Future PK-PD models must accurately represent metastatic NSCLC patients and the presentation of their tumours to be more valid and generalizable to clinical practice.

2.5 Current literature on the exposure-efficacy relationship of erlotinib

There is sparse literature on the relationship between exposure and clinical outcomes at the patient level. Tiseo et al. conducted a prospective cohort study that determined mean serum concentrations observed in metastatic NSCLC patients on erlotinib that corresponded to each response group according to RECIST criteria. (11) A dose-response relationship was evident in the study findings; mean values of erlotinib serum levels after seven days of therapy were 3.44 µmol/L (1.48 mg/L) in patients with progressive disease, 4.00 µmol/L (1.72 mg/L) in those with stable disease, and 5.22 µmol/L (2.24 mg/L) in those with partial response. While this was not statistically significant in the study, it suggests that that treatment response, measured as tumour

shrinkage, may improve with increasing erlotinib serum concentrations. This study also found that patients had greater severity of toxicities with increasing erlotinib serum concentrations. Mean values of erlotinib serum levels after seven days of therapy were 3.00 µmol/L (1.29mg/L) in patients with Grade 0 rash, 3.20 µmol/L (1.38mg/L) in those with Grade 1 rash, 3.77 µmol/L (1.62mg/L) in those with Grade 2 rash and 6.14 µmol/L (2.64mg/L) in those with Grade 3 rash. (11)

Nonetheless, there are some limitations that restrict the clinical utility of these findings. The distribution of the mean serum concentrations in each stratum of treatment response was not reported in this study. Since the coefficients of variation in these response groups ranged from 40-65.7% (11), it is likely that there was substantial overlap among the concentrations of patients in each response group and that clinicians would not likely be successful in reaching optimal efficacy by dosing patients to a mean of serum concentration alone. Furthermore, the serum concentrations were not stratified based on EGFR mutation status of the patients. In the study, 16% of the patients had EGFR mutant NSCLC (exact mutation not specified), 34% of them had EGFR-WT NSCLC and 50% of them had an unknown EGFR status. (11) Because erlotinib is more potent against NSCLC with either exon 19 del or exon 21 L858R mutations compared to EGFR-WT NSCLC, and is known to produce more profound responses in EGFR-mutant patients, it would be worthwhile to determine whether EGFR mutant tumours respond to lower erlotinib serum concentration than EGFR-WT NSCLC tumours. A third key limitation to this study is the use of treatment outcome as RECIST response instead of PFS, the main outcome for evaluating efficacy of anticancer therapies for metastatic NSCLC, when correlating erlotinib exposure to its efficacy. While RECIST response may help to determine whether therapy garners any benefit to patients, it does not provide any insight about the duration of benefit for patients. Since PFS is more clinically relevant as an efficacy outcome to patients and clinicians, future research should focus on correlating erlotinib serum concentrations to PFS. The construction of a model that connects population pharmacokinetics of erlotinib and its pharmacodynamics from the cellular level to the tumour level to its clinical benefit at the population-level will provide insight on this intricate relationship.

2.6 Summary

Erlotinib has revolutionized the treatment of metastatic NSCLC patients and has made headway in improving efficacy in these patients, particularly in patients with tumours that have EGFR- sensitizing mutations. Nonetheless, prognosis in metastatic NSCLC patients continues to be poor, and the variability in outcomes of patients on erlotinib suggest a large potential for more effective use of erlotinib. Unlocking this potential involves understanding the relationship between erlotinib pharmacokinetics, pharmacodynamics at the cellular and tumour levels, as well as constructing a PK-PD model that simulates how this relationship affects erlotinib’s clinical effect on progression-free survival at the population level. Therefore, using current literature on the pharmacokinetics and pharmacodynamics of erlotinib, the writer has developed and validated a PK-PD model of erlotinib that simulates progression-free survival in metastatic NSCLC patients.

Chapter 3

Research Objective and Methodological Steps

3 Research Objective and Methodological Steps 3.1 Objective

The objective of this project was to develop a population-based PK-PD model that simulates progression-free survival in patients with metastatic non-small-cell lung cancer taking erlotinib.

3.2 Methodological Steps

The following steps were completed to accomplish the overall objective:

1) Construct a population-based pharmacokinetic model that accurately describes the pharmacokinetic profile of erlotinib reported in the literature 2) Construct a pharmacodynamic model that accurately portrays the pharmacological action of erlotinib 3) Construct a model that tracks the effect of erlotinib on tumour size over time 4) Integrate the PK, PD and tumour growth components into a single PK-PD model 5) Fit the PK-PD model based on the findings from previous literature that are pertinent to the pharmacokinetics, pharmacodynamics and human clinical trial efficacy data of erlotinib 6) Validate the simulated data of the PK-PD model using phase II and III clinical trial data. 7) Simulate a dose-PFS relationship 8) Determine the minimum effective concentration of erlotinib required to prevent progression 9) Determine the minimum effective concentration of erlotinib required to induce at least a partial response in metastatic NSCLC patients

Chapter 4

Methods

4 Methods 4.1 Overview of model

The pharmacokinetic-pharmacodynamic (PK-PD) model consists of three components – a pharmacokinetic (PK) component, a pharmacodynamic (PD) component and a tumour growth (TG) component. The PK and PD components are required to construct the TG component, which is designed to simulate tumour growth over time. The combination of these three components yields the final PK-PD model that can predict population-based estimates of PFS and best RECIST response in a simulated cohort of patients.

4.2 The pharmacokinetic model component

The PK component describes the pharmacokinetics of erlotinib. Using a one-compartment oral administration model with first order absorption and elimination, it uses population-based values of PK parameters of erlotinib to calculate the total serum concentration of erlotinib over time. The equation of the PK model used to calculate erlotinib concentration over time is

(8)

Table 6 provides a summary of PK parameters (definition, unit and pooled values of mean and variability) used in the model.

Table 6: Parameters of pharmacokinetic component of model Parameter Definition Unit References * F Oral bioavailability of erlotinib N/A 125 Dose ** Admi nistered daily dose of erlotinib mg -1 Ka Absorption rate constant hour 54, 126 t1/2 Half -life hour 4, 104, 127 *** -1 Ke Elimination rate constant hour **** Vd Apparent volume of disribution L 4, 54, 126 t Time from first administered dose of erlot inib hour C(t) Serum concentration of erlotinib over time mg ∙hour/L N/A = non-applicable *References of studies from which population estimates of PK parameters are taken **Standard therapeutic dose of erlotinib ***Calculated from t 1/2 ****Calculated as V d/F

To determine the population-based values of each PK parameter, a literature search was performed to identify all English studies documenting mean values of population-based PK parameters of erlotinib in adult cancer patients with normal hepatic and renal function (Table 7). Studies did not have to assume a certain PK model. The inclusion and exclusion criteria of the literature search are summarized in Table 8.

Table 7: Search strategy of literature research of population pharmacokinetic studies of erlotinib Parameters Comments MeSH terms Exp erlotinib/pk [Pharmacokinetics] Limits Human Adult 18-64 and 65+ Databases used EMBASE, PubMed, Medline References of selected studies were also scanned to ensure completeness of literature search.

Table 8: Inclusion and exclusion criteria of literature search for population pharmacokinetic studies of erlotinib Inclusion criteria Exclusion criteria - Erlotinib used as monotherapy - Patients with impaired renal function - Adult cancer patients only (age 18 and - Patients with impaired hepatic older) function - Mean and standard deviation of half- - Erlotinib taken at other than once life (t 1/2 ), absorption rate constant daily doses (K a), volume of distribution (V d), time - Documenting median and range at peak concentration (Tmax), peak values of PK parameters concentration (Cmax), trough - Not written in English concentration (Cmin), area under the curve (AUC), and/or clearance (Cl/F) provided

The literature search was not intended to be a full systematic review, but only intended to get as much erlotinib PK studies as reasonably possible to represent the data and interpatient variability of the patient population. Therefore, a library scientist and a second reader was not required for this process.

Once the literature search was conducted, articles reporting the pharmacokinetic variables of interest were selected. These parameters (V d, k a and t 1/2 ) were recorded and the average of mean and standard deviation values among the selected studies (weighted by the sample size of each PK study, as well as the coefficients of variation were calculated (Table 9) as population-based estimates of the mean and measure of variability of each PK parameter of the model.

Table 9: Formulae used to determine population-based estimates of pharmacokinetic parameters Parameter Formula

Weighted mean value ( )

Weighted standard deviation value ( )

Coefficient of variation ( )

µi = mean value of study i pi = sample size of value i si = standard deviation of study i

Once the population estimates of mean and variability of V d, k a and t 1/2 were determined, they

were used to calculate population estimates of the T max , Cl/F, AUC, C min and C max . Monte Carlo simulations (Oracle Crystal Ball basic edition, USA) of 1,000,000 iterations were used to produce a mean and standard deviation of T max , Cl/F, AUC, C min and C max . The PK model was evaluated for its ability to accurately simulate the pharmacokinetic profile of erlotinib using internal and external validation processes. Internal validation was performed by comparing the

population estimates with the means, standard deviations and coefficients of variation of V d/F, k a

and t 1/2 documented in each study included in the calculation and comparing calculated population estimates with reported means and measures of variability of other PK parameters

(T max , Cl/F, AUC, C min or C max ) that may be documented in included studies. The PK model was considered to have good internal validity if the ratio between the model PK mean estimates and reported study means was within 75-125%. 25% is considered to be the amount of interpatient variability of erlotinib. (128)

External validation was completed by comparing population estimates with means and measures

of variability of V d/F, k a and t 1/2 , as well as T max , Cl/F, AUC, C min or C max as documented in the meta-analysis of erlotinib population PK conducted by Petit-Jean et al. (129) that reported the values of all population PK parameters pooled from various erlotinib PK studies. The values of this study were not used to calculate the population estimates. Similar to the evaluation of internal validity, the PK component was considered to have good external validity if the ratio between the model PK mean estimates and reported values from Petit-Jean et al (129) was within 75-125%. Since these PK studies used either single or multiple-dose regimens and could yield

different values of PK parameters, a 2-sided Student T-test (with Bonferroni correction of α = 0.05/6 = 0.0083) was performed at α = 0.05 for each PK parameter to determine whether there were any differences in the mean values generated between the single-dose and multiple-dose PK studies. No T-test was conducted for the PK parameters in which the value was only generated from multiple-dose studies. Once internal and external validation processes determined that the PK model accurately described the population-based PK profile erlotinib, of the construction of the PD component began.

4.3 The pharmacodynamics model component

The PD component describes the pharmacodynamics of erlotinib based on its mechanism of action within the cytoplasm of the cell and its potency based on in vitro data. The equation of the PD model quantifies the degree to which erlotinib inhibits tumour growth, which is described by the kill constant (k). This equation,

(9)

is based on a sigmoidal E max equation in which maximal degree of tumour inhibition by erlotinib is achieved with full inhibition of tyrosine kinase domain of EGFR. Table 6 summarises the parameters used in the PD equation.

Table 10: Parameters of pharmacodynamics mode component Parameters Definition k (k r, k s) Kill constant The degree by which erlotinib contributes towards overall tumour growth inhibition by inhibiting proliferation and inducing apoptosis The kill constant for erlotinib- resistant* cells is “k r”. The kill constant for erlotinib-sensitive** cells is “k s”. EC 50 (EC 50,r , EC 50,s ) Half maximal effective c oncentration Concentration at which erlotinib reaches 50% of its maximal killing ability (measured by its kill constant) EC 50,r is the half maximal effective concentration of erlotinib-resistant cells EC 50,s is the half maximal effective concentration of erlotinib-sensitive cells Emax (E max,r , E max,s ) Maximal effective concentration Concentration at which erlotinib reaches 100% of its maximal killing ability 48

(measured by its kill constant) Emax,r is the maximal effective concentration of erlotinib for erlotinib-resistant cells Emax,s is the maximal effective concentration of erlotinib for erlotinib-sensitive cells

[C ss ] Average serum concentration of erlotinib at steady state (mg/L) , as derived from the PK component fr Adjustment factor The proportion of intracellular erlotinib concentration to the total serum erlotinib concentration to intracellular erlotinib concentration The value of f r in the PD equation is 0.00005 *Erlotinib-resistant cells = cells against which erlotinib have low potency **Erlotinib-sensitive cells = cells against which erlotinib have high potency

The assumption was made that erlotinib shifts the balance between cell growth and death in favour of death by simultaneously inducing death and hindering growth of cancer cells. The degree of “shift” in this balance towards death is quantified by the kill constant (k). However, the magnitude of k is dependent on the effect of erlotinib on tumour growth inhibition and the extent of cell exposure to erlotinib. The former is represented by the parameters E max and EC 50. The latter is represented by the intracellular concentration of erlotinib, which is determined by multiplying the serum erlotinib concentration by the adjustment factor (f r), which was estimated

to be 0.00005. The distributions surrounding the EC 50 and E max reflect the presence of intratumoral heterogeneity in susceptibility to growth inhibition caused by erlotinib. Similarly, the variability of serum erlotinib concentration reflects the presence of interpatient variability of

erlotinib exposure. Therefore, EC 50 , E max and [C] are all variables that influence k and erlotinib’s potency as a growth inhibitor of NSCLC cells.

Because erlotinib is more potent in some cells than others, the model divides the cells within the simulated tumour into two groups: a group that is relatively more sensitive to erlotinib (named simply sensitive cells) and a second group that is relatively more resistant to erlotinib (or the resistant cells). The sensitive group consists of cells that are generally more susceptible to be inhibited by erlotinib and were modeled against lung cancer cells with exon 19 del or exon 21 L858R mutations. The resistant group were modelled against exon 20 T790M mutant lung cancer cells to which erlotinib is observed to have generally low potency. Therefore, there is an

EC 50 and E max associated with resistant and sensitive groups of cells.

In order to calculate k r and k s using Equation 9, the values of the variables [C ss ], EC 50,r , EC 50,s ,

Emax,r and E max,s were determined as follows. [C] was calculated using Equation 8. EC 50 values were taken from in vitro studies conducted by Yuza et al and Ohashi et al. (9, 10) Since EC 50 values of NSCLC cells differ based on their susceptibility to erlotinib, EC 50 values of more erlotinib-sensitive cells were referenced from in vitro data of NSCLC cells with exon 19 del and

exon 21 L858R mutations, while EC 50 values were referenced from in vitro data of NSCLC cell lines carrying the exon 20 T790M mutation for cells that were more resistant to erlotinib. Since different studies reported a variety of EC 50 values of different cell lines with the same mutations, these EC 50 values were pooled into a single EC 50 value each for resistant and sensitive cell groups. A standard deviation was calculated based on the variability demonstrated in the values

reported in in vitro studies. (9, 10) E max,r and E max,s values were determined using a fitting process to be explained in a later subsection. Because the concept of k has not been studied previously in in vitro studies, this component could not be externally validated. Nonetheless, the concentration-kill constant curves generated from the PD equation was compared with published concentration-tumour inhibition curves and estimates of EC 50 values reported in the in vitro study conducted by Chmielecki et al. (8) to determine potential validity of simulated data.

4.4 The tumour growth (TG) model component

The TG component of the model tracks tumour size over time by calculating the number of cells present within the tumour over time. The TG equation is

(3)

Table 11 defines each parameter of the equation and Figure 1 summarises the process by which the equation was derived.

Table 11: Parameters of the tumour growth equation Parameters Definition N(t) Total number of tumour cells over time A Initial number of tumour cells x Proportion of resistant cells to the initial number of tumour cells gs, g r Growth constants for sensitive and resistant groups of cells ks, kr Kill constants for sensitive and resistant groups of cells t Time from first administered dose of erlotinib (hours)

Figure 1: Derivation of the tumour growth equation. The first step (blue) denotes the basic exponential growth model. The second step (green) incorporates the use of difference in growth and kill constant to represent the overall growth constant (k 0). The third step (pink) represents the concept of intratumoral heterogeneity in erlotinib susceptibility using a simplified model of two groups of tumour cells with different sensitivities to erlotinib.

The TG component is based on a mathematical function describing exponential cell growth, which is expressed as Equation 7, where

- N(t) is a function of number of cells over time - A is the initial number of cells at time 0

- k0 is the net growth constant, and - t is time in hours

However, in order for the model to reflect the effect of erlotinib in shifting the growth-death balance within the tumour in favour of death, k 0 was deconstructed as the difference of the cells’ growth constant (g), the natural growth constant of the tumour in the absence of drug, and erlotinib’s kill constant (k), calculated from the PD component, to reflect the pharmacological actions of erlotinib on the cancer cells. While g and k are named constants, they are by definition a mean of each tumour’s rate constant within the simulated patient population. This is reflected by the association of a population-based measure of variability, defined by a standard deviation, to each constant. The mean growth and kill constants may differ among cell groups of differing susceptibilities to erlotinib. It is thought that cells that are more resistant to erlotinib may have different growth rates compared to those more sensitive to erlotinib due to the presence of hypoxia and/or the presence of other oncogenic mutation that promote proliferation and metastases independently of EGFR. Furthermore, erlotinib-resistant cells are less likely to be killed compared to erlotinib-sensitive cells due to their intrinsically reduced susceptibility to erlotinib as a growth inhibitor.

Similar to k in the PD equation, the TG component has a separate value of g for resistant and sensitive cells, denoted as g r and g s, respectively. The values of k r and k s were calculated from Equation 9 of the PD model component. The model component employs a distribution for each mean value of g and k for sensitive and resistant groups of cells to account for heterogeneity in their intrinsic proliferation rate and their propensity to die with erlotinib exposure. Furthermore, the proportion of resistant cells to the total number of tumour cells (x) can differ among patients. It is hypothesized that tumours with a greater proportion of resistant cells are intrinsically more resistant to erlotinib as a whole and therefore exhibit less net growth inhibition compared to tumours with a smaller proportion of resistant cells. Similar to the growth constants, the model includes the proportion parameter as a mean with an associated measure of variability as the standard deviation.

To reflect the presence of resistant and sensitive groups of tumour cells, the TG equation is a sum of exponential growth functions – one each for resistant and sensitive cells. By being able to calculate the number of resistant and sensitive cells within the tumour over time, the model is able to detect the minimum size that the tumour reaches in the presence of erlotinib, the time at which the minimal size is reached and when the tumour has progressed. Progression is defined as the point in time when the tumour size has grown by at least 20% of its minimal size. (42) The ability of the model to track the minimal size of the tumour due to growth inhibition allows it to predict best overall response of population by RECIST criteria. Furthermore, the ability of the model to track when progression occurs allows the model to determine time to progression (TTP).

4.5 The fitting process and determining internal validity

In order for the PK-PD model to simulate TTP, the values of each parameter from all three equations were derived by calculation, from a reference of prior literature or through fitting (Table 12).

Table 12: Derivation method for the value of all model parameters Parameters Derivation method Vd/F Pooled from references 4, 35, 93 Ka Pooled from references 35, 93 Ke Pooled from references 4, 55, 94 Dose Reference (6) C(t) Calculation Fu Reference 37 Emax_r, Emax_s Fitting EC50 r, EC50 s Pooled from references 10, 11 kr, k s Calculation gr, gs Fitting x Fitting A Reference based on tumour size from radiologic scans

The parameters that required fitting were x, g r, g s, E max ,r and E max ,s, based on an initial number of cells (A) within the tumour of 5 billion cells. This was based on a tumour size of 1 billion cells, 3 or 1 cm (130), which is the smallest volume that can be detected by PET-CT scans and treated. 53

(131) These parameters were fit in a two-part process based on PFS data from BR.21 (4), which was extracted via determining the data points of the PFS Kaplan-Meier plot using Grab It! XP software (Datatrend Software). BR.21 was selected as the clinical trial of choice for the following reasons. It included PFS for both EGFR-mutant and EGFR-WT metastatic NSCLC patients, which is the patient population that most commonly receives erlotinib in clinical practice for treatment of metastatic NSCLC. It was one of two landmark trials that included a placebo arm instead of an active comparator arm and therefore, PFS data were available for metastatic NSCLC patients who did not receive active treatment, but had similar characteristics to those who received erlotinib. This is important because of the need for the model to determine the characteristics of the tumour and its growth in the absence of erlotinib. Furthermore, it was a large study consisting of 638 patients, which allowed a substantial amount of variability in erlotinib efficacy and tumour growth to be captured. It also included a post-hoc analysis in which the PFS data were stratified by EGFR mutation status, which provided insight into the differences in tumour characteristics and their effects on PFS, and the impact that erlotinib had on tumour growth and PFS among EGFR mutation subgroups. (132)

The data were used instead of TTP due to the lack of availability of TTP clinical trial data and the inability of the model to predict patient mortality from other causes. Therefore, fitting to PFS enabled the simulated TTP data to resemble clinical trial PFS data despite the fact that the TG equation was initially designed to track TTP. Tumour-based parameters associated with tumour growth and composition of resistant cells (g r, g s and x) were fit first, followed by erlotinib-based parameters associated with the potency of erlotinib as a tumour growth inhibitor (E max,r and

Emax,s ). This fitting process is essential in developing internal validity of the model’s ability to predict time to progression and best response.

The process of fitting the tumour-based parameters of the model against PFS clinical data of the placebo group generated the “base case” that uncovered how the simulated tumour grew in the absence of erlotinib exposure and how quickly an event occurred as a result of this tumour growth. An event was defined as the detection of progression by the model. Once the mean, distribution and standard deviation of the tumour-based parameters were established, the values of the erlotinib-based parameters were determined by fitting the model in the same manner

against PFS clinical trial data of the erlotinib trial group with the values of the tumour-based parameters in place. Fitting the placebo model before the erlotinib model, allowed for an initial understanding of the tumour growth characteristics before superimposing the erlotinib-induced effects of tumour growth inhibition.

Before the fitting process began, approximate values of parameters that required fitting were determined to establish a starting point for the fitting process. These calculations were based on the following assumptions:

- Emax was approximately 100 times greater than the value of EC 50 (9, 10) - The initial number of cells (A) is 5 billion cells - The proportion of resistant cells at t = 0 hours approximates 0.01 (133) - The median PFS of EGFR-WT patients is around 2 months on placebo and 3 months on erlotinib 150mg daily (4, 34) - Median PFS of EGFR-mutant (exon 19 or 21) patients is estimated to be 7 months on placebo and 10 months on erlotinib (134) - Objective response rates (ORR) of erlotinib were 10% in EGFR-WT patients (4, 34) and 70% in EGFR-mutant (exon 19 or 21) patients (36, 39)

A series of preliminary sensitivity analyses on each parameter (while keeping the remaining values of the untested parameters constant) were performed to determine the sensitivity of the model to the changes in the value of each parameter, which would determine the order by which each parameter was fit. The order was based on the model’s sensitivity to each parameter, starting with the most sensitive and concluding with the most robust to minimize the amount of error in the fitting of the last parameter.

Once the preliminary values of each parameter and the sequence in which the parameters were fit was established, the fitting process began. In each phase of fitting, the model underwent multiple cycles of fitting using a simulated patient population of 500 (an approximate size of a large oncology clinical trial) generated by Monte Carlo simulations (Oracle Crystal Ball basic edition, USA). The generated survival data set was graphed as a Kaplan-Meier curve. Its fit was compared to the Kaplan-Meier curve from the clinical trial both visually and through calculating a ratio of AUC sim /AUC CT , where AUC sim was the area under the Kaplan-Meier curve generated 55

from the simulated data and AUC CT was the area under the Kaplan-Meier curve reported from the clinical trial. Each cycle of fitting yielded the mean, distribution and standard deviation values of each parameter that corresponds to the best fit, or the curve generated from the model that was most visually superimposable upon the clinical trial curve and had the lowest value of

AUC sim /AUC CT . The sequential fitting of each parameter provided understanding of how each parameter affected the TTP of the patients in the simulation cohort. This allowed for further refinement of the model fit achieved on the last step of the fitting process in which the values of all parameters were adjusted simultaneously based on prior information on how each parameter affected TTP. The simultaneous fitting process aimed to reduce any accumulation of error from the fitting of the last parameter and allowed the model fit to be optimized. As there was no software available to us for model fit optimization in customized mathematical models, this process was done manually. Similar to the sequential fitting process, the best fit was determined based on visual inspection of fit in the shapes of the simulated and clinical trial curve and a calculation of the AUC sim /AUC CT .

The fitting process was initiated with a heterogeneous patient population with metastatic NSCLC of any mutation (noted here as all-comers), which was the population represented in BR.21. (4) Because erlotinib is known to yield higher PFS and RECIST response rates in metastatic NSCLC patients with exon 19 del or exon 21 L858R EGFR mutations compared to EGFR-wild type patients, the fitting process was repeated four times to allow the model to predict PFS data accurately based on EGFR mutation status. The subgroups included in addition to the all-comer group were the EGFR-mutant (exon 19 del/exon 21 L858R), EGFR-mutant (uncommon mutations), EGFR-wild type after progression from first-line treatment (EGFR-WT, 2 nd /3 rd line) and EGFR-WT after response to first-line treatment (EGFR-WT, maintenance). The PFS data from the BR.21 post-hoc study (132) and phase II/III clinical trials that corresponded to the subgroup of interest were used as references to fit the model by subgroup. Specifically, the data from the BR.21 post-hoc study (132), OPTIMAL (39) and EURTAC (36) were used as reference to fit the model for EGFR-mutant (exon 19/21) patients, the BR.21 post-hoc study data (132) for EGFR-mutant (uncommon mutations) patients, data from the BR.21 post-hoc study (132), TITAN (40) and a phase II trial by Kobayashi et al (134) for EGFR-WT, 2 nd /3 rd line patients and the data from SATURN (34) for EGFR-WT maintenance patients. No other trials evaluated the

efficacy of erlotinib in patients with uncommon EGFR mutations or EGFR-WT patients using erlotinib as maintenance therapy. These studies were chosen for internal validation because they were considered to be landmark clinical trials that were often used for the development of guidelines (31, 32) and they studied patients of different regions of the world to account for variations in survival from ethnicities and clinical practices.

4.6 The external validation and applications of the model

Once the fitting was completed for each subgroup and the model was able to simulate PFS data that were comparable to that of clinical trials used in the fitting process, the model underwent external validation. Due to the absence of RECIST data and availability of PFS data from other clinical trials in the uncommon EGFR mutation and EGFR-WT (maintenance) subgroups, external validity of the model was not determined in these subgroups. . Therefore, external validation was only undertaken in EGFR-mutant (exon 19 del/exon 21 L858R) and EGFR-WT (2 nd /3 rd line) patient populations for which other PFS and RECIST response efficacy data exist in other phase II and IV post-marketing clinical trials. Studies that were chosen for external validity had to have at least 100 patients in the erlotinib arm. External validity was evaluated by comparing the fit of PFS Kaplan-Meier curves from the model and clinical trials visually and

through calculation of the AUC sim /AUC CT ratios, as well as comparing the proportions of patients reaching each best RECIST response group (CR, PR, SD and PrD) generated from the model to those reported in the clinical trials.

Once the model’s simulation data were validated, the information was used to explore the relationship between exposure, tumour susceptibility and efficacy. A dose-PFS relationship was first constructed in each patient subgroup to determine the minimum doses required to gain significant increase in survival compared to placebo. The dose-PFS relationship was constructed by simulating a series of PFS datasets in a population of 500 patients, with the conditions of each simulation differing only by the erlotinib dose. Five simulations were generated with one of the following doses: 0, 75, 100, 150 or 300mg. These doses were chosen because 0mg is used as the placebo reference, 150mg is the recommended initial dose, 100mg and 75mg are the doses for the first and second dose reductions, respectively, and 300mg is potentially used for dose escalation for a select patient population to compensate for lower AUC values (135), despite not 57

being used in routine clinical practice or recommended by the product monograph. The median PFS and 95% confidence intervals were determined from each Kaplan-Meier curve. A clinically significant change in median PFS is defined by a change of a month, while a statistically significant change is determined by a log rank test with α = 0.05.

Furthermore, the model was used to determine the minimum serum concentrations required in each subgroup to predict progression and/or ORR, where ORR is the proportion of patients achieving a PR or CR. The minimum Css values were determined by constructing a receiver- operating characteristic (ROC) curve for each patient subgroup and calculating the threshold that is associated with the highest AUC, specificity and sensitivity.

To further characterize the exposure-tumour susceptibility-efficacy relationship, the extent of which variations of the PK or PD components contribute to variability in erlotinib efficacy (PFS or survival) was determined. The extent by which variations of PK or PD components contribute to variations in erlotinib efficacy was determined by evaluating the changes in the survival data when either all of the PK parameters were kept constant (to determine the extent at which PK variability had affected erlotinib efficacy) or all of the PD parameters were constant (to determine how PD variability affected erlotinib efficacy). The median PFS and 95% confidence intervals of the base case (including variability in PK and PD components) was compared with the data from the case with constant PK parameters and those from the constant PD parameters case.

The model was constructed in Microsoft Excel 2007, while the Monte Carlo simulations and the set-up of distributions for each model parameter was made possible using Oracle Crystal Ball. All statistical analyses were performed using R (version 3.0.2).

Chapter 5

Results

5 Results 5.1 The pharmacokinetic component

The literature search to develop the PK component yielded 108 studies that were screened, 5 studies that were included in the resulting model (4, 54, 104, 126, 127) and 7 studies were used (4, 54, 104, 126, 127, 136, 137) for internal validation and 1 meta-analysis was used for external validation (Figure 4).

Figure 4: Flow diagram of the literature search: Screening and selection of erlotinib pharmacokinetic studies for the PK component

After pooling the values of the PK parameters from the included studies (Tables 13 and 14), it was found that the weighted mean and standard deviation values of V d/F, t 1/2 , and k a were 281.57 (44.80) L, 39.46 (34.82) hours, and 1.06 (0.23) hr -1, respectively (Table 14). Furthermore, the pooled mean values of T max, Cl/F, C max , C min , and AUC were 3.15 hours, 4.11 L/hour, 0.982mg/L, 0.980mg/L and 14.59mg*hr/L, respectively (Table 15). Each of these parameters had a log-normal distribution.

Table 13: Summary of values of erlotinib pharmacokinetic parameters reported in the included studies

-1 Reference Sample Erlotinib Single or Vd/F (L) t1/2 (hours) Ka (hours ) size dose multiple Mean SD Mean SD Mean SD (mg) dose 104 12 100 Single N/A 11.5 4.86 N/A 4 3 150 Single 175.36 208.117 20 1.948 4 5 100 Single 61 23.6802 16.14 85.542 4 7 200 Single 175.1 176.3257 35.4 6.2658 4 3 50 Single 117.7 78.5059 16.33 0.9716 54 591 150 Multiple 233 25.164 40.9 N/A 0.949 0.2467 126 80 150 Multiple 89.443 68.2769 N/A 1.86 0.0369 127 6 150 Multiple N/A 27.19 9.07 N/A 127 5 100 Multiple N/A 15.56 8.66 N/A 127 3 50 Multiple N/A 23.6 15.84 N/A Vd/F = apparent volume of distribution adjusted by bioavailability t1/2 = half-life ka = absorption rate constant

Table 14: Summary of all other pharmacokinetic parameters for validation

Reference Sample Erlotinib Single Tmax Cl/F Cmax Cmin AUC size dose or Mean SD Mean SD Mean SD Mean SD Mean SD (mg) multiple dose 104 12 100 Single 4.4 6.2 11.16 3.81 0.804 0.358 N/A N/A 4 3 150 Single 4 0.1384 10.11 1.285 1.136 0.0098 N/A 16.51 1.8194 4 5 100 Single 2.8 0.0305 8.83 0.6799 0.943 0.0062 N/A 13.17 1.5580 4 7 200 Single 4.5 0.1616 3.4 0.0187 1.549 0.0096 N/A 23.35 1.8750 4 3 50 Single 3.17 0.0365 4.98 0.0956 0.427 0.0006 N/A 6.11 0.2139 136 18 150 Multiple N/A 4.387 2.511 N/A N/A N/A 54 591 150 Multiple N/A 3.95 0.1383 1.956 N/A N/A 41.337 N/A 137 76 150 Multiple 3 1.91 N/A 0.872 0.399 0.98 0.368 11.86 5.01 126 80 150 Multiple N/A 3.29 2.2361 N/A N/A N/A 127 6 150 Multiple 1.83 0.41 5.08 4.51 2.3843 0.9324 1.642 1.085 42.6785 20.4348 127 5 100 Multiple 3 2 10.03 8.95 1.0234 0.3201 0.453 0.380 14.6226 7.0072 127 3 50 Multiple 4.33 4.93 3.92 2.36 0.8203 0.3473 0.538 0.286 15.8437 7.9916

Tmax = time at peak concentration Cl/F = clearance adjusted by bioavailability Cmax = peak serum concentration Cmin = trough serum concentration AUC = area under the curve

Table 15: Summary of weighted values of all pharmacokinetic parameters PK Parameter Total sample Weighted Weighted standard Coefficient of size mean deviation variation (%) Vd/F 689 218.57 44.80 20.50 t1/2 635 39.46 34.82 87.03 ka 671 1.06 0.23 21.93 Tmax 133 3.15 12.91 410.133 Cl/F 733 4.11 1.30 31.68 Cmax 115 0.982 0.41 41.63 Cmin 90 0.980 0.45 46.04 AUC 108 14.59 6.73 46.17

Multiple T-tests for heterogeneity with a Bonferroni correction on the PK parameters found that these values were not significantly different from each other and therefore, they could be pooled together (Table 16). Values for k a and C min were not included in the heterogeneity tests because all of the values came from multiple dose studies.

Table 16: T-test of heterogeneity between single-dose and multiple-dose erlotinib pharmacokinetic studies PK parameter Mean 95% confidence p-value Single-dose Multiple- interval dose Vd/F 132.29 161.2215 -603.21, 545.34 0.7586 t1/2 19.8740 26.8125 -23.26, 9.41 0.3397 Tmax 3.774 3.040 -0.80, 2.27 0.2805 Cl/F 7.6960 5.1095 -1.66, 6.83 0.1944 Cmax 0.9718 1.4112 -1.33, 0.45 0.2756 AUC 14.785 25.268 -29.52, 8.56 0.2256

Upon inputting our pooled mean and standard deviation values of V d/F, t 1/2 , and k a into the equation of the PK component, the model calculated values for T max, Cl/F, C max , C min , and AUC were 2.93 hours, 3.79 L/hr, 1.92mg/L and 1.35mg/L, respectively. The calculated steady state concentration (C ss ) for the population had a mean and standard deviation of 1.65 (0.35) mg/L. This value was used in the PD model component for the calculation of k (kill constant). 63

With respect to the internal validity of the model, it was determined that the mean values of Cl/F,

Cmax and AUC from Monte Carlo simulations calculated from the model were more than 75- 125% from the 5 pooled studies (Table 13). Furthermore, the pooled values seemed to have reported a lower value for C max compared to C min.

With respect to the external validity of the model, it was determined that the mean values of

Vd/F, t 1/2 , and k a from Monte Carlo simulations were generally similar by visual inspection to what has been reported in a meta-analysis by Petit-Jean et al (Table 17). (108) The only values in which the ratio fell beyond the range of 75-125% were t 1/2 and C max parameters.

Table 17: Comparison of values of PK parameters from model and other PK studies PK Parameter Model – mean (SD) Pooled values – Petit-Jean et al.– mean mean (SD) (SD) Vd/F (L) 218.57 (44.80) N/A 208 (133) t1/2 (hours) 39.46 (34.82) N/A 21.86 (28.35) -1 ka (hours ) 1.06 (0.23) N/A N/A Tmax (hours) 3.859 (0.9) 3.15 (12.91) 3.69 (3.21) Cl/F (L/hr) 5.927 (3.981) 4.11 (1.30) 4.85 (4.71) Cmax (mg/L) 1.6323 (0.747) 0.982 (0.41) 2.29 (0.84) Cmin (mg/L) 1.1092 (0.9215) 0.980 (0.45) 1.32 (0.95) AUC (mg*hr/L) 41.2316 (37.3268) 14.59 (6.73) 35.76 (15.72)

Using the validated PK equation, it was determined that a single dose of erlotinib can lead to concentrations that vary four-fold among patients (Table 18).

Table 18: Steady state serum concentrations of erlotinib at different doses Dose Median Css (mg/L) Range (mg/L) 75mg 0.833 0.472 – 1.349 100mg 1.121 0.626 – 2.438 150mg 1.699 0.945 – 3.314 300mg 3.308 1.598 – 6.088

5.2 The pharmacodynamic component

The mean and standard deviation EC 50 values of exon 19 del/exon 21 L858R and exon 20 T790M NSCLC cells in in vitro studies were 0.006mg/L (0.0045) and 1.72mg/L (0.397), respectively. (10, 11) Upon fitting the model onto the erlotinib clinical trial data, it was

determined that the E max values for resistant and sensitive cell groups were 90.83 and 0.045 for all-comers, 155.83 and 0.113 for EGFR-mutant (exon19/21) patients, 100.83 and 0.050 for EGFR-mutant (uncommon) patients, 75.83 and 0.045 for EGFR-WT (2 nd /3 rd line) patients, and 2.13 and 0.007 for EGFR-WT (maintenance) patients (Table 19).

Table 19: Summary of E max values of sensitive and resistant cell groups for each patient subgroup population Patient subgroup Sensitive cells Resistant cells Mean Standard Mean Standard deviation deviation EGFR-mutant (exon 0.113 0.01 155.83 5 19/21) EGFR-mutant 0.050 0.01 100.83 5 (uncommon) EGFR-WT (2 nd /3 rd 0.045 0.01 75.83 5 line) EGFR-WT 0.0070 0.01 2.13 5 (maintenance)

With the input of the calculated C ss value, EC 50 value from in vitro studies and E max value from the fitting process into the sigmoidal equation, the k values for resistant and sensitive cell groups were 0.00022 and 0.000031 for all-comers, 0.00037 and 0.000077 for EGFR-mutant (exon 19/21) patients, 0.00024 and 0.000035 for EGFR-mutant (uncommon) patients, 0.00018 and 0.000031 for EGFR-WT (2 nd /3 rd line) patients, and 0.0000051 and 0.0000048 for EGFR-WT (maintenance) patients (Table 20).

Table 20: Summary of k s and k r values calculated from the PD equation

Patient population kr (resistant cells) ks (sensitive cells) All-comers 0.00022 0.000031 EGFR-mutant (exon 19/21) 0.00037 0.000077 EGFR-mutant (uncommon) 0.00024 0.000035 EGFR-WT (2 nd /3 rd line) 0.00018 0.000031 EGFR-WT (maintenance) 0.0000051 0.0000048

By varying the C ss value by a factor of nine, a concentration-kill constant relationship was constructed for the resistant and sensitive groups of cells (Figure 5).

Figure 5: Concentration-kill constant relationships of sensitive and resistant cell groups

The concentration-kill constant relationship using E max and EC 50 values of sensitive and resistant cells were similar to the values reported in the in vitro study by Chmielecki et al. (Table 21). (9) This comparison was not considered to be a formal evaluation of external validity, as the two sigmoidal equations calculated different measures of response. That is, while the model uses a kill constant, Chmielecki et al. uses tumour inhibition instead. (9) Nonetheless, the similarities between the concentration-kill constant curve and the concentration-tumour inhibition curve from in vitro studies provided some confidence that the results were valid.

Table 21 – Comparison of EC 50 and E max values of the model and Chmielecki et al study Model simulation Chmielecki et al EC 50 Emax EC 50 Emax Resistant cells 1.72mg/L 90.83 mg/L >1.72mg/L Not determined (100% T790M cells) Sensitive cells 0.006mg/L 0.045mg/L 0.011mg/L 0.043mg/L (0% T790M)

5.3 The tumour growth component

The model was able to track tumour inhibition and growth through time and differentiate between cases of complete response (CR), partial response (PR), stable disease (SD) and progressive disease (PrD), as per RECIST criteria (Figure 6). In the cases of a CR, PR or SD, an initial phase of tumour size decrease was observed until it reached a minimum, where erlotinib has achieved its best RECIST response. Beyond this minimum, the tumour seems to have developed resistance and begins to grow exponentially until it reaches the point of progression. However, no tumour inhibition was observed in the case of PrD. The growth curve was observed to increase exponentially from treatment initiation. No minimum was achieved in PrD.

Figure 6: Tumour size changes over time based on best response simulated by model Complete response (CR) Partial response (PR)

Stable disease (SD) Progressive disease (PrD)

More patients seemed to have an objective response (either a PR or CR) if they were classified to have either exon 19 or 21 EGFR mutation compared to patients who were EGFR-WT, either using erlotinib as maintenance therapy or as 2 nd or 3 rd line of treatment ( p < 0.0001) (Table 22).

Table 22: Breakdown of best response for each patient subgroup population Patient Complete response Partial response Stable disease Progressive disease subgroup (on Total Progressed Survived Total Progressed Survived Total Progressed Survived Total Progressed Survived erlotinib) EGFR-mutant 29.4% 15.2% 14.2% 21.8% 20.8% 1% 30.8% 30.4% 0.4% 18% 18% 0% (exon 19/21) EGFR-mutant 8.2% 4.4% 3.8% 17.4% 17% 0.4% 40.8% 40.8% 0% 33.6% 33.6% 0% (uncommon) EGFR-WT 7.6% 4.4% 3.2% 17.6% 16.2% 1.4% 60.2% 59.6% 0.6% 14.6% 14.6% 0% (2 nd /3 rd line) EGFR-WT 23.8% 17.6% 6.2% 33.8% 32.2% 1.6% 39.8% 39.6% 0.2% 2.6% 2.6% 0% (maintenance)

Through fitting the model based on the placebo curves, the values of the growth constants, g r and gs, as well as the proportion of resistant cells (x) in each patient subgroup were determined (Table 23).

Table 23: Summary of mean and standard deviation values for gs gr and x for each patient subgroup Patient subgroup Growth constant for Growth constant for Proportion of resistant cells (g r) sensitive cells (g s) resistant cells (x) Mean SD Mean SD Mean SD EGFR-mutant (exon 0.0015 0.0009 0.000081 0.0009 0.002 0.15 19/21) EGFR mutant 0.00155 0.0009 0.00009 0.0009 0.01 0.15 (uncommon) EGFR-WT (2 nd /3 rd 0.0013 0.0009 0.000008 0.0009 0.015 0.15 line) EGFR-WT 0.00185 0.0007 0.000075 0.0001 0.0035 0.15 (maintenance)

Once the growth constants were known, the mean, standard deviation and distributions of g, k and g-k parameters were explored (Table 24). Figure 7 shows the distributions of g, k and g-k parameters for resistant and sensitive cell groups. The three graphs demonstrated that there was still substantial overlap in distributions between resistant and sensitive cells groups in all three parameters despite their mean values differing by at least two. This overlap in distributions of these parameters found some resistant cells becoming more sensitive to erlotinib than the most resistant cells of the sensitive group, and vice versa.

Table 24 - Summary of simulated k r gr-kr, k s and g s-ks values for each patient subgroup population (n= 500) Patient subgroup Sensitive cells Resistant cells ks gs-ks kr gr-kr Mean SD Mean SD Mean SD Mean SD All comers 1.19 x 2.41 x 4.92 x 4.86 x 2.44 x 2.54 x 1.72 x 9.23 x 10 -4 10 -4 10 -5 10 -4 10 -4 10 -4 10 -3 10 -3 EGFR-mutant 2.91 x 5.19 x -1.96 x 1.16 x 3.95 x 4.1 x 1.12 x 9.69 x (exon 19/21) 10 -3 10 -3 10 -4 10 -3 10 -3 10 -3 10 -3 10 -4 EGFR-mutant 1.25 x 2.64 x -3.10 x 7.11 x 2.49 x 2.29 x 1.28 x 9.00 x (uncommon) 10 -4 10 -4 10 -5 10 -4 10 -4 10 -4 10 -3 10 -4 EGFR-WT 1.63 x 4.28 x -1.16 x 3.60 x 5.03 x 1.46 x 1.13 x 9.20 x (2 nd /3 rd line) 10 -5 10 -5 10 -4 10 -4 10 -6 10 -5 10 -3 10 -4 EGFR-WT 2.10 x 6.98 x 4.27 x 2.92 x 5.33 x 1.23 x 9.35 x 1.89 x (maintenance) 10 -5 10 -5 10 -5 10 -4 10 -6 10 -5 10 -4 10 -3

Figure 7: Distributions of g-k values of resistant and sensitive cell groups (2 pages) EGFR-mutant (exon 19/21) EGFR-mutant (uncommon)

EGFR-wild-type (2 nd /3 rd line) EGFR-wild-type (maintenance)

While there was no evaluation of external validity of the TG equation, the pattern of initial therapeutic response, indicated by an initial stage of tumour inhibition, followed by eventual treatment resistance leading to progression that was generated from the TG component closely outlined the typical clinical course of metastatic NSCLC patients on erlotinib. (22) Therefore, given its ability to replicate how tumour size changes over time in clinical practice, the TG component could be concluded to generate simulated data of adequate external validity.

5.4 Simulation and prediction of PFS and response by erlotinib

The model had fit well to each of the patient subgroup populations. Figure 8 shows the comparison of placebo and erlotinib Kaplan-Meier PFS curves for each subgroup, while Table

25 summarises the AUC sim /AUC CT ratios of the placebo and erlotinib curves for each patient subgroup. In general, the model was fit to the clinical trials of each patient subgroup with less than 10% error in AUC, with the exception of the PFS data from the uncommon EGFR mutant patient group, which was fit to clinical trial data within 20% error.

Figure 8: Comparison of model and clinical trial Kaplan-Meier PFS curves for all-comers and each patient subgroup (3 pages) Placebo Erlotinib All-comers

EGFR-mutant (exon 19/21)

EGFR-mutant (uncommon)

EGFR-WT (2 nd /3 rd line)

EGFR-WT (maintenance)

Table 25: Summary of AUC sim /AUC CT ratios of placebo and erlotinib curves for each patient subgroup

Patient subgroup AUC sim /AUC CT ratio of placebo curves AUC sim /AUC CT ratio of erlotinib curves All-comers 1.0098 0.9584 EGFR-mutant (exon 19/21) 0.9926 1.0091* EGFR-mutant (uncommon) 1.0914 0.8017 EGFR-WT (2 nd /3 rd line) 1.0422 1.0026* EGFR-WT (maintenance) 1.0422 1.0162 *AUC sim compared to AUC CT of pooled clinical trials

Through the sequential fitting process of g r, g s and x with the placebo clinical trial data, as well as E max,r and E max,s with the erlotinib clinical trial data, we were able to develop an understanding of how the survival dataset and the changes in the values of the model parameters affect the shape of the Kaplan-Meier curve (Figure 9 and Table 26). The PFS of a simulated patient population could be deconstructed based on the time at which the initial “non-responder” group of patients start to progress (the top “shoulder” of the curve – labelled 1), the rate at which patients within this population progress over time (the middle portion of the curve – labelled 2) and the proportion of patients who do not progress after 50 months of follow-up (the bottom tail part of the curve – labelled 3). It was observed that any conditions favouring the growth of cells or that lessen the susceptibility of tumour cells to erlotinib (e.g. increased g s, g r, x, g-k, and E max,r and E max,s values, and decreased dose) led to a Kaplan-Meier curve that demonstrated a shorter time to progression of the initial non-responders, a steeper decline of patients who progress within the population through time and a smaller proportion of patients who remain without progression upon the end of the follow-up period. The inverse conditions predictably led to a Kaplan-Meier curve that showed a longer time to progression of initial non-responders, a gentler decline of patients progressing over time and a greater proportion of patients who did not progress at the end of the follow-up period. It was noted that parameters for resistant versus sensitive groups of cells may have subtly different effects on the PFS dataset. While g r and g s are both growth constants that have the same effects on the overall patterns of PFS, it seemed that g s had more effect on how quickly the tumour reaches its minimal size from its initial size while g r seemed to affect how quickly the tumour regrows from its minimal size to progression.

Figure 9: Breakdown of a PFS Kaplan-Meier curve

1. Time at which the initial “non-responder” group of patients start to progress

2. Rate at which patients progress over 3. Proportion of time patients who do not progress after last follow -up

Table 26: Summary of the effects of PK parameters on the shape of the PFS Kaplan-Meier curve Parameter Effect on curve Trends shape Dose 2 and 3 ↑ dose = ↑ people have not progressed after 50 months (tail of curve is higher) ↑ dose = ↓ rate of progression in patients ( ↓ steepness of middle portion of curve) gr, g s 1, 2 and 3 ↑g = ↓ time to progression in initial non-responder patients ( ↓ distance from y-axis to shoulder of curve) ↑g = ↑ rate of progression in patients ( ↑ steepness of middle portion of curve) ↑g = ↓ people who do not progress after last follow-up date (tail of curve is lower) Emax,r , 1, 2 and 3 ↑Emax = ↑people without progression after last follow-up date Emax,s (tail of curve is higher) ↑Emax = ↓rate of progression in patients ( ↓ steepness of middle portion of curve) ↑Emax = ↑time to progression in initial non-responder patients (↑ distance from y-axis to shoulder of curve) x 1, 2 and 3 ↑x = ↓ time to progression in initial non-responder patients ( ↓ distance from y-axis to shoulder of curve) ↑x = ↑ rate of progression in patients ( ↑ steepness of middle portion of curve) ↑x = ↓ people who do not progress after last follow-up date (tail of curve is lower) g-k 1,2, and 3 ↑g –k = ↓ time to progression in initial non-responder patients (↓ distance from y-axis to shoulder of curve) ↑g –k = ↑ rate of progression in patients ( ↑ steepness of middle portion of curve) ↑g –k = ↓ people who do not progress after last follow-up date (tail of curve is lower)

In the external validation process, the model was observed to reproduce similar curves to those from other clinical trials of the patient subgroups in question (Figure 10). The AUC sim /AUC CT ratios of the erlotinib curves for EGFR-mutant and EGFR-WT groups were 0.887 and 0.833, respectively.

Figure 10: External validation of model – Comparison of Kaplan-Meier PFS curves from model and other clinical trials EGFR-mutant (exon 19/21)

EGFR-WT (2 nd /3 rd line)

With respect to the comparison of RECIST criteria from model simulations and clinical trials (Table 27 and Figure 11), the model simulated RECIST response rates for each level of response that had a similar trend to that observed in BR.21 for all-comers on erlotinib despite the rates being statistically significantly different than those in BR.21. In the EGFR-mutant (exon 19/21) patient subgroup, the model had estimated more patients achieving CR and fewer patients with PR compared to OPTIMAL and EURTAC. Furthermore, the model had estimated more patients achieving a SD and fewer patients with PrD in the EGFR-WT (2 nd /3 rd line) patients compared to TITAN or the phase II trial conducted by Kobayashi et al.

Table 27: Comparison of RECIST response rates of model simulation and clinical trial data Patient subgroup Complete Partial Stable Progressive p-value response response disease disease All-comers Model 38/500 88/500 301/500 73/500 P < 0.0001 (4.2%) (8.8%) (33.2%) (53.8%) BR.21 3/392 35/392 192/392 162/392 (0.84%) (9.9%) (43.5%) (45.8%) EGFR-mutant (exon 19/21) Model 147/500 109/500 154/500 90/500 (18%) (29.4%) (21.8%) (30.8%) OPTIMAL 2/82 (2.4%) 66/82 11/82 3/82 (3.7%) P < 0.0001 (80.5%) (13.4%) EURTAC 2/74 (2.7%) 48/74 18/74 6/74 (8.1%) P < 0.0001 (64.9%) (24.3%) EGFR-WT (2 nd /3 rd line) Model TITAN 0/203 (0%) 16/203 54/203 133/203 P < 0.0001 (7.9%) (26.6%) (65.5%) Kobayashi 1/29 (3.4%) 4/29 (13.8%) 8/29 (27.6%) 16/29 (55.1%) P < 0.0001

Figure 11: Comparison of RECIST response rates between model and clinical trial data (2 pages) All-comers

EGFR-mutant (exon 19/21)

EGFR-WT (2 nd /3 rd line)

5.5 Applications of model

Figure 12 shows a dose-PFS relationship with daily doses spanning from 0 to 300mg of erlotinib for each of the patient subgroups. Each relationship was observed to yield increasing PFS with increasing dose. The differences in PFS among the doses were more prominent in EGFR-mutant (exon 19/21) patients in which responses to erlotinib were more profound compared to those in EGFR-WT or EGFR-mutant (uncommon) patient groups. In the context of dose reductions, there was a statistically, but not clinically significant differences in PFS from 150mg to 100mg daily in the EGFR-WT (2 nd /3 rd line) patient subgroup (from 8.7 to 8.23 months); however, there was a statistically and clinically significant decrease in PFS by 1.7 months if the dose had been reduced from 150mg to 75mg daily. The PFS of patients on 75mg daily was not found to be statistically significantly different from that of patients on placebo (6.54 months (95% CI = 6.00 – 7.16) vs. 7.02 months (95% CI = 6.44-7.65)). Similarly, there was a statistically, but not clinically significant decrease in PFS from 6.04 months to 5.69 months with a dose reduction from 150mg to 100mg, but statistically and clinically significant decrease from 6.04 months to 5.08 months if the dose had been reduced from 150mg to 75mg daily in patients with uncommon EGFR mutations. The PFS from the patients on the 75mg dose is borderline statistically significantly larger than that of patients on placebo (5.08 months (95% CI = 4.72-5.52) vs. 4.45

months (95% CI = 4.11-4.74)). There was no difference in PFS among patients who were taking 150mg, 100mg or 75mg daily. In contrast with the rest of the subgroups, it was found that the first dose reduction from 150mg to 100mg daily yielded a clinically and statistically significant difference in PFS from 10.24 months to 8.55 months. A second dose reduction to 75mg daily did not cause a significant reduction in PFS (8.75 months (95% CI = 8.03-9.58)) from that observed with the 100mg dose (8.55 months (95% CI = 7.78-9.31)). Nonetheless, a PFS of 8.75 months was still clinically and statistically significantly higher than the PFS in patients on placebo (6.89 months (95% CI = 6.41-7.36)).

Figure 12: Dose-PFS relationship of erlotinib in all patient subgroups (2 pages) EGFR-mutant (exon 19/21) EGFR-mutant (uncommon mutations)

EGFR -WT (2 nd /3 rd line) EGFR -WT (maintenance)

Upon testing how much PK or PD variability contributes towards the variability in erlotinib efficacy (Figure 13), it was found that erlotinib efficacy was mostly driven by variation in tumour susceptibility (PD variability) and less from variation in erlotinib exposure (PK variability). Changes in erlotinib exposure contributed to a change in hazard ratio that ranged from 5 to 35% from the reference simulation data; however, changes in tumour susceptibility led to more remarkable changes in hazard ratio ranging from 280-725% from the reference simulation data. This was a consistently observed trend in all patient subgroups.

Figure 13: Extent of PK and PD variability on variability of erlotinib efficacy (5 pages)

Extent of PD variability Extent of PK variability All-comers

EGFR-mutant (exon 19/21)

EGFR-mutant (uncommon)

EGFR-WT (2 nd /3 rd line)

EGFR-WT (maintenance)

Upon evaluating the ability of the model to predict progression, it was determined that the AUC values of each ROC curve ranged from 0.232 to 0.515 (Table 28).

Table 28: Area under the curve (AUC) of a receiver-operating characteristic (ROC) curve to determine the ability of steady state serum erlotinib concentrations (C ss ) in predicting the occurrence of progression in a patient EGFR-mutant EGFR-mutant EGFR-WT EGFR-WT (exon 19/21) (uncommon) (2 nd /3 rd line) (maintenance) AUC 0.2320815 0.371005 0.5153765 0.4325815

Furthermore, the AUC values for ROC curve that evaluates the model’s ability to predict best RECIST response ranged from 0.548 to 0.690 throughout all metastatic NSCLC subgroups on erlotinib (Figure 14 and Table 29). While the thresholds for predicting progression were not determined due to poor performance of the model in predicting progression, the thresholds for predicting at least 30% tumour shrinkage are 1.64mg/L for EGFR-mutant (exon 19/21) patients, 1.53mg/L for EGFR-mutant (uncommon) patients, 1.58mg/L for EGFR-WT patients using 2nd /3 rd line erlotinib, and 2.22mg/L for EGFR-WT patients using maintenance erlotinib (Table 29).

Figure 14: Receiver-operating characteristic (ROC) curves of C ss predicting for at least a 30% reduction in tumour size (objective response) EGFR-mutant (exon 19/21) EGFR-mutant (uncommon mutations)

EGFR-WT (2 nd /3 rd line) EGFR-WT (maintenance)

Table 29: Minimum serum erlotinib concentration at steady state (C ss ) required to gain objective response (at least 30% reduction in tumour size) in each patient subgroup AUC (95% CI) Sensitivity Specificity Threshold ORR EGFR-mutant 0.557 (0.487, 0.586 0.578 1.64mg/L (exon 19/21) 0.626) EGFR-mutant 0.6902 (0.659, 0.628 0.671 1.53mg/L (uncommon) 0.721) EGFR-WT 0.607 (0.523, 0.698 0.507 1.58mg/L (2 nd /3 rd line) 0.692) EGFR-WT 0.548 (0.385, 0.71) 0.135 1 2.22mg/L (maintenance)

Chapter 6

Discussion

100

6 Discussion 6.1 The pharmacokinetic component

The PK model accurately predicted the population pharmacokinetic profile of erlotinib with excellent internal and external validity. It was able to reproduce values similar to those reported in other studies that were not selected from the literature search, which implied that it is likely able to accurately simulate erlotinib pharmacokinetics of patients in clinical practice. It also reflected the high degree of variability in each parameter of erlotinib’s PK profile, which has been well documented in various studies. (54, 129)

A log-normal distribution was selected for each of the parameters, as it is traditionally the type of distribution chosen for PK parameters of medications. (138) Furthermore, this distribution type best explains the presence of disproportionately large values and the absence of equally disproportionately small values in those parameters. This pattern has been highlighted in some erlotinib PK studies (54, 104, 136), which have consistently reported a positive skew in the distribution of PK parameters. Despite the fact that skewed distributions are best described using median and range, the reasons for using mean and standard deviation values for the PK component were two-fold. Crystal Ball, the software used for establishing the distributions of each parameter and simulating patient populations, only allowed the input of mean and standard deviation values despite it accurately depicting the distribution in its simulations. Furthermore, mean and standard deviations are the only measures of central tendency and variability that are

101

considered to be additive; median and range values of different studies cannot be added and thus pooled together into a population-based number.

In particular, the literature search yielded a pooled half-life of erlotinib with a coefficient of variation of 87%, which implies that a significant amount of interpatient variability in erlotinib exposure may be accounted by the half-life. In metastatic NSCLC patients, there are a number of factors that may contribute to variability in half-life, including changes in protein binding (influenced by interpatient variability in cachexia and levels of inflammation contributed by the cancer and treatment), polymorphisms in CYP450 function, and differences in the degree of inflammation and their inhibitory effects on CYP450 function. Variability in half-life, in addition to variation in V d and k a leads to considerable variability in the values of AUC, C max and

Cmin , which was reflected by the coefficients of variation of AUC, C max and C min , which were reported to be 46.17%, 41.63% and 46.04%, respectively. It is this variability that may have led to differences in the model population estimates of half-life and the half-life values reported in the meta-analysis by Petit-Jean et al. (129)

The coefficient of variation of T max was also found to be high at 410%. In addition to differences in absorption among patients, this variation is likely artificially inflated by variations in the time when the first concentration is measured in the included PK studies. For example, Rakhit et al. had started measurements of plasma concentrations of erlotinib 15 minutes after the erlotinib dose. (104) In contrast, Hidalgo et al. had designed their PK study such that the first plasma concentration is taken at 2 hours after erlotinib administration (4), which could lead to very different estimates of T max from those of Rakhit et al.

102

The error in the measurement of T max will also affect the accuracy of C max . The C max was noted to be smaller than the C min value in the 5 pooled studies because some studies had measured their

Cmax values before or after the actual T max based on the schedule of blood draws from their study designs. This imprecision would artificially lower the value of the C max . For instance, Hidalgo

et al. (4) had measured concentrations every 2 hours and likely reported a C max at 2 hours before the actual T max at 3 hours. In contrast, the C min is much less affected by the variability in timing of blood draws, as the blood draws are typically performed at 24 hours right before the next daily dose is administered. This is the reason for the good agreement between model population

estimates of C min and the C min values reported in the meta-analysis of Petit-Jean et al. (129)

While the T-test showed no significant difference between the PK values from single- versus multiple-dose studies, the values in the single-dose studies tend to be smaller than those observed in multiple-dose studies. This trend is likely due to the lack of drug accumulation from single doses that is otherwise seen in multiple doses of erlotinib. Without drug accumulation, there would be lower levels of erlotinib exposure, as reflected by lower values of AUC, C max and C min . Calculations using these values yielded smaller apparent values of volumes of distribution with the same dose, lower rates of clearances and lower values of the elimination constant. Since single dose studies had a larger weight on the pooled PK values compared to multiple dose studies, the pooled values also tended to be smaller in magnitude compared to those calculated in the model after multiple doses. However, because the values from Petit-Jean et al were derived from many multiple-dose studies, the values were more comparable to those generated from the model. (129)

103

The variability noted among the values of each PK parameter led to wide variation in the calculated values of the serum erlotinib concentration at steady state. There was a substantial amount of overlap in serum concentration among the doses, which was also noted in other PK studies (54, 104, 129). This variability in erlotinib exposure may account for some of the variability in erlotinib efficacy. It was found that variability in erlotinib exposure led to a change of PFS that was up to 35% in some subgroups, which is a statistically significant change. This reinforced the conclusions of Tiseo et al. that increasing plasma concentrations were associated with improved best responses in metastatic NSCLC patients. (11) These findings collectively suggest a need to establish a therapeutic drug monitoring system that ensures consistency in exposures of erlotinib and therefore, the benefits of prolonging PFS and achieving good response in each metastatic NSCLC patient.

6.2 The pharmacodynamic component

The concentration-kill constant relationships generated from the PD component demonstrated that erlotinib was 287 times more potent against sensitive cells with exon 19 del or exon 21 L858R mutations compared to those that are resistant with an exon 20 T790M mutation. This is similar to the observations from previous in vitro studies, where growth inhibition was much more profound in sensitive cells compared to resistant cells at the same tumour concentration of erlotinib. (8) The EC50 values used for the concentration-kill constant relationship were also similar to those reported in the study by Chmielecki et al., which implies that the PD equation is representative of reality. While the model was not designed to look at the EC50 value of erlotinib on EGFR-WT cells, it is likely that erlotinib has intermediate potency against EGFR-

104

WT cells in which its concentration-kill constant curve would fall in between those of resistant and sensitive cell groups, given its documented EC50 of 40nM (0.0172mg/L)(69), which falls in between 0.006 and 1.72mg/L.

The concentration-kill constant relationships for both sensitive and resistant cell groups were generated using an E max sigmoidal model. This model is appropriate for describing the pharmacodynamics of erlotinib for the following reasons. There are a finite number of tyrosine kinase domains of EGFR on each NSCLC cell targeted by erlotinib. Therefore, the reversible binding of erlotinib on these tyrosine kinase domains is considered to be saturable if sufficiently high erlotinib concentrations are reached within the cytoplasmic space of the NSCLC cell. Furthermore, the sigmoidal model is a well-recognized and frequently employed mathematical model used to describe receptor-ligand binding dynamics. The findings generated from the PD equation replicated the data reported in previous in vitro studies on tumour inhibition by erlotinib. (8)

However, there was a discrepancy in the magnitude of concentrations required to saturate EGFR targets and reach E max and the concentrations postulated to be achieved in the cytoplasmic space of NSCLC cells through current therapeutic doses of erlotinib. Our model demonstrated that extremely high concentrations in the tumour microenvironment of around 0.37mg/L for sensitive cells and 314mg/L for resistant cells are required for the maximum k value to be reached. Given a dose of 150mg yielding an C ss of 1.65mg/L, with only 5% of it being unbound and able to penetrate into the intratumoural environment and an adjustment factor of 0.00005, the approximate cytoplasmic concentrations are likely only 0.0000041mg/L, which is 3 magnitudes smaller than the E max of sensitive cells and 6 magnitudes smaller than the E max of resistant cells

105

from the model. Furthermore, dose limiting toxicities, such as rash or diarrhea, prevent the therapeutic dose from being escalated by multiple magnitudes in metastatic NSCLC patients to reach the same extent of tumour inhibition noted in in vitro studies. Therefore, the degree of tumour inhibition that resulted in the PFS and response rates reported in clinical trials may be significantly lower than that achieved in in vitro studies. This may explain why complete response are very rarely observed, even in patients with EGFR-mutant (exon 19/21) NSCLC and that the vast majority of the tumour with cells of intermediate or low susceptibility to erlotinib would not be killed or inhibited.

6.3 The tumour growth component

The model performed well in outlining the typical clinical course observed in metastatic NSCLC patients on erlotinib. It simulated the presence of clinical improvement observed 6 to 8 weeks after treatment initiation, which is approximately when the first follow-up CT scan is scheduled for patients. Furthermore, it replicated the extent of tumour shrinkage observed in patients with all four types of responses by RECIST criteria. These trends were reliably replicated regardless of the initial tumour size that is realistically seen in clinical practice from 1 to 13cm 3 (death thought to occur when the tumour reaches around 13cm 3). (139, 140)

The ability of the TG equation to replicate tumour growth through time is based on the theory that the tumour was initially sensitive to therapy, but is transformed into a resistant tumour as the sensitive cells are slowly replaced by a growing proportion of erlotinib-resistant cells over time ultimately leading to the event of progression. Erlotinib is able to kill a significant number of

106

highly sensitive tumour cells while the rest of the tumour that is relatively more resistant against treatment continues to grow. However, as the reservoir of erlotinib-susceptible cells decrease, the growth inhibition from erlotinib will slow down to the proliferation rate from the erlotinib- resistant cells. As both processes reach equilibrium, the minimal size of the tumour would be reached, where best response by RECIST criteria can be measured. Once the rate of growth inhibition becomes slower than the tumour proliferation rate over time, the pendulum of the growth-kill balance swings in the opposite direction and the tumour is observed to grow with an increasing proportion of resistant cells, eventually becoming resistant to erlotinib. The continuously changing growth-kill interaction results in the characteristic U-shaped curve that is observed in patients with a CR, PR, or SD (Figure 3). In the case of progressive disease, the proliferation rate is always faster than the rate of growth inhibition by erlotinib; therefore, the U- shaped curve appears lopsided and the tumour size rises through time exponentially even at the time of treatment initiation (Figure 3). The shape of this curve depends on multiple parameters, as reflected by the equation of the TG component. Tumours that experience a higher degree of inhibition at the initial stage, a smaller minimal size at the time of best response, and have a more indolent course towards progression were found to have cells with greater susceptibility and more indolent growth (reflected by a smaller difference in g-k) and a lower proportion of resistant cells (x).

The parameters of the TG component also support the concept that tumour susceptibility to any medication, including erlotinib, should be portrayed as a continuous spectrum, bordered by the extremes of total sensitivity and total resistance on either side, instead of as a binary variable. In general, the mean values of g-k for resistant cells were around a magnitude higher than the g-k values in sensitive cells. This reinforces the observation that resistant cells are generally less

107

susceptible to erlotinib compared to sensitive cells. In both EGFR-mutant subgroups and EGFR- WT patients taking erlotinib as 2 nd /3 rd line therapy, the g-k value for sensitive cells was negative, which implies that the potential for growth inhibition (k) was generally higher than the potential for proliferation (g) in the presence of erlotinib. However, the standard deviations were of the same magnitude as the mean, suggesting that there is large variability of g-k values among the patients. This was reflected by the substantial overlap in the tails of the distributions of g-k values of resistant and sensitive cell groups. It is likely that there are positive g-k values (a sign of resistance) in the extremes of the sensitive cell group and negative g-k values (a sign of sensitivity) in the extremes of the resistant cell group. This finding has two implications. It indicates the enormous amount of intratumoral heterogeneity in susceptibility that is typically seen in clinical practice. It also implies how a tumour that is generally perceived to have a certain susceptibility to erlotinib may have a tiny group of cells that may have a degree of susceptibility that is different from the average of its population.

Nonetheless, the idea of susceptibility assumes the presence of the drug. In the case where the tumour is exposed to no erlotinib, tumour growth inhibition does not exist, as the value of the kill constant would be 0. Therefore, in patients on placebo, the tumour is modelled as two different groups of cells proliferating at different natural rates.

6.4 The simulation and prediction of PFS data

The model simulated clinical trial data accurately within an error margin of 10%. The model simulated less accurately for patients with uncommon EGFR mutations. This is likely due to the

108

large random error of the PFS survival data in this subgroup that the simulation would not be able to replicate, which is reflected by the block-like staircase, as opposed to a smooth, appearance of the Kaplan-Meier PFS curve. This error is likely associated with small sample sizes in this patient group in the BR.21 study; there were only 7 patients in the placebo group and 13 patients in the erlotinib group.

The model also accurately predicted how efficacious erlotinib is in the real world in different patient subgroups. It simulated within 11.3% and 16.7% error of the PFS data of EGFR-mutant and EGFR-WT patient groups, respectively, observed in clinical practice. These errors were similar to those reported in the internal validation process, implying that the model simulated PFS data of adequate external validity.

In contrast to its ability in predicting PFS well, the model had mixed accuracy in predicting best RECIST response rates in patient subgroups. While the model predicted RECIST response well in the all-comer group, it was less accurate in predicting RECIST response rates in the specific patient subgroups where the response data was available. It was observed that the model tended to overestimate the number of patients with a CR and underestimate the proportion of patients with a PR for the EGFR-mutant (exon 19/21) patients. Furthermore, the model overestimated the proportion of SD patients and underestimated the proportion of PrD patients in the EGFR- WT (2 nd /3 rd line) group. The reason for this discrepancy in predicting RECIST response may originate from the absence of a stochastic equation in the model to simulate tumour growth over time, which implies that the model is unable to simulate the transition of sensitive cells into resistant cells over time. Therefore, the model likely underestimated the number of resistant

109

cells present in the tumour over time and tended to overestimate the best RECIST response reached by the population despite the fact that it was able to predict its PFS over time accurately.

6.5 Applications of the model

Since the model was proven to provide accurate simulations of PFS in multiple patient subgroups, it can be used for multiple applications. The dose-PFS relationship developed from this model showed the loss in PFS resulting from dose reductions in metastatic NSCLC patients. This was particularly true in the EGFR-mutant (exon 19/21) patients who typically see a large increase in PFS while on erlotinib compared to placebo, as their PFS can decrease from 10.24 to 8.55 months if they required a dose reduction from 150mg to 100mg daily. While the difference in PFS was not prominent between 150mg and 100mg in both groups of EGFR-WT patients, the difference became clinically significant if the dose was halved from 150 to 75mg. Furthermore, this reduction in PFS led to a PFS that was not clinically different from that observed in patients on placebo. Therefore, clinicians must be cognizant that dose reductions intended to reduce toxicity may be detrimental to patient outcome. Clinicians must seek improved approaches of toxicity management without dose reductions or other methods that may jeopardize potential valuable increases in PFS.

The model also provided some insight in the degree of which PD or PK variability contributes to variability in erlotinib efficacy within each patient subgroup. Despite the observation that different doses of erlotinib yield different PFS within the same metastatic NSCLC patient subgroup, it is evident that most of the variation in erlotinib efficacy was driven by variability in

110

tumour susceptibility and that variations in erlotinib exposure do not fully explain variations in erlotinib efficacy among patients. This may explain why patients experiencing a certain level of RECIST response may have a wide variety of serum concentrations of erlotinib. (11) While this model had included a basic PD component that simulates the PD of erlotinib at a preliminary level that sufficiently predicts erlotinib efficacy in a patient population, it is likely that further understanding and more complex modelling of the PD of erlotinib is required to better understand the different factors to tumour susceptibility that influence erlotinib efficacy among patients and to improve our accuracy in predicting erlotinib efficacy in individual patients.

6.6 Limitations of the model

Despite the promising results in model performance there is much room for improvement of our model. The main limitation of this model’s predictive ability lies in the absence of a stochastic model in simulating tumour growth. As a result, the model is not able to simulate the transition of sensitive cells to resistant cells over time, which may enable the model to be more accurate in predicting best RECIST responses within the patient population. Another limitation associated with the mechanics of the model is the use of an adjustment factor to account for intracellular erlotinib concentration in the PD component. However, without studies to uncover how erlotinib penetrates into the tumour and enters the intracellular space, one can only speculate what this process may be and how much of the erlotinib in the serum reaches the intracellular space. Therefore, additional research studying tumour penetration of erlotinib is required to externally validate the adjustment factor of the model.

111

In addition, the model lacks the ability to predict mortality of patients. The model was originally designed to predict progression of metastatic NSCLC and time to progression. In order for the model to yield efficacy data that are more relevant to practice, given the high mortality rates of metastatic NSCLC patients, the model’s generated TTP data was fit instead to predict PFS instead. Future research would be needed to determine other covariates that affect mortality of these patients and how they may be related to the tumour growth inhibitory effects of erlotinib.

The model does not predict for the probability of the patient benefiting by relief of NSCLC- related symptoms or having decreased quality as a result of experiencing erlotinib-induced toxicities. These outcomes are crucial to the treatment of metastatic NSCLC patients, as disease symptom and toxicity management are crucial to the improvement of quality of life, which is often the primary goal of therapy (141), and is suggested to be predictive of longer survival times in metastatic NSCLC patients. (142)

Finally, this model is limited in its ability to predict population estimates of PFS and is currently unable to predict estimates of PFS of a single patient. As a result, this model cannot be used in clinical practice to adjust erlotinib exposure in a single patient based on his or her tumour susceptibility to erlotinib and current projection of RECIST response and PFS. Further research in specific tumour susceptibility- or patient-related factors significantly affecting erlotinib efficacy is required to convert this population-based PK-PD model into a patient-based model used for therapeutic drug monitoring (TDM) in the clinical setting.

112

6.7 Future research

Future research is required to improve the current PK-PD model’s ability to describe tumour growth over time and predict RECIST response of metastatic NSCLC patients on erlotinib. Specifically, this model will benefit in the development of a stochastic model describing tumour growth in the presence and absence of erlotinib and further research on factors affecting tumour susceptibility to erlotinib and penetration of erlotinib into the tumour that would allow the development of a more complex PD model that better describes both processes. Furthermore, future research is needed to transition this current PK-PD model into a patient-based prediction model. Besides improving the mechanics of this current model, it also involves fitting this model using patient-level instead of trial-level study data. Finally, a relationship between erlotinib exposure, biological susceptibility and erlotinib-induced toxicity must be built in order for a therapeutic range of erlotinib to be established in a variety of metastatic NSCLC patients. This relationship is required to determine the upper limit of the therapeutic range of erlotinib in order for efficacy and safety to be optimized for each NSCLC patient.

Once the relationship between toxicity, efficacy, PK and PD of erlotinib, patient- and tumour- related factors and PFS has been established, research aimed to build a TDM program can commence. This would likely involve a series of prospective cohort studies that recruit metastatic NSCLC patients taking erlotinib for each of the three possible indications of its use and externally validating the model’s ability to retrospectively predict each patient’s PFS and RECIST response based on his or her serum erlotinib concentrations, tumour histology and characteristics, and patient-related factors that are known to affect treatment efficacy. Technologies that allow for point-of-care determination of serum erlotinib concentration will

113

need to be developed to allow real-time dose adjustments to optimize serum erlotinib concentrations to the ideal therapeutic range for each patient. Use of these technologies prevent significant time lags between the blood sample and the measurement of serum erlotinib concentration and any potential decreases in treatment efficacy associated with the lag. Randomized controlled trials will need to be conducted to determine the impact of this real-time TDM program in improving efficacy and toxicity in metastatic NSCLC patients on erlotinib.

The success of a therapeutic drug monitoring program for erlotinib in improving patient outcomes will allow similar models and TDM programs to be constructed for other TKIs. The mechanics of this model is anticipated to be generalizable to other TKIs for the following reasons. TKIs are all orally administered with a frequency that ranges from once to four times daily. Their PK profiles can likely be modelled based on a one- or two-compartment model (143-145) and are typically associated with high interpatient variability. (146) All of their pharmacological actions are also based on receptor-ligand binding in the intracellular space of malignant cells and have similar inhibitory effects on tumour growth, regardless of the site from which the solid tumour originates. To tailor the model to each TKI, the model must be fit according to its in vitro , population PK and clinical trial data to ensure that the model can accurately PFS of patient populations receiving that TKI. Because the model is anticipated to be generalizable to every TKI for the treatment of any solid organ tumour, the application of this model can extend far beyond the treatment of metastatic NSCLC to other solid organ cancer sites. After further refinement of this current model, future research can be focused on applying the mechanics of this model to construct similar PK-PD models and TDM programs of all other TKIs used in solid organ tumours.

114

115

Chapter 7

Conclusion

116

7 Conclusion

In summary, a population-based PK-PD model was developed to accurately simulate PFS in multiple subgroups of metastatic NSCLC patients treated with erlotinib. The corresponding values of each model parameter have been included in Table 26.

Table 26: Summary values of model parameters for all patient populations All-comers EGFR- EGFR-mutant EGFR-WT EGFR-WT mutant (exon (uncommon) (2 nd /3 rd line) (maintenance) 19/21) gs 0.00019 0.000081 0.00009 0.000008 0.000075 gr 0.002 0.0015 0.00155 0.0013 0.00185 Emax,r 90.83 155.83 100.83 75.83 2.13 EC 50,r 1.72 1.72 1.72 1.72 1.72 Emax,s 0.045 0.1133 0.050 0.045 0.00704 EC 50,s 0.006 0.006 0.006 0.006 0.006 X 0.01 0.002 0.01 0.0115 0.0035 -5 gs-ks 0.0000492 -0.000196 -0.0000310 -0.000012 4.27 x 10 -4 gr-kr 0.00172 0.000969 0.0009 0.001131 9.35 x 10 Free fraction 0.05 0.05 0.05 0.05 0.05

The model determined that a daily erlotinib dose of 100mg in EGFR-WT (maintenance and 2nd /3 rd line erlotinib) and EGFR-mutant (uncommon) patients and 75mg in EGFR-mutant (exon 19/21) patients were likely the minimum effective doses with significantly higher PFS compared to patients on placebo. Despite significant overlap in concentrations among patients, it is likely that a median Css of 1.64mg/L in EGFR-mutant (exon 19/21), 1.53mg/L in patients with

117

uncommon EGFR mutations, 1.58mg/L in EGFR-WT patients on 2 nd /3 rd line erlotinib and 2.22mg/L in EGFR-WT patients on maintenance therapy is the minimum serum concentration required to reach an ORR (partial and complete response). Given the similarities of these minimum effective concentrations, 2.5mg/L may be used as the minimum effective concentration for any indication of erlotinib. An exposure-efficacy relationship is likely not sufficient in explaining the wide variation of erlotinib efficacy in metastatic NSCLC patients; multiple factors related to tumour susceptibility likely contribute significantly to the benefits that patients receive while being treated with erlotinib. Further research is required to determine what these factors may be. While there is indeed much room to improve the mechanics of this model, it remains as an excellent reminder to clinicians about the potential negative consequences on the PFS and by extension, survival of patients each time a dose reduction is considered for the management of toxicities.

118

119

References

1. Canadian Cancer Society’s Advisory Committee on Cancer Statistics. Canadian Cancer Statistics 2015. Toronto, ON: Canadian Cancer Society; 2015. 2. Howlander N, Noone AM, Krapcho M, Garshell J, Miller D, Altekruse SF et al, editors. SEER Cancer Statistics Review, 1975-2012. Bethesda, MD: National Cancer Institute; 2015 [cited 2015 May 17]. Available from: http://seer.cancer.gov/csr/1975_2012/ 3. Hidalgo M, Siu LL, Nemunaitis J, Rizzo J, Hammond LA, Takimoto C et al. Phase I and pharmacologic study of OSI-774, an epidermal growth factor receptor tyrosine kinase inhibitor, in patients with advanced solid malignancies. Journal of Clinical Oncology. 2001; 19(13):3267-3279. 4. Shepherd FA, Pereira JR, Ciuleanu T, Tan EH, Hirsh V, Thongprasert S et al. Erlotinib in previously treated non-small-cell lung cancer. New England Journal of Medicine. 2005; 353:123-132. 5. Product Monograph: Tarceva. Mississauga, ON: Hoffmann-La Roche Limited. 2005 [updated 2013 Dec 5, cited 2015 May 15]. Available from: http://www.rochecanada.com/content/dam/internet/corporate/rochecanada/en_CA/documents/Res earch/ClinicalTrialsForms/Products/ConsumerInformation/MonographsandPublicAdvisories/Tarc eva/Tarceva_PM_E.pdf 6. Hirsh V. Managing treatment-related adverse events associated with EGFR tyrosine kinase inhibitors in advanced non-small-cell lung cancer. Current Oncology. 2011;18(3):126-138. 7. Melosky B and Hirsh V. Management of common toxicities in metastatic NSCLC related to anti-lung cancer therapies with EGFR-TKIs. Frontiers in Oncology. 2014; 4. doi:10.3389/fonc.2014.00238 8. Chmielecki J, Foo J, Oxnard GR, Hutchinson K, Ohashi K, Somwar R et al. Optimization of dosing for EGFR-mutant non-small cell lung cancer with evolutionary cancer modeling. Science Translational Medicine. 2011; 3(90): 90ra59. doi: 10.1126/scitranslmed.3002356

120

9. Yuza Y, Glatt KA, Jiang J, Greulich H, Minami Y, Woo MS et al. Allele-dependent variation in the relative cellular potency of distinct EGFR inhibitors. Cancer Biology and Therapy. 2007; 6(5): 661-667. 10. Ohashi K, Sequist LV, Arcila ME, Moran T, Chmielecki J, Lin Y et al. Lung cancers with acquired resistance to EGFR inhibitors occasionally harbor BRAF gene mutations but lack mutations in KRAS, NRAS, MEK1. Proceedings of the National Academy of Sciences of the United States of America. 2012; 109(31): E2127-E2133. 11. Tiseo M, Andreoli R, Gelsomino F, Mozzoni P, Azzoni C, Bartolotti M et al. Correlation between erlotinib pharmacokinetics, cutaneous toxicity and clinical outcomes in patients with advanced non-small cell lung cancer (NSCLC). Lung Cancer. 2014; 83:265-271. 12. Wu Q, Li M, Li H, Deng C, Li L, Zhou et al. Pharmacokinetic-pharmacodynamic modeling of the anticancer effect of erlotinib in a human non-small-cell lung cancer xenograft mouse model . Acta Pharmacologica Sinica. 2013; 34:1427-1436. 13. Herbst RS, Heymach JV and Lippman SM. Lung Cancer. New England Journal of Medicine. 2008; 359:1367-1380. 14. Sun S, Schiller JH and Gazdar AF. Lung cancer in never smokers – a different disease. Nature Reviews. 2007; 7:778-790. 15. Samet JM, Avila-Tang E, Boffetta P, Hannan LM, Olivo-Marston S, Thun MJ et al. Lung cancer in never smokers: Clinical epidemiology and environmental risk factors. Clinical Cancer Research. 2009; 15(18): 5626-5645. 16. Mitsudomi T and Yatabe Y. Epidermal growth factor receptor in relation to tumor development: EGFR gene and cancer. FEBS Journal. 2010; 277:301-308. 17. Cooper WA, Lam DCL, O’Toole SA and Minna JD. Molecular biology of lung cancer. Journal of Thoracic Disease. 2013; 5(S5):S479-S490. doi:10.3978/j.issn.2072.1439- 2.013.08.03. 18. Dahse R and Kosmehl H. Detection of drug-sensitizing EGFR exon 19 deletion mutations in salivary gland carcinoma. British Journal of Cancer. 2008; 99(1): 90-92. 19. Lovly C, Horn L and Pao W. EGFR Exon 19 Deletion in Non-Small Cell Lung Cancer [Internet]. Nashville: My Cancer Genome; 2015 [updated 2015 Mar 5]. Available from: http://www.mycancergenome.org/content/disease/lung-cancer/egfr/21/

121

20. Lovly C, Horn L and Pao W. EGFR c.2573T>G (L858R) Mutation in Non-Small Cell Lung Cancer [Internet]. Nashville: My Cancer Genome; 2015 [updated 2015 Mar 5, cited 2015 Jul 25]. Available from: http://www.mycancergenome.org/content/disease/lung- cancer/egfr/5/ 21. Inukai M, Toyooka S, Ito S, Asano H, Ichihara S, Soh J et al. Presence of epidermal growth factor receptor gene T790M mutation as a minor clone in non-small cell lung cancer. Cancer Research. 2006; 66:7854-7858. 22. Lovly C, Horn L and Pao W. EGFR c.2369C>T (T790M) Mutation in Non-Small Cell Lung Cancer [Internet]. Nashville: My Cancer Genome; 2015 [updated 2015 Mar 5, cited 2015 Mar 5]. Available from: http://www.mycancergenome.org/content/disease/lung- cancer/egfr/4/ 23. Kobayashi S, Boggon TJ, Dayaram T, Janne PA, Kocher O, Meyerson M et al. EGFR mutation and resistance of non-small-cell lung cancer to gefitinib. New England Journal of Medicine. 2005; 352:786-792. 24. Pao W, Miller VA, Politi KA, Riely GJ, Somwar R, Zakowski MF et al. Acquired resistance of lung adenocarcinomas to gefitinib or erlotinib is associated with as second mutation in the EGFR kinase domain. PLoS Medicine. 2005; 2(3):e73. doi: 10.1371/journal.pmed.0020073. 25. Paez JG, Janne PA, Lee JC, Tracy S, Greulich H, Gabriel S et al. EGFR mutations in lung cancer: correlation with clinical response to gefitinib therapy. Science. 2004; 304:1497-1500. 26. Lindeman MI, Cagle PT, Beasley MB, Chitale DA, Dacic S, Giaccone G et al. Molecular testing guideline for selection of lung cancer patients for EGFR and ALK tyrosine kinase inhibitors: Guideline from the College of American Pathologists, International Association for the Study of Lung Cancer, and Association for Molecular Pathology. Journal of Thoracic Oncology. 2013; 8(7):823-859. 27. Shi Y, Au JS, Thongprasert S, Srinivasan S, Tsai CM, Khoa MT et al. A prospective, molecular epidemiology study of EGFR mutations in Asian patients with advanced non- small-cell lung cancer of adenocarcinoma histology (PIONEER). Journal of Thoracic Oncology. 2014; 9(2):154-162.

122

28. Boch C, Kollmeier J, Roth A, Stephan-Falkenau S, Misch D, Gruning W et al. The frequency of EGFR and KRAS mutations in non-small cell lung cancer (NSCLC): routine screening data for central Europe from a cohort study. BMJ Open. 2013; 3:e0025060. doi:10.1136/bmjopen-2013-002560. 29. Skov BG, Hogdall E, Clementsen P, Krasnik M, Larsen KR, Sorensen JB et al. The prevalence of EGFR mutations in non-small cell lung cancer in an unselected Caucasian population. Acta Pathologica, Microbiologica et Immunologica Scandinavica. 2014; 123:108-115. 30. Leighl NB, Rekhtman N, Biermann WA, Huang J, Mino-Kenudson M, Ramalingam SS et al. Molecular testing for the selection of patients with lung cancer for epidermal growth factor receptor and anaplastic lymphoma kinase tyrosine kinase inhibitors: American Society of Clinical Oncology Endorsement of the College of American Pathologists/International Society for the Study of Lung Cancer/Association of Molecular Pathologists Guideline. Journal of Clinical Oncology. 2014; 32. doi:10.1200/JCO.2014.57.3055. 31. Masters GA, Temin S, Azzoli CG, Giaccone G, Baker Jr S, Brahmer JR et al. Systemic therapy for stage IV non-small-cell lung cancer: American Society of Clinical Oncology clinical practice guideline update. Journal of Clinical Oncology. 2015; 13. doi:10.1200/JCO.2015.62.1342 32. Ellis PM, Coakley N, Feld R, Kuruvilla Ung YC; Lung Disease Site Group. Use of the epidermal growth factor receptor inhibitors gefitinib (Iressa ®), erlotinib (Tarceva ®), Afatinib, Dacomitinib or Icotinib in the treatment of non-small-cell lung cancer: a clinical practice guideline. Toronto, ON: Cancer Care Ontario; 2014. Program in Evidence- based Care Evidence-based Series No.: 7-9 Version 2. 33. U.S. Food and Drug Administration. FDA approves new drug for the most common type of lung cancer drug shows survival benefit [Internet]. Washington, D.C: U.S. Department of Health & Human Services; 2004 Nov 19 [cited 2015 May 28]. Available from: http://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/2004/ucm108378.htm

123

34. Cappuzzo F, Ciuleanu T, Stelmakh L, Cicenas S, Szczescna A, Juhasz E et al. Erlotinib as maintenance treatment in advanced non-small-cell lung cancer: a multicentre, randomised, placebo-controlled phase 3 study. Lancet Oncology. 2010; 11: 521-529. 35. U.S. Food and Drug Adminstration. Erlotinib [Internet]. Washington, D.C: U.S. Department of Health & Human Services; 2010 Apr 19 [cited 2015 May 28]. Available from: http://www.fda.gov/AboutFDA/CentersOffices/OfficeofMedicalProductsandTobacco/CDER/ucm 209058.htm 36. Rosell R, Carcereny E, Gervais R, Vergnenegre A, Massuti B, Felip E et al. Erlotinib versus standard chemotherapy as first-line treatment for European patients with advanced EGFR mutation-positive non-small-cell lung cancer (EURTAC): a multicentre, open- label, randomised phase 3 trial. Lancet Oncology. 2012; 13:239-246. 37. U.S. Food and Drug Adminstration. Erlotinib [Internet]. Washington, D.C: U.S. Department of Health & Human Services; 2013 May 15 [cited 2015 May 28]. Available from: http://www.fda.gov/Drugs/InformationOnDrugs/ApprovedDrugs/ucm352317.htm 38. Exceptional Access Program. EAP Reimbursement Criteria for Frequently Requested Drugs. Toronto, ON: Ministry of Health and Long-Term Care. 2015 Aug 1. 39. Zhou C, Wu Y, Chen G, Feng J, Liu Z, Wang C et al. Erlotinib versus chemotherapy as first-line treatment for patients with advanced EGFR mutation-positive non-small-cell lung cancer (OPTIMAL, CTONG-0802): a multicentre, open-label, randomized, phase 3 study. Lancet Oncology. 2011; 12:735-742. 40. Ciuleanu T, Stelmakh L, Cicenas S, Miliauskas S, Grigorescu AC, Hillenbach C et al. Efficacy and safety of erlotinib versus chemotherapy in second-line treatment of patients with advanced, non-small-cell lung cancer with poor prognosis (TITAN): a randomised multicentre, open-label, phase 3 study. Lancet Oncology. 2012; 13:300-308. 41. U.S. Food and Drug Administration. Guidance for Industry: Clinical Trial Endpoints for the Approval of Cancer Drugs and Biologics. U.S. Department of Health and Human Services. 2007 [cited on 2015 Jun 20]. Available from http://www.fda.gov/downloads/Drugs/Guidances/ucm071590.pdf

124

42. Eisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, Sargent D, Ford R et al. New response evaluation criteria in solid tumours: Revised RECIST guideline (version 1.1). European Journal of Cancer. 2009; 45:229-247. 43. Soria JC, Massard C and Le Chevalier T. Should progression-free survival be the primary measure of efficacy for advanced NSCLC therapy? Annals of Oncology. 2010; 21(12):2324-2332. doi:10.1093/annonc/mdq204. 44. Villaruz LC and Socinski MA. The clinical viewpoint – definitions, limitations of RECIST, practical considerations of measurement. Clinical Cancer Research. 2013; 19(10): 2629-2636. 45. European Medicines Agency. Guideline on the evaluation of anticancer medicine products in man. European Medicines Agency. 2012 [cited on 2015 Jun 20]. Available from: http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guideline/2013/01/WC5001 37128.pdf 46. Jemal A, Bray F, Center MM, Ferlay J, Ward E and Forman D. Global Cancer Statistics. CA: A Cancer Journal for Clinicians. 2011; 61:69-90. 47. Le Tourneau C, Lee JJ, and Siu LL. Dose escalation methods in phase I cancer clinical trials. Journal of National Cancer Institute. 2009; 101:708-720. 48. Marshall JL. Maximum-tolerated dose, optimum biologic dose, or optimum clinical value: dosing determination of cancer therapies. Journal of Clinical Oncology. 2012; 30(23):2815-2816. 49. Beumer JH. Without therapeutic drug monitoring, there is no personalized cancer care. Clinical Pharmacology & Therapeutics. 2013; 93(3): 228-230. 50. Mould DR, Walz C, Lave T, Gibbs JP and Frame B. Developing exposure/response models for anticancer drug treatment: special considerations. Clinical Pharmacology and Therapeutics: Pharmacometrics and Systems Pharmacology. 2015; 4, e16;doi:10.1002/psp4.16 51. Csajka C and Verotta D. Pharmacokinetic-pharmacodynamic modelling: history and perspectives. Journal of Pharmacokinetics and Pharmacodynamics. 2006; 33(3):227- 279.

125

52. Rosenbaum S. Basic pharmacokinetics and pharmacodynamics. Hoboken, N.J.: John Wiley & Sons; 2011. 53. Huang W. PHA5127 - Pharmacokinetics. Lecture presented at; 2010; Yuanpei University of Medical Technology. 54. Lu J, Eppler SM, Wolf J, Hamilton M, Rakhit A, Bruno R et al. Clinical pharmacokinetics of erlotinib in patients with solid tumors and exposure-safety relationship in patients with non-small cell lung cancer. Clinical Pharmacology and Therapeutics. 2006; 80(2): 136-145. 55. Lu J, Eppler SM, Lum BL, Hamilton M, Rakhit A and Gaudreault J. A population pharmacokinetic (PK) for erlotinib (E), a small molecule inhibitor of the epidermal growth factor receptor (EGFR). Journal of Clinical Oncology. 2005; 23(16S):2032. 56. Scheffler M, Di Gion P, Doroshyenko O, Wolf J and Fuhr U. Clinical pharmacokinetics of tyrosine kinase inhibitors: focus on 4-anilinoquinazolines. Clinical Pharmacokinetics. 2011; 50(6): 371-403. 57. Anderson GD. Children versus adults: pharmacokinetic and adverse-effect differences. Epilepsia. 2002; 43(s3):53-59. 58. Ginsberg G, Hattis D, Sonawane B, Russ A, Banati P, Kozlak M et al. Evaluation of child/adult pharmacokinetic differences from a database derived from the therapeutic drug literature. Toxicological Sciences. 2002; 66:185-200. 59. Fernandez E, Perez R, Hernandez A, Tejada P, Arteta M and Ramos JT. Factors and mechanisms for pharmacokinetic differences between pediatric population and adults. Pharmaceutics. 2011; 3:53-72. 60. Bonate PL, Floret S and Bentzen C. Population pharmacokinetics of APOMINE: A meta-analysis in cancer patients and healthy males. British Journal of Clinical Pharmacology. 2004; 58(2):142-155. 61. Perez-Ruixo JJ, Piotrovskij V, Zhang S, Hayes S, De Porre P and Zannikos P. Population pharmacokinetics of tipifarnib in healthy subjects and adult cancer patients. British Journal of Clinical Pharmacology. 2006; 62(1):81-96. 62. Verbeeck RK. Pharmacokinetics and dosage adjustment in patients with hepatic dysfunction. European Journal of Clinical Pharmacology. 2008; 64:1147-1161.

126

63. Fry DW. Mechanism of action of erbB tyrosine kinase inhibitors. Experimental cell research. 2003; 284: 131-139. 64. Scaltriti M and Baselga J. The epidermal growth factor receptor pathway: a model for targeted therapy. Clinical Cancer Research. 2006; 12(18): 5268-5272. 65. Huether A, Hopfner M, Sutter AP, Schuppan D and Scherubl H. Erlotinib induces cell cycle arrest and apoptosis in hepatocellular cancer cells and enhances chemosensitivity towards cytostatics. Journal of Hepatology. 2005; 43(4):661-669. 66. Levi E, Mohammad R, Kodali U, Marciniak D, Reddy S, Aboukameel A et al. EGF- Receptor related protein causes cell cycle arrest and induces apoptosis of colon cancer cells in vitro and in vivo . Anticancer Research. 2004; 24: 2885-2892. 67. Gong Y, Somwar R, Politi K, Balak M, Chmielecki J, Jiang X et al. Induction of BIM is essential for apoptosis triggered by EGFR kinase inhibitors in mutant EGFR-dependent lung adenocarcinoma. PLOS medicine. 2007; 4(10): e294. doi:10.1371/journal.pmed.0040294 68. Bethune G, Bethune D, Ridgway N, Xu Z. Epidermal growth factor receptor (EGFR) in lung cancer: an overview and update. Journal of Thoracic Disease. 2010; 2: 48-51. 69. Li D, Ambrogio L, Shimamura T, Kubo S, Takahashi M, Chirieac LR et al. BIBW2992, an irreversible EGFR/HER2 inhibitor highly effective in preclinical lung cancer models. Oncogene. 2008; 27(34):4702-4711. 70. Lalonde, R.L. Pharmacodynamics. In: Burton ME, Shaw LM, Schentag JJ and Evans WE, editors. Applied Pharmacokinetics and Pharmacodynamics: Principle of Therapeutic Drug Monitoring, 4th ed. Baltimore: Lippincott Williams & Wilkins; 2006 71. Benzekry S, Lamont C, Beheshti A, Tracz A, Ebos JML, Hlatky L et al. Classical mathematical models for description and prediction of experimental tumor growth. PLoS Computational Biology. 2014; 10(8):e1003800. doi:10.1371/journal.pcbi.1003800 72. Felmlee MA, Morris ME, and Mager DE. Mechanism-based pharmacodynamics modeling. Methods Mol Biol. 2012; 929:583-600. 73. Tannock IF, Lee CM, Tunggal HK, Cowan DSM and Egorin MJ. Limited penetration of anticancer drugs through tumor tissue: a potential cause of resistance of solid tumors to chemotherapy. Clinical Cancer Research. 2002; 8:878-884.

127

74. Lankheet NAG, Schaake EE, Burgers SA, van Pel R, Beijnen JH, Huitema ADR et al. Concentrations of erlotinib in tumor tissue and plasma in non-small-cell lung cancer patients after neoadjuvant therapy. Clinical Lung Cancer. 2015; 16(4):320-324. 75. Yamasaki F, Johansen MJ, zhang D, Krishnamurthy S, Felix E, Bartholomeusz C et al. Acquired resistance to erlotinib in A-431 epidermoid cancer cells require down- regulation of MMAC1/PTEN and up-regulation of phosphorylated Akt. Cancer Research. 2007; 67(12):5779-5788. 76. Gerlee P. The model muddle: in search of tumour growth laws. Cancer Research. 2013; 73(8):2407-2411. 77. Kahm M, Hasenbrink G, Lichtenberg-Fraté H, Ludwig J and Kschischo M. grofit: Fitting Biological Growth Curves with R. Journal of Statistical Software. 2010; 33(7). 78. Chasin L, Mowshowitz D. Exponential growth. Lecture presented at; 2000; Columbia University. 79. Shah DK, Haddish-Verhane N and Betts A. Bench to bedside translation of antibody drug conjugates using a multiscale mechanistic PK/PD model: a case study with brentuximab-vedotin. Journal of Pharmacokinetics and pharmacodynamics. 2012; 39:643-659. 80. Rehman M and Pedersen SA. Validation of simulation models. Journal of Experimental & Theoretical Artificial Intelligence. 2012; 24(3):351-363. 81. Soo RA, Kawaguchi T, Loh M, Ou SH, Shieh MP, Cho BC et al. Differences in outcome and toxicity between Asian and Caucasian patients with lung cancer treated with systemic therapy. Future Oncology. 2012; 8(4):451-462. 82. Radzikowska E, Glaz P and Roszkowski K. Lung cancer in women: age, smoking, histology, performance status, stage, initial treatment and survival. Population-based study of 20 561 cases. Annals of oncology. 2002; 13:1087-1093. 83. Oken MM, Creech RH, Tormey DC, Horton J, Davis TE, McFadden ET et al. Toxicity and Response Criteria of the Eastern Cooperative Oncology Group. American Journal of Clinical Oncology. 1982; 5:649-655. 84. Kawaguchi T, Takada M, Kubo A, Matsumura A, Fukai S, Tamura A et al. Performance status and smoking status are independent favorable prognostic factors for survival in

128

non-small cell lung cancer: a comprehensive analysis of 26,957 patients with NSCLC. Journal of Thoracic Oncology. 2010; 5(5):620-630. 85. Schiller JH, Harrington D, Belani CP, Langer C, Sandler A, Krook J et al. Comparison of four chemotherapy regimens for advanced non-small-cell lung cancer. New England Journal of Medicine. 2002; 346:92-98. 86. Parsons A, Daley A, Begh R and Aveyard P. Influence of smoking cessation after diagnosis of early stage lung cancer on prognosis: systematic review of observational studies with meta-analysis. British Medical Journal. 2010; 340:b5569. 87. Temel JS, Greer JA, Muzikansky A, Gallagher ER, Admane S, Jackson VA et al. Early palliative care for patients with metastatic non-small-cell lung cancer. New England Journal of Medicine. 2010; 363:733-742. 88. Groot MM, Vernooij-Dassen MJFJ, Verhagen SCA, Crul BJP and Grol RPTM. Obstacles to the delivery of primary palliative care as perceived by GPs. Palliative Medicine. 2007; 21:697-703. 89. Lynch TJ, Bell DW, Sordella R, Gurubhagavatula S, Brannigan BW, Harris PL et al. Activating mutations in the epidermal growth factor receptor underlying responsiveness of non-small-cell lung cancer to gefiitnib. New England Journal of Medicine. 2004; 350(21):2129-2139. 90. Takano T, Fukui T, Ohe Y, Tsuta K, Yamamoto S, Nokihara H et al. EGFR mutations predict survival benefit from gefitinib in patients with advanced lung adenocarcinoma: a historical comparison of patients treated before and after gefitinib approval in Japan. Journal of Clinical Oncology. 2008; 26(34):5589-5595. 91. Health Quality Ontario. Epidermal growth factor receptor mutation (EGFR) testing for prediction of response to EGFR-targeting tyrosine kinase inhibitor (TKI) drugs in patients with advanced non-small-cell lung cancer: an evidence-based analysis. Ontario Health Technology Assessment Series. 2010; 10(24):1-48. 92. Saijo N, Niitani H, Tominaga K, Eguchi K, Koketsu H, Fujino T et al. Comparison of survival in nonresected well differentiated and poorly differentiated adenocarcinoma of the lung. Journal of Cancer Research and Clinical Oncology. 1980; 97(1):71-79. 93. Barletta JA, Yeap BY, and Chirieac LR. The prognostic significance of grading in lung adenocarcinoma. Cancer. 2010; 116(3):659-669.

129

94. Kletzl H, Giraudon M, Ducry PS, Abt M, Hamilton M and Lum BL. Effect of gastric pH on erlotinib pharmacokinetics in healthy individuals: omeprazole and ranitidine. Anti- cancer Drugs. 2015; 26:565-572. 95. Ling J, Fettner S, Lum BL, Riek M and Rakhit A. Effect of food on the pharmacokinetics of erlotinib, an orally active epidermal growth factor receptor tyrosine- kinase inhibitor, in healthy individuals. Anti-cancer Drugs. 2008; 19:209-216. 96. Routledge PA. Factors contributing to variability in drug pharmacokinetics. II. Distribution. Journal of clinical and hospital pharmacy. 1985; 10:15-23. 97. Del Ferraro C, Grant M, Koczywas M and Dorr-Uyemura LA. Management of anorexia- cachexia in late stage lung cancer patients. Journal of hospice and palliative nursing. 2012; 14(6): doi: 10.1097/NJH.0b013e31825f3470 98. Locatelli F and Pozzoni P. Chronic kidney disease in the elderly: is it really a premise for overwhelming renal failure? Kidney International. 2006; 69:2118-2120. 99. O’Riordan S. Chronic kidney disease and older patient – implications of the publication of the Part 2 of the National Service Framework for Renal Services. Age Aging. 2005; 34(6): 546-548. 100. Fournier T, Medjoubi-N,N and Porquet D. Alpha-1-acid-glycoprotein. Biochimica et Biophysica Acta. 2000; 1482:157-171. 101. Lamba J, Lin YS, Schuetz EG and Thummel KE. Genetic contribution to variable human CYP3A-mediated metabolism. Advanced Drug Delivery Reviews. 2012; 64:256- 269. 102. Ingelman-Sundberg M, Sim SC, Gomez A and Rodriguez-Antona C. Influence of cytochrome P450 polymorphisms on drug therapies: Pharmacogenetic, pharmacoepigenetic and clinical aspects. Pharmacology & Therapeutics. 2007; 116: 496-526. 103. Hustert E, Haberl M, Burk O, Wolbold R, He Y, Klein K et al. The genetic determinants of the CYP3A5 polymorphism. Pharmacogenetics. 2001; 11:773-779. 104. Rakhit A, Pantze MP, Fettner S, Jones HM, Charoin J, Rick M et al. The effects of CYP3A4 inhibition on erlotinib pharmacokinetics: computer-based simulation (SimCYP) predicts in vivo metabolic inhibition. Eur J Clin Pharmacol. 2008; 64:31-41.

130

105. Hamilton M, Wolf JL, Drolet DW, Fettner SH, Rakhit AF, Witt K et al. The effect of rifampicin, a prototypical CYP3A4 inducer, on erlotinib pharmacokinetics in healthy subjects. 106. Hamilton M, Wolf JL, Rusk J, Beard SE, Clark GM, Witt K et al. Effects of smoking on the pharmacokinetics of erlotinib. Clinical Cancer Research. 2006; 12(7):2166-2171. 107. Hughes AN, O’Brien MER, Petty J, Chick JB, Rankim E, Wall PJ et al. Overcoming CYP1A1/1A2 mediated induction of metabolism by escalating erlotinib dose in current smokers. Journal of Clinical Oncology. 2009; 27:1220-1226. 108. Harvey RD and Morgan ET. Cancer, inflammation, and therapy: Effects on cytochrome P450-mediated drug metabolism and implications for novel immunotherapeutic agents. Clinical Pharmacology & Therapeutics. 2014; 96(4):449- 457. 109. Morgan ET. Regulation of cytochrome P450 by inflammatory mediators: why and how? Drug Metabolism and Disposition. 2001; 29(3):207-212. 110. Ling J, Johnson KA, Miao Z, Rakhit A, Pantze MP, Hamilton M et al. Metabolism and excretion of erlotinib, a small molecule inhibitor of epidermal growth factor receptor tyrosine kinase, in healthy male volunteers. Drug Metabolism and Disposition. 2006; 34(3):420-426. 111. Lopez-Rios F, Angulo B, Gomez B, Mair D, Martinez R, Conde E et al. Comparison of molecular testing methods for the detection of EGFR mutations in formalin-fixed paraffin-embedded tissue specimens of non-small cell lung cancer. Journal of Clinical Pathology. 2013; 66:381-385. doi:10.1136/jclinpath-2012-201240. 112. De Bruin EC, McGranahan N, Mitter R, Salm M, Wedge DC, Yates L et al. Spatial and temporal diversity in genomic instability processes defines lung cancer evolution. Science. 2014; 346 (6206):251-256. 113. Taniguchi K, Okami J, Kodama K, Higashiyama M and Kato K. Intratumor heterogeneity of epidermal growth factor receptor mutations in lung cancer and its correlation to the response to gefitinib. Cancer Science. 2008; 99(5):929-935. 114. Park S, Holmes-Tisch AJ, Cho EY, Shim YM, Kim J, Kim HS et al. Discordance of molecular biomarkers associated with epidermal growth factor receptor pathway

131

between primary tumors and lymph node metastasis in non-small cell lung cancer. Journal of Thoracic Oncology. 2009; 4(7): 809–815. 115. Schmid K, Oehl N, Wrba F, Pirker R, Priker C and Filipits M. EGFR/KRAS/BRAF mutations in primary lung adenocarcinomas and corresponding locoregional lymph node metastases. Clinical Cancer Research. 2009; 15(14): 4554- 4560. 116. Wang F, Fang P, Hou, DY, Leng ZJ, Cao LJ. Comparison of epidermal growth factor receptor mutations between primary tumors and lymph nodes in non-small cell lung cancer: a review and meta-analysis of published data. Asian Pacific Journal of Cancer Prevention. 2014; 15(11):4493-4497. 117. Hanahan D and Coussens LM. Accessories to the crime: functions of cells recruited to the tumor microenvironment. Cancer Cell. 2012; 21:309-322. 118. Junttila MR and de Sauvage FJ. Influence of the tumour micro-environment heterogeneity on therapeutic response. Nature. 2013; 501:346-354. 119. Seoane J, and De Mattos-Arruda L. The challenge of intratumour heterogeneity in precision medicine. Journal of Internal Medicine. 2014; 276:41-51. 120. Chen Z, Fillmore CM, Hammerman PS, Kim CF and Wong KK. Non-small-cell lung cancers: a heterogeneous set of diseases. Nature Reviews. 2014; 14:535-546. 121. Wu Q, Li M, Li H, Deng C, Li L, Zhou et al. Pharmacokinetic-pharmacodynamic modeling of the anticancer effect of erlotinib in a human non-small-cell lung cancer xenograft mouse model . Acta Pharmacologica Sinica. 2013; 34:1427-1436. 122. Wong SY and Hynes RO. Lymphatic or hematogenous dissemination: How does a metastatic tumor cell decide? Cell cycle. 2006; 5(8):812-817. 123. Zhou CH, Yang SF and Li PQ. Human lung cancer cell line SPC-A1 contains cells with characteristics of cancer stem cells. Neoplasma. 2012; 59(6):685-692. 124. Van der Worp HB, Howells DW, Sena ES, Porritt MJ, Rewell S, O’Collins V et al. Can animal models of disease reliably inform human studies? PLoS Medicine. 2010; 7(3):e1000245. doi:10/1371/journal.pmed.1000245 125. Frohna P, Lu J, Eppler S, Hamilton M, Wolf J, Rakhit A et al. Evaluation of the absolute oral bioavailability and bioequivalence of erlotinib, an inhibitor of the epidermal

132

growth factor receptor tyrosine kinase, in a randomized, crossover study in healthy subjects. Journal of Clinical Pharmacology. 2006; 46(3):282-290. 126. Rudin CM, Liu W, Desai A, Karrison T, Jiang X, Janisch L et al. Pharmacogenomic and pharmacokinetic determinants of erlotinib toxicity. Journal of Clinical Oncology. 2008; 26:1119-1127. 127. Yamamoto N, Horiike A, Fujisaka Y, Murakami H, Shimoyama T, yamada Y et al. Phase I dose-finding and pharmacokinetic study of the oral epidermal growth factor receptor tyrosine kinase inhibitor Ro50-8231 (erlotinib) in Japanese patients with solid tumors. Cancer Chemotherapy and Pharmacology. 2008; 61:489-496. 128. De Wit D, Guchelaar H, den Hartigh J, Gelderblom H and van Erp NP. Individualized dosing of tyrosine kinase inhibitors: are we there yet? Drug Discovery Today. 2015; 20(1):18-36. 129. Petit-Jean E, Buclin T, Guidi M, Quoix E, Gourieux B, Decosterd LA et al. Erlotinib: Another candidate for the therapeutic drug monitoring of targeted therapy of cancer? A pharmacokinetic and pharmacodynamics systemic review of literature. Therapeutic Drug Monitoring. 2015; 37:2-21. 130. Del Monte U. Does the cell number 10 9 still really fit one gram of tumor tissue? Cell Cycle. 2009; 8(3):505-506. doi:10.4161/cc.8.3.7608 131. Schrevens L, Lorent N, Dooms C and Vansteenkiste. The role of PET scan in diagnosis, staging, and management of non-small cell lung cancer. The Oncologist. 2004; 9:633-643. 132. Tsao M, Sakurada A, Cutz J, Zhu C, Kamel-Reid S, Squire J et al. Erlotinib in lung cancer – molecular and clinical predictors of outcome. New England Journal of Medicine. 2005; 353:133-144. 133. Goldie J, Coldman AJ, and Lai A. Mathematical models of multidrug resistance. In: Kellen JA, editor. Reversal of Multidrug Resistance in Cancer. Boca Raton, FL: CRC press; 1994. p61-69 134. Kobayashi T, Koizumi T, Agatsuma T, Yasuo M, Tsushima J, Kubo K et al. A phase II trial of erlotinib in patients with EGFR wild-type advanced non-small-cell lung cancer. Cancer Chemotherapy and Pharmacology. 2012; 69:1241-1246.

133

135. Hughes AN, O’Brien MER, Petty J, Chick JB, Rankin E, Woll PJ et al. Overcoming CYP1A1/1A2 mediated induction of metabolism by esc alating erlotinib dose in current smokers. Journal of Clinical Oncology. 2009; 27:122-1226. 136. Tan AR, Yang X, Hewitt SM, Berman A, Lepper ER, Saparreboom A et al. Evaluation of biologic end points and pharmacokinetics in patients with metastatic breast cancer after treatment of erlotinib, an epidermal growth factor receptor tyrosine kinase inhibitor. Journal of Clinical Oncology. 2004; 22(15):3080-3090. 137. Raizer JJ, Abrey LE, Lassman AB, Chan g SM, Lamborn KR, Kuhn JG et al. A phase I trial of erlotinib in patients with nonprogressive glioblastoma multiforme postradiation therapy, and recurrent malignant gliomas and meningiomas. Neuro- Oncology. 2010; 12(1):87-94. 138. Mizuta E and Tsubotani A. Preparation of mean drug concentration-time curves in plasma. A study on the frequency distribution of pharmacokinetic parameters. Chem Pharm Bull. 1985; 33(4):1620-1632. 139. Detterbeck FC and Gibson CJ. Turning gray: the natural history of lung cancer over time. Journal of Thoracic Oncology. 2008; 3:781-792. 140. Geddes DM. The natural history of lung cancer: a review based on rates of tumour growth. British Journal of Disease of the Chest. 1979;73:1–17. 141. Gralla RJ. Quality-of-life considerations in patients with advanced lung cancer: effect of topotecan on symptom palliation and quality of life. Oncologist. 2004; 9 Suppl 6:14-24. 142. Ediebah DE, Coens C, Zikos E, Quinten C, Ringash J, King MT et al. Does change in health-related quality of life score predict survival? Analysis of EORTC 08975 lung cancer trial. British Journal of Cancer. 2014; 110(1):2427-2433. 143. Wang S, Zhou Q and Gallo JM. Demonstration of the equivalent pharmacokinetic/pharmacodynamics dosing strategy in a multiple-dose study of gefitinib. Molecular Cancer Therapeutics. 2009; 8(6):1438-1447. 144. Yamazaki S, Vicini P, Shen Z, Zou HY, Lee J, Li Q et al. Pharmacokinetic/pharmacodynamics modeling of crizotinib for anaplastic lymphoma

134

kinase inhibitor and antitumor efficacy in human tumor xenograft mouse models. Journal of Pharmacology and Experimental Therapeutics. 2012; 340(3):549-557. 145. Freiwald M, Schmid U, Fleury A, Wind S, Stopfer P and Staab A. Population pharmacokinetics of afatinib, an irreversible ErbB family blocker, in patients with various solid tumors. Cancer Chemotherapy and Pharmacology. 2014; 73(4): 759-770. 146. Josephs DH, Fisher DS, Spicer J and Flanagan RJ. Clinical pharmacokinetics of tyrosine kinase inhibitors: implications for therapeutic drug monitoring. 2013; 35:562- 587.