Abstract:
Background:
Lung cancer is one of the major health burdens with persistent high incidence and poor survival.
Treatment with Tyrosine Kinase Inhibitors (TKIs) that target Epidermal Growth Factor Receptor
(EGFR) significantly improves disease outcomes for non-squamous Non-Small Cell Lung Cancer
(NSCLC) patients with EGFR mutation. A better understanding is needed of the epidemiology of lung
cancer in relation to EGFR mutation and strategies to facilitate identification of EGFR mutation-positive
patients who may be eligible for EGFR targeted treatment. Thus, this PhD research aimed
to contribute knowledge to this area of research by 1) estimating the incidence of EGFR mutation-specific
non-squamous NSCLC, 2) investigating factors associated with survival of EGFR mutation-specific
non-squamous NSCLC, and 3) estimating the probabilities of EGFR mutation positivity for
individual patients using a predictive model to assist in treatment decision-making where tissue
biopsy results were unavailable.
Methods:
Data sources: This research was based on all patients who were diagnosed with non-squamous non-small
cell lung cancer (NSCLC) in northern New Zealand between 1 February 2010 and 31 July 2017
(n=3815). The patients were identified and their demographic and clinico-pathological information
was collected from the population-based New Zealand Cancer Registry (NZCR). The NZCR data were
then linked with TestSafe clinical information sharing database, laboratory reports, individual
patient medical records, pharmaceutical records and mortality records to obtain EGFR mutation
test results, further clinical, pathological and treatment information.
Statistical analysis: 1) The incidence rates were calculated for EGFR mutation-positive and negative
non-squamous NSCLC, in subgroups of patients by age, sex, ethnicity, and smoking status. The
incidence rates were age-standardised using the WHO standard population, and were also
estimated taking into account incomplete EGFR mutation testing. 2) Factors associated with overall
survival were investigated using Cox regression analysis. The association estimates were observed
in terms of four separate survival models based on EGFR mutation status and metastasis status at
diagnosis. 3) The EGFR mutation predictive model was developed using multivariable logistic
regression and validated internally and externally. The model validity was assessed by the fit
between observed and predicted mutation probabilities. The model discriminative ability between
positive and negative status was assessed by observing the Area Under Curve (AUC). The model
performance was assessed by relating the predicted probabilities to the duration remaining on
EGFR-TKI treatment and overall survival from the start of EGFR-TKI therapy in a group of patients
with unknown EGFR mutation status.
Results:
Of the total 3815 non-squamous NSCLC patients, 45% were tested for EGFR mutations; 22.5% of
those tested were EGFR mutation-positive.
1) The population-based age-standardised rate (ASR) of EGFR mutation-positive NSCLC was 5.05
(95%CI 4.71-5.39) per 100,000 person-years, showing that the disease risk was 1.5 times higher for
females than males; approximately 3.5 times higher for Pacific Peoples and Asians, and 2 times
higher for Māori compared with New Zealand Europeans; and only 1.25 times higher for ever-smokers
(i.e. current smokers and ex-smokers) than never-smokers. The ASR of EGFR mutation-negative
NSCLC was 17.39 (16.75-18.02) per 100,000 person-years overall, showing contrasting
patterns of relationships among subgroups compared to EGFR mutation-positive disease: the risk
for EGFR mutation-negative disease was higher for males; higher for Māori and Pacific Peoples but
lower for Asians; and much higher for ever-smokers compared to their respective counterparts.
Standardised Incidence Ratios (SIRs) by sex, ethnicity and smoking, for both diseases, remained
similar to those based on tested patients, when accounting for incomplete EGFR mutation testing.
2) The median overall survival times were significantly different between EGFR mutation-negative
and positive groups: 0.8 years versus 2.79 years. Metastasis at diagnosis had a large impact on
overall survival (hazard ratio (HR) 3.6 in EGFR mutation-negative and 3.3 in positive groups). In
subgroup analyses by EGFR mutation status and metastasis, females had lower survival than males
only if they were EGFR mutation-positive; Māori had lower survival than New Zealand Europeans
only if the disease was metastatic, and tumour site had significant effects only in patients without
metastasis. The remaining factors such as older age, higher ECOG performance status and being a
current smoker showed lower overall survival consistently in all subgroups.
3) The EGFR mutation predictive model used three major predictors – sex, ethnicity and smoking
status, presented as a nomogram, to estimate EGFR mutation probabilities for individual patients.
The model demonstrated a good fit between predicted and observed values in both development
and validation groups. The model showed an AUC of 0.75, and the probability cut-point of 0.2
resulted 63% sensitivity and 79% specificity in the validation group. The model predictions
significantly corresponded to the outcomes of patients with unknown EGFR mutation status.
Conclusions:
EGFR mutation-positive and EGFR mutation-negative non-squamous NSCLC represented two
distinct diseases, showing different risk factors and survival outcomes. The EGFR mutation-positive
disease showed population incidence rates that were similar to other major cancers, e.g., gastric
cancer. The population-based risk of EGFR mutation positive lung cancer was significantly higher
for females than males; and for Māori, Pacific Peoples and Asian than for New Zealand Europeans;
but the risk differed only slightly with the smoking history. Assessing overall survival of nonsquamous
NSCLC showed that EGFR mutation status and metastasis status at diagnosis
independently affected overall survival and modified the effects of other factors on overall survival.
The EGFR mutation predictive model developed in this research showed good internal and external
validity as well as good performance in the independent group of patients, and may be useful in
clinical practice to estimate the EGFR mutation probabilities for individual patients.