A predictive model to estimate the probability of EGFR mutation(s) in New Zealand patients with non-squamous NSCLC


Year:

Session type:

Theme:

Phyu Aye1,Sandar Tin Tin1,Mark McKeage1,Prashannata Khwaounjoo1,Arier Lee1,Alana Cavadino1,Mark Elwood1
1The University of Auckland

Abstract

Background

Activating mutations in the Epidermal Growth Factor Receptor (EGFR) gene confer sensitivity to EGFR tyrosine kinase inhibitors (TKI) in patients with non-small cell lung cancer (NSCLC). EGFR mutation testing is now recommended for all patients with non-squamous NSCLC to guide treatment decision making. However, it can be challenging to obtain sufficient tumour tissue for testing in many patients.  We therefore developed and validated a predictive model to estimate the probability of EGFR mutation(s) in New Zealand patients.

Method

We identified a population-based cohort of 1669 non-squamous NSCLC patients diagnosed in northern New Zealand between 1 Feb 2010 and 31 Jul 2017 from the New Zealand Cancer Registry, and obtained information on EGFR mutation testing from TestSafe, a clinical information sharing service. We developed a multivariable logistic regression model based on 1176 patients, diagnosed between 1 Mar 2014 and 31 July 2017, and depicted the predicted probabilities of EGFR mutation positivity in a nomogram. We assessed the model’s calibration and discrimination in the remaining cohort diagnosed earlier (n=493). To optimise the model’s clinical applicability, we are currently undertaking further analyses by including an independent NZ cohort (n=129), who are non-tested TKI-treated.

Results

The EGFR mutation positivity was 23.0% in the model development cohort and 22.1% in the validation cohort. Significant predictors were sex (female OR=1.6, p=0.007 compared to male), ethnicity (Asian OR=2.7, p=<0.001, and Pacific OR=1.8, p=0.016 compared to NZ European), and smoking (non-smokers OR=7.1, p<0.001 and ex-smokers OR=2.0, p=0.008 compared to current smokers). Calibration in quintiles showed all mean predicted probabilities were within the 95% confidence intervals of observed probabilities. Area under the ROC curve (AUC) of 0.71 shows a reasonably good discrimination. (The results from further analyses will be available at presentation.)

Conclusion

Our model was reasonably accurate in estimating the probability of EGFR mutation positivity in NZ patients with non-squamous NSCLC.