Epidermal growth factor receptor (EGFR) tumour genotyping is crucial to guide treatment decisions regarding the use of EGFR tyrosine kinase inhibitors in nonsmall cell lung cancer (NSCLC). However, some patients may not be able to obtain tumour testing, either because tissue is limited and/or tests are not routinely offered. Here, we aimed to build a model-based nomogram to allow for prediction of the presence of EGFR mutations in NSCLC.
We retrospectively collected clinical and pathological data on 3,006 patients with NSCLC who had their tumours genotyped for EGFR mutations at five institutions worldwide. Variables of interest were integrated in a multivariate logistic regression model.
In the 2,392 non-Asian patients with lung adenocarcinomas, the most important predictors of harbouring EGFR mutation were: lower tobacco smoking exposure (OR 0.41, 95% CI 0.37–0.46), longer time interval between smoking cessation and diagnosis (OR 2.19, 95% CI 1.71–2.80), advanced stage (OR 1.58, 95% CI 1.18–2.13), and papillary (OR 4.57, 95% CI 3.14–6.66) or bronchioloalveolar (OR 2.84, 95% CI 1.98–4.06) histologically predominant subtype. A nomogram was established and showed excellent discriminating accuracy: the concordance index on an independent validation dataset was 0.84.
As clinical practices transition to incorporating genotyping as part of routine care, this nomogram could be highly useful to predict the presence of EGFR mutations in lung adenocarcinoma in non-Asian patients when mutational profiling is not available or possible.
- epidermal growth factor receptor mutations
- lung cancer
- prediction score
- tyrosine kinase inhibitor
Nonsmall cell lung cancer (NSCLC) is the most frequent cause of cancer-related death worldwide, accounting for >1 million deaths per year . NSCLC is primarily comprised of three different types: squamous cell carcinoma, large cell carcinoma and adenocarcinoma. Adenocarcinoma now comprises >50% of all cases of lung cancer in the USA and Western Europe [2, 3].
Somatic mutations in the epidermal growth factor receptor (EGFR) gene are present in a subset of lung adenocarcinomas. Biological data regarding EGFR mutations have been extensively reviewed elsewhere . Briefly, EGFR mutations primarily occur in the tyrosine kinase domain, mostly involving deletions in exon 19 and a recurrent point mutation (L858R) in exon 21 [5, 6]. These mutations lead to constitutive activation of the receptor, independent of ligand binding, and are associated with increased sensitivity to the specific EGFR tyrosine kinase inhibitors (TKIs) gefitinib and erlotinib . Clinically, response rates to these agents are >70% in tumours harbouring EGFR activating mutations [7–9]. Recent randomised phase III trials (Iressa Pan-Asia Study (IPASS), West Japan Thoracic Oncology Group 3405 and North-East Japan Study Group) have found that, for patients with EGFR mutant tumours, first-line gefitinib leads to a longer progression-free survival compared with standard platinum-based doublet chemotherapy [7–9]. Thus, genotyping of EGFR is crucial to guide treatment decisions regarding the use of EGFR TKIs and is becoming a standard recommendation in the pre-treatment work-up of patients with lung adenocarcinoma [10–12].
Unfortunately, in many countries, routine access to mutation testing has been hampered by a lack of well-established molecular diagnostic laboratories or by economic and/or regulatory issues. For example, from 2008 to 2009, only 16% and 11% of metastatic adenocarcinoma cases were genotyped for EGFR activating mutations in France and Spain, respectively. Moreover, many patients with metastatic NSCLC are diagnosed with needle biopsies, limiting the amount of tissue available for mutation testing. In IPASS, tumour specimens were collected for only 66% of patients and EGFR mutation status could ultimately be assessed in only 64% of those cases . In these settings, the value of an accurate prediction tool for the presence of EGFR mutations in patients’ tumours, based upon clinical characteristics, is indisputable.
Statistical prediction models are widely used for cancer outcome prediction. Among those, nomograms create user-friendly graphical representations of complex models to generate the probability of an event based on the individual profile of each patient . Here, we integrated multiple clinical and pathological factors from a large, well-annotated international cohort of genotyped lung tumours to develop a multivariate logistic regression model and an associated nomogram to predict the presence of EGFR activating mutations.
PATIENTS AND METHODS
We retrospectively collected clinical and pathological data on 3,006 patients with NSCLC, including 2,856 with adenocarcinomas, who had their tumours genotyped for EGFR activating mutations at five institutions worldwide: 1,990 adenocarcinoma patients were from Memorial Sloan-Kettering Cancer Center (MSKCC) (New York, NY, USA); 297 were from the Dana-Farber Cancer Institute (DFCI) (Boston, MA, USA); 269 were from Massachusetts General Hospital Cancer Center (MGH) (Boston, MA); 213 were from the Cancer Hospital of Fudan University (CHFU) (Shanghai, China); and 87 were from the National Taiwan University Hospital (NTUH) (Taipei, Taiwan) . The Catalan Institute of Oncology (CIO) (Hospital Germans Trias I Pujol, Badalona, Spain) also contributed data on 217 patients (198 adenocarcinomas) who were ultimately not included in the analyses because of missing data on key variables, such as smoking history.
Starting in 2004 at NTUH, DFCI, MGH and MSKCC, 2005 at CIO, and 2007 at CHFU, EGFR genotyping was systematically and prospectively performed for all consecutive, cytologically or histologically proven adenocarcinoma tumours with available material for testing, without any selection based on additional clinical or pathological criteria. Therefore, patients included in our analysis are representative of all patients treated at the five institutions included.
Data were collected in July 2009 under local institutional review board (IRB)-approved protocols. The present study was approved by each institution IRB.
For this analysis, we considered only drug-sensitive EGFR activating mutations, i.e. kinase domain mutations associated with sensitivity to EGFR TKIs: exon 19 deletions, exon 21 mutation (i.e. L858R) and other point mutations (G719A/C/S and L861Q). Tumours harbouring other EGFR mutations were excluded. Direct sequencing was the standard method used for EGFR genotyping (exons 18–21) at DFCI, MGH, CIO, CHFU and NTUH. At MSKCC, EGFR mutational analysis was performed both using direct sequencing and PCR-based, mutation-specific assays to detect the two most common EGFR mutations, i.e. exon 19 deletions and L858R point mutations . We also retrieved the KRAS mutational status of NSCLC tumours, if available. KRAS mutations were assessed using direct sequencing .
Clinical and pathological data
Clinical and pathological variables of interest included: age, sex, race/ethnicity, histological type (adenocarcinoma, squamous cell carcinoma or large cell carcinoma), and for adenocarcinoma tumours, histologically predominant subtype , tumour stage according to the 6th American Joint Committee on Cancer TNM (tumour, metastasis and node) system that was standard at time the data were generated , and smoking variables (smoking status, smoking quantity (measured in pack-yrs) for smokers and time interval since quitting for former smokers). Patients were categorised as never-smokers (<100 lifetime cigarettes), former smokers (quit >1 yr prior to diagnosis) or current smokers (continued smoking within 1 yr of diagnosis). Pathological review was performed at each institution. We also retrieved results from routine immunohistochemical studies performed with an antibody against thyroid transcription factor (TTF)-1, a marker of terminal respiratory unit adenocarcinoma, which is more likely to harbour EGFR mutation . However, because data were lacking, we could not include this covariate in the final model.
Analyses were performed on patients diagnosed with NSCLC adenocarcinoma, separately for non-Asian and Asian patients, resulting in one model for each of the two populations. The Chi-squared and two-sample paired t-tests were used to compare the univariate association between clinical parameters and the presence of EGFR or KRAS mutations, for categorical and continuous variables, respectively. The methodology used to develop the prediction nomogram was previously described [13, 20]. Briefly, for each model developed, we randomly divided the available data from patients diagnosed with lung adenocarcinoma into a training set (including two-thirds of the patients), used to develop a prediction model, and a validation set (including the remaining one-third). A model-building exercise was performed to develop multivariate logistic regression models that used available clinical and pathological factors to predict the likelihood of an EGFR mutation. All variables that were univariately associated with EGFR status at a level of α<0.2 were candidates for inclusion in the multivariate model, and were retained in the model if they remained significantly associated with the outcome. Those variables with a markedly skewed distribution were log-transformed; the linearity assumption was further checked for continuous and ordinal variables and, if the functional form of the relationship with the outcome suggested, they were modelled using quadratic polynomials. Levels of the categorical variables were combined if they had similar effects. Interactions between all combinations of two variables were tested but were not included in the final model because their effect did not reach a level of significance of α<0.05 and their addition to the model did not improve the model’s predictive accuracy by >1%.
The discriminatory ability of the model was quantified using the concordance index, which is numerically equivalent to the area under the receiver operating characteristic curve. Concordance index indicates the probability that, in a randomly selected pair in which one patient has and the other patient does not have EGFR mutation, the model correctly predicts a higher probability of mutation for the first patient. In practice, concordance index ranges from 0.5 (no discriminatory ability, equivalent to a coin toss) to 1.0 (perfect discrimination) .
Calibration was assessed by plotting the probabilities predicted by the logistic model versus the actual probability. Since the individual actual probability is either 1 or 0 (for patients with and without EGFR mutation, respectively), calibration plots were built using local regression, a method commonly used for data smoothing that relies on approximating each data point with a low-degree polynomial using the neighbouring values. A plot along the 45° line will correspond to a model in which the predicted probabilities are identical with the actual probabilities, therefore indicating perfect calibration. Both discrimination and calibration of the prediction models were calculated on the independent validation dataset.
To protect against the influence of the initial random split, the analysis was repeated in 10 randomly split training and validation datasets, using 10-fold cross-validation, refitting the model to omit one-tenth of the observations to obtain prediction for the omitted tenth. Statistical calculations were performed using SAS version 9.2 (SAS Institute Inc., Cary, NC, USA), and the Design and locfit packages in R (The R Foundation for Statistical Computing, Vienna, Austria).
EGFR mutations in lung adenocarcinomas of non-Asian patients
We focused our analysis on the largest population in our cohort, namely non-Asian patients (including white and black, and Hispanic and non-Hispanic), with lung adenocarcinoma. To maximise the chances of building an effective model, we included only cases for which all smoking variables were available (n=2,392). 1,579 (66%) patients were randomly included in a training dataset and the remaining 813 (34%) patients formed a validation dataset. Demographic, clinical and pathological characteristics of the patients are presented in table 1. Overall, 604 (25%) tumours harboured an EGFR activating mutation, corresponding to exon 19 deletion in 54% of cases and L858R substitution in 37% of cases.
Patients with EGFR mutant tumours were more likely to be females (72% versus 62% of patients; p<0.001) and to be diagnosed with advanced-stage disease (55% versus 45%; p<0.001) compared with the patients with EGFR wild-type tumours. There was no significant difference between the two groups with respect to age (64±12 versus 65±11 yrs; p=0.370) or race/ethnicity (94% of patients were white in both groups). 55% of the patients with EGFR mutant tumours were never-smokers, compared with 19% of those with EGFR wild-type tumours (p<0.001). For former smokers, median time interval since quitting was 27 yrs among EGFR mutant and 20 yrs among EGFR wild-type patients.
Tumours with EGFR mutations were more likely to harbour papillary (30% versus 12% of patients; p<0.001) and bronchioloalveolar predominant histological subtypes (25% versus 17%; p<0.001), and less likely to exhibit acinar (18% versus 39%; p<0.001) or solid architecture (9% versus 14%; p=0.002). TTF-1 expression was more frequent in EGFR mutant tumours (97% versus 84% of tested tumours; p<0.001). Histological subtype was not otherwise specified (NOS) or could not be assessed for 18% of adenocarcinomas, because the diagnosis was made either on small-size tumour tissue samples (22% of NOS cases) or cytological specimens (78% of NOS cases).
The results of a multivariate model predicting the probability of EGFR mutation in non-Asian patients with lung adenocarcinoma are presented in table 2. The most important predictors of harbouring EGFR activating mutation were: smoking history, measured by lower total tobacco smoking exposure (log-transformed pack-yrs; OR 0.41, 95% CI 0.37–0.46) and longer time interval between tobacco smoking cessation and NSCLC diagnosis (OR 2.19, 95% CI 1.71–2.80); advanced stage at diagnosis (OR 1.58, 95% CI 1.18–2.13), and papillary (OR 4.57, 95% CI 3.14–6.66) or bronchioloalveolar (OR 2.84, 95% CI 1.98–4.06) histologically predominant subtype. These predictors were not different for L858R mutations and exon 19 deletions.
Based upon this analysis, we generated a nomogram to predict the presence of activating EGFR mutation in tumours (fig. 1). The concordance index calculated for the independent validation dataset was 0.84 (95% CI 0.80–0.86), indicating excellent accuracy in discriminating EGFR mutant versus EGFR wild-type cases in this group of patients. In addition, the calibration plot (fig. 2) indicates that the nomogram-predicted probabilities compared very well with actual probabilities. Moreover, additional internal validation indicated no change between the original concordance index and that based on 10-fold cross-validation.
Influence of KRAS mutations on the nomogram
We tested the effect of KRAS mutational status on the nomogram. KRAS mutations are mutually exclusive with EGFR mutations in lung adenocarcinoma and KRAS mutant tumours represent a subset of EGFR wild-type tumours that do not respond to EGFR TKIs [16, 22]. Given the higher frequency of KRAS mutations in NSCLC tumours from non-Asian patients and technical issues that may make KRAS genotyping easier, KRAS mutations have been used by some institutions as a negative surrogate marker for the presence of EGFR mutations. KRAS mutational status was available for 82% of cases; 17% of genotyped tumours harboured KRAS mutations. In our cohort, we did not identify clinical or pathological variables that were significantly associated with the presence of KRAS mutations. A multivariate model restricted to patients who had KRAS wild-type tumours yielded very similar results to the model described above. The prediction accuracy of this model on the independent validation dataset was 0.83.
EGFR mutations in lung adenocarcinomas of East Asian patients
We analysed separately a group of 464 East Asian patients with lung adenocarcinoma for whom all smoking covariates were available (table 3). Variables that were significantly associated with the presence of activating EGFR mutation in the tumour were: female sex (66% versus 46% of patients; p<0.001), never-smoker status (79% versus 57%; p<0.001) and lower tobacco quantity for former/current smokers (24±21 versus 39±24 pack-yrs; p<0.001). Similar to what was observed in the group of non-Asian patients, EGFR mutant tumours were more likely to harbour papillary (20% versus 8% of patients; p<0.001) or bronchioloalveolar (22% versus 12%; p=0.01) histologically predominant subtypes, and less likely to harbour acinar (43% versus 54%; p=0.02) or solid (1% versus 7%; p=0.010) features.
The results of a multivariate model are presented in table 4. Because of the small number of patients in each subgroup, we combined histologically predominant subtype categories as follows: bronchioloalveolar carcinoma/papillary versus acinar/solid/NOS. Lower pack-year smoking history (OR 0.63, 95% CI 0.53–0.75), early stage (OR 2.27, 95% CI 1.32–3.85) and bronchioloalveolar carcinoma/papillary predominant histology (OR 2.73, 95% CI 1.51–4.95) remained significantly associated with a higher probability of EGFR mutation. This model had a prediction accuracy of 0.75 on the training dataset; however, its accuracy was unsatisfactory when validated on an independent dataset (concordance index was 0.64, 95% CI 0.55–0.73). Notably, in our dataset, only 29% of Asian patients were smokers, resulting in a lack of power of smoking variables to estimate the risk of mutation, especially in the lower-risk group.
Tumour EGFR genotype has become an important variable in the evaluation of patients with lung cancer. However, in many countries, barriers such as tissue and test accessibility still exist that prohibit patients from obtaining tumour EGFR mutation status. Here, we integrated clinical and pathological factors from thousands of patients from the USA and East Asia, to develop a logistic regression-based model to better predict the presence of EGFR mutations in NSCLCs, should mutation testing not be available. The nomogram has excellent discrimination properties in non-Asian patients with a concordance index >80%. To our knowledge, this cohort represents the largest dataset analysed to date for comprehensive clinical characteristics associated with EGFR mutations.
Other studies have identified clinical and pathological characteristics associated with the presence of EGFR activating mutations in lung cancer [6, 17, 23–27]. Consistent with previous results, we found, through multivariate analysis, that smoking history, including never-smoker status, lower tobacco consumption in smokers and longer smoking discontinuation before NSCLC diagnosis in former smokers, were all associated with the presence of EGFR activating mutation [17, 23–27]. We also observed an association between bronchioloalveolar and papillary histological subtypes [5, 17, 27]. Contrary to our analyses, previous studies did not identify advanced tumour stage as a significant predictor of harbouring EGFR mutation. Unidentified selection or technical bias may exist in the datasets we analysed, as EGFR mutations are rather known to occur early in the lung carcinogenesis process .
Overall, compared with prior reports, this study was the first, to our knowledge, to incorporate multiple clinical variables, including smoking history and histological subtype, as well as sex, stage of disease and age, into a user-friendly predictive nomogram for EGFR mutations. The additional variables allow for higher accuracy, and the use of test and validation sets further substantiates the validity of the model.
Our nomogram does have some limitations. First, our analyses focused on adenocarcinoma tumours, limiting our analysis of other histologies. However, the majority of EGFR mutations are found in adenocarcinomas . Secondly, the proposed nomogram does not apply to Asian patients, who had very different distributions of the predictive variables (smoking history and stage) and considerably higher prevalence of EGFR mutations (61% compared with 25% in non-Asian patients) in our dataset. Probably due to these issues, attempts to build a common nomogram that applies to both non-Asian and Asian patients resulted in a predictive instrument with unsatisfactory predictive accuracy. Thirdly, any prediction instrument inherently incorporates a certain degree of uncertainty  and individual predictions remain imperfect. Our data come from large academic centres, and the patient population in these institutions may be different from that observed in other medical settings; in general, caution should be exercised if the nomogram is used to predict the likelihood of EGFR mutation for patients coming from populations with a different case-mix and a different prevalence of mutation than those studied here.
In IPASS, inclusion criteria were based on clinical characteristics associated with the presence of EGFR mutation in the tumour, including never- or light smoker status and adenocarcinoma histology . In the whole intent-to-treat population, for whom subsequent genotyping showed a 59.7% rate of EGFR mutation in tumours, progression-free survival was longer in the gefitinib arm than in the chemotherapy arm (hazard ratio 0.74, 95% CI 0.65–0.85; p<0.0001) . Based on these data, the threshold for considering the chance of EGFR mutation, as predicted by our nomogram, to be clinically significant for recommending EGFR TKI, may be 60%. The use of EGFR TKIs in patients with a moderate probability of a EGFR mutant tumour based on the nomogram may not be recommended, as EGFR TKI treatment resulted in a poorer outcome in EGFR wild-type tumours compared with standard chemotherapy in the reported trials [7, 29]. Finally, prospective validation of independent datasets will have to be conducted; inclusion of additional Asian patients and non-adenocarcinoma tumours, as well as response to EGFR TKI treatment, would increase the clinical significance of our results.
Ultimately, despite the statistically accurate value of our nomogram to predict the presence of EGFR activating mutation in lung adenocarcinoma from non-Asian patients, EGFR tumour genotyping should be obtained when possible. Trials that have included “clinically enriched” populations, i.e. patients more likely to present with EGFR mutant tumours, did not achieve improvements in outcome in the intent-to-treat population [7, 30], while trials that studied patients with known EGFR mutant tumours did [8, 9]. However, tumour genotyping is not and will not be feasible in all cases. In such instances, our nomogram can then be used to prioritise the use of EGFR TKIs in patients with lung adenocarcinoma.
We thank M.G. Kris (Memorial Sloan-Kettering Cancer Center, New York, NY, USA) for helpful comments, M. Butaney (Dana-Farber Cancer Institute, Boston, MA, USA) for DFCI data assembly and Y.L. Lin (National Taiwan University Hospital, Taipei, Taiwan) for NTUH data retrieval.
Statement of Interest
Statements of interest for H. Ji, M. Ladanyi and G. Riely can be found at www.erj.ersjournals.com/site/misc/statements.xhtml
- Received January 19, 2011.
- Accepted June 16, 2011.
- ©ERS 2012