Prediction model for tumor volume nadir in EGFR-mutant NSCLC patients treated with EGFR tyrosine kinase inhibitors

Abstract

Purpose:

In patients with advanced NSCLC and oncogenic driver mutations treated with effective targeted therapy, a characteristic pattern of tumor volume dynamics with an initial regression, nadir, and subsequent regrowth is observed on serial CT scans. We developed and validated a linear model to predict the tumor volume nadir in EGFR-mutant advanced NSCLC patients treated with EGFR tyrosine kinase inhibitors (TKI).

Methods:

Patients with EGFR-mutant advanced NSCLC treated with EGFR-TKI as their first EGFR-directed therapy were studied for CT tumor volume kinetics during therapy, using a previously validated CT tumor measurement technique. A linear regression model was built to predict tumor volume nadir in a training cohort of 34 patients, and then was validated in an independent cohort of 84 patients.

Results:

The linear model for tumor nadir prediction was obtained in the training cohort of 34 patients, which utilizes the baseline tumor volume before initiating therapy (V₀) to predict the volume decrease (mm³) when the nadir volume (V_p) was reached: V₀ - V_p = 0.717*V₀ – 1347 (P=2*10⁻¹⁶; R²= 0.916). The model was tested in the validation cohort, resulting in the R² value of 0.953, indicating that the prediction model generalizes well to another cohort of EGFR-mutant patients treated with EGFR-TKI. Clinical variables were not significant predictors of tumor volume nadir.

Conclusion:

The linear model was built to predict the tumor volume nadir in EGFR-mutant advanced NSCLC patients treated with EGFR-TKIs, which provide an important metrics in treatment monitoring and therapeutic decisions at nadir such as additional local abrasive therapy.

INTRODUCTION

Molecular targeted therapy has become the mainstream of the treatment of advanced non-small-cell lung cancer (NSCLC) with specific genomic driver mutations.^1–4 In patients with advanced NSCLC harboring sensitizing mutations of epidermal growth factor receptor (EGFR) treated with EGFR-tyrosine kinase inhibitors (EGFR-TKIs), their tumors initially show marked shrinkage, followed by further subsequent decrease leading to the nadir (the smallest tumor burden since the initiation of therapy) which is most commonly noted between 6–12 months of therapy. After reaching the nadir, the tumors typically start to grow back slowly and eventually progress, due to the development of different mechanisms of acquired resistance.^5–8

Imaging plays a key role in diagnosing and monitoring advanced NSCLC in the setting of precision oncology.^8–12 Notably, tumor volume dynamics of advanced NSCLC during effective targeted therapy can be reliably assessed using CT tumor volumetry using clinical chest CT scans with high inter- and intraobserver agreements.^{13, 14} The pattern of initial response, nadir, and subsequent regrowth is consistently noted in patients with EGFR mutations and anaplastic lymphoma kinase (ALK) rearrangements that have effective targeting agents.^{8, 15, 16} Moreover, cohorts of patients with specific oncogenic driver mutations treated with corresponding effective targeted therapy (i.e., EGFR-mutant patients treated with EGFR-TKI) demonstrate reproducible tumor dynamics patterns,^{5–7, 14, 17} which potentially can inform the anticipated metrics of tumor volume dynamics at the initiation of therapy.

Tumor volume nadir, defined as the smallest tumor volume after the initiation of therapy, is one of the important metrics of the tumor volume dynamics during targeted therapy, as it defines the maximal tumor shrinkage and it also serves as the start point to determine the rate of recurrent tumor growth.^5–7 The identification of tumor nadir is also important because it can help to guide the treatment decisions at the nadir, for the possibilities including additional agents to targeting agents, or the application of local ablative therapy such as radiotherapy targeted to the residual tumor at nadir using stereotactic radiation therapy (SBRT) techniques.^18–30 Intuitively, ablative therapy may be most effectively applied when tumor burden is at its nadir during EGFR-TKI therapy. However, the identification of the nadir is mostly retrospective, and no robust model has been established to predict the tumor nadir during EGFR-TKI therapy for EGFR-mutant NSCLC. Developing a nadir prediction model based on the objective assessment of tumor volume burden may further guide the combination therapy with EGFR-TKI and local ablative therapy.

Building on the experiences of using CT tumor volumetry to characterize the tumor dynamics during targeted therapy for lung cancer,^5–7 the purpose of the present study is to develop and validate a prediction model for the tumor volume nadir during EGFR-TKI therapy for patients with EGFR-mutant advanced NSCLC.

MATERIALS AND METHODS

Patients

Patients with advanced NSCLC harboring sensitizing mutations of EGFR treated with EGFR-TKI, erlotinib, gefitinib, or afatinib, as their first EGFR-directed therapy between 2003–2017 were studied for their CT tumor volume kinetics. Our datasets had two cohorts; the first cohort was a training cohort and included 34 patients, and another cohort as a validation cohort included 84 patients. All patients had at least one dominant measurable lung lesion (≥10 mm in the longest diameter) on baseline chest CT.^{6, 16} All patients also had initial tumor volume decrease noted on at least two CT scans before reaching the nadir, and had subsequent tumor growth after the nadir while on EGFR-TKI.^{6, 16} Patients in the training cohort have been previously evaluated for the initial tumor volume decrease during EGFR-TKI therapy⁵, and patients in the validation cohort were newly identified and evaluated with CT volume analysis.

Chest CT scans and medical records of these patients were retrospectively reviewed following the institutional review board approval. Demographics and clinical characteristics of the patients, including age, gender, race, smoking history, tumor histology, stage at diagnosis were collected from the medical records.

CT tumor volume measurements and volume tracking

The tumor volume measurements of dominant lung lesions (one lesion per patient) on the baseline CT and on all follow-up CT scans during EGFR-TKI therapy were performed by thoracic radiologists (MN for the training cohort and TH for the validation cohort), using a previously validated technique on the volume analysis workstation (Vitrea; Vital Images, Minnetonka, MN), as done before.^{5–7, 13} In patients with more than one measurable lung lesion, the largest lung lesion was selected as a dominant lesion based on the longest diameter of the lesion, as in the prior studies.^{5–7, 13, 15} Follow-up CT scans during therapy were performed every 8 weeks in patients treated in clinical trials of EGFR-TKIs, and at the discretion of the treating providers for patients who received treatment off protocol, however, these non-trial patients were also scanned approximately at every 8 weeks in general, according to the CT scan interval used in the clinical trials.^{5, 31}

The workflow for tumor volume measurements for advanced NSCLC treated with molecular targeted therapy have been previously published.^{5–7, 13, 16, 32} In brief, the axial chest CT images were displayed with a real-time interactive navigation, and a reader manually selects a small region of interest within a tumor by a mouse click, determining a seed point. The software automatically segments the lesion from the surrounding lungs and adjacent structures, using a three-dimensional seed-growing algorithm. The reader then visually assessed the automatically segmented tumor contours, and if needed, manually adjusts the contour to generate the final tumor contour. After segmentation and manual correction, tumor volume (measured in mm³) was automatically calculated by the software.^{5–7, 13, 16, 32} The intra- and interobserver variability of the tumor volume measurements using this technique in advanced NSCLC patients has previously demonstrated a high reproducibility with interobserver concordance correlation coefficients (CCC) of 0.990.¹³ The longitudinal tumor volume data for each patient were evaluated in the analytic module, which automatically detects the tumor nadir and provide graphical display of tumor volume dynamics as published before.¹⁷

Statistical analysis

For each patient, the baseline tumor volume before initiating therapy (V₀) and the nadir volume (V_p) were obtained from the serial volume data analyzed by the module. The linear model was fitted to predict the volume decrease at the nadir (V₀ − V_p). R² value was used to evaluate the performance of the fitted linear model on the training and validation sets. Pearson’s chi-square test was implemented on the residuals of fitted model on the validation set as a goodness-of-fit test. Multivariable analyses were also performed adjusting for the demographics and clinical variables, to assess their impact on the nadir prediction model.

RESULTS

Clinical characteristics and imaging metrics of the training and validation cohorts

The demographics and clinical characteristics of patients in the training set (n=34) and in the validation set (n=84) are described in Table 1. There were no significant differences in the demographics and clinical characteristics between the training and validation cohorts (P>0.11; Table 1). All patients had advanced NSCLC at the time of initiation of EGFR-TKI. In the training set, 30 patients had stage IV disease at diagnosis and 4 patients had disease recurrence after an initial diagnosis of stage I-III NSCLC. In the validation set, 78 patients had stage IV disease at diagnosis and 6 patients had disease recurrence after an initial diagnosis of stage I-III NSCLC (P=0.47).

Table 1.

Demographics, clinical characteristics, and imaging metrics for the training cohort (n=34) and testing cohort (n=84)

		Training Cohort (n=34)	Testing Cohort (n=84)	P value
Demographics and Clinical Variables
Age	Median [range]	62 [35,78]	62 [38,87]	0.26
Sex	Male	3	19	0.11
	Female	31	65
Race	White	30	68	0.42
	Other	4	16
Smoking	Never	18	51	0.53
	Current/Former	16	33
Pathology	Adeno	31	82	0.14
	Other	3	2
Stage at Diagnosis	IV	30	78	0.47
	I – III *	4	6
EGFR-TKI	Erlotinib	31	82	0.14
	Other	3	2
Imaging Metrics
Baseline Volume (V 0 ; mm 3 )	Median [range]	22179.85 [4544, 172692]	17159.65 [1730, 243819]	0.21
Nadir Volume (V p ; mm 3 )	Median [range]	5021.25 [609.5, 35259.5]	4840.6 [176.1,45921.5]	0.56
Time to Nadir (months)	Median [range]	7.7 [3.2, 22.7]	6.3 [2.4, 37.1]	0.22

^*These patients originally presented with stage I-III disease at diagnosis, and treated but later recurred and thus had advanced NSCLC, which was then treated with EGFR-TKI.

In the training cohort, the median baseline volume (V₀) was 22179.85 mm³, and the median nadir volume (Vp) was 5021.25 mm³, with the median time to nadir from the initiation of therapy of 7.7 months. In the validation cohort, the median baseline volume (V₀) was 17159.65 mm³, and the median nadir volume (Vp) was 4840.6 mm³, with the median time to nadir from the initiation of therapy of 6.3 months. The representative case of tumor volume segmentation at baseline and at nadir is shown in Figure 1. There were no significant differences in these imaging metrics between the training and validation cohorts (P>0.21).

Figure 1.

The representative case of tumor volume changes on CT in a 55-year-old woman with stage IV advanced NSCLC harboring EGFR mutation treated with erlotinib. The baseline CT (A) prior to initiating therapy demonstrated a dominant lung tumor in the right upper lobe, measuring 14847.3 mm³, with smaller metastatic nodules in both lungs. The patient started erlotinib therapy and reached the tumor nadir at 11.3 months when the dominant lesion measured 2521.8 mm³ (B). Note the disappearance of other smaller nodules in both lungs (B).

The nadir prediction model in the training and validation sets

In the training cohort of 34 patients, the baseline tumor volume before initiating therapy (V₀) was used to fit a linear model predicting the volume decrease (mm³) when the nadir volume (V_p) was reached. The following linear model was obtained:

$$V_0-V_p=0.717*V_0–1347\left(\text{P}=2*{10}^{-16};{\text{R}}^2=0.916{\text{ for linear coefficient on V}}_0\right)$$

The multivariable analyses were performed adjusting for the demographics and clinical variables listed in Table 1 to assess their impact on the model, however, these variables were not shown to be significant predictors of nadir volume (P>0.18), and thus were not included in the model.

The linear model was tested in the validation cohort of 84 patients, resulting in the R² value of 0.953 (Fig. 2). The result indicates that the prediction model generalizes well to these patients with EGFR-mutant advanced NSCLC treated with EGFR inhibitors.

Figure 2.

Scatter plot of baseline volume (V₀) and volume decrease at nadir (V₀ − V_p), fitted in a linear model in training (n=34) and validation (n=84) sets.

DISCUSSION

The present study demonstrated that the linear regression model can predict the tumor volume nadir based on the baseline tumor volume at the time of initiation of therapy for EGFR-mutant advanced NSCLC patients treated with EGFR-TKI. The prediction model was independent of the demographics and clinical variables, and was reproduced well in the validation cohort. The model reflects the homogeneous tumor dynamics of this genomically defined cohort treated with effective targeted therapy, and provides an additional metrics to guide treatment decisions based on CT tumor volume analysis.

The linear model predicted the tumor volume decrease at the nadir (V_p) compared to the baseline (V₀) as V₀ − V_p = 0.717*V₀ − 1347, with R² values of 0.916 in the training set and 0.953 in the validation set, demonstrating that this linear model was generalizable in the EGFR-mutant patients treated with the conventional EGFR-TKIs including erlotinib, gefitinib, and afatinib that were used in the cohorts. The model also indicates that the nadir volume (V_p) can be predicted as 28.3% of the baseline volume (V₀) plus the intercept of 1347 mm³. The persistent tumor burden at the nadir, consisting of about 30% of the baseline volume plus approximately 1 cm³, may reflect the drug-resistant “persister” cell populations³³, which is noted as a persistent “core” of the tumor that remains on CT scan at the time of maximal response to therapy. This persistent tumor burden at the nadir prevents otherwise very effective therapy from achieving complete response (CR) by RECIST. It is often the site of tumor regrowth after the disease starts to recur due to the development of acquired resistance, eventually leading to progressive disease (PD).

A number of therapeutic approaches have been investigated when EGFR-mutant patients start to experience tumor regrowth and disease progression during EGFR-TKI therapy, such as adding other systemic therapy including chemotherapy, immunotherapy, or other targeting agents.^{18, 19, 26, 28} Alternatively, local ablative therapy using radiotherapy or surgery was also explored in EGFR-mutant patients treated with EGFR-TKI when tumor progression was limited to one or few sites that may be locally controlled.^{22–25, 29, 30} A recent randomized phase III trial compared the first-line EGFR-TKI alone versus first-line EGFR-TKI plus upfront stereotactic radiotherapy to all sites at diagnosis in patients with previously untreated EGFR-mutant oligometastatic NSCLC. Addition of upfront stereotactic radiotherapy resulted in significantly longer PFS and OS.²⁹ Another study retrospectively evaluated 36 patients metastatic EGFR-mutant NSCLC treated with first-line gefitinib who developed oligoprogression and received high-dose hypofractionated radiotherapy while continuing EGFR-TKI,³⁰ which provided additional time of disease control, with the median time from focal progression to further progression of disease or death being 6.3 months.³⁰ These data provide a rationale for local ablative treatment such as SBRT for oncogene-driven NSCLC progressing on TKI while continuing the original TKI therapy. A phase 2 trial is ongoing to integrate SBRT with systemic targeted therapy in stage IV oncogene-driven NSCLC for treatment of residual disease after initial response to molecular targeting agents (NCT02314364). In the trial, SBRT is performed within 6–12 months from the initiation of targeted therapy, which roughly corresponds to the median time to nadir in patients treated with EGFR-TKI but does not utilize quantitative volume threshold to trigger SBRT. The nadir prediction model obtained in the present study will allow more precise prediction of the tumor nadir in each patient based on the baseline tumor volume, which may contribute to optimize the timing of SBRT by indicating the observed tumor volume on a follow-up scan during TKI therapy has reached the expected nadir for the patient.

Another potential value of the nadir prediction model is to help better plan follow-up CT scans during targeted therapy and optimize treatment monitoring. Currently, advanced NSCLC patients treated with targeted therapy often undergo follow-up CT scans for the evaluation of tumor response and progression with a fixed interval, most commonly every 8 weeks. The 8-week interval is based on the follow-up interval used in the prior clinical trials of EGFR-TKI,^{5, 31, 34} and is regardless of the trajectory of tumor burden dynamics in each patient after starting therapy. The precise prediction of tumor volume dynamics may help to reduce unnecessarily frequent scans and decrease radiation exposure when tumors are relatively stable at or around the nadir, while alerting a need to go back to serial scanning when tumors are at the risk to demonstrate regrowth. This is particularly pertinent with a further increase of efficacy of targeting agents, including osimertinib with the median PFS of 18.9 months in EGFR-mutant patients³⁵ and alectinib with the median PFS of 34.8 months for ALK-positive patients, which will prolong the follow-up period on therapy.³⁶ Further investigations are needed to guide timing of therapy response imaging, including the prediction of the time to nadir and prediction of the risk of tumor regrowth and progression, which may require an integration of both imaging and non-imaging markers. The prediction of tumor volume nadir on serial CT scans is an important step to provide a key metrics for such strategy.

The limitations of the present study include a retrospective design with cohorts treated at the single institution. EGFR-TKIs included erlotinib, gefitinib, and afatinib, which are the conventional agents that have been used for this genomically defined patient cohort, and does not include newer third-generation agents such as osimertinib. Further studies are needed to prospectively validate the model and to determine if a similar model can be built in osimertinib-treated EGFR-mutant patients and in patients with other genomic drivers such as ALK-rearranged patients treated with ALK inhibitors. The present study included the first and second-generation EGFR-TKIs that have been clinically used in the past decade with the mature follow-up data. Osimertinib was approved as the first-line treatment for EGFR-mutant advanced NSCLC in 2018, and will require some time before the data accumulate with sufficient follow-up. The current study focused on the volume nadir, because the prior studies have addressed the initial volumetric response and its association with survival,^{5, 15} and the tumor growth rates after nadir while on TKI have also been previously defined and validated.^{6, 7, 16} Tumor volume analysis was performed using the measurement of a dominant lung lesion for each patient, following the approach used in the prior studies that demonstrated the prognostic value of tumor volume analysis of one dominant lung lesion in advanced NSCLC patients receiving precision therapy including EGFR-TKIs and ALK-TKIs.^{5–7, 13, 15} Additionally, in most cases of EGFR-mutant NSCLC patients treated with effective EGFR-TKIs, smaller metastatic sites often disappear or shrink significantly and become unmeasurable, while the dominant lung mass remains and represents most of the remaining volume burden at the nadir. Follow-up chest CT scan interval was 8 weeks in general for our cohort treated at the tertiary cancer center, but may vary according to the practice setting and may potentially affect the nadir volume; however, typical tumor volume trajectories in EGFR-mutant patients treated with EGFR-TKIs often demonstrate a period of tumor stability at/around nadir for at least a few months or longer.

In conclusion, the linear regression model was able to predict the tumor volume nadir in EGFR-mutant advanced NSCLC patients treated with EGFR-TKI, which may help to guide additional therapeutic options such as local ablative therapy in these patients to prolong survival. Given the similar patterns of tumor volume dynamics in patients treated with oncogenic driver mutations treated with effective targeted therapy, the approach can be expanded to the cohorts treated with newer EGFR-TKI and other genomically-defined cohort of NSCLC patients.