Real-world relationship of early end points to survival end points in patients with resectable non-small-cell lung cancer

Lung cancer is the leading cause of cancer death in USA, with 80–85% of cases being non-small-cell lung cancer (NSCLC). At initial diagnosis, 26% of patients present with stage I disease, 8.3% with stage II, 27.6% with stage III, and 38.1% with stage IV disease [1]. For many patients with non-metastatic NSCLC, surgery with curative intent has been the gold standard for many years. Despite surgery, 30 to 55% of patients with NSCLC eventually develop recurrence within 2–5 years and die of their disease [2–6]. Patients with stage IB-IIIA NSCLC still represent a population with a high unmet need, with a 5-year survival rate ranging from 36 to 66% [7].

Historically, neoadjuvant or adjuvant chemotherapy with or without radiotherapy have been recommended for patients at higher risk of recurrence [8,9]. These treatment strategies have been associated with slower recurrence and modestly prolonged survival among patients with stage IB-IIIA disease [10–13]. However, both strategies have only shown approximately 5% overall survival (OS) absolute risk reduction at 5 years compared with surgery alone [7].

Typically, the investigation of improvement in outcomes for patients with resectable lung cancers has been challenging, limiting the development and approval of new drugs. Phase III trials with conventional primary end points (e.g., OS) can take up to 10 years to complete the accrual and required follow-up [7,14]. Over the years, these trials have struggled to accrue, leaving several underpowered for their event-free survival (EFS) and OS end points [7]. In addition, these large and lengthy trials may be confounded by the use of additional post-operative interventions based on local clinical practice, including chemotherapy or radiotherapy to address residual disease. Hence, there is a need for early or intermediate end points that both isolate the effect of novel neoadjuvant interventions and expedite development and approval of promising treatments for patients with resectable NSCLC.

Over the last decade, pCR and MPR have been proposed as early end points for prediction of long-term clinical benefit, such as EFS and OS among patients with NSCLC, and are increasingly being included in neoadjuvant clinical trials [15–18]. Prior research has found that pathologic response is predictive of OS and EFS [19,20]. However, heterogeneous definitions of pCR, low pCR rates and infrequent reporting of MPR in clinical trials present a major challenge while evaluating the outcomes in patients with NSCLC [15,17,21,22].

The purpose of this study is to generate real-world data characterizing the relationship between early end points and survival-based outcomes and to describe the patient profiles and neoadjuvant treatment patterns in patients with resectable NSCLC (stage IB [tumor ≥4 cm]-IIIA) [23,24] treated in the US community oncology setting.

Materials & methods

Study design & population

This is a retrospective cohort study including patients with a diagnosis of stage IB (tumor ≥4 cm), II, or IIIA NSCLC receiving neoadjuvant treatment at practices within The US Oncology Network that utilize the iKnowMed™ electronic health records (iKM EHR). The US Oncology Network consists of approximately 1400 physicians in more than 500 sites of care across 40 states in the USA, representing approximately 12% of US patients newly diagnosed with cancer [25]. Data were obtained via programmatic database abstraction from the EHR and supplemented with chart review. Vital status was confirmed with data from the Limited Access Death Master File (LADMF) and chart review. Eligible patients were ≥18 years of age with a diagnosis of stage IB (tumor ≥4 cm)-IIIA NSCLC¹ or stage II–IIIB² who initiated neoadjuvant treatment between 1 January 2008 and 31 January 2019 (defined as the patient identification period) with an intention of surgery and with ≥2 visits or a record of death after the index date (date of neoadjuvant treatment initiation) through 31 July 2019, which was the end of the study observation period. Patients were followed through the last visit, death or end of the study period, whichever came first. Patients were excluded if they were enrolled in clinical trials for non-metastatic NSCLC at any time during the study period or if they were treated for a documented primary cancer other than NSCLC during the study period. Patients with documented EGFR mutations or ALK translocation at diagnosis were excluded. Patients with EGFR mutations/ALK translocations are different from patients without. These differences include patient demographics, clinical characteristics and outcomes. In particular, given the adoption and efficacy of EGFR and ALK therapies in the metastatic/recurrent setting, outcomes for these patients would differ from wild-type patients, which could confound the analysis.

Statement of ethics

Institutional Review Board approval was obtained for the study. All data were handled in compliance with the Health Insurance Portability and Accountability Act of 1996 and the Health Information Technology for Economic and Clinical Health Act.

Study outcomes

The study explored the following outcomes: pCR, MPR, EFS and OS. These outcomes were analyzed for the overall study population and by type of neoadjuvant therapy (chemotherapy only, chemoradiotherapy). pCR was defined as no viable tumor cells in resected tumor tissue and lymph node tissue as reported in the pathology report or physician notes from the chart review of the index surgery, pathologic stage of pT0 pN0 or ypT0 ypN0. MPR was defined as ≤10% residual tumor tissue in the resected lung specimen. Patients with no surgery and with no information on pCR were classified as non-pCR. pCR was calculated for all patients who received neoadjuvant treatment with an intention of surgery. The as-treated approach is important, as isolating only patients that undergo surgery would introduce bias and overinflate pathologic response rates, excluding several real-world situations where patients may initiate treatment but discontinue (e.g. toxicity) or not be able to make it to surgery (e.g. tumor may progress while awaiting resection) and have worse outcomes than going straight to surgery.

Progression or recurrent disease was defined by the availability of provider-recorded evidence of recurrence or progression in a scan report or progress note.

EFS was estimated as the interval from the index date until the earliest date of progression of disease precluding surgery, progression or recurrence after surgery or death due to any cause. Patients who did not undergo surgery for any reason other than progression were considered to have had an event at progression or death. Patients without relapse, recurrence, progression or death were censored on the study end date or the last visit date, whichever occurred first, available in the database.

OS was defined as the time interval between the index date and the date of death as documented in the LADMF or the iKM EHR database. Patients who did not die were censored on the study end date or the last visit date available in the database, whichever occurred first.

Statistical analysis

Descriptive analyses were conducted to summarize the baseline patient characteristics (clinical and treatment) for the study cohort population. The baseline period was defined as a 6-month period prior to the index date. These characteristics were presented for the overall study population and by type of neoadjuvant therapy (chemotherapy only, chemoradiotherapy). Continuous variables were described by means, standard deviations, medians and ranges, and categorical study variables were reported as frequencies (counts and percentages). Kaplan-Meier survival curves were generated for the outcome of OS/EFS for patients with and without pCR/MPR. An unadjusted Cox analysis was used to calculate HRs with 95% CIs for the association between pCR/MPR and OS/EFS in the overall population. A Cox proportional hazards model controlling for demographic and clinical characteristics (e.g., age, gender, disease stage, histology, site) was then conducted to describe the association between pCR and OS. The variables with a p-value less than or equal to 0.20 in Cox univariate analyses were included in the multivariable analyses. Landmark analyses were performed according to an event status at 3, 6, 12 and 24 months to identify survival outcomes, and unadjusted HR for OS was calculated. Additionally, EFS was used as a time-dependent covariate in a Cox regression on OS. As a patient-level analysis of the association between EFS and OS, event time versus survival time was plotted for each participant and the overall patient-level correlation between EFS and OS was calculated (Kendall's Tau rank correlation).

Results

Study population

A total of 858 eligible patients with NSCLC were identified in the structured dataset. After further inclusion/exclusion criteria were applied during the chart review, 425 patients were eligible for this analysis (overall population); among these patients, 278 and 147 were treated with chemoradiotherapy and chemotherapy, respectively (Figure 1).

Demographics & baseline characteristics

The median ages of patients treated with neoadjuvant chemotherapy and chemoradiotherapy were 67 years (range: 41, 88) and 66 years (range: 27, 89), respectively (Table 1). Most patients among both treatment cohorts had stage IIIA disease at index. Among the patients treated with chemotherapy, most patients had stage III disease (67.3%), followed by stage IIA (16.3%), stage IIB (15.0%) and stage IB (1.4%). Among the patients treated with chemoradiotherapy, most patients had stage III disease (65.5%), followed by stage IIB (21.9%), IIA (9.0%) and IB (3.6%). The proportion of patients with an ECOG PS score of 0 or 1 were numerically similar between patients treated with chemotherapy and chemoradiotherapy. The median (range) follow-up was 18.8 months (0.4, 135.6) among these patients. The most common comorbidities overall were hypertension (12.0%), ischemic disease (6.4%), pulmonary disease/asthma (6.4%) and diabetes (5.4%). More than 42% of the patients had at least 24 months of follow-up.

Treatment characteristics

Among the patients treated with neoadjuvant chemotherapy, the median (range) duration of treatment was 9.1 (0.1, 28.3) weeks or 3 (1, 15) cycles (Table 2). All except for one patient (gemcitabine) received platinum-based therapies; 53.7% received carboplatin-based and 45.6% received cisplatin-based. The most common neoadjuvant regimens were paclitaxel + carboplatin (34.7%), cisplatin + pemetrexed (17.7%) and carboplatin + pemetrexed (8.8%).

Among the patients treated with neoadjuvant chemoradiotherapy, the median (range) duration of treatment was 6.3 (1.0, 32.9) weeks or 3 (1, 16) cycles (Table 2). All except for two patients (docetaxel and paclitaxel monotherapies) received platinum-based therapies; 59.0% received carboplatin-based and 40.3% received cisplatin-based. The most common neoadjuvant regimens were paclitaxel + carboplatin (56.8%), cisplatin + etoposide (35.9%), docetaxel + cisplatin (9.5%) and gemcitabine + cisplatin (9.5%). Nearly all (95%) patients received radiation therapy concurrent with neoadjuvant chemotherapy. The median (range) dose was 8.4 Gy (0.67, 71.20).

Demographics & clinical characteristics by surgery status

A total of 225 (52.9%) patients in the overall study population underwent a surgical resection. Patients who underwent surgical resection were younger, with a median (range) age of 64 years (27, 89) versus 68 years (41, 87). They were also less likely to be current or former smokers (64.0 vs 72.5%) or have an ECOG PS score of 1 or 2 (60.8 vs 72.5%). Patients who underwent surgical resection included a higher proportion who had nonsquamous histology (57.3 vs 46.5%).

Among the patients who did not undergo surgical resection (n = 200), the most common reasons (among ≥5% of those patients) were physician recommendation (n = 72, 36.0%), disease progression (n = 55, 27.5%), patient preference (n = 14, 7.0%) and death (n = 14, 7.0%).

Pathologic complete response

Of the 425 patients in the overall population, 48 (11.3%) patients achieved a pCR. The pCR rates among patients initiated on neoadjuvant chemotherapy and chemoradiotherapy were 4.8 and 14.7%, respectively. In the overall population, 184 patients (43.3% of overall) died during the study observation period: 16 (33.3%) and 168 (44.6%) among patients with and without pCR, respectively. The median OS (95% CI) among patients with a pCR was 92.9 months (60.5–125.1) compared with 41.0 months (30.6–53.6) among patients without a pCR (Figure 2A). A total of 264 (62.1%) patients had an EFS event (progression/death) during the study observation period with 22 (45.8%) and 242 (64.2%) among patients with and without pCR, respectively. The median (95% CI) EFS among patients with a pCR was 64.5 months (25.6-NR) compared with 13.0 months (10.8–15.2) among patients without a pCR (Figure 2B). Compared with no pCR, pCR was shown to be associated with better OS and EFS by univariate and multivariable Cox analysis (Table 3). In subgroup analyses, pCR was associated with OS in patients with stage III (HR = 0.46, 95% CI: 0.22–0.95 in patients with pCR vs not) and with EFS in patients with stages I/II and III (HR = 0.43, 95% CI: 0.24–0.78 in patients with stage III, HR = 0.51, 95% CI: 0.28–0.94 in patients with stages I/II). However, in patients treated with chemotherapy, the risk of an event was numerically lower in patients who achieved a pCR versus those who did not, but this was not statistically significant (HR = 0.22, 95% CI: 0.03–1.59 for OS, HR = 0.26, 95% CI: 0.06–1.05 for EFS; data not shown). In a sensitivity analysis of only patients who underwent resection, the association between EFS and pCR remained significant (HR = 0.61; 95% CI: 0.38–0.98) in the multivariable analysis, but the association between OS and pCR was not as strong in the univariate analysis (HR = 0.81; 95% CI: 0.47–1.41).

Major pathologic response

Of the 425 patients in the overall population, 80 (18.8%) patients achieved an MPR. The MPR rates among patients initiated on neoadjuvant chemotherapy and chemoradiotherapy were 8.8 and 24.1%, respectively. The median OS (95% CI) among patients with an MPR (total n = 80; 25 events, 31.3%) was 92.9 months (59.0–125.1) compared with 37.3 months (28.2–52.0) among patients without an MPR (total n = 345; 159 events, 46.1%; Figure 2C). The median EFS (95% CI) among patients with a MPR (total n = 80; 39 events, 48.8%) was 37.8 (25.2–NR) months compared with 12.9 (9.9–14.4) months among patients without MPR (total n = 345; 225 events, 65.2%; Figure 2D). MPR was shown to be associated with better OS and EFS by univariate and multivariable Cox analysis (Table 4). Among patients treated with chemoradiotherapy compared with chemotherapy as the reference, MPR was also associated with better OS (HR = 0.48; 95% CI 0.30–0.77; p = 0.0024) and EFS (HR = 0.506; 95% CI 0.347–0.738; p = 0.0004). Among patients who had undergone resection, in a univariate Cox analysis, MPR did not have a significant association with OS (HR = 0.84; 95% CI: 0.52–1.37; p = 0.4908) or with EFS (HR = 0.76; 95% CI: 0.51–1.11; p = 0.1527). In subgroup analyses, MPR in stages I/II and III was associated with OS (HR = 0.42, 95% CI: 0.21–0.86 and HR = 0.52, 95% CI: 0.30–0.89, respectively) and EFS (HR = 0.46, 95% CI: 0.27–0.80 and HR = 0.48, 95% CI: 0.31–0.74, respectively), and although the risk for an event was numerically lower for patients treated with chemotherapy who achieved an MPR compared with patients who did not achieve MPR, it was only statistically significant for EFS (HR = 0.59, 95% CI: 0.22–1.63 for OS; HR = 0.37, 95% CI: 0.15–0.90 for EFS; data not shown).

EFS & OS correlation

Landmark analysis/time-dependent analysis

In landmark analyses at the different time points (6, 12 and 24 months), patients without an EFS event were shown to have significantly better OS than those with an EFS event (HR = 0.34; 95% CI: 0.23–0.50; HR = 0.24; 95% CI: 0.16–0.36; and HR = 0.29; 95% CI: 0.17–0.50, respectively; Figure 3A, B & C respectively). This was also evident in a Cox analysis where the event was considered as a time dependent variable (unadjusted HR = 0.51; 95% CI: 0.38–0.59; Supplementary Table 1).

Correlation analysis

The correlation between EFS and OS was tested using the Kendall's Tau rank correlation method. Considering all patients who initiated neoadjuvant therapy, the correlation between EFS and OS was 0.66 (p < 0.001) in all patients (n = 425) and 0.72 (p < 0.001) for patients who underwent a surgical resection (n = 225). As there was no established threshold for interpreting Kendall's Tau (that we are aware of), we noted that EFS was moderately correlated with OS according to Chan [26] (Supplementary Figure 1).

Discussion

In this real-world study of 425 patients with resectable NSCLC treated with neoadjuvant chemotherapy or chemoradiotherapy in USA, achieving pCR or MPR was associated with significantly longer survival. More specifically, among the overall population, patients who achieved a pCR or MPR had approximately half the mortality risk compared with those who did not achieve a pCR or MPR.

These results were in line with large meta-analyses that found an association between pathologic response and survival in patients receiving neoadjuvant chemotherapy or chemoradiotherapy, increasing the confidence of a true association between pCR and longer-term efficacy end points [19,20]. The association between pCR and EFS was also observed in the CheckMate 816 trial, a phase III clinical trial comparing neoadjuvant immunotherapy (nivolumab + chemotherapy vs chemotherapy alone) where patients who experienced a pCR had longer EFS compared with those without a pCR (HR = 0.13, 95% CI: 0.05–0.37) [27,28].

Our study populations were generally similar to the studies included in the meta-analysis (resectable NSCLC treated with neoadjuvant chemotherapy ± radiotherapy). However, caution should be exercised when comparing the two, given the differing data sources (real-world study vs literature review including various study designs such as cohorts, single-arm trials and randomized controlled trials [RCT]). It is encouraging to see similar trends of improved survival outcomes for the patients with NSCLC who achieved a pathologic response (relative to those who did not) after neoadjuvant treatment from a literature review, after neoadjuvant immunotherapy in a phase III RCT, and now in a real-world study. Although all patients had the intention to undergo surgery, approximately 50% of the patients who were initiated on neoadjuvant therapy did not undergo surgery in this real-world study. Physician recommendation, disease progression and patient preference were the most common reasons for not proceeding with surgery among these patients.

In this study, the landmark analysis considering EFS events at different landmark points showed a decreased risk for death among patients who did not have an EFS event. The landmark method, which avoids the immortal time bias in the estimation of the OS by re-baselining patients and accounting for outcomes at a landmark time point, has been proposed as an appropriate method to estimate the survival based on the responses at the different points in time [30–34]. When utilizing a time-dependent Cox analysis, we found that EFS was moderately correlated with OS (according to Chan) [26]. Given the attenuation bias because of censoring of survival, the results of the correlation may be underestimated due to lack of visibility of events occurring after the follow-up period. A meta-analysis of patients treated with neoadjuvant therapy for resectable NSCLC also found a correlation between EFS with OS [35].

Acknowledging the time bias carried by analyses of survival by response, in the context of surrogacy and association of end points, these summaries of EFS/OS by pCR/MPR status ensure that the full EFS/OS definition information (including events occurring before surgery) is used in the assessment, including early events. Due to the limited size of this study (N = 425), it is challenging to reliably explore the various clinical subgroups of interest, particularly in the context of a limited number of patients achieving pCR with chemotherapy/chemoradiotherapy. In the sensitivity analysis of only patients undergoing surgical resection, the magnitude of the association between pCR/MPR and EFS/OS was not as strong as in the overall population, though the HR point estimates for EFS and OS were both <1. These results should be interpreted with caution, as the sample size was small and selection of those patients was biased by events occurring after the index date.

Of note, the study identification period for this real-world study spans over 10 years, during which therapeutic options for management of advanced NSCLC have evolved significantly. The use of staging methods such as PET/CT scanning and endobronchial ultrasound biopsy might have changed [36,37]. Immunotherapy and targeted therapy have changed the landscape of lung cancer management, leading to improved survival rates. Further, data from phase III studies in populations with non-metastatic, resectable NSCLC are still emerging in the neoadjuvant and adjuvant settings [27,38]. The associations between early end points and survival in this study are based on chemotherapy- and chemoradiotherapy-based neoadjuvant treatments, which have shown only modest survival benefit versus no treatment in this setting (∼5% OS absolute risk reduction at 5 years) [10]. With newer advances such as immunotherapy, larger increases in pCR rates in RCTs are expected. When more trial data become available in patients treated with immunotherapy in the neoadjuvant setting, it will be useful to conduct a meta-analysis to show whether a trial-level association holds true as well in this patient population.

Strengths/limitations

This study has several strengths to be highlighted. The study investigated a large representative population of patients with NSCLC receiving care in the US community oncology setting. The iKM data include the documentation of diagnostic, clinical, treatment characteristics and outcomes data outside clinical trials. Therefore, the study could capture the real-world experiences of patients with cancer with up to 10 years of follow-up. The most widely used definition of pCR (0% viable tumor in the primary tumor and nodes) was used in this study to minimize definition heterogeneity in the primary analyses. Different networks may follow different treatment protocols. However, The US Oncology Network utilizes National Comprehensive Cancer Network guidelines, which is a common standard across the USA.

As with any database study, there are limitations to be considered when interpreting data analyzed from an EHR database. While the iKM database contains a wealth of information about community-based oncology, it may be limited in various aspects. First, the iKM system is used for clinical practice purposes, not for research purposes. As such, associations but not causality can be detected. Second, not all community oncology practices are included in the iKM dataset. Furthermore, not all The US Oncology Network clinics utilize the full EHR/iKM capabilities. Those practices that participate in the Network may be different from those that do not participate in some way, such as the patient population that is seen or the prescribing practices of the physicians. Third, the EHR data do not capture information about services and procedures provided outside of the practice if not documented in the EHR; therefore, results may not be generalized to the US population without further evaluation. Some variables of interest may not be as complete across the entire population.

Although surgery was planned in all patients receiving neoadjuvant treatment, approximately 50% of the patients did not eventually undergo surgery, thus reducing the number of patients with available data for study end points.

As these are real-world data, potential heterogeneity in pathologic definitions and analyses between pathologists and study sites may exist. Particularly for this study, it was important to standardize the definition of pCR to the highest degree possible across the study population. The study defines pCR as ypT0N0, which was the highest available standard based on the data collected, and included all treated patients with resectable intent, so to avoid biasing the study with events occurring after the index date. The pathologists who provide services to The US Oncology Network follow the College of American Pathologists guidelines for consistent identification and reporting of cancer, which acts as a standardization process. However, comprehensive data on residual tumor were not available for the lymph nodes and hence, the definition of MPR had to be modified from what was included in the study protocol. Also, while pathology reports were used to determine pathologic response rate, the scoring methods used for assessment were not provided.

Conclusion

In this real-world study among patients initiating chemotherapy or chemoradiotherapy as neoadjuvant therapy with an intention of surgery, achieving pCR or MPR was associated with significantly better OS and EFS compared with those patients without a pCR or MPR. In addition, patients with longer EFS had longer OS. These results suggest that pathologic response and EFS may have utility as candidate surrogate end points, which would have great impact on allowing clinicians to evaluate early response to a treatment and on shortening drug development timelines to get effective drugs to patients faster.