Overall Survival and Biomarker Analysis of Neoadjuvant Nivolumab Plus Chemotherapy in Operable Stage IIIA Non-Small-Cell Lung Cancer (NADIM phase II trial).

PURPOSE: Neoadjuvant chemotherapy plus nivolumab has been shown to be effective in resectable non-small-cell lung cancer (NSCLC) in the NADIM trial (ClinicalTrials.gov identifier: NCT03081689). The 3-year overall survival (OS) and circulating tumor DNA (ctDNA) analysis have not been reported.
METHODS: This was an open-label, multicenter, single-arm, phase II trial in which patients with stage IIIA NSCLC, who were deemed to be surgically resectable, were treated with neoadjuvant paclitaxel (200 mg/m2 once a day) and carboplatin (area under curve 6) plus nivolumab (360 mg) once on day 1 of each 21-day cycle, for three cycles, followed by adjuvant nivolumab monotherapy for 1 year (240 mg once every 2 weeks for 4 months, followed by 480 mg once every 4 weeks for 8 months). The 3-year OS and ctDNA analysis were secondary objectives of the trial.
RESULTS: OS at 36 months was 81.9% (95% CI, 66.8 to 90.6) in the intention-to-treat population, rising to 91.0% (95% CI, 74.2 to 97.0) in the per-protocol population. Neither tumor mutation burden nor programmed cell death ligand-1 staining was predictive of survival. Conversely, low pretreatment levels of ctDNA were significantly associated with improved progression-free survival and OS (hazard ratio [HR], 0.20; 95% CI, 0.06 to 0.63, and HR, 0.07; 95% CI, 0.01 to 0.39, respectively). Clinical responses according to RECIST v1.1 criteria did not predict survival outcomes. However, undetectable ctDNA levels after neoadjuvant treatment were significantly associated with progression-free survival and OS (HR, 0.26; 95% CI, 0.07 to 0.93, and HR, 0.04; 95% CI, 0.00 to 0.55, respectively). The C-index to predict OS for ctDNA levels after neoadjuvant treatment (0.82) was superior to that of RECIST criteria (0.72).
CONCLUSION: The efficacy of neoadjuvant chemotherapy plus nivolumab in resectable NSCLC is supported by 3-year OS. ctDNA levels were significantly associated with OS and outperformed radiologic assessments in the prediction of survival.

INTRODUCTION

Lung cancer is a devastating disease, being the leading cause of cancer deaths worldwide.[1] Nevertheless, immunotherapy-based treatments have dramatically improved outcomes and become established as a major modality for the treatment of metastatic non–small-cell lung cancer (NSCLC).[2-5] Yet, its role in earlier stages needs to be established. In this regard, we previously published the results from the primary analysis of the NADIM trial, in which patients with resectable stage IIIA NSCLC were treated with neoadjuvant nivolumab plus chemotherapy, which showed a progression-free survival (PFS) at 24 months of 77.1% in the intention-to-treat (ITT) population.[6] In addition, a pathologic complete response (pCR) rate of 63.4% was reported. These are unprecedented results that outperform outcomes with the standard-of-care preoperative chemotherapy.[7] Consequently, currently, there is intense research ongoing focused on the efficacy of chemoimmunotherapy in the neoadjuvant setting. Without long-term survival data available, methodologies for the early measurement of treatment efficacy are of particular interest. Scoring approaches for pathologic response assessments for neoadjuvant immunotherapy in NSCLC have been reported.[8] Specifically, Cottrell et al[9] proposed quantitative immune-related pathologic response criteria on the basis of the histologic features of the regression bed in the tumors. Similarly, Stein et al[10] have proposed a pan-tumor immune-related pathologic response score system. Yet, its capacity to predict long-term survival has not been established.

CONTEXT

Key Objective

Our objectives were to evaluate the long-term clinical benefit of neoadjuvant nivolumab plus chemotherapy in operable stage IIIA non–small-cell lung cancer (NSCLC) and to assess the utility of circulating tumor DNA (ctDNA) as an early surrogate end point for treatment efficacy.

Knowledge Generated

Overall survival at 36 months was 81.9% in the intention-to-treat population, rising to 91.0% in the per-protocol population. Additionally, we report for the first time a significant association between ctDNA levels after neoadjuvant chemoimmunotherapy and survival outcomes in operable NSCLC. Indeed, ctDNA outperformed clinical responses, assessed on computed tomography scans and according to RECIST criteria v.1.1, in the prediction of survival.

Relevance

The efficacy of neoadjuvant chemotherapy plus nivolumab in resectable NSCLC is supported by unprecedentedly high survival rates. Overall survival was almost three times that reported in the historical series. Our data support the usefulness of ctDNA as an early surrogate end point in the context of neoadjuvant treatment.

Here, we report the results of the planned secondary end point of 3-year overall survival (OS) of the NADIM trial. Finally, we evaluate the prognostic value of the circulating tumor DNA (ctDNA) and compare its capacity to predict long-term survival with classical survival surrogates.

METHODS

Study Design and Participants

This is an open-label, multicenter, single-arm phase II trial. The trial was conducted in accordance with the precepts established in the Declaration of Helsinki, Good Clinical Practice guidelines, and all applicable regulatory requirements. The study Protocol (online only) was approved by the ethics committee of Hospital Puerta de Hierro and the Spanish Agency of Medicines and Medical Devices. Full details of the NADIM trial (ClinicalTrials.gov identifier: NCT03081689) have been published elsewhere.[6] Briefly, eligible patients included patients age ≥ 18 years, with operable stage IIIA NSCLC (American Joint Committee on Cancer seventh edition criteria) and an Eastern Cooperative Oncology Group performance status of 0 or 1. Patients were treated with neoadjuvant intravenous paclitaxel (200 mg/m2 once a day) and carboplatin (area under the curve 6; 6 mg/mL per min) plus nivolumab (360 mg) once on day 1 of each 21-day cycle, for three cycles before surgical resection, followed by adjuvant intravenous nivolumab monotherapy for 1 year (240 mg once every 2 weeks for 4 months, followed by 480 mg once every 4 weeks for 8 months). The primary end point was PFS at 24 months and it has been previously published.[6] Secondary end points included 3-year OS and the analysis of tissue and plasma biomarkers.

Peripheral blood and tissue from all patients were prospectively collected. Plasma samples were collected before and after neoadjuvant treatment. Post-treatment plasma samples were collected before surgery in all cases. Informed consent for the collection of research samples was obtained.

Procedures

Patients were assessed every 21 days for clinical response. Computed tomography (CT) scans were performed locally before and after neoadjuvant treatment, after surgery, every 3 months during the first-year follow-up, every 4 months during the second year of follow-up, and every 6 months thereafter. The tumor response to neoadjuvant treatment was evaluated by comparing before and after neoadjuvant treatment CT scans, and according to RECIST version 1.1.

The pathologic response was locally assessed in the pulmonary resection specimen (lobectomies, bilobectomies, or pneumonectomies) according to the pathologist of each of the 18 participating hospitals. A second evaluation was performed by two independent pathologists with 100% agreement. The number of sections reviewed for pathologic response assessment ranged from 8 to 28 (median 10, mean 12). In all cases, the pathologist was blinded to the patient's identity and outcome. pCR was defined as the absence of any viable tumor cell in the resected lung specimen and all regional lymph nodes. Major pathologic response was defined as the presence of 10% or fewer tumor cells in the primary tumor, and incomplete pathologic response was considered when there were 10% or more viable tumor cells present in the primary tumor.

Next-generation sequencing analysis of formalin-fixed paraffin-embedded and plasma samples is described in the Data Supplement (online only). Briefly, DNA from formalin-fixed paraffin-embedded samples was sequenced using the Oncomine Tumor Mutation Load Assay. Likewise, cfDNA, from plasma samples, was analyzed using the Oncomine Pan-Cancer Cell-Free Assay kit.

Mutant allele fraction (MAF) was defined as the number of mutant molecules at a specific nucleotide location relative to the sum of total DNA molecules (mutant plus wild-type). A cutoff of MAF ≥ 0.1% was established as the limit of detection.

Statistical Analysis

Median follow-up time was estimated by the reverse Kaplan-Meier (KM) method.[11] Estimation of the median follow-up and the ratio of the expected variance of S(t) to the current variance of S(t) at 36 and 42 months were used to quantify data maturity.[12]

OS was defined as the time from the start of neoadjuvant treatment to death from any cause. PFS was defined as the time between the start of neoadjuvant treatment and disease progression, as assessed by RECIST criteria v1.1, or death from any cause, whichever occurred first. Patients who were alive or without the event at the end of follow-up were censored at the time of the last contact. PFS and OS were assessed in the ITT population, which included all patients who received neoadjuvant treatment, and in the per-protocol (PP) population, which included all patients who underwent tumor resection and received at least one cycle of adjuvant therapy (Data Supplement). Cox proportional-hazards models were used to determine the association of each of the study variables with survival outcomes. The models were adjusted by surgery. Two patients died of COVID-19 disease, which represents a competing event for cause-specific mortality. Thus, competing risk analysis was also performed. Specifically, cumulative incidence functions and subhazard ratios on the basis of the Fine and Gray approach are presented to estimate the risks of progression and cancer-related death.

To avoid potential bias in the association of PFS and OS with response to treatment assessed by pathologic response, radiologic response, and ctDNA detection after neoadjuvant treatment, the landmark analysis approach[13] was used with the landmark chosen as the date of the end of neoadjuvant treatment.

The discrimination ability for each model was evaluated using Harrell's concordance index (C-index).[14,15] The C-index can take values from 0 to 1, with higher values indicating better discrimination. A value of 0.5 corresponds to no better discrimination than by chance. Likelihood ratio statistics of tumor response to treatment assessed by CT scans and ctDNA were also evaluated after accounting for surgery status. Models were first conditioned on one predictor, and then the significance of the other was tested. P values of < .05 were considered to be statistically significant.

Role of Funding Source

The study was sponsored by the Spanish Lung Cancer Group.[16] The study funders had no role in study design, data collection, data analysis, data interpretation, or writing of the manuscript.

RESULTS

Clinical Outcomes

The demographic characteristics of the patients at baseline have been reported previously.[6] All patients (N = 46) were stage IIIA. Regarding nodal status, nine (19.6%) patients were N0, three (6.5%) patients were N1, and 34 (73.9%) were N2. The median follow-up time was 38.0 months (95% CI, 36.7 to 40.7), with 94% maturity at 36 months and 90% maturity at 42 months. There were no events (death or disease progression) during neoadjuvant treatment. Among the ITT population (N = 46), 41 patients underwent tumor resection and 37 patients, constituting the PP population, received subsequent adjuvant therapy (90.2% of the planned population; Data Supplement). Of these, 29 (78.4%) patients completed the adjuvant treatment (14-17 cycles), eight (21.6%) patients received between three and 13 cycles of adjuvant therapy. There was no interruption of nivolumab administration in any of the 17 cycles. KM curve according to completion of adjuvant therapy is presented in the Data Supplement. Details of the patients who did not undergo surgery (n = 5) or did not receive adjuvant treatment (n = 4) are available in the Data Supplement.

At the time of data cutoff (March 2021), disease progression had been diagnosed in 12 patients and nine deaths had been recorded. Three of these deaths were of patients who did not undergo surgery and had disease progression, four were of patients who underwent surgery and had disease progression, and two were of patients diagnosed as being disease-free after surgery but who died of COVID-19 disease. The Data Supplement shows the cumulative incidence function curves for each cause of death. The median PFS and the median OS were not reached in the ITT or PP population (Fig 1). The median time to progression for patients who had progressive disease was 19.4 months (P25-P75: 10.6-25.1 months; Data Supplement).

FIG 1.

Kaplan-Meier curves for (A) PFS and (B) OS in the ITT population (N = 46). ITT, intention-to-treat; OS, overall survival; PFS, progression-free survival.

PFS at 36 and 42 months in the ITT population was 69.6% (95% CI, 54.1 to 80.7) in both cases. Similarly, PFS at 36 and 42 months in the PP population was 81.1% (95% CI, 64.4 to 90.5) in both cases. OS at 36 and 42 months in the ITT population was 81.9% (95% CI, 66.8 to 90.6) and 78.9% (95% CI, 63.1 to 88.6), respectively. Likewise, OS at 36 and 42 months in the PP population was 91.0% (95% CI, 74.2 to 97.0) and 87.3% (95% CI, 69.3 to 95.1), respectively.

In univariate Cox regression analyses, no statistically significant associations were noted between the baseline characteristics of the patients and PFS or OS, except for Eastern Cooperative Oncology Group performance status (0 v 1) and the tumor lesion size (maximum diameter), which were both associated with inferior OS (hazard ratio [HR], 4.91; 95% CI, 1.01 to 23.80; and HR, 1.03; 95% CI, 1.01 to 1.06, respectively; Data Supplement). Finally, patients who underwent surgery had significantly improved OS (HR, 0.14; 95% CI, 0.04 to 0.59; Data Supplement).

Treatment-related adverse events (AEs) during neoadjuvant treatment have been reported previously.[6] Of note, any of them were associated with surgery delays or deaths. There was no intraoperative or in-hospital mortality either at 30 or 90 days after surgery. AEs of grade 1 or 2 during adjuvant treatment were noted in 27 (73.0%) patients. The most common grade 1 or 2 AE was fatigue that was noted in 10 patients (27.0%; Data Supplement). AEs of grade 3 or 4 during adjuvant treatment were notified in five (13.5%) patients, being the most common toxicity increased lipase which was reported in four (10.8%) patients (Data Supplement). No long-term toxicities were noted.

Baseline Biomarkers

Of the 46 patients included in the trial, 35 (76.1%) had a biopsy sample available for next-generation sequencing analysis and 29 (63.0%) had valid data for tumor mutation burden (TMB) assessment. Similarly, programmed cell death ligand-1 (PD‐L1) data were available for 28 (60.9%) samples. In total, 43 pretreatment plasma samples were collected.

The expression of PD-L1 in tumor cells was not associated with improved PFS or OS (Data Supplement). Similarly, TMB assessment was not associated with survival outcomes (Data Supplement; Table 1).

TABLE 1.

HR and Corresponding 95% CI According to Each Biomarker (TMB, PD-L1, and ctDNA levels at baseline)

Baseline ctDNA was detected in 30 of 43 (69.8%) of the pretreatment plasma samples (Data Supplement). ctDNA levels at baseline were significantly associated with tumor size (maximum diameter; Data Supplement).

To explore the prognostic value of the amount of ctDNA at baseline, for each positive plasma sample, we calculated the sum of MAFs for all detected mutations. Different MAF thresholds were evaluated (Data Supplement), and 1% MAF was selected (Table 1). In the multivariate analysis, patients with low ctDNA levels (< 1% MAF), at baseline, had significantly improved PFS and OS than patients with high ctDNA levels (adjusted HR, 0.20; 95% CI, 0.06 to 0.63; P = .006; and adjusted HR, 0.07; 95% CI, 0.01 to 0.39; P = .002 for PFS and OS, respectively; Fig 2; Data Supplement).

FIG 2.

Kaplan-Meier curves for (A) PFS and (B) OS by ctDNA levels at baseline, using a cutoff of < 1% MAF. ctDNA, circulating tumor DNA; HR, hazard ratio; MAF, mutant allele fraction; OS, overall survival; PFS, progression-free survival; ref, reference category.

Tumor Response to Treatment Assessment: Comparative Analysis of Different Surrogates for the Prediction of Long-Term Survival

Tumor response to treatment was evaluated by CT scans in all patients (N = 46), the pathologic response was assessed in all patients who underwent surgery (n = 41), and a plasma sample collected after neoadjuvant treatment but before surgery was available in 40 cases (Data Supplement).

According to RECIST v1.1 criteria, two (4.3%) patients had a complete response, 33 (71.7%) had a partial response, and 11 (23.9%) showed stable disease. Regarding pathologic response, 34 (82.9%) patients had a major pathologic response, including 26 (63.4%) patients who showed pCR, and seven (17.1%) had an incomplete response.

Radiologic response according to CT scans did not show any association with PFS or OS (P = .698 for PFS and 0.848 for OS). Likewise, pCR was not significantly associated with survival (P = .111 for PFS and 0.102 for OS; Table 2). However, when treating COVID-19 deaths as competing risk events, pCR (but not radiologic response) identified patients with improved PFS (adjusted subHR, 0.23; 95% CI, 0.06 to 0.86; P = .030 for PFS and adjusted subHR: not estimable for OS because of lack of events). Of note, two of the 26 patients diagnosed as having pCR were deceased. Both patients died of COVID-19 disease and did not show disease progression according to CT scans during the study. ctDNA dynamics at the individual level and according to pathologic response are available in the Data Supplement.

TABLE 2.

Prognostic Value of Tumor Response to Treatment Assessments on the Basis of CT Scans, Pathologic Evaluation, and ctDNA (landmark analysis)

Improved PFS and OS were observed for patients with undetectable ctDNA (limit of detection established at 0.1% MAF) after neoadjuvant treatment (adjusted HR, 0.26; 95% CI, 0.07 to 0.93; P = .038; and HR, 0.04; 95% CI, 0.00 to 0.55; P = .015 for PFS and OS, respectively; Table 2; Fig 3; Data Supplement). As mentioned, 13 of 43 patients had undetectable ctDNA at baseline. Adjusted HRs remained significant when excluding patients who were ctDNA-negative at baseline (Data Supplement).

FIG 3.

Kaplan-Meier curves for (A) PFS and (B) OS according to ctDNA detection after neoadjuvant treatment and Kaplan-Meier curves for (C) PFS and (D) OS by clinical response assessed on CT scans (landmark approach). Among patients who had undetectable ctDNA after neoadjuvant treatment,[17] five were diagnosed as having progression disease. All of these patients (n = 5) underwent surgery. Regarding pathology assessments, two of them were diagnosed as having pCR, one as having major pathologic response and two were diagnosed as incomplete pathologic response. One of the patients showing undetectable ctDNA after treatment but incomplete pathologic response died, representing the unique death event among patients with undetectable ctDNA after treatment. Among patients with ctDNA detection after treatment (n = 13), two patients did not undergo surgery, three patients showed an incomplete pathologic response, one patient showed a major pathologic response, and seven patients had pCR. Of these, two patients showed progressive disease despite having pCR. CR, complete response; CT, computed tomography; ctDNA, circulating tumor DNA; HR, hazard ratio; OS, overall survival; pCR, pathologic complete response; PFS, progression-free survival; PR, partial response; ref, reference category; SD, stable disease.

To evaluate the ability of each survival surrogate to discriminate between deceased and nondeceased patients and between progressed and nonprogressed patients, we calculated Harrell's C-index. The adjusted C-index to predict OS of ctDNA (0.82) was higher than the C-index for the RECIST criteria (0.72; Table 2).

Finally, we investigated whether the prognostic information provided by radiologic responses can improve by adding ctDNA information. In our hands, ctDNA added a significant degree of prognostic information to the radiologic response in terms of OS (P = .003; Data Supplement).

DISCUSSION

The main objective of any neoadjuvant study should be to contribute to the cure of the patients and increase their OS.[18] Our study shows an OS of 81.9% at 3 years in the ITT population and 91.0% in the PP population. These are unprecedentedly high survival rates in patients with stage IIIA NSCLC and have not been reported in prior studies evaluating neoadjuvant approaches.[7,19] Importantly, data maturity was 94% at 36 months. Consistent with this, a clear plateau in the KM curves for OS and PFS was observed (Fig 1). Moreover, the median time to progression was 19.4 months in patients who showed progression disease, exceeding that of the overall follow-up from the previous series.[19] We previously reported a pCR rate of 63.4% and a major pathologic response rate of 82.9%.[6] Similarly, preliminary data from the CheckMate 816 randomized phase III trial showed that neoadjuvant nivolumab plus chemotherapy increased the pCR rate compared with chemotherapy alone (24.0% v 2.2%; odds ratio: 13.94; 99% CI, 3.49 to 55.75).[20] Nevertheless, pathologic responses have not always resulted in prolonged OS. In this way, despite neoadjuvant treatments with chemoradiotherapy demonstrating significant benefit in terms of pathologic response rates compared with chemotherapy alone, they did not have any impact on the PFS or OS.[21]

The marked difference between the current standard of care and the NADIM-based treatments is shifting our perspective on stage IIIA NSCLC from being a lethal disease to one where it may be considered potentially curable. Accordingly, there are a significant number of ongoing clinical trials addressing the role of chemoimmunotherapy in the neoadjuvant setting. It should be acknowledged that the development of novel neoadjuvant strategies for resectable NSCLC has been hampered by a lack of surrogate end points that can be measured much faster than the end points they are meant to predict.[22] Currently, we continue to lack surrogate end points for immunotherapy-based treatment efficacy that accurately predict long-term survival. Although major pathologic response has been proposed as a surrogate end point in neoadjuvant trials for resectable NSCLC,[23] the hitherto accepted definition of major pathologic response as ≤ 10% of residual viable tumor in NSCLC regardless of histologic subtype is under debate. Several alternative approaches have been proposed so far,[8-10,24] yet its capacity to predict long-term survival has not been reported. In our study, all patients diagnosed as having pCR were alive at data cutoff, except for two patients who died of COVID-19 disease. A recent study from the International Neoadjuvant Melanoma Consortium supports the role of pCR as an early surrogate end point for recurrence-free survival and OS.[25] In this regard, it appears that pCR is a distinct biological entity being associated with specific microenvironmental features.[26,27]

In our study, neither TMB nor PD-L1 staining predicted long-term survival. Similar results have been obtained in the metastatic setting where none of these biomarkers have proved to be predictive for chemoimmunotherapy.[28] Specifically, the KEYNOTE-189 trial demonstrated that the addition of pembrolizumab to chemotherapy, as a first-line treatment, significantly improved both PFS and OS in NSCLC patients with metastatic disease, regardless of status.[17] Similarly, Rothschild et al[29] reported that no significant association was found between PD-L1 expression and major pathologic response or nodal downstaging in patients with NSCLC treated with neoadjuvant chemotherapy followed by durvalumab. However, it should be acknowledged that in our study, the sample size for PD-L1 and TMB analyses was rather low, which makes any association difficult to demonstrate as being statistically significant.

In our study, patients with low ctDNA levels (MAF < 1%) at baseline had significantly improved PFS and OS than patients with high pretreatment ctDNA levels. Currently, there is not a standardized methodology to quantify ctDNA. We hypothesize that the sum of MAFs from all detected mutations would better recapitulate the status of disease as different tumor lesions may harbor different somatic mutations. Although this approach would be dependent on the total number of genes included in the panel as well as TMB, it is plausible to think that it may be less limited by tumor heterogeneity than measuring only the mutation at the highest MAF. Anyhow, similar results were obtained when using maximum MAF (Data Supplement). This approach has been used by other researchers.[30] Consistent with our findings, numerous reports have shown that the baseline ctDNA level is a prognostic factor in a wide range of patients with lung cancer.[31-33] Indeed, it has been proposed to incorporate ctDNA levels in a modified TNM staging system.[34]

Tumor response to treatment according to RECIST criteria was not associated with survival questioning the usefulness of radiologic response as a survival surrogate or even PFS as a trial end point when evaluating the efficacy of immunotherapy-based treatments. On the contrary, undetectable ctDNA at the end of neoadjuvant treatment clearly identified patients with improved OS. Although our analysis is exploratory, the notable effect size (HR, 0.04) prompts us to postulate ctDNA as being a pivotal surrogate for long-term OS in the immunotherapy field, with a similar prognostication capacity as pCR. Prior studies have noted an association between ctDNA dynamics and response to immunotherapy in NSCLC,[35] but the NADIM set is larger and the associations are more robust. Likewise, ctDNA clearance was associated with pCR in the CheckMate 816 trial.[20] In addition, it is well established that conventional imaging cannot always reliably predict long-term OS in patients undergoing immunotherapy and it has been shown that immunotherapy-based treatments can significantly improve OS rather than PFS.[2,36,37]

Longer follow-up has not revealed any signs of any unexplained or unexpected toxicities or deaths. There have been two deaths in the context of the COVID-19 pandemic, both involving patients without active tumor disease. This is an expected outcome, considering the high morbidity and mortality of this infection in patients with lung cancer.[38]

In our study, adjuvant nivolumab was administered for up to 12 months. It remains to be determined how much this treatment contributes to the OS. In this regard, adjuvant atezolizumab following lung resection and adjuvant chemotherapy has been shown to extend disease-free survival in patients with NSCLC,[39] although this has not been replicated in other tumors.[40]

In conclusion, here we report mature OS data, with more than 3 years of follow-up, in patients with resectable stage IIIA NSCLC treated with neoadjuvant chemoimmunotherapy. Survival time was almost three times that reported in the historical series, in which the 3-year OS did not exceed 30%.[19] Pretreatment ctDNA levels were significantly associated with survival but not classical biomarkers such as TMB or PD-L1 staining. Finally, undetectable ctDNA levels after neoadjuvant treatment outperformed radiologic responses assessed according to RECIST criteria v 1.1 in the prediction of OS.