Acquired resistance represents a bottleneck to epidermal growth factor receptor (EGFR) tyrosine kinase inhibitor (TKI) treatment in lung cancer. Our study aimed to explore the efficacy of antiangiogenic‐based therapy in osimertinib‐resistant NSCLC patients and assess the efficacy of anlotinib in vitro study.

Background Large cell neuroendocrine carcinoma (LCNEC) is a rare high-grade neuroendocrine carcinoma of the lung. Little is known about the differences between the pure and combined LCNEC subtypes, and thus we conducted this study to provide more comprehensive insight into LCNEC. Methods We reviewed 221 patients with pure LCNEC (P-LCNEC) and 120 patients with combined LCNEC (C-LCNEC) who underwent pulmonary surgery in our hospital to compare their clinical features, driven genes’ status ( EGFR/ALK/ROS1/KRAS/BRA F), and adjuvant chemotherapy regimens. Propensity score matching (PSM) was applied to reduce selection bias. Results The P-LCNEC group included a higher proportion of males and smokers than the C-LCNEC group. Furthermore, the C-LCNEC group had higher incidences of visceral pleural invasion (VPI), EGFR mutation and ALK rearrangement compared with the P-LCNEC group. Expression of neuroendocrine markers (CD56, CGA, and SYN) and recurrence patterns were not significantly different between the two groups. The P-LCNEC group had better disease-free survival (DFS) and overall survival (OS) compared with the C-LCNEC group (median DFS: 67.0 vs. 28.1 months, p = 0.021; median OS: 72.0 vs. 45.0 months, p = 0.001), which was further confirmed by the PSM method ( p = 0.004 and p < 0.001, respectively). Adjuvant chemotherapy was also an independent factor for DFS and OS. Subgroup analysis found that regardless of whether it was for the entire LCNEC group or the P- and C-LCNEC subtypes, the small cell lung cancer (SCLC) regimens presented with superior survival compared with the non-small cell lung cancer (NSCLC) regimens. Conclusion P-LCNEC was associated with more favorable prognosis compared with C-LCNEC. SCLC-based adjuvant chemotherapy was more appropriate for LCNEC patients than NSCLC-based regimens, regardless of whether they were the pure or combined LCNEC subtypes. C-LCNEC patients may be the potential beneficiary of targeted therapy.

AbstractBackgroundLocal consolidative therapy (LCT) has emerged as a treatment option in patients with oligometastatic non‐small cell lung cancer (NSCLC) undergoing chemotherapy or targeted therapy. However, the current literature lacks evidence as to whether LCT improves survival in NSCLC patients receiving immunotherapy. Our study aimed to assess whether LCT combined with pembrolizumab ± chemotherapy could improve the survival of patients with synchronous oligometastatic NSCLC.MethodsPatients with NSCLC, without EGFR or ALK genetic aberrations, who were treated with first‐line pembrolizumab ± chemotherapy, were included in the study. Survival analysis of the LCT and non‐LCT groups was compared.ResultsA total of 231 patients were included in the study. The median follow‐up time was 15.24 months. Median progression‐free survival (PFS) and overall survival (OS) of the entire cohort were 12.00 and 23.43 months, respectively. Of the 231 patients included, 76 patients received LCT combined with pembrolizumab ± chemotherapy (LCT group) while 155 patients received pembrolizumab ± chemotherapy alone (non‐LCT group). Of note, the PFS of the LCT and non‐LCT groups was 13.97 and 10.08 months (p = 0.016), respectively. The OS were 30.67 and 21.97 months (p = 0.011), respectively. The PFS and OS were significantly improved with LCT for patients with brain or lung metastases but not bone metastases. No significant increase in treatment‐related toxicity was observed in the LCT group.ConclusionsThe present study shows that LCT to metastatic sites is an option for consideration in patients with synchronous oligometastatic NSCLC during first‐line pembrolizumab treatment, with significantly improved PFS and OS compared with systemic treatment alone.

Abstract Background Epidermal growth factor receptor (EGFR) mutations were frequently found with concomitant genetic alterations in lung adenocarcinoma (LUAD). This study aimed to investigate the profile of concomitant alterations of EGFR‐mutant LUAD ≤3 cm in size and its prognostic effect on recurrence. Methods From January 2018 to December 2018, patients with resected LUAD ≤3 cm in size in Shanghai Chest Hospital were identified. All patients underwent capture‐based targeted next‐generation sequencing (NGS) with a panel of 68 lung cancer‐related genes and were found with EGFR mutation. Clinicopathological and molecular characteristics and recurrence‐free survival (RFS) were analyzed. Results A total of 637 patients were enrolled in this study. The top three frequent co‐mutational genes were TP53 (179 of 637, 28.1%), PIK3CA (27 of 637, 4.2%), and ATM (22 of 637, 3.5%). The most common amplified genes were EGFR (37 of 637, 5.8%), followed by CDK4 (37 of 637, 5.8%) and MYC (12 of 637, 2.0%). Only TP53 mutation and EGFR amplification were adverse prognostic factors for RFS (all p < 0.001) in univariate analysis. Multivariable analysis further demonstrated that TP53 mutation and EGFR amplification were independent risk factors for RFS [(hazard ratio (HR) 2.07, 95% confidence interval (CI) 1.07–4.00, p = 0.030; HR 3.09, 95% CI 1.49–6.40, p = 0.002, respectively]. Conclusions Concomitant TP53 mutation and EGFR amplification were poor prognostic factors for RFS in patients with EGFR‐mutant resected LUAD. Our findings provide valuable understanding of the impact of concurrent alterations and implication for better implementation of precision therapy for patients.

Objective Several trials have shown that pembrolizumab plus chemotherapy was more effective in patients with advanced non-small-cell lung cancer (NSCLC) than chemotherapy monotherapy. However, whether pembrolizumab plus chemotherapy is still a better choice for first-line treatment in elderly patients (≥75 years old) remain unknown. We retrospectively compared the efficacy and safety of these two treatments in elderly patients. Patients and Methods We collected data of 136 elderly patients with advanced NSCLC who were treated with pembrolizumab plus chemotherapy or chemotherapy monotherapy in our hospital from 2018 to 2020. We compared the progression-free survival (PFS) and overall survival (OS) of patients and analyzed which subgroups might benefit more significantly from pembrolizumab plus chemotherapy. Results In total population, pembrolizumab plus chemotherapy showed superior PFS and OS than chemotherapy monotherapy (PFS: 12.50 months vs. 5.30 months, P<0.001; OS: unreached vs. 21.27 months, P=0.037). Subgroup analysis showed patients with positive PD-L1 expression, stage IV, good performance score (ECOG-PS <2), fewer comorbidities (simplified comorbidity score <9) or female patients had demonstrated a more evident OS benefit in pembrolizumab plus chemotherapy. In terms of safety, the pembrolizumab plus chemotherapy group had higher treatment discontinuation (26% vs. 5%). Conclusions Elderly patients using pembrolizumab plus chemotherapy achieved longer PFS and OS, but were more likely to discontinue due to adverse effects, so disease stage, PD-L1 expression, ECOG-PS and comorbidities should be considered when selecting first-line treatment.

It is challenging for reinforcement learning (RL) algorithms to succeed in real-world applications. Take financial trading as an example, the market information is noisy yet imperfect and the macroeconomic regulation or other factors may shift between training and evaluation, thus it requires both generalization and high sample efficiency for resolving the task. However, directly applying typical RL algorithms can lead to poor performance in such scenarios. To derive a robust and applicable RL algorithm, in this work, we design a simple but effective method named Ensemble Proximal Policy Optimization (EPPO), which learns ensemble policies in an end-to-end manner. Notably, EPPO combines each policy and the policy ensemble organically and optimizes both simultaneously. In addition, EPPO adopts a diversity enhancement regularization over the policy space which helps to generalize to unseen states and promotes exploration. We theoretically prove that EPPO can increase exploration efficacy, and through comprehensive experimental evaluations on various tasks, we demonstrate that EPPO achieves higher efficiency and is robust for real-world applications compared with vanilla policy optimization algorithms and other ensemble methods. Code and supplemental materials are available at https://seqml.github.io/eppo.

Ensemble learning, which can consistently improve the prediction performance in supervised learning, has drawn increasing attentions in reinforcement learning (RL). However, most related works focus on adopting ensemble methods in environment dynamics modeling and value function approximation, which are more essentially supervised learning tasks of the RL regime. Moreover, considering the inevitable difference between RL and supervised learning, the conclusions or theories of the existing ensemble supervised learning cannot be directly adopted to policy learning in RL. Adapting ensemble method to policy learning has not been well studied and still remains an open problem. In this work, we propose to learn the ensemble policies under the same RL objective in an end-to-end manner, in which sub-policy training and policy ensemble are combined organically and optimized simultaneously. We further theoretically prove that ensemble policy learning can improve exploration efficacy through increasing entropy of action distribution. In addition, we incorporate a regularization of diversity enhancement over the policy space which retains the ability of the ensemble policy to generalize to unseen states. The experimental results on two complex grid-world environments and one real-world application demonstrate that our proposed method achieves significantly higher sample efficiency and better policy generalization performance.

BackgroundSeveral oncogenic drivers in non-small cell lung cancer (NSCLC) are considered actionable with available or promising targeted therapies. Although targetable drivers rarely overlap with each other, there were a minority of patients harboring co-occurring actionable oncogenic targets, whose clinical characteristics and prognosis are not yet clear.MethodsA total of 3,077 patients with NSCLC who underwent molecular analysis by NGS were included, and their demographic and clinical data were retrospectively collected.ResultsOur study found that the frequency of NSCLC patients harboring co-occurring potentially actionable alterations was approximately 1.5% (46/3077); after excluding patients with EGFR-undetermined mutations, the incidence was 1.3% (40/3077); 80% (37/46) harbored both EGFR mutations and other potentially actionable drivers such as MET amplification (21.6%; 8/37) and alterations in ERBB2 including mutations (27%; 10/37) and amplification (21.6%; 8/37); other combinations of potentially actionable drivers including alterations in ERBB2, KRAS, MET, ALK, and RET were also identified. Additionally, de novo MET/ERBB2 amplification in patients harboring EGFR-mutant NSCLC treated with first-generation EGFR tyrosine kinase inhibitors (TKIs) was associated with shorter PFS (p < 0.05). The efficacy of TKIs in NSCLC patients harboring other co-occurring potentially actionable drivers varied across different molecular subtypes.ConclusionsApproximately 1.5% of NSCLCs harbored co-occurring potentially actionable oncogenic drivers, commonly involving EGFR mutations. Co-occurring actionable targets may impact the efficacy of TKIs; therefore, future clinical trials in these patients should be anticipated to tailor the combination or sequential treatment strategies.

OBJECTIVES:Combined large cell neuroendocrine carcinoma (C-LCNEC) is pulmonary large cell neuroendocrine carcinoma (LCNEC) mixed with other components, such as adenocarcinoma (AD) and squamous cell carcinoma (SCC). This study aimed to describe the distinct features between C-LCNEC with different components and explore the treatment strategy.METHODS:We retrospectively collected data of 96 C-LCNEC patients who underwent surgical resection. Propensity score matching was used to balance baseline characteristics of LCNEC combined with AD (LCNEC/AD) and LCNEC combined with SCC (LCNEC/SCC).RESULTS:In our final cohort, 71 (74%) were LCNEC/AD, while 25 (26%) were LCNEC/SCC. LCNEC/AD was more likely to occur in female, younger adults, with visceral pleural invasion and with driver gene expression. However, there was no significant difference in disease-free survival and overall survival between the 2 groups (before matching: P = 0.79 and P = 0.85; after matching: P = 0.87 and P = 0.48), while adjuvant chemotherapy (P = 0.019 and P = 0.043) was an independent predictor. C-LCNEC patients of stage II or III receiving adjuvant chemotherapy had longer disease-free survival and overall survival (P = 0.054 and P = 0.025), and the benefit of etoposide-based chemotherapy was greater than the other regimens (P = 0.010 and P = 0.030). EGFR and ALK mutations were present in 28% (17/60) and 7% (4/60) of C-LCNEC patients, respectively, and they responded well to targeted therapy.CONCLUSIONS:LCNEC/AD was the most common type of C-LCNEC, and there were many differences between different combined components. Adjuvant chemotherapy, especially etoposide-based chemotherapy, was a beneficial option for resected C-LCNEC.Subj collection: 152.

Anlotinib is a novel multi-target tyrosine kinase inhibitor approved by NMPA (China National Medical Products Administration) and its anti-tumor vascular targets include VEGFR, PDGFR, c-Kit and FGFR. This phase II study aims to further evaluate the efficacy and safety of anlotinib plus etoposide and carboplatin in treatment-naïve patients with extensive-stage small cell lung cancer (ES-SCLC).

OBJECTIVES:More and more encouraging evidence revealed that immunotherapy could improve clinical outcomes in patients with previously treated non-small cell lung cancer (NSCLC) with epidermal growth factor receptor (EGFR) variations. However, immunotherapy is still a controversy for NSCLC patients with EGFR mutation.METHOD:In this retrospective analysis, we compared the clinical efficacy of pembrolizumab monotherapy (PM), pembrolizumab combined with chemotherapy (P+C) and pembrolizumab combined with anlotinib (P+A) in NSCLC patients with EGFR mutation who had failed on EGFR-TKI and platinum-based chemotherapy.RESULT:Eighty-six patients were included in this study. The overall median progression free survival (PFS) was 3.24 months. Multivariate analysis suggested that EGFR L858R and combined therapy were positive prognostic factors of PFS. The overall median OS was 12.28 months. Multivariate analysis found that high PD-L1 expression (≥50%) and combined therapy seemed to be positive prognostic factors of OS. Among the population, 32 patients received PM, 26 patients received P+C and 28 patients received P+A. Up to Jan 30, 2021, the median progression-free survival was 1.5 months in the PM group, 4.30 months in the P+C group and 3.24 months in the P+A group. The median OS were 7.41, 14.92 and 15.97 months, respectively. The ORR were 3.1%, 23.1% and 21.4%.CONCLUSION:The addition of chemotherapy or antiangiogenic therapy to pembrolizumab resulted in significantly longer PFS, OS and ORR than pembrolizumab alone in our study. EGFR L858R might be a positive prognostic factor of PFS and high PD-L1 expression might be a positive prognostic factor of OS.

OBJECTIVES:Pembrolizumab plus platinum-based chemotherapy and pembrolizumab monotherapy (PM) both become standard of care in patients with advanced non-small-cell lung cancer (NSCLC) and a programmed death ligand 1 (PD-L1) tumor proportion score (TPS) greater than 50%. This study aimed to figure out the better treatment choice.METHOD:In this retrospective analysis, we compared the clinical efficacy of PM and PC as first-line treatment in NSCLC patients with a PD-L1 ≥50% and negative for genomic alterations in the EGFR and ALK genes.RESULT:Among the population, 115 patients received PC, and 91 patients received PM. Up to Dec 30, 2020, median follow-up was 17.13 months. The median progression-free survival (PFS) rates of PC and PM were 12.37 and 9.60 months (HR: 0.44, p < 0.001), respectively. The median overall survival (OS) rates were NE and 28.91 months (HR: 0.40, p = 0.005), respectively. Subgroup analysis found that the PFS benefit of PC was evident in most subgroups excepting patients with brain metastasis. The 1-year overall survival rates of PC and PM were 89.3% and 76.1%, respectively. The ORR was 61.7 and 46.9% (p = 0.004), respectively.CONCLUSION:In patients with previously untreated, PD-L1 ≥50%, advanced NSCLC without EGFR or ALK mutations, the addition of pembrolizumab to standard platinum-based chemotherapy seems to be the preferred treatment, which needs to be validated by further prospective trials.

Offline reinforcement learning (RL) tasks require the agent to learn from a pre-collected dataset with no further interactions with the environment. Despite the potential to surpass the behavioral policies, RL-based methods are generally impractical due to the training instability and bootstrapping the extrapolation errors, which always require careful hyperparameter tuning via online evaluation. In contrast, offline imitation learning (IL) has no such issues since it learns the policy directly without estimating the value function by bootstrapping. However, IL is usually limited in the capability of the behavioral policy and tends to learn a mediocre behavior from the dataset collected by the mixture of policies. In this paper, we aim to take advantage of IL but mitigate such a drawback. Observing that behavior cloning is able to imitate neighboring policies with less data, we propose \textit{Curriculum Offline Imitation Learning (COIL)}, which utilizes an experience picking strategy for imitating from adaptive neighboring policies with a higher return, and improves the current policy along curriculum stages. On continuous control benchmarks, we compare COIL against both imitation-based and RL-based methods, showing that it not only avoids just learning a mediocre behavior on mixed datasets but is also even competitive with state-of-the-art offline RL methods.

This paper is concerned with robust learning to simulate (RL2S), a new problem of reinforcement learning (RL) that focuses on learning a high-fidelity environment model (i.e., simulator) for serving diverse downstream tasks. Different from the environment learning in model-based RL, where the learned dynamics model is only appropriate to provide simulated data for the specific policy, the goal of RL2S is to build a simulator that is of high fidelity when interacting with various policies. Thus the robustness (i.e., the ability to provide accurate simulations to various policies) of the simulator over diverse corner cases (policies) is the key challenge to address. Via formulating the policy-environment as a dual Markov decision process, we transform RL2S as a novel robust imitation learning problem and propose efficient algorithms to solve it. Experiments on continuous control scenarios demonstrate that the RL2S enabled methods outperform the others on learning high-fidelity simulators for evaluating, ranking and training various policies.

Offline Reinforcement Learning (RL) aims at learning effective policies by leveraging previously collected datasets without further exploration in environments. Model-based algorithms, which first learn a dynamics model using the offline dataset and then conservatively learn a policy under the model, have demonstrated great potential in offline RL. Previous model-based algorithms typically penalize the rewards with the uncertainty of the dynamics model, which, however, is not necessarily consistent with the model error. Inspired by the lower bound on the return in the real dynamics, in this paper we present a model-based alternative called DROP for offline RL. In particular, DROP estimates the density ratio between model-rollouts distribution and offline data distribution via the DICE framework [45], and then regularizes the modelpredicted rewards with the ratio for pessimistic policy learning. Extensive experiments show our DROP can achieve comparable or better performance compared to baselines on widely studied offline RL benchmarks.

Atezolizumab, an immune checkpoint inhibitor, has been approved for use in clinical practice in non-small cell lung cancer (NSCLC) patients, but potential biomarkers for response stratification still need further screening. In the present study, a total of 399 patients with high-quality ctDNA profiling results were included. The mutation status of ubiquitin-like conjugation (UBL) biological process genes (including ABL1 , APC , LRP6 , FUBP1 , KEAP1 , and TOP2A ) and clinical information were further integrated. The results suggested that the patients with the clinical characteristics of male or history of smoking had a higher frequency of UBL mutation positivity [UBL (+)]; the patients who were UBL (+) had shorter progression-free survival (PFS) (1.69 vs. 3.22 months, p = 0.0007) and overall survival (8.61 vs. 16.10 months, p < 0.0001) than those patients with UBL mutation negativity [UBL (–)]; and more promising predictive values were shown in the smoker subgroup and ≤ 3 metastasis subgroup. More interestingly, we found the predictor has more performance in TP53 -negative cohorts [training in an independent POPLAR and OAK cohorts ( n = 200), and validation in an independent MSKCC cohort ( n = 127)]. Overall, this study provides a predictor, UBL biological process gene mutation status, not only for identifying NSCLC patients who may respond to atezolizumab therapy but also for screening out the potential NSCLC responders who received other immune checkpoint inhibitors.

Computer games are an extremely popular but overlooked workload. Cloud-gaming has been one of the biggest buzzwords in the gaming industry throughout 2020. The rapid growth of the video gaming industry and the diverse set of popular video games available today raises increasing concern to properly understand its I/O characteristics to improve their performance and design better gaming servers and consoles. To the best of our knowledge, this is the first attempt to systematically measure, quantify, and characterize the organization of game data into files, back-end storage access patterns, and the performance of gaming workloads. We explore the I/O behavior of 14 recent and famous games, producing a series of observations coming from measurements done on a real setup.

Objectives Pulmonary large-cell neuroendocrine carcinoma (LCNEC) and small-cell lung cancer (SCLC) are both classified as pure and combined subtypes. Due to the low incidence and difficult diagnosis of combined LCNEC (C-LCNEC) and combined SCLC (C-SCLC), few studies have compared their clinical features and prognosis. Materials and Methods We compared the clinical features, mutation status of driver genes (EGFR, ALK, ROS1, KRAS, and BRAF), and prognosis between C-LCNEC and C-SCLC. Univariate and multivariate Cox regression analyses were applied for survival analysis. Results We included a total of 116 patients with C-LCNEC and 76 patients with C-SCLC in the present study. There were significant differences in distribution of smoking history, tumor location, pT stage, pN stage, pTNM stage, visceral pleural invasion (VPI), and combined components between C-LCNEC and C-SCLC (P<0.05 for all). C-SCLC was more advanced at diagnosis as compared to C-LCNEC. The incidence of EGFR mutations in C-LCNEC patients was higher than C-SCLC patients (25.7 vs. 5%, P=0.004). We found that tumor size, pN stage, peripheral CEA level, and adjuvant chemotherapy were independently prognostic factors for DFS and OS in C-LCNEC patients, while peripheral NSE level, pT stage, pN stage, VPI and adjuvant chemotherapy were independently associated with DFS and OS for C-SCLC patients (P<0.05 for all). Propensity score matching with adjustment for the confounders confirmed a more favorable DFS (P=0.032) and OS (P=0.019) in patients with C-LCNEC in comparison with C-SCLC patients upon survival analysis. Conclusions The mutation landscape of driver genes seemed to act in different way between C-SCLC and C-LCNEC, likely by which result in clinical phenotype difference as well as better outcome in C-LCNEC.