This integrated analysis of a phase 1/2 study (NCT03046992) evaluated the efficacy and safety of lazertinib, a third-generation EGFR tyrosine kinase inhibitor (TKI), in patients with advanced EGFR T790M-positive NSCLC after previous EGFR TKI therapy. Adults with EGFR mutation-positive NSCLC that progressed after prior EGFR-directed TKIs received once daily oral lazertinib 240 mg continuously until disease progression. Prior TKIs to treat T790M-positive NSCLC were prohibited. Primary endpoints were safety and objective response rate (ORR). Secondary endpoints included progression-free survival, overall survival, and intracranial ORR. A total of 78 patients received lazertinib 240 mg at 17 centers in South Korea. Among patients with T790M-positive tumors at baseline (N = 76), one (1.3%) had a complete response and 41 (53.9%) had partial responses, giving an ORR of 55.3% (95% confidence interval [CI]: 44.1-66.4). Median progression-free survival was 11.1 months (95% CI: 5.5-16.4). Median overall survival was not reached (median follow-up = 22.0 mo). In patients with measurable intracranial lesions (n = 7), one (14.3%) had a complete intracranial response and five (71.4%) had partial responses, giving an intracranial ORR of 85.7% (95% CI: 59.8%-100.0%). The most common treatment-emergent adverse events were rash (37.2%), pruritus (34.6%), and paresthesia (33.3%); most were mild to moderate in severity. Serious drug-related adverse events occurred in three patients (gastritis, pneumonia, pneumonitis). The major mechanism of resistance was EGFR T790M loss. Lazertinib 240 mg/d has a manageable safety profile with durable antitumor efficacy, including brain metastases, in patients with advanced T790M-positive NSCLC after previous EGFR TKI therapy.

Head and neck squamous cell carcinoma (HNSCC) is a deadly disease in which precision medicine needs to be incorporated. We aimed to implement next-generation sequencing (NGS) in determining actionable targets to guide appropriate molecular targeted therapy in HNSCC patients. Ninety-three tumors and matched blood samples underwent targeted sequencing of 244 genes using the Illumina HiSeq 2500 platform with an average depth of coverage of greater than 1,000×. Clinicopathological data from patients were obtained from 17 centers in Korea, and were analyzed in correlation with NGS data. Ninety-two of the 93 tumors were amenable to data analysis. TP53 was the most common mutation, occurring in 47 (51%) patients, followed by CDKN2A (n=23, 25%), CCND1 (n=22, 24%), and PIK3CA (n=19, 21%). The total mutational burden was similar between human papillomavirus (HPV)-negative vs. positive tumors, although TP53, CDKN2A and CCND1 gene alterations occurred more frequently in HPV-negative tumors. HPV-positive tumors were significantly associated with immune signature-related genes compared to HPV-negative tumors. Mutations of NOTCH1 (p=0.027), CDKN2A (p < 0.001), and TP53 (p=0.038) were significantly associated with poorer overall survival. FAT1 mutations were highly enriched in cisplatin responders, and potentially targetable alterations such as PIK3CA E545K and CDKN2A R58X were noted in 14 patients (15%). We found several targetable genetic alterations, and our findings suggest that implementation of precision medicine in HNSCC is feasible. The predictive value of each targetable alteration should be assessed in a future umbrella trial using matched molecular targeted agents.

Accurate prediction of non-small cell lung cancer (NSCLC) prognosis after surgery remains challenging. The Cox proportional hazard (PH) model is widely used, however, there are some limitations associated with it. In this study, we developed novel neural network models called binned time survival analysis (DeepBTS) models using 30 clinico-pathological features of surgically resected NSCLC patients (training cohort, n = 1,022; external validation cohort, n = 298). We employed the root-mean-square error (in the supervised learning model, s- DeepBTS) or negative log-likelihood (in the semi-unsupervised learning model, su-DeepBTS) as the loss function. The su-DeepBTS algorithm achieved better performance (C-index = 0.7306; AUC = 0.7677) than the other models (Cox PH: C-index = 0.7048 and AUC = 0.7390; s-DeepBTS: C-index = 0.7126 and AUC = 0.7420). The top 14 features were selected using su-DeepBTS model as a selector and could distinguish the low- and high-risk groups in the training cohort (p = 1.86 × 10⁻¹¹) and validation cohort (p = 1.04 × 10⁻¹⁰). When trained with the optimal feature set for each model, the su-DeepBTS model could predict the prognoses of NSCLC better than the traditional model, especially in stage I patients. Follow-up studies using combined radiological, pathological imaging, and genomic data to enhance the performance of our model are ongoing.