Abstract Background: Co-mutations, PD-L1 and TILs are key NSCLC biomarkers. We applied deep learning to a multimodal patient cohort to identify prognostic patterns integrating morphology, mutations, and clinical features. Methods: 367 NSCLC patients from 18 Hellenic Cooperative Oncology Group-affiliated centers were retrospectively assessed for PD-L1 status (Dako 22C3 pharmDx), TILs (H&E slides), and somatic pathogenic variants with a 38-gene next-generation sequencing (NGS) panel. Whole slide images (WSI) were digitized by an optical microscope scanner. Ten pathology foundation models were benchmarked for predicting mutation, co-mutation, PD-L1 and TILs status. Mutated genes with >5% prevalence were considered for mutation and co-mutation endpoints (TP53, KRAS, STK11, PTEN, EGFR). A vision transformer model was trained on WSI features to predict endpoints and evaluate AUROC. Kaplan-Meier analysis assessed prognostic relevance of models and top feature tiles from model attention maps provided morphological explainability. The STAMP digital pathology pipeline supported feature extraction and model training. Results: Single mutation models yielded AUROC scores of 0.6-0.85, with STK11 prediction from HOptimus1 features highest. Co-mutation models produced AUROC scores of 0.69-0.77 with EGFR-TP53 prediction from Uni2 features the best. The KRAS-TP53 co-mutation model (AUROC 0.69, Uni2) showed significant separation in overall survival curves (p=0.05) between classes. Best-performing PD-L1 and TIL models also demonstrated significant survival separation (p=0.005 and p=0.05). Conclusion: Findings demonstrate the potential of pathology foundation models to derive complex clinically-relevant prognostic models for NSCLC with multimodal explainability. Citation Format: Sanddhya Jayabalan, Konstantinos Efthymiadis, Alexia Eliades, Kyriaki Papadopoulou, Abraham Pouliakis, Elena Fountzilas, Sofia Lampaki, Mattheos Bobos, Anna Goussia, Soultana Meditskou, Konstantinos Kyritsis, Helena Linardou, George Pentheroudakis, Dimitrios Bafaloukos, Dimitrios Pectasides, Epaminondas Samantas, Zunamys I. Carrero, George Fountzilas, Jakob N. Kather. Deep learning integration of molecular and histopathological data for prognostic stratification in non small cell lung cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1448.

Background The superiority of paclitaxel/ramucirumab over alternative therapeutic regimens in patients with gastric cancer has yet to be defined. Our aim was to evaluate whether second-line treatment with paclitaxel/ramucirumab is superior compared with other therapies. Patients and methods Retrospective real-world data from patients with advanced adenocarcinoma of the stomach, gastroesophageal junction, or distal esophagus, treated at Departments of Medical Oncology, affiliated with the Hellenic Cooperative Oncology Group (HeCOG), were collected. All patients had received at least 2 months of second-line treatment. The primary endpoint was progression-free survival 1 (PFS1). Results From March 2015 to March 2023, 179 patients received second-line treatment (median age 61.3 years). Of those, 77 (43%) received paclitaxel/ramucirumab, 21 (11.7%) irinotecan/5-fluorouracil (5-FU)/leucovorin, 16 (8.9%) docetaxel, and 65 (36.3%) other treatments. The efficacy of paclitaxel/ramucirumab was assessed by histological subtype: diffuse, intestinal, and mixed. For diffuse histology, the adjusted hazard ratio (aHR) for PFS1 was 1.03 [95% confidence interval (CI) 0.50-2.16] and for overall survival (OS) was 1.71 (95% CI 0.79-3.68). For intestinal histology, the aHR for PFS1 was 0.53 (95% CI 0.28-1.01) and for OS was 0.44 (95% CI 0.22-0.88), indicating a statistically significant OS benefit. Mixed histology showed no significant differences in PFS1 (aHR 1.00, 95% CI 0.23-4.37) or OS (aHR 1.10, 95% CI 0.32-3.82). Toxicity, dose reduction, and discontinuation rates were similar between paclitaxel/ramucirumab and other regimens. Conclusions Second-line treatment with paclitaxel/ramucirumab was independently associated with OS compared with other regimens in patients with advanced intestinal-type gastric cancer. Identification of the most effective treatment for advanced gastric cancer remains a challenge.

Background: Dose-dense sequential (dds) chemotherapy has changed the clinical outcome of patients with early breast cancer (BC). To investigate the impact of dose intensity (DI) in the adjuvant setting of BC, this observational trial (HE 10/10) was conducted assessing the long-term survival outcome, safety and toxicity of a currently widely used chemotherapeutic regimen. In addition, the prognostic significance of tumor infiltrating lymphocytes (TILs) and infiltrating CD8+ lymphocytes were also evaluated in the same cohort. Patients and methods: Totally, 1054 patients were prospectively enrolled in the current study with 1024 patients being eligible, while adequate tissue was available for 596 of them. TILs, CD8+ lymphocytes in intratumoral areas in contact with malignant cells (iCD8), CD8+ lymphocytes in tumor stroma (sCD8) as well as the total number of CD8+ lymphocytes within the tumor area (total CD8) were assessed by immunohistochemistry. Results: Within a median follow-up of 125.18 months, a total of 200 disease-free survival (DFS) events (19.5%) were reported. Importantly, the 10-year DFS and OS rates were 78.4% (95% CI 75.0–81.5) and 81.7% (95% CI 79.0–84.1), respectively. Interestingly, higher CD8+ T cells as well as TILs in the tumor microenvironment were associated with an improved long-term survival outcome. Conclusions: In conclusion, this study confirms the significance of dds adjuvant chemotherapeutic regimen in terms of long-term survival outcome, safety and toxicity as well as the prognostic significance of TILs and infiltrating CD8+ lymphocytes in BC patients with early-stage disease.

Supplementary Figure 6. Posttreatment CD163 (by QIF) increase in three patients with concurrent increase in the CD163 transcripts. Representative images from a patient’s tumor tissue sample showing pre and post treatment (Durvalumab-Olaparib Arm) CD163 expression. Nuclei (blue), Cytokeratin (green), CD163 (yellow) and CSF1R (red).

Purpose: We conducted a phase II randomized noncomparative window of opportunity (WOO) trial to evaluate the inhibition of cellular proliferation and the modulation of immune microenvironment after treatment with olaparib alone or in combination with cisplatin or durvalumab in patients with operable head and neck squamous cell carcinoma (HNSCC). Experimental Design: Forty-one patients with HNSCC were randomized to cisplatin plus olaparib (arm A), olaparib alone (arm B), no treatment (arm C) or durvalumab plus olaparib (arm D). The primary endpoint was to evaluate the percentage of patients in each arm that achieved a reduction of at least 25% in Ki67. Secondary endpoints included objective response rate (ORR), safety, and pathologic complete response (pCR) rate. Paired baseline and resection tumor biopsies and blood samples were evaluated for prespecified biomarkers. Results: A decrease in Ki67 of at least 25% was observed in 44.8% of treated patients, as measured by quantitative immunofluorescence. The ORR among treated patients was 12.1%. pCR was observed in 2 patients. Two serious adverse events occurred in 2 patients. Programmed death ligand 1 (PD-L1) levels [combined positive score (CPS)] were significantly higher after treatment in arms A and D. Expression of CD163 and colony-stimulating factor 1 receptor (CSF1R) genes, markers of M2 macrophages, increased significantly posttreatment whereas the expression of CD80, a marker of M1 macrophages, decreased. Conclusion: Preoperative olaparib with cisplatin or alone or with durvalumab was safe in the preoperative setting and led to decrease in Ki67 of at least 25% in 44.8% of treated patients. Olaparib-based treatment modulates the tumor microenvironment leading to upregulation of PD-L1 and induction of protumor features of macrophages. Significance: HNSCC is characterized by defective DNA repair pathways and immunosuppressive tumor microenvironment. PARP inhibitors, which promote DNA damage and “reset” the inflammatory tumor microenvironment, can establish an effective antitumor response. This phase II WOO trial in HNSCC demonstrated the immunomodulatory effects of PARP inhibitor–induced DNA damage. In this chemo-naïve population, PARP inhibitor–based treatment, reduced tumor cell proliferation and modulated tumor microenvironment. After olaparib upregulation of PD-L1 and macrophages, suggests that combinatorial treatment might be beneficial. Synopsis: Our WOO study demonstrates that preoperative olaparib results in a reduction in Ki67, upregulation of PD-L1 CPS, and induction of protumor features of macrophages in HNSCC.

Patient and tumor characteristics at baseline

Supplementary Figure 4. Relative fold change (2-ΔΔCq) of PD-L1 in respect to Β2Μ (reference gene) in the CTC fraction for individual samples of HNSCC patients before (blue) and after (orange) therapy.

Supplementary Figure 5. Map showing the distribution of mutations per gene per tumor. Light and dark purple, green and orange indicate pre- and post-treatment samples for patients treated with cisplatin and olaparib, olaparib and durvalumab and olaparib only, respectively. Blues correspond to samples before and after second biopsy/surgery for patients who did not receive treatment.

Supplementary Figure 2. Differentially expressed signatures in Responders relative to no-Responders. Response is based on physical examination, pathology or imaging. Responders had higher scores in Inflammatory Chemokines and Exhausted CD8 Signatures, as well as PD-1, Cytotoxicity and CD45. The significance (p-value, P and adjusted p-value, Padj) is represented relative to Fold Change (FC) in the x-axis.

Supplementary Figure 1. Differentially expressed genes (Left) and gene signatures (Right) Cisplatin-Olaparib (A, B), Olaparib (C,D) and Durvalumab-Olaparib (E, F) arms in pre- and post-treatment samples. The significance (p-value, P and adjusted p-value, Padj) is represented relative to Fold Change (FC) in the x-axis.

Background. Third-generation aromatase inhibitors (AIs) are the mainstay of treatment in hormone receptor (HR)-positive breast cancer. Even though it is considered to be a well-tolerated therapy, AI-induced musculoskeletal symptoms are common and may be accused for treatment discontinuation. Recently, selective cyclin-dependent kinase 4 and 6 (CDK4/6) inhibitors changed the therapeutic setting, and currently, ribociclib, palbociclib, and abemaciclib are all approved in combination with nonsteroidal AIs in patients with ER-positive, HER2-negative advanced or metastatic breast cancer. This systematic review aims to identify the frequency of aromatase inhibitor-associated musculoskeletal syndrome (AIMSS) in the adjuvant setting in patients under AI monotherapy compared to patients under combination therapy with AIs and CDK4/6 inhibitors and demonstrate the underlying mechanism of action. Methods. This study was performed in accordance with PRISMA guidelines. The literature search and data extraction from all randomized clinical trials (RCTs) were done by two independent investigators. Eligible articles were identified by a search of MEDLINE and ClinicalTrial.gov database concerning the period 2000/01/01–2021/05/01. Results. Arthralgia was reported in 13.2 to 68.7% of patients receiving AIs for early-stage breast cancer, while arthralgia induced by CDK4/6 inhibitors occurred in a much lower rate [20.5–41.2%]. Bone pain (5–28.7% vs. 2.2–17.2%), back pain (2–13.4% vs. 8–11.2%), and arthritis (3.6–33.6% vs. 0.32%) were reported less frequently in patients receiving the combination of CDK4/6 inhibitors with ET. Conclusions. CDK4/6 inhibitors might have a protective effect against joint inflammation and arthralgia occurrence. Further studies are warranted to investigate arthralgia incidence in this population.

The details for the assays and image analysis are summarized in the Supplementary Materials and Methods.

Differentially expressed genes (A) and gene signatures (B) in pre-olaparib–based treatment relative to post-olaparib–based treatment samples. The significance (P value, P and adjusted P value, Padj) is represented relative to fold change (FC) in the x-axis. Annotated markers are highlighted with orange (increased) or green (decreased). C, “All Signatures” heat map shows relatedness among signature scores for each sample. D, Subgroup analysis looking at the 4 patients with CR (1), PR (2), and pCR (1) revealed an increase in “Genes Within Tumor Inflammation Signature (TIS)” score posttreatement.

Supplementary Table 1a. Representativeness of Study Participants Supplementary Table 1b. Nanostring IO360 panel

Supplementary Table 5. Incidence of adverse events by maximum grade in the safety population.

A, Waterfall plot of Ki67 change (measured by QIF) compared with pretreatment values. B, Comparison of Ki67 distribution assessed by QIF pretreatment and posttreatment/second biopsy or surgery within study groups. Wilcoxon matched pairs signed-rank P values were calculated for each treatment group; Cisplatin-Olaparib (C-O), P = 0.266; Olaparib (O), P = 0.004; Control, P = 0.875; Durvalumab-Olaparib (D-O), P = 0.461.

Table 4a. Fold-change values for each signature between Response and No Response based on physical examination, pathology or imaging Table 4b. Fold-change values for significantly differentially expressed genes between Response and No Response based on physical examination, pathology or imaging

Supplementary Table 2. Signature scores with significanlty changes pre- and post-Olaparib based treatment