{
  "abstract": "Background Smoking influences host biology from phenotype to molecular levels through complex, interdependent mechanisms. Tobacco smoke and inflammation-derived reactive oxygen species disrupt cellular homeostasis and increase genomic mutational burden. 1–3 Cancer patients with smoking-related comorbidities often experience poorer outcomes due to both systemic comorbidity and cellular-level damage.1 In the context of immune checkpoint inhibitor (ICI) therapy, smoking and smoking-associated mutational signatures have been shown to correlate with therapeutic response.4–6 However, many existing studies are limited by small-to-moderate cohort sizes and cancer-specific scopes. It remains unclear how these findings scale across larger populations and diverse cancer types. Notably, continued smoking after a cancer diagnosis has emerged as a potentially modifiable factor affecting treatment outcomes.7 8 Methods We queried TriNetX, a large, real-world, dataset of approximately 90 million electronic health records across the United States. We identified all patients with any cancer type treated with ICIs and stratified them by smoking status before ICI, as well as continued versus quit-smoking status afterward. Primary outcomes included one- and five-year overall survival post-ICI. Subgroup analyses were conducted by cancer type. We performed unsupervised phenome-wide clustering to examine high-dimensional relationships between smoking, cancer types, and patient phenotypes. Supervised machine learning algorithms, including multivariable logistic regression, support vector machines, and neural networks, were used to predict five-year overall survival, followed by feature importance analysis.Results Across all cancer types, smoking was associated with decreased five-year overall survival (hazard ratio: 1.15; 95% CI: 1.11–1.36) after matching. Smoking was also associated with worse outcomes in head and neck, skin, and breast cancers ( figure 1). Paradoxically, continued smoking after ICI initiation was associated with improved five-year survival in subsets of head and neck, lung, skin, and male genital cancers (figure 2). Machine learning models achieved AUROCs approaching 0.80, demonstrating strong predictive potential of EHR-based phenotyping for ICI outcomes.Conclusions Smoking is associated with decreased long-term survival in cancer patients receiving ICI therapy, particularly in epithelial cancers with direct exposure to smoke or its metabolites. However, unexpected survival benefits in continued smokers suggest underlying biological or behavioral confounders warranting further investigation. Smoking status is a critical feature of the patient phenome and contributes meaningfully to machine learning models predicting ICI outcomes.References Desrichard A, Kuo F, Chowell D, et al. Tobacco smoking-associated alterations in the immune microenvironment of squamous cell carcinomas. JNCI: Journal of the National Cancer Institute. 2018;110:1386–1392.Yang H, Ma W, Sun B, et al. Smoking signature is superior to programmed death-ligand 1 expression in predicting pathological response to neoadjuvant immunotherapy in lung cancer patients. Translational Lung Cancer Research. 2021;10:3807–3822.Andersson BA, Lofgren S, Lewin F, et al. Impact of cigarette smoking and head and neck squamous cell carcinoma on circulating inflammatory biomarkers. Oncology. 2020;98:42–47.Li M, Zhao L-Y. Smoking signature as a biomarker for immunotherapy. Translational Lung Cancer Research. 2022;11:122–123.Liu F, Han Z, Lu J, et al. Development and validation of a tobacco smoking-related index for predicting overall survival and immunotherapy response in bladder cancer. Environ Sci Pollut Res Int. 2023;30:68701–68715.Yin X, Wang H, Li R, et al. Tobacco exposure primes the secretion of CCL21 positively associated with tertiary lymphoid structure and response to immunotherapy. J Immunother Cancer. 2023;11.Rink M, Furberg H, Zabor EC, et al. Impact of smoking and smoking cessation on oncologic outcomes in primary non-muscle-invasive bladder cancer. European Urology. 2013;63:724–732.Hohl SD, Matulewicz RS, Salloum RG, et al. Integrating tobacco treatment into oncology care: reach and effectiveness of evidence-based tobacco treatment across national cancer institute-designated cancer centers. Journal of Clinical Oncology. 2022Abstract 1088 Figure 1Smokers versus non-smokersAbstract 1088 Figure 2Continued-smoking versus quit-smoking status",
  "authors": [
    {
      "affiliations": [
        "Thomas Jefferson University, Philadelphia, PA, USA"
      ],
      "name": "Teresa Duong"
    },
    {
      "affiliations": [
        "University of Iowa, Iowa City, IA, USA"
      ],
      "name": "Zachary Urdang"
    },
    {
      "affiliations": [
        "Thomas Jefferson University, Philadelphia, PA, USA"
      ],
      "name": "Pablo Llerena"
    },
    {
      "affiliations": [
        "Vanderbilt University Medical Center, Nashville, TN, USA"
      ],
      "name": "Ramez Philips"
    },
    {
      "affiliations": [
        "Thomas Jefferson University, Philadelphia, PA, USA"
      ],
      "name": "Christophe McNair"
    },
    {
      "affiliations": [
        "Thomas Jefferson University, Philadelphia, PA, USA"
      ],
      "name": "Jennifer Johnson"
    },
    {
      "affiliations": [
        "Thomas Jefferson University, Philadelphia, PA, USA"
      ],
      "name": "Ubaldo Martinez-Outschoorn"
    },
    {
      "affiliations": [
        "Thomas Jefferson University, Philadelphia, PA, USA"
      ],
      "name": "Joseph M Curry"
    }
  ],
  "title": "1088 Cancer immune checkpoint therapy efficacy and machine learning outcome predictions as stratified by smoking status and cancer type – a multi-national database study",
  "uid": "d332e85a-743e-5c9f-9535-c8e7b7ecd0b9"
}