Lung Cancer

Lung cancer is the single largest cause of cancer death in the United States and the cancer for which screening evidence is cleanest: low-dose CT screening of eligible smokers and former smokers reduces mortality by roughly 20% in the published trials. But determining who is eligible depends on accurate pack-year exposure history - and pack-year history is almost always buried in unstructured clinical narratives, not the EHR's structured smoking-status field. That gap between "evidence says screen eligible patients" and "the EHR can't reliably tell you who is eligible" is the problem Dr. Osterman has worked on for more than a decade. It was also the work that got him started.

SHAPES (2015-2017) and the YIA. Dr. Osterman built the Smoking History and Pack-Year Extraction System (SHAPES) during his medical-oncology fellowship and biomedical-informatics MS at Vanderbilt - a natural-language-processing pipeline that reads clinical notes and reconstructs granular tobacco exposure (pack-years, duration, quit dates) from free text. The early work was presented at international meetings starting in 2015 (Quantifying Tobacco Exposure Using Clinical Notes and Natural Language Processing, IASLC Targeted Therapies 2016) and earned him the Conquer Cancer Foundation Young Investigator Award in 2016. The first downstream application followed almost immediately: an EHR-wide gene-environment interaction study using SHAPES-extracted smoking data to evaluate lung-cancer risk (Osterman, Wei, Mize & Denny, ASCO 2016 abstract 1524).

Lung cancer screening implementation. Once an extraction pipeline exists, the question becomes whether it actually moves screening forward. A series of Vanderbilt collaborations with the lung-cancer-screening community examined that translation in practice. Two examples: Women screened for breast cancer are dying from lung cancer: An opportunity to improve lung cancer screening (Sandler, Haddad, Paulson, Osterman & Scott, Journal of Medical Screening, 2021) - mining the existing breast-cancer screening cohort for women who would also qualify for lung-cancer screening but weren't being offered it. And Identification and Characterization of Avoidable Hospital Admissions in Patients With Lung Cancer (Lander, Li, Huang, Cass, Iams et al., JNCCN, 2023) - on the care-delivery side, identifying admissions that could be prevented with better outpatient management.

SmokeBERT (2025) - the language-model successor. SHAPES was a careful hand-engineered NLP pipeline; ten years later the technique of choice is transformer-based language modeling applied to the same problem. Heng Tan, MD - a Hematology-Oncology fellow Dr. Osterman mentored - led the development of SmokeBERT, a BERT-based smoking-history extraction model that significantly outperforms the original SHAPES on reconstruction accuracy and generalization. The work was published as SmokeBERT and Beyond: Bridging Clinical Narratives and Structured Smoking Data To Improve Lung Cancer Screening (Tan & Osterman, JCO Clinical Cancer Informatics, 2025), with Dr. Osterman as senior author. The arc is exactly the one this site keeps coming back to: a fellow inheriting a problem that's been around for a decade, applying the current generation of tools to it, and pushing the field forward. See AI in oncology for the broader thread on LLMs in clinical contexts, and clinical informatics education for the trainee pipeline that produced the work.

Immune checkpoint inhibitor prediction in lung cancer. Lung cancer is the single largest clinical application of immune checkpoint inhibitors, and is therefore the highest-yield context for the GE HealthCare Digital Precision Oncology work on predicting ICI efficacy and toxicity from real-world data. The cohorts that powered the flagship Lippenszky et al. JCO CCI 2024 paper draw heavily on Vanderbilt lung-cancer patients, and the predictive framework is most clinically useful where it is most clinically needed: at the point of choosing whether to start an ICI in a patient with metastatic NSCLC.

Bedside research from active clinical practice. Dr. Osterman's lung-cancer clinical practice continually surfaces problems beyond screening that the same data infrastructure can address. A notable example: in collaboration with thoracic oncologist Dr. Wade Iams and the maternal-fetal-medicine team at Vanderbilt, Dr. Osterman co-authored Two Uncomplicated Pregnancies on Alectinib in a Woman With Metastatic ALK-Rearranged NSCLC (Weidenbaum, Cann, Osmundson, Iams & Osterman, JTO Clinical and Research Reports, 2022) - a clinical case report documenting a young woman with metastatic ALK-rearranged NSCLC who carried two pregnancies to term while continuing targeted therapy with alectinib. The paper is the kind of bedside-driven contribution that only happens when an active clinician is paying attention to atypical cases and willing to publish them. Adjacent NSCLC-care work continues through the Vanderbilt-Ingram Thoracic Oncology research community.

Clinical practice and appointments. Dr. Osterman maintains an active medical-oncology clinical practice at Vanderbilt-Ingram Cancer Center, with a clinical focus on lung cancer. New-patient appointments and second opinions can be requested through the Vanderbilt Health physician directory.

The through-line: lung cancer is where the data-infrastructure thesis got tested first. If clinical informatics can solve the smoking-history extraction problem cleanly enough to move population-level screening, it can solve almost any structured-data extraction problem in oncology. And the active clinical practice is what keeps the rest of the work honest. See also cancer data standards (mCODE) and clinical genomics in the EHR for the standards-and-substrate work that the lung cancer research connects to.