The Han Lab In the News

A Framework Combining LLMs and Rule-Based Harmonization for Smoking History Extraction from Clinical Notes

A new study from the Han Lab, published in npj Digital Medicine, presents a framework that integrates large language models (LLMs) with rule-based harmonization algorithms to extract and correct detailed smoking history variables from unstructured clinical notes — resolving temporal inconsistencies such as implausible transitions across longitudinal records. The framework was validated against manually curated records, achieving 96.6–97.5% accuracy across key smoking history fields in a cohort of 4,792 lung cancer patients across academic and community-based settings. This work enables scalable, automated surveillance for lung cancer screening eligibility and advances the application of AI-driven methods in population health research.

Smoking Duration–Based Screening Criteria Show Improved Performance and Equity Across Racial and Ethnic Groups

A new study led by Chloe Su, published in Annals of Internal Medicine, evaluates the eligibility and prognostic performance of smoking duration–based versus pack-year–based U.S. national lung cancer screening criteria across racial and ethnic groups. Using data from diverse cohorts, the study finds that smoking duration–based criteria improve screening eligibility and prognostic equity among historically underrepresented populations compared to current pack-year–based guidelines. This study reveals that smoking duration–based screening criteria identify high-risk individuals more equitably across racial and ethnic groups compared to traditional pack-year thresholds, with implications for reducing disparities in early detection and improving guideline design.

Stanford Cancer Institute's Cancer Data Science Core: Drs. Summer Han and Allison Kurian Highlighted for Clinical Decision Support Work

Neurosurgery featured the contributions of Drs. Summer Han and Allison Kurian to clinical decision support (CDS) systems at Stanford, recognizing the Cancer Data Science Core's interdisciplinary work at the intersection of data science, AI, and clinical care. The feature underscores how the team's methodological innovations are being translated into tools that support clinical decision-making across departments.

Lung Cancer Screening Guidelines Perpetuate Racial Disparities, Stanford-Led Study Finds

Current national guidelines that rely on age and smoking exposure to recommend people for lung cancer screening are disproportionally failing minority populations including African Americans, according to a new study led by researchers at Stanford Medicine. An alternative risk-based method that incorporates additional information including family history and other health problems such as previous cancer diagnoses does a better job of eliminating disparities among races, the study found.

Risk Model-Based Lung Cancer Screening More Cost-Effective Than USPSTF Recs

Risk model-based screening for lung cancer accounting for personal risk may be more cost-effective than age- and smoking history-based screening recommended by the U.S. Preventive Services Task Force (USPSTF), according to a cost-effectiveness analysis.