Stephanie Ji Chen

All Publications

• Evaluation of Large Language Models for Structured Data Extraction From Interstitial Lung Disease Clinical Notes: Comparative Study. Journal of medical Internet research Chen, S. J., Maddali, M. V., Langlotz, C., Bluethgen, C., Chen, J., Raj, R. 2026; 28: e90547

  Abstract

  BACKGROUND: Most clinically relevant data are in unstructured clinical notes, which are verbose and imprecise, making structured data extraction a costly bottleneck for screening patients for studies or maintaining health care registries. This challenge is particularly pronounced in interstitial lung disease (ILD) and requires significant human effort to interpret notes and determine classification to create an ILD registry. Large language models (LLMs) have the potential to significantly reduce this cost and effort.OBJECTIVE: We aim to compare the performance of various LLMs for structured data extraction from unstructured ILD clinic notes. Our primary aim was to evaluate LLM extraction of binary structured data (yes/no answers) from clinical notes regarding key ILD clinical questions. A secondary analysis evaluated select LLMs for the extraction of multiclass data to determine ILD classification.METHODS: We used 12 different LLMs to extract binary answers to 10 ILD clinical questions from the most recent clinic notes of 100 ILD clinic patients. We additionally used 2 LLMs (gpt-oss-20b and gpt-oss-120b) to extract multiclass data regarding ILD classification. Prompts were created with the assistance of ChatGPT (OpenAI) and refined with an iterative approach by testing on a prompt engineering cohort of 10 ILD clinic patient notes. Ground truth was established by consensus among 3 ILD physicians. LLM performance was evaluated using accuracy, precision, recall, and F1-scores.RESULTS: LLMs processed each interface call of a clinical note-prompt combination in 1-2 seconds, with estimated costs ranging from less than US $0.001 to US $0.11 (or approximately US $0.05 to US $10.50 per clinical note accounting for 10 runs and 10 binary prompts) depending on the model. Out of the 12 LLMs assessed, 7 models (Claude 3.5 Sonnet [Anthropic], GPT-4o, gpt-oss-20b, gpt-oss-120b, o1, o1-mini, and o3-mini [OpenAI]) performed at human-level accuracy, similar to that of the 3 ILD clinicians (96.2%). A total of 5 LLMs performed significantly worse than humans (Holm-adjusted P≤.003 for all). gpt-oss-120b, o1, and o3-mini models achieved the highest F1-scores of all the evaluated LLMs. There was no significant difference in model accuracy among the top tier models (Claude 3.5 Sonnet, gpt-oss-20b, gpt-oss-120b, o1, o1-mini, and o3-mini), though GPT-4o achieved significantly lower accuracy than o1 (Bonferroni-adjusted P=.04). Multiclass data extraction using gpt-oss-120b and gpt-oss-20b demonstrated lower accuracy when compared to its corresponding binary data extraction (91.1% and 88.0%, respectively). There was no significant difference in accuracy between gpt-oss-120b and gpt-oss-20b for multiclass extraction.CONCLUSIONS: Multiple LLMs consistently achieved human-level accuracy in extracting structured binary data from ILD clinical notes, while being orders of magnitude faster and cheaper. Multiclass data extraction was possible but associated with a lower accuracy. LLMs are promising tools that can be used for clinical data extraction to improve clinical research efficiency.

  View details for DOI 10.2196/90547

  View details for PubMedID 42361337

• Macroscopic Pulmonary Fat Embolism Secondary to Intraosseous Line Placement: A Case Report. Case reports in critical care Chen, S. J., Sanchez, A. C., Hua, S. T., Murala, J. N., Kaushik, C. J., Banga, A., Dhillon, G. S. 2026; 2026: 7302079

  Abstract

  Intraosseous access is commonly used for vascular access in emergent settings. It is generally thought to be well tolerated with minimal complications. We report the first case of clinically significant macroscopic pulmonary fat embolism secondary to intraosseous access.A 67-year-old woman developed profound hypotension and severe biventricular dysfunction shortly after intraosseous access and resuscitation. She required emergent venoarterial extracorporeal membrane oxygenation and Impella for cardiopulmonary support. Computed tomography revealed a new macroscopic pulmonary fat embolism compared with prior imaging 1 day prior. Aspiration thrombectomy was successfully performed with significant improvement in hemodynamics. Unfortunately, she suffered an anoxic brain injury during resuscitation and was ultimately transitioned to comfort care.Although intraosseous access is often considered a safe procedure, this case highlights the need for awareness of this rare but serious and potentially lethal complication. Treatment of pulmonary fat embolism is often supportive; however, aspiration thrombectomy has a potential therapeutic role in macroscopic cases.

  View details for DOI 10.1155/crcc/7302079

  View details for PubMedID 42273236

  View details for PubMedCentralID PMC13247423

• Successful GH Treatment of Hepatopulmonary Syndrome in Panhypopituitarism-related Advanced Liver Disease. JCEM case reports Chen, S., Diaz-Lankenau, R., Kwong, A., Chang, J., McAvoy, J., Lai, Y. K. 2026; 4 (4): luaf307

  Abstract

  Hepatopulmonary syndrome (HPS) is a known pulmonary vascular complication of chronic liver disease. In rare circumstances, HPS has been described in the context of panhypopituitarism. An underlying mechanism of panhypopituitarism-related liver injury is thought to stem from GH deficiency, leading to steatohepatitis from augmented lipid deposition within hepatocytes. Although liver transplantation remains the definitive treatment for HPS, resolution of panhypopituitarism-related HPS following GH replacement therapy has been occasionally described. These successful cases uniformly showed hepatic steatosis on biopsy that resolved after GH replacement, suggesting GH may effectively reverse the pathological process before permanent damage occurs. We present the first reported case of panhypopituitarism-related HPS successfully treated with GH replacement in the presence of significant liver fibrosis without steatosis. This case highlights the sustained therapeutic efficacy of GH even in advanced liver disease and adds to the limited literature regarding successful treatment of HPS, especially in the context of panhypopituitarism, without liver transplantation.

  View details for DOI 10.1210/jcemcr/luaf307

  View details for PubMedID 41909159

  View details for PubMedCentralID PMC13019524

• Automated Artificial Intelligence Detection of Early or Under-diagnosed Interstitial Lung Disease by Computed Tomography in the COPDGene Trial. Respiratory medicine Chen, S. J., Kalra, A., Muelly, M., Reicher, J., Callahan, S., Scholand, M. B., Kulkarni, T. 2025: 108545

  Abstract

  Diagnostic delays are common in interstitial lung disease (ILD) and there is a need for improved detection methods for early clinical ILD detection. ScreenDx is an artificial intelligence tool that assesses computed tomography (CT) scans for interstitial lung findings compatible with ILD. We investigated the ability of ScreenDx to identify ILD cases in the COPDGene dataset that were initially undiagnosed to assess the tool's performance in detecting early or undiagnosed ILD.The COPDGene trial was a NIH registered study assessing genetic factors in COPD. ILD was an exclusion criterion, however some ILD patients were unintentionally included initially and subsequently re-labeled by investigators. These patients were selected as "positives" for the study. Additional COPD and control patients were randomly selected from the dataset as "negatives" for the study to achieve a target ILD prevalence of ∼1-2% for the cohort. ScreenDx is a deep learning model designed to detect features of ILD on CT. CT scans from the study cohort were processed through ScreenDx, testing for sensitivity and specificity.At the previously selected and optimized threshold, ScreenDx demonstrated a sensitivity of 84.8% (95th CI: 68.1 - 94.9%) and specificity of 98.0% (95th CI: 97.3 -98.5%) for automatically detecting ILD in the study cohort.ScreenDx successfully detected 84.8% of clinically significant ILD that were underdiagnosed and intended to be excluded from the COPDGene trial, while maintaining high specificity. It holds promise as an efficient method for identifying early or under-diagnosed ILD automatically.

  View details for DOI 10.1016/j.rmed.2025.108545

  View details for PubMedID 41314436