Dissertation
Enhancing cancer symptom detection in electronic health records using artificial intelligence
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Spring 2025
DOI: 10.25820/etd.008020
Abstract
This dissertation investigates the application of artificial intelligence (AI), including machine learning (ML), deep learning (DL), and large language models (LLMs), to advance cancer symptom detection and prediction. By addressing the complexities of unstructured clinical notes, this research bridges gaps in existing methodologies and contributes to improving personalized symptom management. The study emphasizes dual-level analyses: document-level modeling to identify broad symptom patterns and token-level processing to perform detailed symptom profiling. A systematic review identifies prevailing trends, challenges, and opportunities in ML and DL-based symptom prediction. Building on these findings, the research develops Symptom-BERT, a domain-specific model fine-tuned on cancer-related EHRs. This model achieves superior internal validation (F1-score: 0.933, AUC: 0.929) and robust external validation with synthetic notes (F1-score: 0.831, AUC: 0.834). Comparative analysis of fine-tuning and prompt-based learning approaches highlights trade-offs in computational efficiency, precision, and adaptability. This demonstrates that fine-tuning excels in detecting physical symptoms with consistent documentation. In contrast, prompt learning offers flexibility for broader applications. The dissertation underscores the role of AI in bridging structured and unstructured data, enhancing predictive accuracy, and supporting proactive healthcare interventions. Key contributions include improved psychological symptom detection, scalable AI frameworks for unstructured clinical text, and insights into multimodal data integration for future research. Despite challenges such as data bias, resource intensity, and interpretability, the findings advance symptom science and establish a foundation for integrating AI-driven methods into clinical workflows. These contributions offer significant implications for oncology and other medical domains, enabling more accurate, efficient, and patient-centered healthcare.
Details
- Title: Subtitle
- Enhancing cancer symptom detection in electronic health records using artificial intelligence
- Creators
- Nahid Zeinali
- Contributors
- Stephanie Gilbertson White (Advisor)Weiguo Fan (Committee Member)Alaa Albashayreh (Committee Member)Juan Pablo Hourcade (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Informatics
- Date degree season
- Spring 2025
- DOI
- 10.25820/etd.008020
- Publisher
- University of Iowa
- Number of pages
- viii, 89 pages
- Copyright
- Copyright 2025 Nahid Zeinali
- Grant note
- Disclosure: This work was supported by the Betty Irene Moore Fellowship for Nurse Leaders and Innovators; College of Nursing, University of Iowa, Center for Advancing Multimorbidity Science (CAMS) NINR (National Institute for Nursing Research) P20 1P20NR018081; Holden Comprehensive Cancer Center, University of Iowa, National Cancer Institute (NCI) P30 P30CA086862; Institute for Clinical and Translational Science, CTSA University of Iowa UL1TR002537; and Iowa Health Data Resource (IHDR) and the University of Iowa https://strategicplan.uiowa.edu/publicprivate-partnership-p3/p3-program-supportstrategicpriorities/p3-proposals-funded-fy-2022. This study utilized ChatGPT, a language model developed by OpenAI, for total transparency to create an external dataset that externally validates our findings.
- Language
- English
- Date submitted
- 03/05/2025
- Description illustrations
- Illustrations, tables, graphs, charts
- Description bibliographic
- Includes bibliographical references (pages 73-89).
- Public Abstract (ETD)
Cancer significantly impacts patients' physical and mental well-being, making accurate detection and monitoring of symptoms critical for improving quality of life and healthcare outcomes. This dissertation focuses on advancing the detection of cancer symptoms by harnessing cutting-edge AI technologies, including ML, DL, and LLMs. By leveraging these innovative methods, this research enhances the understanding and prediction of symptoms documented in electronic health records (EHRs), particularly in unstructured clinical notes, which often pose challenges due to their variability and complexity. The study introduces Symptom-BERT, a domain-specific AI model that detects cancer symptoms at broad (document-level) and granular (token-level) scales. It also explores fine-tuning and prompt-based learning approaches to adapt AI models to the nuanced language of clinical care. These methods significantly improve the detection of physical symptoms like pain and fatigue and psychological symptoms like anxiety and depression, bridging critical gaps in current healthcare technologies. This research integrates insights from structured data and unstructured text to establish a comprehensive framework for identifying symptoms that require personalized management. This integration can potentially reduce the burden on healthcare providers, streamline clinical workflows, and enable more personalized, proactive patient care. Moreover, the findings offer a roadmap for applying AI to address similar challenges in other medical domains, such as chronic diseases and mental health conditions. This dissertation represents a step forward in symptom science, paving the way for AI-driven healthcare solutions that transform clinical decision-making and enhance the lives of cancer patients.
- Academic Unit
- IDGP in Informatics
- Record Identifier
- 9984830729402771
Metrics
2 Record Views