Revolutionizing Health Record Analysis with Natural Language Processing: A Case Study on Indolent Systemic Mastocytosis
In a rapidly evolving healthcare landscape, traditional methods of manually reviewing health records are becoming obsolete. Enter large language models (LLMs) and natural language processing (NLP)—innovative tools that promise to transform how we analyze patient data, particularly in complex cases like indolent systemic mastocytosis (ISM).
The Challenge of Manual Data Review
The sheer volume of electronic medical records (EMRs) can be daunting. Dr. Thanai Pongdee, codirector of the Population Health Research Program at Mayo Clinic Health System, notes, “There are millions of clinical records that we have in our own institutional EMR. Manually mining those records takes an enormous amount of time.” Traditional reliance on International Classification of Diseases (ICD) codes can be limiting, often missing critical, nuanced patient data.
To bridge this information gap, the Mayo Clinic’s team turned to NLP—an automated method capable of efficiently sifting through vast datasets to identify pertinent patient characteristics without the laborious effort of manual searching.
A Deep Dive into the NLP Approach
In a retrospective cohort study, Dr. Pongdee and colleagues tapped into de-identified data from the Mayo Clinic EMR database, covering patients with ISM from January 1, 2005, to June 30, 2023. The study aimed to capture a holistic view of ISM, including patient demographics, symptom burden, and healthcare resource utilization.
The final dataset comprised 203 ISM patients, matched with 2030 control patients based on various characteristics such as age, sex, and comorbidities. The innovative use of NLP allowed researchers to accurately assess a plethora of unstructured clinical notes in conjunction with structured EMR data.
Comprehensive Patient Data Landscape
The analysis revealed key demographics: the mean age of ISM patients was 51.4 years, with a significant female predominance of 66.5%. Notably, the symptoms were varied and included allergic reactions, lymphadenopathy, and fatigue, with patients reporting an average of 10.6 distinct symptoms at baseline, which increased to 13.3 over time.
Data drawn from the study illustrates the extensive health challenges faced by these patients. For example, 66% underwent some form of biopsy, with skin and bone marrow biopsies being the most common. Furthermore, the study highlighted the importance of KIT testing, where 68% of patients were tested and 54% showed KIT detection.
The Power of NLP in Data Collection
The most striking aspect of this study was NLP’s ability to efficiently capture a wide array of data elements, far surpassing traditional methods. By intelligently parsing through clinical notes, the NLP model facilitated the creation of comprehensive clinical encounter summaries enriched with precise data labels.
This automated functionality not only saved time but also improved the accuracy of recorded patient information. For instance, patients with ISM had significantly higher utilization rates of healthcare resources compared to matched controls, indicating a broader impact on the healthcare system.
Insights on Comorbidities and Health Resource Utilization
The study’s findings also demonstrated a notable prevalence of various comorbidities among ISM patients, including osteoporosis, diabetes, and irritable bowel syndrome. The data indicated how these conditions amplified the challenges faced by ISM patients, making effective management even more critical.
Healthcare resource utilization was significantly higher in the ISM cohort, with a mean annual outpatient visit rate of 7.5 compared to 2.1 in the control group. This discrepancy underscores the necessity for effective, immediate interventions for managing ISM symptoms, which often require frequent healthcare visits and medications.
Transforming Clinical Practice with AI
Dr. Pongdee’s insights illuminate the potential of NLP technologies in modern clinical practice. “The model performed quite well. It was much more time-efficient to query all of the symptoms that it could see. It took a global view and evaluated all the symptoms reported in the EMR,” he emphasizes.
By automating the analysis of vast patient datasets, healthcare providers can focus on patient care rather than getting bogged down in paperwork. This shift not only enhances clinical workflows but also minimizes the chance of overlooking critical, patient-reported data.
References
- Pongdee T, Powell D, Weis T, et al. Assessing real-world natural history of indolent systemic mastocytosis: retrospective matched cohort study from Mayo Clinic Electronic Health Records. J Allergy Clin Immunol. 2025;155(suppl 2):AB312. doi:10.1016/j.jaci.2024.12.960
- Clay B, Bergman HI, Salim S, Pergola G, Shalhoub J, Davies AH. Natural language processing techniques applied to the electronic health record in clinical research and practice – an introduction to methodologies. Comput Biol Med. 2025:188:109808. doi:10.1016/j.compbiomed.2025.109808
This exploration of NLP in the context of ISM emphasizes the critical intersection between technology and healthcare, promising a future where patient care is streamlined, efficient, and more insightful than ever.