Navigating Diagnostic Challenges in Lymphoma: Patient Narrative Analysis Using Large Language Models
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Accurate diagnosis is a recognized challenge in lymphoma, yet little is known about how patients navigate their treatment journey when experiencing diagnostic error.
Methods
We retrospectively analysed patient narratives from house086.com, the largest Chinese online lymphoma forum (2011-2025). Posts related to potential diagnostic errors were retrieved and processed using an AI-based pipeline integrating HTML parsing, optical character recognition, and a large language model. Structured case profiles captured demographics, diagnostic history, barriers, facilitators, and patient-reported outcomes, with quality control applied. A blinded human review was conducted for preliminary validation. Logistic regression examined factors associated with favorable outcomes, defined as diagnostic delay of ≤ 2 months and absence of severe harms.
Results
Among 1656 unique cases, the median age was 47 years; 28.9% reported female and 32.8% male; 53% of posts were authored by family members. Psychological distress was reported in 88.5%, disease progression in 37.2%, and treatment-related harm in 16.6%. Unfavorable outcomes were strongly associated with clinician-related issues (68.5%; OR 0.26, 0.16-0.42), case complexity (67.4%; OR 0.30, 0.19-0.48), specialist access barriers (36.4%; OR 0.48, 0.30-0.76), and ≥3 diagnostic encounters (OR 0.44, 95% CI 0.19-1.00). Specialist input (60.0%; OR 2.02, 1.34-3.07) was the strongest facilitator, while patient self-advocacy (12.9%; OR 0.55, 0.33-0.93) and family support (12.0%; OR 0.54, 0.30-0.96) were more common in difficult cases. A patient journey map illustrated how barriers and facilitators interacted across care stages.
Conclusion
This large-scale patient narrative analysis highlights clinician-related issues, case complexity, and specialist access barriers as major contributors to patient-reported diagnostic challenges, while specialist input facilitates resolution. AI-driven text mining of online forums enables scalable, patient-centred insights into diagnostic pathways and points to opportunities for digital tools to improve overall patient experience.
