Researchers have demonstrated a breakthrough in fibromyalgia diagnosis using artificial intelligence, showing that large language models can effectively distinguish the condition from other chronic pain disorders through analysis of patients' pain descriptions.
The proof-of-concept study, conducted by researchers at the University of Bari "Aldo Moro" in Italy, achieved an impressive 87% accuracy in differentiating fibromyalgia cases from other chronic pain conditions using advanced language analysis techniques.
AI Model Performance and Methodology
The research team, led by Vincenzo Venerito and Florenzo Iannon, employed the Mistral-7B-Instruct-v0.2 language model to analyze transcribed patient responses about pain and sleep patterns. The study included 80 patients - 40 with confirmed fibromyalgia according to the 2016 American College of Rheumatology Criteria and 40 with other chronic pain conditions.
Using prompt engineering specifically designed to detect fibromyalgia-associated language nuances, the system achieved:
- 87% accuracy
- 92% precision
- 84% recall
- 82% specificity
- 0.86 area under the receiver operating characteristic curve (AUROC)
These results significantly outperformed the non-prompted (ablated) approach, which showed lower performance across all metrics (P <.001).
Key Language Patterns and Misclassification Analysis
The AI system's attention weight analysis revealed particular sensitivity to specific words associated with fibromyalgia, including:
- Terms describing widespread pain ("everywhere")
- Distinctive pain patterns ("leopard-spot" pain)
- Associated symptoms ("exhaust," "depressed")
- Characteristic pain qualities ("electric," "burning")
The system did face some challenges, misclassifying 10 patients (25%) with various conditions, including axial spondyloarthritis, subacromial bursitis, and psoriatic arthritis. Notably, subacromial bursitis due to calcifying tendinopathy showed a higher likelihood of misclassification (OR 29.57, 95% CI 2.70-323.69).
Clinical Implications and Future Directions
"By dissecting and interpreting subjective information, such as patient-generated data from social media platforms, online forums and electronic health records, sentiment analysis can provide a unique lens through which to examine the complexity of emotions in FM," explained the investigators.
The researchers emphasize that while these results are promising, the technology is intended to complement rather than replace clinical assessment. They recommend further validation in larger prospective cohorts and suggest that optimizing model interpretability and integration with patient-reported outcomes could enhance the clinical impact of this diagnostic tool.
This development comes at a time when technology-based interventions for fibromyalgia are showing increasing promise, including recent success with digital acceptance and commitment therapy in the PROSPER-FM trial, highlighting a growing trend toward digital solutions in both diagnosis and management of fibromyalgia.
