Belangrijkste concepten
A novel patient-side disease prediction model, PoMP, that leverages patient narratives including textual descriptions and demographic data to enable early disease detection and facilitate seamless healthcare access.
Samenvatting
The paper introduces Personalized Medical Disease Prediction (PoMP), a novel approach that predicts diseases based solely on patient-provided health narratives, including textual descriptions and demographic information. This is in contrast to existing disease prediction methods that heavily rely on clinical data such as laboratory tests and medical imaging, which are typically only available after a patient consults a healthcare professional.
PoMP employs a two-tiered classification architecture. It first predicts the broad disease category and then narrows down to the specific disease within that category. This hierarchical approach leverages the inherent structure of disease classifications.
The authors collected a comprehensive dataset, Haodf, from a leading online doctor consultation platform in China. The dataset includes patient narratives across six prevalent disease categories with varying risk levels, as well as the corresponding diagnoses made by doctors.
Extensive experiments on the Haodf dataset demonstrate the effectiveness of PoMP. Compared to various pre-trained language models, PoMP achieves state-of-the-art performance in 6 out of 7 evaluation scenarios, with significant improvements in both category prediction and disease prediction. The authors also conduct an ablation study to highlight the importance of incorporating demographic information in addition to textual descriptions for accurate disease prediction.
PoMP represents a significant advancement in making disease prediction more accessible and tailored to patient needs, thereby enhancing the efficiency of healthcare communication. By empowering patients to gain a clearer understanding of their potential health conditions, PoMP can facilitate timely connections with appropriate medical specialists, reducing the time and effort spent navigating the healthcare system.
Statistieken
Patients with Coronary Heart Disease (CHD) have an average age of 60.5 years.
Patients with Lung Cancer have an average age of 67.2 years.
The Haodf dataset contains a total of 29,326 patient records across 6 disease categories, with an average of 481.4 tokens per patient narrative.
Citaten
"PoMP enables rapid comprehension of potential health conditions for individuals and seamless connections with doctors specializing in relevant medical disciplines."
"PoMP presents a promising approach and introduces the possibilities in patient-side disease prediction."