Abstract
BACKGROUND: Influenza-like illness (ILI) is a syndromic diagnosis characterized by symptoms such as fever, cough, and sore throat, and may be caused by various respiratory pathogens, including influenza viruses, adenoviruses, parainfluenza viruses, and respiratory syncytial viruses. In Traditional Chinese Medicine (TCM), ILI is treated using syndrome differentiation and individualized herbal prescriptions. However, current prescription recommendations often rely on expert experience, with limited systematic analysis of clinical patterns. This study applies data mining techniques to analyze ILI-related TCM prescriptions, aiming to describe diagnostic features, prescription patterns, and herb compatibility. METHODS: Electronic medical records from April to December 2023 were collected from a TCM clinic, comprising ILI cases with documented therapeutic outcomes. Unstructured TCM inquiry text was transformed into structured diagnostic features using a rule-based keyword extraction method, encompassing 22 categories and over 70 specific indicators. Prescription data were analyzed using the Apriori algorithm with permutation-based significance testing to identify statistically significant herb combinations. Complex network analysis was applied to visualize and examine the global structure of herb compatibility. RESULTS: A total of 457 electronic medical records of influenza-like illness (ILI) were analyzed, yielding 105 distinct herbal ingredients with detailed characterization of thermal properties and flavor profiles. Cold- and warm-natured herbs predominated, with sweet, bitter, and acrid flavors most frequently observed. Using a rule-based extraction method, 22 diagnostic feature categories encompassing over 70 specific indicators-such as Personal Information, Mental State, Facial complexion and tongue coating, Fever, Pain, Sweating Condition, Nasal Discharge, Cough, Bowel Movement, Urination Status, Thirst-were systematically identified from TCM inquiry texts. Association rule mining initially yielded 101 statistically significant rules, of which 77 persisted under a stricter criterion, highlighting the robustness of both core and extension patterns in the treatment of influenza-like illness. The core structure of Maxing Shigan Decoction was highly stable across different herb combinations. Several extension modules, including Jiegeng, Houpo, Zhishi, and Guizhi, frequently co-occurred with the core, supporting the flexible application of classical prescriptions. Permutation testing, Z-score estimation, and effect size analysis confirmed the statistical robustness of these associations, and complex network analysis revealed a highly connected core comprising key herbs, emphasizing the stability and modularity of traditional TCM prescriptions. CONCLUSION: This study systematically analyzed the diagnostic features and prescription patterns of ILI in TCM using data mining methods. The results identified Maxing Shigan Decoction as the core prescription, around which corresponding formulations and combinations were developed, quantitatively supporting classical prescription principles. Despite limitations in data volume, the findings demonstrate that integrating data-driven analysis with traditional knowledge can promote prescription standardization, enhance clinical decision-making, and contribute to the modernization of TCM within multidisciplinary healthcare contexts.