iFlytek Medical has developed understanding and reasoning capabilities of AI-driven medical records that enable comprehensive medical record comprehension, disease reasoning, intelligent coding, and the expansion of knowledge and rule databases for medical insurance, innovatively achieving AI-based integrated medical record intelligent risk control. The intelligent risk control system for medical insurance relies on iFlytek Medical's advanced AI technologies such as natural language understanding, medical knowledge graphs, deep reasoning of medical records, and big data disease profiling. Based on patient’s comprehensive medical records in the entire medical treatment process, expense settlement, hospital inventory, and other data, we have independently developed four core capabilities: AI + medical insurance compliance audit, AI + diagnosis and treatment rationality audit, AI + payment method supervision, and AI + fraud prevention and insurance supervision. We provide relevant organizations with proactive reminders, real-time warnings, post-event supervision, and analytical services. Our system actively identifies various instances of improper use of medical insurance funds, such as non-compliance with regulations, excessive treatment, low coding with high billing, and substitution of procedures. This helps reduce improper expenditures and significantly enhances the efficiency of fund utilization.

I. Medical Record Comprehension Capacity:

(1) Extraction of non-structured medical entities in medical records achieves an accuracy of 97.8%, covering 60 common medical entity types such as diseases, symptoms, signs, test results, body parts, and more.

(2) Extraction of entity relationships in non-structured medical documents achieves an accuracy of 97.69%, including 136 common entity relationships such as surgery-body parts, symptoms, and durations.

(3) Parsing at the segment level of non-structured medical documents to extract medical event segments achieves an accuracy of 96.7%, covering 97 types of key events such as symptom descriptions, treatment processes, examinations, physical examinations, surgical situations, and patient conditions.

II. Disease Reasoning Capacity:

(1) Consistency reasoning in the main diagnostic and treatment process achieves an accuracy of 97.82% in detecting errors.

(2) Predicting the reasonableness of other diagnoses based on medical records, test results, medical orders, and other information achieves an accuracy of 94.59% in detecting errors.

III. Intelligent Coding Capability:

Discrepancies between diagnostic/surgical operation names and national coding versions are detected with accuracies of 91.02% and 96.51%, respectively.

 (Quoted from internal test data)

IV. Expansion of Knowledge and Rule Databases:

(1) A disease knowledge base used for medical insurance audit is available, covering more than 1000 diseases and over 50 disease attributes.

(2) The rule database covers 30320 rules in 132 categories, including Western medicine, traditional Chinese patent medicines, Chinese herbal decoction pieces, medical services, medical consumables, diseases, surgeries, tests, examinations, etc.