Adverse reaction attribution analysis method and system based on multi-modal constraint decoding

By employing a multimodal constraint decoding-based adverse reaction attribution analysis method, combined with large language models and knowledge graphs for causal inference, the problems of underreporting, lag, and false positives in adverse drug reaction monitoring have been solved, enabling accurate quantification and safety early warning of adverse drug reactions.

CN121938659BActive Publication Date: 2026-07-10HANGZHOU YI YAO INFORMATION TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-03-30
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

Existing technologies for adverse drug reaction monitoring suffer from problems such as underreporting, lag, subjective bias, high false positive rates, and uncontrollable model outputs, especially lacking effective methods for spatiotemporal causal association analysis of multi-source heterogeneous data.

Method used

By constructing a multimodal constraint decoding method for adverse reaction attribution analysis, we extract features using a pre-trained natural language processing model, combine a large language model with a dynamic knowledge graph for causal inference, apply decoding constraints, calculate quantitative causal association scores, and perform hybrid simulation verification by combining population pharmacokinetics and historical case retrieval.

Benefits of technology

It enables precise quantification and safety early warning of adverse drug reactions, reduces false positive rates, ensures that attribution results are consistent with pharmacological mechanisms, and provides reliable adjustment plans.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121938659B_ABST
    Figure CN121938659B_ABST
Patent Text Reader

Abstract

The application relates to an adverse reaction attribution analysis method and system based on multi-modal constraint decoding. The method fuses disease course semantics and physiological space-time characteristics by constructing a three-dimensional splicing vector containing text, numerical value and time offset characteristics. A logic bias parameter is used to constrain the output layer of a large language model, converting uncontrolled text generation into quantitative causal calculation based on probability expectation, thereby avoiding model hallucinations. Combined with hybrid simulation verification of population pharmacokinetics and historical case retrieval, the attribution result conforms to the pharmacological mechanism, and accurate quantification and safety warning of adverse reaction risk are realized.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Adverse drug reaction event identification method and system based on multivariate knowledge mixed retrieval enhancement

    CN120767012A

  • Machine learning model with constrained output token vocabulary

    US20250356125A1