Scholarlink

In personalized medicine, accurately modeling patient health trajectories is crucial due to inherent temporal dependencies in Electronic Health Records (EHRs), which often contain episodic and irregularly timed data. To address these challenges, we developed AMITA (Adaptive Multi-Way Interpretable Time-Aware LSTM), an enhanced LSTM model. AMITA incorporates frequency measurements and the most recent observations into its framework, improving the predictive modeling of patient illnesses. It also adapts LSTM cell states to manage irregular timings effectively by utilizing elapsed times and a frequency-based decay factor. This enables a deeper understanding of the impact of medical interventions over time, enhancing the model’s ability to capture the dynamics of health status. Our model’s effectiveness has been empirically proven through studies conducted on two real-world clinical datasets, showcasing its capability to handle the complexities and irregularities of EHR data while providing actionable insights for healthcare providers.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting Patient Health Outcomes with AMITA from Irregular Time Series

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mamadou Ben Hamidou Cissoko,
  • Vincent Castelain,
  • Nicolas Lachiche

摘要

In personalized medicine, accurately modeling patient health trajectories is crucial due to inherent temporal dependencies in Electronic Health Records (EHRs), which often contain episodic and irregularly timed data. To address these challenges, we developed AMITA (Adaptive Multi-Way Interpretable Time-Aware LSTM), an enhanced LSTM model. AMITA incorporates frequency measurements and the most recent observations into its framework, improving the predictive modeling of patient illnesses. It also adapts LSTM cell states to manage irregular timings effectively by utilizing elapsed times and a frequency-based decay factor. This enables a deeper understanding of the impact of medical interventions over time, enhancing the model’s ability to capture the dynamics of health status. Our model’s effectiveness has been empirically proven through studies conducted on two real-world clinical datasets, showcasing its capability to handle the complexities and irregularities of EHR data while providing actionable insights for healthcare providers.