Dual-Attention BiLSTM for Interpretable Forecasting of Treatment Toxicities
Eric Ababio Anyimadu1, Xinhua Zhang2, Clifton David Fuller3
1University of Iowa, Iowa City, IA 52242 USA.
... IEEE-EMBS International Conference on Biomedical and Health Informatics. IEEE-EMBS International Conference on Biomedical and Health Informatics
|May 22, 2026
Summary
This study introduces an attention-enhanced Bidirectional Long Short-Term Memory (Bi-LSTM) model for forecasting cancer symptom trajectories using patient-reported outcomes (PROs). The novel approach improves prediction accuracy and clinical interpretability for personalized oncology care.
Area of Science:
- Oncology
- Artificial Intelligence
- Data Science
Background:
- Longitudinal patient-reported outcomes (PROs) are vital for tracking cancer symptom progression and treatment effectiveness.
- Deep learning models like Bidirectional Long Short-Term Memory (Bi-LSTM) can forecast symptom trajectories but often lack interpretability and fail to capture nuanced symptom-time dynamics.
- Existing models struggle with uniform input weighting, limiting their ability to highlight clinically significant symptom-time interactions.
Purpose of the Study:
- To develop an attention-enhanced Bi-LSTM model for improved forecasting of symptom severity in oncology.
- To enhance the interpretability of deep learning models applied to longitudinal PRO data.
- To identify informative symptom-time interactions for personalized cancer care.
Main Methods:
- Proposed an attention-enhanced Bi-LSTM architecture with dual attention mechanisms (item and temporal levels).
- Applied dual attention to selectively weigh informative symptom-time interactions, addressing uniform input weighting limitations.
- Evaluated the model on a longitudinal PRO dataset from a major cancer center.
Main Results:
- The attention-enhanced Bi-LSTM model demonstrated superior performance in predicting 12-month symptom severity compared to conventional Bi-LSTM.
- The model provided clinically meaningful insights into symptom evolution over time.
- Enhanced forecasting accuracy and interpretability were achieved through selective emphasis on key symptom-time interactions.
Conclusions:
- Attention-based temporal modeling offers a promising approach for enhancing predictive accuracy and interpretability in oncology PRO analysis.
- The developed model supports personalized and timely decision-making in cancer care by better understanding symptom trajectories.
- This work underscores the potential of advanced AI techniques to refine patient monitoring and treatment strategies in oncology.
Related Concept Videos
01:16Therapeutic Drug Monitoring: Overview and Classification
Therapeutic Drug Monitoring (TDM) is a clinical practice that measures specific drug levels in a patient's blood at designated intervals to ensure the drug concentration stays within a therapeutic range. This monitoring is crucial for optimizing individual dosage regimens, enhancing therapeutic efficacy, and minimizing drug-related toxicity. TDM is vital for drugs with narrow therapeutic windows, significant variability in pharmacokinetics, and a clear correlation between plasma levels and...
01:29Therapeutic Drug Monitoring: Affecting Factors
Therapeutic Drug Monitoring (TDM) is the clinical practice of measuring specific drug levels in a patient's blood or body tissues to manage and optimize therapy. TDM is crucial for drugs with narrow therapeutic windows, like warfarin and phenytoin, where incorrect doses can lead to treatment failure or severe side effects. This monitoring ensures the dosage administered is within a safe and effective range. The factors affecting therapeutic drug monitoring include:Patient-Specific Factors:a.
01:26Therapeutic Drug Monitoring: Drug Analysis Methods
Therapeutic Drug Monitoring (TDM) is a clinical practice that measures specific drug levels in a patient's blood or body tissues to tailor drug therapy effectively. This monitoring is critical for managing drugs with narrow therapeutic indices like digoxin and phenytoin, ensuring they are both safe and effective. For instance, monitoring theophylline levels in asthma patients involves precision and sensitivity to adjust doses according to individual responses to therapy, ensuring efficacy and...
01:11Blinding
Blinding is a commonly used method of not telling participants which treatment a subject is receiving. Blinding is a critical part of a randomized control trial or RCT. It reduces the bias that affects the results. In an RCT, blinding is used in the form of a placebo. A placebo effect occurs when untreated subjects falsely believe they have received the treatment and report improved symptoms. A placebo or a dummy treatment is administered to subjects to negate the bias caused by such an effect.
01:11Pharmacogenetics of Drug Targets: β₂-Adrenergic Receptors, Apo E, Thymidylate Synthase
Genetic polymorphisms in drug targets have emerged as critical determinants of interindividual variability in drug response and toxicity. Pharmacogenomic investigations increasingly focus on identifying these variations to personalize and optimize therapeutic interventions. A drug target may be a receptor, enzyme, or signaling protein involved in pharmacologic responses or disease-related pathways. While early pharmacogenetic studies focused primarily on drug metabolism, current research...
01:25Determination of Multiple Dosing Parameters: Loading and Maintenance Doses
A loading dose is an essential pharmacological strategy to rapidly achieve the target plasma drug concentration necessary for an immediate therapeutic effect. This approach is especially critical for drugs characterized by slow absorption or extended half-lives, where delaying therapeutic plasma levels could compromise treatment outcomes. By administering a loading dose, clinicians ensure a prompt onset of drug action, even for agents with complex pharmacokinetic profiles.Achieving steady-state...