RecallRisk-BERT: A Multi-Task Framework for Post-Report Medical Device Recall Triage

2026-06-25 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Medical Devices & Health Technology · Depth: Advanced, quick

Summary

RecallRisk-BERT is a multi-task framework designed to automate post-report medical device recall triage, severity assessment, and root-cause interpretation. Addressing the challenge of increasing FDA recall records, this framework utilizes 54,165 FDA medical device recall records spanning from 2002 to October 2025. The system integrates PubMedBERT-based textual representations of recall narratives with embedding-based structured features like product code and medical specialty. RecallRisk-BERT simultaneously predicts recall severity (Class I/II/III) and one of nine consolidated root-cause categories. In single-task severity prediction, a LightGBM-based text-tabular configuration achieved an accuracy of 0.963, macro-F1 of 0.856, and ROC-AUC of 0.974. The multi-task RecallRisk-BERT significantly surpassed the single-task PubMedBERT baseline, demonstrating strong consistency between model-derived risk rankings and observed root-cause severity patterns (rho = 0.983, p = 1.936e-6).

Key takeaway

For Machine Learning Engineers developing regulatory compliance systems, you should consider multi-task text-tabular models for improved recall triage. This approach, exemplified by RecallRisk-BERT, significantly enhances the simultaneous prediction of recall severity and root-cause categories. You can leverage combined textual and structured data to achieve high accuracy and F1-scores, streamlining post-report analysis. Implement similar BERT-based architectures when joint classification of related attributes is critical for decision support.

Key insights

Multi-task text-tabular models effectively triage medical device recalls by jointly predicting severity and root-cause.

Principles

• Combining text and tabular data enhances recall prediction.
• Multi-task learning improves over single-task baselines.
• Automated triage supports regulatory decision-making.

Method

RecallRisk-BERT combines PubMedBERT for recall narratives with embeddings for structured features to simultaneously predict recall severity (3 classes) and root-cause (9 classes).

In practice

• Apply text-tabular models for regulatory compliance.
• Use multi-task learning for related classification tasks.
• Integrate BERT-based models for narrative analysis.

Topics

• Medical Device Recalls
• Multi-Task Learning
• Text-Tabular Models
• PubMedBERT
• Regulatory Compliance
• Root Cause Analysis

Best for: NLP Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• TRIAGE: Dialectical Reasoning for Explainable Risk Prediction on Irregularly Sampled Medical Time Series with LLMs — TRIAGE uses dialectical LLM reasoning to provide calibrated, continuous risk scores and interpretable rationales for medical time series.
• Predicting Post-Traumatic Epilepsy from Clinical Records using Large Language Model Embeddings — LLM embeddings from clinical records enhance early prediction of Post-Traumatic Epilepsy when fused with tabular data.
• Medication-Aware Financial Exploitation Detection for Alzheimer's Patients Using Edge-Aware Interaction Risk Modeling — Integrating medication adherence with financial transaction monitoring significantly enhances detection of financial exploitation in Alzheimer's patients.
• MADE: A Living Benchmark for Multi-Label Text Classification with Uncertainty Quantification of Medical Device Adverse Events — MADE is a living MLTC benchmark for medical device adverse events, emphasizing reliable uncertainty quantification.
• Text Knows What, Tables Know When: Clinical Timeline Reconstruction via Retrieval-Augmented Multimodal Alignment — Multimodal alignment of clinical text and EHR data significantly improves temporal precision in patient timeline reconstruction.
• Language Models as Interfaces, Not Oracles: A Hybrid LLM-ML System for Pediatric Appendicitis — Hybrid AI systems improve clinical decision support by using LLMs for feature extraction and validated ML models for reliable, auditable prediction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.