Identification of Cancer Recurrence through Thai-English Electronic Medical Records Utilizing Sentence Embeddings
Clinical Scorecard: Identification of Cancer Recurrence through Thai-English Electronic Medical Records Utilizing Sentence Embeddings
At a Glance
| Category | Detail |
|---|
| Condition | Cancer Recurrence Detection |
| Key Mechanisms | Sentence-bidirectional encoder representations from transformers (SBERT) models |
| Target Population | Patients with breast, colorectal, cervical, and head and neck cancers |
| Care Setting | Multicentre oncology hospitals in Thailand |
Key Highlights
- Developed and validated SBERT models for cancer recurrence detection in Thai-English EMRs
- MetBERT achieved highest AUPRC for locoregional versus no recurrence and locoregional versus distant recurrence
- Bilingual-SBERT demonstrated robust performance during external validation
- Low AUPRC values indicate extreme class imbalance in recurrence prevalence
- Models suitable for clinical integration as a screening tool for cancer registry workflows
Guideline-Based Recommendations
Diagnosis
- Utilize SBERT models for detecting cancer recurrence in EMRs
Management
- Implement bilingual-SBERT as a screening tool for prioritizing high-probability records
Monitoring & Follow-up
- Regularly validate model performance with external datasets
Risks
- Consider the impact of class imbalance on model performance
Patient & Prescribing Data
Patients with breast, colorectal, cervical, and head and neck cancers
Models can streamline the identification of recurrence, reducing manual workload for registrars
Clinical Best Practices
- Integrate sentence embedding frameworks into clinical workflows
- Ensure continuous external validation of models in diverse clinical settings
- Address data completeness and accuracy challenges in EMRs
Related Resources & Content
