Clinical Scorecard: Extraction of Geriatric Syndromes from Discharge Summaries: Development of a Novel Dataset, Annotation Framework, and Preliminary Insights

At a Glance

Key Highlights

• Development of a manually annotated corpus for geriatric syndromes.
• Evaluation of multiple pretrained transformer-based models for syndrome extraction.
• Best performance observed for frailty, falls, and delirium in model evaluations.
• Challenges identified in extracting low-frequency geriatric syndromes and context-specific labels.
• Document-level aggregation improved model performance by reducing local errors.

Guideline-Based Recommendations

Diagnosis

• Utilize NLP to identify and classify geriatric syndromes from discharge summaries.

Management

• Incorporate findings from NLP extraction into patient-oriented clinical research and service planning.

Monitoring & Follow-up

• Regularly assess the performance of NLP models in extracting geriatric syndromes.

Risks

• Be aware of the challenges posed by linguistic complexity and data sparsity in geriatric syndrome extraction.

Patient & Prescribing Data

Older adults with complex clinical conditions.

Understanding geriatric syndromes can inform discharge planning and long-term care needs.

Clinical Best Practices

• Implement detailed annotation frameworks for accurate syndrome identification.
• Consider context and historical mentions in clinical narratives for improved extraction accuracy.
• Utilize document-level aggregation to enhance model performance.

Related Resources & Content

• Source Article