A phrase-based questionnaire-answering approach for automatic initial frailty assessment based on clinical notes

Area of Science:

• Gerontology
• Medical Informatics
• Natural Language Processing

Background:

• Frailty is a significant geriatric syndrome impacting elderly quality of life and increasing mortality risk.
• Current frailty assessments often require in-person consultations, posing challenges for rural or immunocompromised populations.
• Automated assessment can improve early diagnosis and resource allocation for geriatric care.

Purpose of the Study:

• To introduce an automated method for initial frailty assessment using electronic health records.
• To develop a system that extracts relevant information from clinical notes to complete frailty questionnaires.
• To reduce the need for in-person consultations and facilitate early identification of frail individuals.

Main Methods:

• Utilized a phrase-based query expansion technique with Unified Medical Language System (UMLS) ontology to identify key frailty indicators.
• Developed a method to retrieve pertinent clinical notes for automated frailty assessment.
• Employed a dataset of elderly patients' clinical notes for evaluation of automated assessment and question-answering tasks.

Main Results:

• Demonstrated the effectiveness of incorporating phrases from clinical notes for automated frailty assessment.
• Showcased the potential of the automated system to accurately complete the Tillburg Frailty Indicator (TFI) questionnaire.
• Validated the approach for automating frailty assessment and question-answering tasks using real-world clinical data.

Conclusions:

• Automated frailty assessment using clinical notes is feasible and beneficial for early detection.
• The proposed method enhances diagnostic efficiency and conserves clinical resources.
• This approach is particularly valuable for remote patient monitoring and during public health crises like the COVID-19 pandemic.