UMLS-Based Question-Answering Approach for Automatic Initial Frailty Assessment.

Frailty is the most problematic multidimensional geriatric syndrome among elderly population that leads to poor quality of life and increased risk of death. Adverse effects include an increased risk of hospitalisation and institutionalisation, poorer outcomes of post-hospitalisation, and higher mortality rates. A questionnaire-based frailty assessment is an effective way to achieve early diagnosis of frailty. However, most of the existing frailty assessment tools require face-to-face consultation. For elderly patients living in rural areas are more likely to struggle to access healthcare than a patient living in an urban or suburban area, and they have higher chance of catching diseases due to frequent hospital visits as most of them are vulnerable due to being immunocompromised. An automatic initial frailty assessment approach can minimise the impact of mentioned disadvantages and save clinical resources by avoiding unnecessary manual assessments. The objective of this paper is to propose an automatic initial frailty assessment approach which can quickly identify potential patients that require further frailty assessment by using patient's relevant clinical notes to answer Tillburg Frailty Indicator (TFI) questionnaire automatically. A phrase-based query expansion method is proposed to identify the most relevant phrases to the frailty assessment questionnaire based on UMLS ontology. The research shows the advantages of using UMLS based concepts as features in automatic initial frailty assessment using clinical notes. The research enables clinician to assess frailty automatically using medical data, reduces the frequency of face-to-face consultations and hospital visits, which is extremely beneficial during unusual or unexpected times such as COVID-19 pandemic.