Phase 1 – Aggregate terms form other ontologies and validated clinical tools

To extract frailty concepts from existing instruments, five members of the research team reviewed the specific items from 14 instruments chosen by the number of times they were cited and expert recommendation [20]: (1) Physical Frailty Phenotype (PFP, also called CHS frailty phenotype) [2]; (2) SF-36 [48]; (3) FIM [49]; (5) Clinical Frailty Scale [50]; (6) Brief Frailty Instrument [62]; (6) the Barthell Index [51]; (7) Health Assessment Questionnaire (HAQ) [52]; (8) PSMS [53]; (9) Katz ADL [54]; (10) Duke Activity Index [55]; (11) RDRS [56]; (12) FACIT [57]; (13) NYHA [58]; (14) Deficit Accumulation Index (DAI, also called Frailty Index) [63]. Each person reviewed each individual item from each instrument. Terms from the World Health Organization’s International Classification of Functioning, Disability and Health (ICF) were also included in the analysis because the instruments varied in their levels of abstraction, their scopes, and uniqueness.

At this step we added an ontology entry for comorbid condition count. Comorbid conditions are an important indicator of frailty. However, they are not the focus of our investigations. We focused on frailty-specific indicators in order to identify core frailty concepts in clinical documents.

The concept list was expanded by the findings of the interviews of cardiologist and cardiac surgeons described above.

Finally, we included terms extracted from manual note review by members of the research team with clinical experience. These reviews were in preparation for NLP topic modeling by Shao, et al., (2016). They reviewed clinical notes and social media posts [64].