LAUSR

Introduction The impact of frailty surges, as the prevalence increases with age and the population age is rising. Frailty is associated with adverse health outcomes and increased healthcare costs. Many validated instruments to detect frailty have been developed. Using these in clinical practice takes time. Automated estimation of the probability of being frail using routinely collected data from hospital electronic health records (EHRs) would circumvent that. We aim to identify potential predictors that could be used as features for modeling algorithms on the basis of routine hospital EHR data to incorporate in an automated tool for estimating the probability of being frail. Methods PubMed (MEDLINE), CINAHL Plus, Embase, and Web of Science will be searched. The studied population consists of older people (≥65 years). The first step is searching articles published ≥2018. Second, we add two published literature reviews (and the articles included therein) [Bery 2020; Bouillon, 2013] to our search results. In these reviews, articles on potential predictor variables in frailty screening tools were included from inception until March 2018. The goal is to identify and extract all potential predictors of being frail. Domain experts will be consulted to evaluate the results. Discussion The results of the intended study will increase the quality of the developed algorithms to be used for automated estimation of the probability of being frail in secondary care. This is a promising perspective, being less labor-intensive compared to screening each individual patient by hand. Also, such an automated tool may raise awareness of frailty, especially in those patients who would not be screened for frailty by hand because they seem robust. Conclusion The identified potential predictors of being frail can be used as evidence-based input for machine learning based automated estimation of the probability of being frail using routine EHR data in the near future.