Search

Introduction: Frailty is a major health concern associated with aging and with an increased vulnerability to adverse outcomes. The rising health care costs associated with fragile patients represents a threat to National Healthcare Systems due to the limited resources. One approach to this challenge is to consider the development of artificial intelligence (AI) based tools aimed at the redesigning care for frail older people, stratified by Rockwood grade [1]. This study aimed to the development of the SmartChronic platform that integrates a set of data-driven tools to both proactively identify frail older patients at risk of non-programmed hospital readmission and predict hospital clinical pathways for optimal use of limited healthcare resources. Methodology: Retrospective data of the present study came from the electronic healthcare record of La Fe Health Department (Valencia, Spain) that has deployed a Real-World Data analysis platform based on SAS® Software Analytics infrastructure, effectively organized reaching stage 6 in the eight-stage (0-7) EMRAM maturity model. Currently, the data lake layer includes structured and semi-structured records updated daily, coming from several information systems involving clinical activity, and is composed by the aggregation of 22 DataMart comprising 750 million rows, 84 tables and 4.064 columns. Gradient Boosted Trees has been used to develop the 30-day hospital readmission risk model and logistic regression for the development of the pathway prediction model. The study was approved in 2020 by the Ethical Clinical Research Committee of La FE Hospital (Reference number 2020-138-1). Results: The 30-day hospital readmission risk model shows the value as a risk percentage, with reasonable value of 0.665 for the area under the ROC curve. Regarding the pathway prediction model, the algorithm implemented can be applied iteratively at evolutive states of the clinical pathway. These states are determined by a sequence of transition events, that once validated, establish the starting point of a new step of processing. This way, we have developed a novel framework for modelling patient flow through various hospital services. The Accuracy value of the model is 0.40. Both models are integrated into a global SmartChronic platform. Conclusions: A moderately discriminative real-time 30-day readmission and patient pathway predictive models can aid the development, implementation, description and evaluation of integrated care programs for frailty patients with chronic diseases. The successful development of these tools required the establishment of an interdisciplinary collaboration among academia, industry and healthcare providers. Implications for applicability: This technology allows studying big database of patients, contributing to the development of the concept of personalized medicine and to optimize integrated care. Platforms such as SmartChronic would also improve the decision-making of health policy makers. This study has been carried out in a single hospital, it would have to be analyzed if the results can be ex-trapolated to other hospitals. Funding: This research was funded by the Valencian Innovation Agency, Generalitat Valenciana (INNEST/2020/47). [ABSTRACT FROM AUTHOR]