#2004 Artificial intelligence in optimizing nephrology outpatient managment: result of a pilot study

Abstract

Abstract Background and Aims Process mining (PM) has been defined “a new research frontier for process reorganization in health care”. It is an approach of Artificial intelligence (AI) comprising a wide range of methods and techniques for analyzing processes from data collected during routine activity. This is the first application of PM to the clinical management of outpatients, although there are experiments in the literature in administrative processes or emergency room admissions. We evaluated the effectiveness of these AI methods for monitoring and improving outpatient processes in our Nephrology Department and highlighting their potential and critical issues. Method The experimentation included two phases. Phase1: Identification, collection, preprocessing of clinical data and construction of the dataset in the XES standard (Fig. 1). Adherence to the standard will make it possible to be “compliant” with all PM tools and to implement different types of analysis. In addition, observation flows and variables, particularly demographic and clinical variables (age, sex, BP, BMI, CKD stage, age of transplantation, major comorbidities), reception/end visit times, and detection of exams are defined. Phase2: Analysis of the dataset with different PM models: Process Map for process/flow extraction from the data (gap analysis), Process Key Performance Indicator for time and load analysis, Process Simulator for dynamic flow simulation and bottleneck detection, Process Conformance for compliance analysis and deviations from process standards. Results In phase 1 we collected a dataset of 46 outpatients. In phase2 PM tools analysis provided time and load values related to the correct execution process (Fig. 2). Critical issues emerged such as noncompliance with the expected procedure: about 60 percent of patients bypassed admission, and in some the clinical assessments expected in a proper outpatient visit were not detected (e.g., BP was not detected in 9% and BMI in 20%). The simulation also revealed bottlenecks at several points in the process and timing anomalies. The results analysis enabled the identification of action points on which to act to improve management. On the same time, the analysis brought out some situations that the method flagged as abnormal. For instance, some patients diverged from the ideal flow (indicated by AI) and the divergence was related to specific clinical conditions that the model could not identify. Conclusion This pilot study highlights the enormous benefits of AI techniques such as PM to monitor and optimize outpatient management and performance. However, the success of AI is related to the synergy between technology and physician, which helps in the interpretation of analysis data.

Ähnliche Arbeiten