Prediction model of outpatient flow based on behaviour data of outpatients in a Chinese tertiary hospital

Abstract Objective Outpatients at Chinese tertiary hospitals are always over-crowded due to the “walk in” mode and all of the services provided within the hospital, which include patient interviews, lab workups, imaging examinations and prescription fill. The aim of this study was to build a model to predict sequential patterns of the services that the patients sought to use and then provide advice for what services they can choose and in which order to avoid long waiting times. Method Data collected from outpatient information systems were used to construct a data warehouse. Using the Hadoop distributed platform, outpatient data were processed and analysed using SparkR. The algorithms used included exploratory data, correlation analysis, and machine learning algorithms to analyse the patient flow data. Results Approximately 2 hundred thousand qualified records were used for the training set, and 89 thousand records were used for the test set. A prediction model for patient flow was built to predict a patient’s selection from the patients who utilized more than one service in a single outpatient visit. This model can predict the patient’s behaviour of filling prescriptions before going to other services (lab tests or imaging examination), with accuracy rates of 80.94 and 73%, respectively. Diagnosis classification, insurance type, gender and the other three attributes were considered key factors affecting the patient’s selection. Conclusions This model calculates the selection likelihood of each patient after seeing a doctor and then estimates the number of patients waiting at the in-hospital pharmacy, laboratory and radiology services within a time interval (e.g., half an hour). In addition, it can be used as a guide for outpatient services in Chinese tertiary hospitals after further optimization.