e-Medical test recommendation system based on the analysis of patients’ symptoms and anamneses

Abstract

This paper demonstrates an e-medical test recommendation system based on the analysis of patients’ symptoms and anamneses. The exact test selection for a specific patient can be time consuming and error-prone due to the huge amount of information to be considered like: the number of tests, patients, long working hours, exceptional cases, etc. The redundant or missing tests can cause serious loss of money, time and more seriously delay in the initiation of the therapy. The study aims to provide a fast and cost effective system for the medical experts and patients. The data are collected from the patient records of a private hospital, preserving anonymity, from all departments. Only the internal medicine department data are utilized. The patients’ age, gender and the words used in the anamneses and symptoms as plain text are the input for the system. The texts are analyzed and various methods have been applied for selecting the effective words for recommending a specific medical test. These terms, along with the demographic information, are used as the features of the well-known machine learning algorithms of WEKA [5], namely Sequential Minimal Optimization (SMO), J48, Random-Forest (RF), Bagging (Bagg), ADTree (ADT) and AdaBoostMl (ABoost). The number of medical tests that are applicable in the hospitals is too high, therefore only 20 most frequently required ones are selected. The promising results of the study indicated that the symptoms given as plain text can be efficiently utilized by the experts for medical test selection.