Evaluating the DI@L-log System on a Cohort of Elderly, Diabetic Patients: Results from a Preliminary Study Lesley-Ann Black 1 , Michael McTear 1 , Norman Black 1 , Roy Harper 2 , and Michelle Lemon 2 1 Faculty of Engineering, University of Ulster, Jordanstown, Northern Ireland 2 Diabetes Specialist Care, Ulster Hospital, Dundonald, Northern Ireland {la.black,mf.mctear,nd.black}@ulster.ac.uk Abstract DI@L-log is an automated medical spoken dialogue system designed to enable patients to regularly communicate health data to the point-of-care over the telephone. In order to investigate the performance of the system, a preliminary evaluation was conducted using 5 novice hypertensive diabetic patients from the Ulster Hospital in Northern Ireland. The purpose of the study was to assess several factors which need to be addressed when designing spoken dialogue systems for elderly, disabled users. We examined the performance of the system, the interaction preferences of the user and their usage patterns, the level of user satisfaction, and the impact of the system on the patient’s health over time. A revised version of the system, modified to take account of findings from this study, is currently undergoing testing with a larger group of subjects. 1. Introduction Spoken dialogue systems provide access to services and information that have traditionally only been available to people with personal computers and internet access [1]. This paper presents results from a study that examined the use of a spoken dialogue system, known as DI@L-log [2], by elderly, infirm, hypertensive diabetes patients. The system was designed to take account of the difficulties that these patients might experience with an automated system and to provide support and feedback [3]. DI@L-log is a task-oriented system with a female persona that plays a role similar to a nurse. The system acts as an automated “voice logbook” that replaces the more passive paper logbook currently used by diabetes patients to record home monitored data. The system collects and analyses the patient data and communicates the results automatically to doctors at the hospital, alerting them to any abnormal developments and thus enabling them to manage and reduce complications associated with the disease more effectively. 2. Overview of the DI@L-log system DI@L-log consists of a voice user interface, voice browser, web server, backend database, and graphical user interface. The system is written in VoiceXML using Voxpilot’s online development environment (Voxbuilder) and is hosted on the Voxpilot Website [4]. The voice browser, located in Monaco, acts as a gateway to the backend system to which it sends data as SQL queries using the Apache Tomcat Web server and JSP. Figure 1 illustrates the components of the system and the call management. When the patient places a call from their own home in the UK, it is transferred to the VXML interpreter in Monaco. A request is then submitted back to the DI@L-log files residing on the Ulster Hospital server in the UK. Figure 1: Call flow architecture DI@L-log enables patients to speak their vital signs health data (weight, blood sugar and blood pressure) using speech or DTMF (telephone keypad). Data is retrieved from the backend system during a call and at the end of the call the database is updated with the newly elicited readings. The system provides feedback by comparing the new data with previous call and target readings. The graphical user interface, which is not the focus of the present paper, allows patient trends and alerts generated by the system to be viewed and manipulated by doctors at the hospital if individual or multiple readings were regarded as too high. This decision support feature is modelled on medical protocols for blood glucose, blood pressure, and the individual’s weight target, based on the body mass index (BMI). The decision support component aims to act as an intervention strategy in order to mitigate complications associated with diabetes and co-existing hypertension. The following is an extract from an interaction between a patient and the system: System: Welcome to the Dialogue Interactive Service. You can say help or exit at any time. Please now enter your four digit PIN. Patient: One nine seven five System: If your number is 1 9 7 5 say "yes", or press "1". To change, simply say "no," or press 2. Patient: Yes System: One moment. I’m looking up your details… Ok Joe, are you ready to give your measurements? Patient: Yes System: Thank you. What is your weight today? Patient: I’m twelve stone and 6 pounds. System: Please confirm your weight is twelve stone and six pounds. Say "yes", or press 1. Otherwise, say "no" or press "2". Patient: Yes System: Super. And what is your blood sugar? Patient: Six point two Patient: (uses keypad) 6*2 System: Did you say your blood sugar is six point two? Patient: Yes