Using machine learning to improve patient outcomes in a mobile health system for persons with diabetes - Université des Antilles Access content directly
Conference Poster Year : 2019

Using machine learning to improve patient outcomes in a mobile health system for persons with diabetes


Introduction mHealth or mobile health refers to the practice of medicine and public health supported by mobile phones and wireless devices. Over the past decade, many mHealth systems have been developed to provide a range of services targeting a wide range of health conditions. Mobile health has a tremendous potential to improve access to healthcare in developing countries due to mobile penetration rates. MediNet, a mobile health system developed at the University of the West Indies (Mohan and Sultan, 2010), provides remote monitoring services for persons suffering from diabetes and cardiovascular disease. In MediNet, data flows from measuring devices (e.g., blood glucose meter and blood pressure monitor) in a person’s home or work environment to his/her mobile phone using Bluetooth. It then travels through a wireless telecommunications network to the Internet where it is stored securely on a Web server. The system successfully underwent small-scale testing in Trinidad and Tobago in 2009-2010. Feedback in a Mobile Health System In a remote monitoring system such as MediNet, the data collected by the system is used to provide feedback to the person being monitored based on a set of rules. For example, if the person's blood glucose or blood pressure level is “too high” or “too low”, the system recommends certain courses of action to the person. If the person’s blood glucose level is still outside the range after a certain timeframe, a message is sent to the person’s health care provider so that an appropriate intervention can be made. In building the system, the values for "too high" and "too low" were assigned by a medical doctor based on medical practice and were applied across the board to all the patients participating in the study. However, results from the testing of MediNet indicate that the threshold levels for each person may differ for various reasons and may often differ by wide margins. Analysis of test data from the MediNet system also revealed some interesting insights. For example, the blood glucose levels of certain persons were often higher on weekends and on public holidays (Sultan and Mohan, 2013). Data Stored in a Mobile Health System A mobile health system such as MediNet can generate and store a considerable amount of data over time. For example, persons with diabetes are expected to take two measurements each day; each measurement consists of a single value representing the person’s blood glucose level. The system also records a person’s blood pressure; each measurement consists of three values representing the systolic, diastolic, and pulse values. Other mHealth systems can generate even larger amounts of data. For example, Cámara et al (2017) developed a system to screen persons for obstructive sleep apnea; this system collects data from a number of sources including a microphone, a pulse oximeter, and built-in smartphone sensors. Over time, the data in a mobile health system can become so large that it is termed “big data”. It has been suggested that the application of big data analysis techniques to mHealth (e.g., machine learning) can bring value to patient outcomes (Weiler, 2016). Predictive Analysis Using Machine Learning We are presently investigating how machine learning techniques can be used to analyze data stored in a mobile health monitoring system to improve patient outcomes. In particular, supervised learning is being used where data over a certain period (e.g., a year) together with additional input (e.g., whether additional treatment or a visit to a doctor was required) is used to train the system. When local highs or lows occur, the system does not go into “emergency mode” since it would have learned that such variations in the past were not cause for concern. If indeed, for a particular person, the highs or lows resulted in emergency treatment or a visit to a doctor, the system can advise the person beforehand to follow a certain diet or take appropriate exercise. The system is also trained with data obtained during weekends and public holidays. Based on the training data, the system can make predictions which can guide the feedback provided to users. For example, knowing that a particular person is not likely to submit readings over a weekend or knowing that a person’s blood glucose may show certain patterns over a holiday period, the system can make appropriate recommendations before the event occurs. One caveat of this research is that the techniques being developed cannot be applied to real data until the mobile health system has been used for a sufficient period of time to generate the data which will be used to train and test the machine learning algorithms. References Cámara, M.A., Castillo, Y., Blanco-Almazán, D., Estrada, L., and Raimon, J. (2017). mHealth Tools for Monitoring Obstructive Sleep Apnea Patients at Home: Proof-of-Concept. In Proc. 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2017), Seogwipo, South Korea, July 11-15, 2017, pp. 1555-1558. Mohan, P., and Sultan, S. (2010). Staying Connected in a Mobile Healthcare System: Experiences from the MediNet Project. International Journal on Human Computer Interaction, 2, 6. Sultan, S., and Mohan, P. (2013). Transforming Usage Data into a Sustainable Mobile Health Solution. Electronic Markets, 23, 1, pp. 63-72. Weiler A. (2016). mHealth and Big Data Will Bring Meaning and Value to Patient-reported Outcomes, mHealth, 2, 2.  
No file

Dates and versions

hal-02502774 , version 1 (09-03-2020)


  • HAL Id : hal-02502774 , version 1


P. Mohan. Using machine learning to improve patient outcomes in a mobile health system for persons with diabetes. Caribbean Science and Inovation Meeting 2019, Oct 2019, Pointe-à-Pitre (Guadeloupe), France. ⟨hal-02502774⟩


142 View
0 Download


Gmail Facebook X LinkedIn More