*Kenneth Jung; Jasmine Zia, MD
Poster Presentation – Practice Track
Saturday, Sept 29, 2012: 1:20 PM – 2:20 PM – LK Lower Lobby
← Back to Medicine X 2012 Proceedings
Irritable bowel syndrome (IBS) is a chronic functional disorder that is characterized by episodic abdominal pain associated with disturbed bowel habits. It is estimated to affect up to 52 million people in the United States and cost $10 billion annually, excluding the costs of prescriptions and over-the-counter drugs. It is one of the top ten reasons patients seek primary care and accounts for a third of all gastroenterology consultations. Despite this, the pathophysiology of IBS remains poorly understood, and treatment is problematic. Medications for IBS have limited marginal therapeutic gains over placebo of 7 to 15%. Thus, the therapeutic emphasis for IBS has shifted to diet and lifestyle modifications since both food and stress are known triggers of IBS symptoms. This shift has placed the burden of care largely on the shoulders of patients, who are typically asked to keep detailed food, stress and symptom logs which they and their care providers can use to determine triggers, and thus personalized behavioral and diet modifications. This can be a frustrating and inefficient process, in part because patient logs are often incomplete or incorrect – it is difficult for patients to actively collect accurate data over a long period of time.
The lack of complete, reliable and structured data renders the application of sophisticated statistical pattern recognition techniques problematic. We are working to make this process more objective and efficient by using commercial off-the-shelf wearable sensors (for electro-dermal activity, heart rate, and motion) coupled to smart phones and algorithms for the detection of activities of daily living which are relevant to the self-management of IBS – specifically, food intake, bowel movements, and psychological stress – to augment data collection to make it more pervasive, passive, and objective. To this end, we have identified two key aspects of daily food and symptom logs that can be augmented by current technology. First, studies have shown that the majority of food and symptom diary entries are entered long after the fact, and consequently suffer in terms of accuracy and completeness. We will attempt to address this by automatically detecting relevant activities in real time, allowing the system to provide timely prompts to users to enter relevant data in the moment. Second, entering the detailed composition of food is a laborious task that is not necessarily made easier by the highly constrained interfaces presented by current mobile applications. We plan to address this problem by moving the burden of entering composition data away from the user by crowdsourcing through services such as Mechanical Turk.
The first phase of the project uses manually labeled data collected from dominant side wrist and hip mounted accelerometers to train strong classifiers to recognize “primitives” of the relevant activities of daily living – for example, the motion of bringing your hand from a plate of food to your moth. The second phase of this work is focused on combining the output of these classifiers, along with contextual information such as time of day and location to automatically segment, or label, time series of data with relevant labels (e.g. eating, experiencing psychological stress) using dynamic probabilistic graphical models. Ultimately, we hope to deploy these algorithms in real time systems in which sensors are linked to Bluetooth LE enabled smart phones for testing in non-patient populations before testing in IBS patients in the context of augmenting a clinically proven outpatient treatment protocol developed at the University of Washington. Concurrently, we will use focus groups of patients and care providers to guide the design of the system with respect to usability and clinical utility of the data.
This system has the potential to provide patients and their care providers with high quality, objective information about their food intake, stress response and symptoms. In doing so, we hope to enable patients to better manage their symptoms, improving their quality of life and reducing the overall burden of IBS on the health care system. Furthermore, a compendium of such data, compiled from a large population, would be a valuable resource for extracting population level relationships between triggers and associated symptoms, elucidating the pathophysiology of IBS.