IntroductionRelated WorkMobile Phones as Social SensorsLink Between Physical Symptoms, Behavior Changes and StressMethodologyMobile Sensing PlatformDevice SelectionProximity Detection (Bluetooth)Approximate Location (802.11 WLAN)Communication (Call and SMS Records)Daily Survey LauncherBattery ImpactUser Privacy ConsiderationsBackend Post-Processing and SQL DatabaseOpen Source AvailabilityDataset CharacteristicsDate RangeMobile Phone Sensor DataPre-Experiment Surveys (Baseline Labels)Mobile Phone Daily SurveysAnalysisMobile Behavioral FeaturesTotal CommunicationLate night and Early Morning CommunicationCommunication DiversityPhysical Proximity Entropy with Other ParticipantsPhysical Proximity Entropy with Other Participants Late Night and Early MorningPhysical Proximity Entropy for Bluetooth Devices Excluding Experimental ParticipantsWLAN Entropy based on University WLAN APsWLAN Entropy based on external WLAN APsBehavioral Effects of Low Intensity Symptoms (Runny Nose, Sore Throat and Cough)Behavior Effects of Higher-Intensity Symptoms (Fever and Influenza)Behavioral Effects of Stress and Mental Health SymptomsSymptom Classification using Behavioral FeaturesTemporal Flux Between Behavior, Stress and Physical SymptomsThe Phase Slope Index (PSI) MethodResultsConclusionAcknowledgementsREFERENCESSocial Sensing for Epidimiological Behavior ChangeAnmol Madan, Manuel Cebrian, David Lazer†and Alex PentlandMIT Media Lab and Northeastern University†Cambridge MAanmol, manuel, [email protected]; [email protected] important question in behavioral epidemiology and pub-lic health is to understand how individual behavior is af-fected by illness and stress. Although changes in individualbehavior are intertwined with contagion, epidemiologists to-day do not have sensing or modeling tools to quantitativelymeasure its effects in real-world conditions.In this paper, we propose a novel application of ubiquitouscomputing. We use mobile phone based co-location andcommunication sensing to measure characteristic behaviorchanges in symptomatic individuals, reflected in their totalcommunication, interactions with respect to time of day (e.g.late night, early morning), diversity and entropy of face-to-face interactions and movement. Using these extracted mo-bile features, it is possible to predict the health status of anindividual, without having actual health measurements fromthe subject. Finally, we estimate the temporal informationflux and implied causality between symptoms, behavior andmental health.Author KeywordsSocially aware mobile phones, epidemiology, reality mining.General TermsAlgorithms, Design, Documentation, Experimentation, Mea-surement.INTRODUCTIONFace-to-face interactions are the primary medium for prop-agation of airborne contagious disease [5]. An importantquestion in behavioral epidemiology and public health is tounderstand how individual behavior patterns are affected byphysical and mental health symptoms. Epidemiologists cur-rently do not have access to sensing and modeling capa-bilities to quantitatively measure behavioral changes expe-rienced by symptomatic individuals in real-world scenarios[10]. Such research requires simultaneously capturing symp-tom reports, mobility patterns and social interactions amongstPermission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, orrepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.UbiComp ’10, Sep 26-Sep 29, 2010, Copenhagen, Denmark.Copyright 2010 ACM 978-1-60558-843-8/10/09...$10.00.individuals continuously over long-term duration. In this pa-per, we propose a novel application of ubiquitous comput-ing, to better understand the link between physical respira-tory symptoms, influenza, stress, mild depression and auto-matically captured behavioral features. This is an importantproblem in several different ways.Quantitatively understanding how people behave when theyare infected would be a fundamental contribution to epi-demiology and public health, and can inform treatment andintervention strategies, as well as influence public policy de-cisions. On one hand, clinical epidemiology has accurateinformation on the evolution of the health of individuals overtime but lacks realistic social interaction as well as spatiotem-poral data [6]. On the other hand, current trends in theoreti-cal epidemiology model the rate of infection in a populationwhose behavior is stationary over time and do not accountfor individual changes [9]. For instance, if a person infectedwith influenza continues his habitual lifestyle instead of iso-lating himself, he could pose a bigger risk to others in prox-imity. Based on our analysis and results, policymakers canrecommend social interventions that minimize such risk.On the modeling front, epidemiological models like SIS orSIR commonly assume that movement and interaction pat-terns for individuals are unchanged when they are infected,primarily due to absence of empirical evidence. However,our results show that this is not a correct assumption, as thereare evident variations in behavior of symptomatic individu-als that can be measured using mobile sensing. Account-ing for these dynamics of behavior can be used to createrealistic models of disease propagation in spatial epidemiol-ogy. From the individuals perspective, predicting likelihoodof symptoms from behavior could lead to a possible early-warning system and intervention by medical experts.In this paper, we describe experimental work that illustratesthe use of co-location and communication sensors in mo-bile phones to characterize the change in face-to-face in-teractions and individual trajectories in the contagion pro-cess. The experimental context consists of residents of anundergraduate dormitory for two months, from February toApril 2009. Individuals were surveyed on a day-to-day basisfor symptoms of contagious diseases like common colds, in-fluenza and gastroenteritis. We find that there are character-istic changes in behavior when individuals are sick, reflectedin automatically captured features like their total communi-cation, communication patterns with respect to time of day(e.g. late night, early morning), diversity of their networkand entropy of
View Full Document
Unlocking...