1 Data Analytics and Disaster Preparedness and ResponseKaren Smilowitz Industrial Engineering and Management Sciences October 15, 2014

2 Presentation overviewPreparedness and Response Overview Logistical challenges and the impact of data analytics The Northwestern Initiative of Humanitarian and Non-profit Logistics Expanding medical preparedness for mass gathering events through data analytics Volunteer engagement in the age of analytics: A case study with American Red Cross, Greater Chicago Region Analytics for Social Good

3 Phases of Humanitarian Logistics Tomasini, R., Van Wassenhove, L.N., Humanitarian Logistics. Palgrave Macmillan, Basingstoke.

4 Applying Data Analytics to Humanitarian Logistics“… there’s no question that the global emergency relief system has significant shortcomings. Governed for decades more by rules of thumb than research, it’s still more art than science. Humanitarian supply chains are generally less efficient and the people running them less well trained than their commercial and military counterparts.” - Wired 2010 Commercial Applications Analytics approaches Humanitarian Applications Analytics approaches ?

5 Applying Data Analytics to Humanitarian Logistics… Data Look Different Photos from Dr. Jennifer Chan, Feinberg School of Medicine

6 Key Challenges of Disaster Preparedness and ResponseQuick distribution of essential supplies is crucial Efficient and equitable distribution of supplies is needed Significant uncertainty exists in almost every aspect Needs/location of beneficiaries; state of infrastructure; supply availability; magnitude of disaster Multiple players in relief operations create collaboration challenges Humanitarian agencies often have limited information about the region and limited technological support

7 Key Challenges of Disaster Preparedness and ResponseQuick distribution of essential supplies is crucial Efficient and equitable distribution of supplies is needed Significant uncertainty exists in almost every aspect Needs/location of beneficiaries; state of infrastructure; supply availability; magnitude of disaster Multiple players in relief operations create collaboration challenges Humanitarian agencies often have limited information about the region and limited technological support

8 Modeling implications: Quick distribution of essential suppliesTask: Deliver supplies from depot at A to beneficiaries at B, C and D starting at midnight. Objective: Minimize travel time Objective: Minimize last arrival A B C D 2 hours 4 hours 8 hours A B C D A B C D Arrive at B at 2 am Depart A at midnight Arrive at C at 6 pm Arrive at D at 10 am Arrive at B at 2 am Depart A at midnight Arrive at C at 6 am Arrive at D at 2 pm Travel time: 20 hours Latest arrival: 6 pm Travel time: 22 hours Latest arrival: 2 pm In humanitarian logistics, new objective functions must be considered Adapted from Campbell, Vandenbussche, Hermann, “Routing For Relief Efforts”. Trans. Sci. (2008).

9 Challenges of Humanitarian LogisticsQuick distribution of essential supplies is crucial Efficient and equitable distribution of supplies is needed Significant uncertainty exists in almost every aspect Needs/location of beneficiaries; state of infrastructure; supply availability; magnitude of disaster Multiple players in relief operations create collaboration challenges Humanitarian agencies often have limited information about the region and limited technological support

10 Humanitarian Guidelines and Allocation of GoodsThe Code of Conduct for the International Red Cross and Red Crescent Movement and NGOs in Disaster Relief Neutrality and Impartiality in distribution are key guidelines Sphere Project Establishes minimum standards of relief, including: Water, Sanitation, Hygiene, Food Standards, Shelters, Health Beyond these guidelines: Difficult Choices, Interpretation, Leeway IFRC: Reach the most vulnerable populations: Agreements with local leaders to maintain order and fairness MSF: Sudan 1998: Feeding centers admitted only the most malnourished Griekspoor et al. (2001): less selective care would have reduced mortality

11 Challenges of Humanitarian LogisticsQuick distribution of essential supplies is crucial Efficient and equitable distribution of supplies is needed Significant uncertainty exists in almost every aspect Needs/location of beneficiaries; state of infrastructure; supply availability; magnitude of disaster Multiple players in relief operations create collaboration challenges Humanitarian agencies often have limited information about the region and limited technological support

12 Significant uncertainty exists in almost every aspectUncertainty in humanitarian logistics is significant, pervasive and hard to mitigate Comparing with commercial settings, The nature of uncertainty is different in humanitarian logistics The mechanisms to mitigate uncertainty are different

13 Uncertainty – Significant, Pervasive and Hard to Mitigate“There are no good maps. The way to learn where to go is to drive… road blocks and debris move around on a daily basis and it’s never the same.” Process of “Swimming” – take roads and hope they are open Many organizations operate without armed escorts: drivers in potential danger. Alter routes each day for security reasons Unreliable vehicles – maintenance and break down issues Difficult to do pre-planning – how can one predict the location and magnitude of a disaster? Needs of beneficiaries may be unknown early in the relief effort Unreliable drivers can fail to show up. Hurricane Rita: truckers evacuated the area they were supposed to supply

14 Modeling implications: Uncertainty in routing modelsAddressing uncertain travel times in a commercial setting Model as a distribution of travel times on each arc Choose an objective function to minimize expected travel time Recourse actions: Pay drivers overtime Hire additional drivers A B C D 2 hours mean = 2 hours 4 hours 8 hours mean = Issues in humanitarian logistics: Nature of uncertainty: uncertainty in travel time may not fit a nice distribution Recourse actions: due to safety, there may be a hard constraint on route length; hiring additional drivers may not be feasible

15 Challenges of Humanitarian LogisticsQuick distribution of essential supplies is crucial Efficient and equitable distribution of supplies is needed Significant uncertainty exists in almost every aspect Needs/location of beneficiaries; state of infrastructure; supply availability; magnitude of disaster Multiple players in relief operations create collaboration challenges Humanitarian agencies often have limited information about the region and limited technological support

16 Multiple players in relief operations“Dozens or even hundreds of groups swarm into disaster zones, tripping over one another, duplicating efforts, and competing for trucks, fuel, and food.” - Wired 2010 Organizations have different logistics capability, size, authority, organizational structure, political position and level of experience in the disaster relief environments, all of which are potential obstacles for collaboration In 2004 Asian Tsunami, over 700 agencies provided assistance to the impacted region Over 300 agencies provided assistance for the 2001 India earthquake Estimates of up to 10,000 agencies engaged in relief activities after 2010 earthquake in Haiti

17 Modeling implications: Challenges of collaborationA large NGO is transporting goods in a region from an airport A to a refugee camp at D. The vehicle has extra capacity, and the NGO would like to offer this capacity to a smaller NGO in need of transportation from A to D. A great opportunity to collaborate, but what if … the smaller NGO fails to show up at the departure time, should the large NGO wait? goods destined to the smaller NGO are held up in customs, should the large NGO wait? the large NGO receives more donations than expected and no longer has capacity, should the large NGO rescind the offer? Issues in humanitarian logistics: Contracts, penalty fees, incentive structures in commercial settings are often not applicable, but there must be some way to allocate risk

18 Challenges of Humanitarian LogisticsQuick distribution of water, food, medication and other essential supplies is crucial Efficient and equitable distribution of supplies is needed Significant uncertainty exists in almost every aspect Needs/location of beneficiaries; state of infrastructure; supply availability; magnitude of disaster Multiple players in relief operations create collaboration challenges Humanitarian agencies often have limited information about the region and limited technological support

19 Limited, but increasing, information and technological supportRelief org may have temporary presence in region Hire local drivers and vehicles who are familiar with region. Creates decentralized, highly heterogeneous fleets: Physical, social, or political limitations in range and capacity Unreliable or unavailable communication with drivers – recourse in routes made difficult Kosovo crisis (1998-9): Technology applications, such as integrated circuit smartcards, mobile satellite communication systems, the Internet, and geographical data tools (GIS and GPS) Somali crisis (1992–3): Telephone & radio-based communication methods Haiti earthquake (2010): Major telecommunications carriers provided free access to wireless; Groups such as Google.org, Crisis Commons, CrisisMappers building online infrastructure and providing centralized repositories for “crowd-sourced” data UN Joint Logistics Centre offers GIS products and services, including maps, geographic data, and up-to-date database on transport infrastructure that can be accessed and utilized by relief organizations during relief operations FEMA is using GPS technologies to track vehicles and relief supplies in real time

20 What is in the future? Models to address uncertainty Nature of uncertainty and available mechanisms to address uncertainty in disaster relief differ significantly from commercial settings Increases in the availability of real-time mapping information and communication technology in disaster settings Integrate these new capabilities with models Even with these advances, there is still a need for models focus on low-tech implementations and rule-of-thumb policies Metrics are needed to balance equity, efficacy and cost

21 Presentation overviewHumanitarian Logistics Overview Logistical challenges and the impact of data analytics The Northwestern Initiative of Humanitarian and Non-profit Logistics Expanding medical preparedness for mass gathering events through data analytics Volunteer engagement in the age of analytics: A case study with American Red Cross, Greater Chicago Region Analytics for Social Good

22 Expanding Medical Preparedness for Mass Gathering Events Through Data AnalyticsBank of America Chicago Marathon 45,000 registered runners 1.7 million spectators ~1000 medical instances over a 5-7 hour span The Chicago Model Holistic approach to mass participation event planning and management Bring together all major organizations to coordinate preparation and response Comprehensive medical tracking system to monitor medical coverage in real time Joint work with NU students, George Chiampas, Mike Nishi, CEM; Sanjay Mehrotra, IEMS; Jennifer Chan, Feinberg

23 Medical Patient Tracking System (MPTS)… On the Course Handheld devices capture relevant data from patient encounters. Information includes patient, chief complaint, time, diagnosis, treatment provided

24 NU/CEM joint partnershipLeverage and expand the highly dynamic and responsive organizational structure and information systems of the Chicago Model Improve medical preparedness and response for mass participation events Optimize decision-making for mass participation events using data analytics and operations research methodologies Develop new robust and dynamic logistics and resource allocation models.

25 Data Analytics for Medical PreparednessCreate the most comprehensive medical study of large-scale marathons. Pre-2014 Chicago Marathon Integrate data streams from prior events to characterize the spatial and temporal distribution of injuries 2014 Chicago Marathon and beyond Develop a framework for data analytics for mass gathering events Combine with data from other events

26 Key Data Streams Course Data Field Tracking Data Health DataWeather Data Description: Course geographic information Example Variables: Course GIS data Medical Services Aid stations Medical tents Ambulance locations Mile and Kilometer markers Timing mats City road data Description: Timing information along the course Example Variables: Corral assignment by runners seen in medical facilities Start time by corral 5km runner counts Description: Medical records from patients seen at the event Example Variables: Age Gender Medical location Check in and out time Chief complaint Diagnosis Discharge/transfer Description: Information along the course and nearby National Weather Service locations Example Variables: Forecasted and observed temperature Relative humidity Wind speed Wet globe bulb temperature Event Alert System color codes

27 Use descriptive analytics to understand injury trends…… Within One Facility … Over Time Finish line 2013: Balbo Medical Tent Admits 25 20 15 10 5 # of admitted 9:30 am 10:30 am 11:30 am 12:30 am 1:30 pm 2:30 pm

28 Use descriptive analytics to understand injury trends…… Within One Facility … Over Time … By Area Finish line 2013: Balbo Medical Tent Admits 25 20 15 10 5 # of admitted ICU GC POD UC 9:30 am 10:30 am 11:30 am 12:30 am 1:30 pm 2:30 pm

29 Use descriptive analytics to understand injury trends…… Within One Facility … Over Time … By Area … With Finisher Count Finish line 2013: Balbo Medical Tent Admits 25 20 15 10 5 # of admitted 5000 2000 1500 # of finishers 1000 ICU GC POD UC 500 9:30 am 10:30 am 11:30 am 12:30 am 1:30 pm 2:30 pm

30 Use descriptive analytics to understand injury trends…… Within One Facility … Over Time … By Area … With Finisher Count … With Weather Finish line 2013: Balbo Medical Tent Admits 25 20 15 10 5 # of admitted 5000 5000 2000 4000 49.9 58.7 60.5 60.2 61.9 54.9 55.3 60.1 59.0 61.0 59.7 1500 # of finishers 1000 ICU GC POD UC 500 WBGT at finish 9:30 am 10:30 am 11:30 am 12:30 am 1:30 pm 2:30 pm

31 Use descriptive analytics to understand injury trends…… Across Multiple Years 2013 2012

32 Next Steps in Data AnalyticsData merge Merge of 2014 data with prior events Merge with other key data streams Merge with data from other marathons Improved decision making For preparedness For real-time decision making

33 Using Data Analytics to Improve Resource Allocation… Before the Event Simulate the arrival of patients to aid stations to analyze various bed-medical team configurations Total Wait Time (hours) vs Scenarios Due to limited time data, the arrival rate of patients to an Aid Station is a scaled down version of the overall arrival rates of runners passing the nearest distance marker. (In Minutes) 7B 2D 7B 3D 7B 4D 8B 2D 9B 2D Avg. Wait Time 3.6 3.4 2.9 2.8

34 Using Data Analytics to Detect ‘Atypical’ Patterns… During the Event Provide medical volunteers with expected arrival patterns given race day conditions (e.g., temperature) Detect deviations from historical trends in real time Use this information for needed allocation shifts of resources and medical volunteers

35 Data visualization tool debuted at 2014 Shamrock Shuffle

36 Volunteer Engagement in the Age of Analytics: A Case Study with American Red Cross, Greater Chicago Region Situation ARCGC responds to approximately 1,200 disasters per year; using volunteer responders, dispatched by volunteers The two primary objectives of ARCGC are to respond quickly to clients and provide a meaningful experience for volunteers Varying levels of ARCGC responder volunteer engagement lead to unpredictable response effectiveness Goals Utilize multiple data sources to model the drivers of a) volunteer engagement, then b) response effectiveness Develop recommendations for ARCGC to recruit, retain and dispatch volunteers Joint work with NU students Tessa Swanson and Andy Fox; Jim McGowan, ARCCGR

37 Analysis Framework: Connecting the Data

38 Volunteer Training JourneyProspective volunteers pass through multiple checkpoints to become an ARCGC Responder Median: 22 days Range: days Median: 24 days Range: days Median: 8 days Range: days Median: 19 days Range: days END TO END 426 Volunteers Referred to ARCGC 80 Volunteers Called for Dispatch ___________________________________________________________________________ 19% Engagement Timeline: 2.5 months

39 Volunteer ParticipationVolunteer Participation in the Call Out process is heavily skewed to about 50 volunteers These 50 volunteers received between 30 and 233 calls in one year

40 Predicting Dispatch ResponseBuild a model that will predict volunteer engagement, measured as a positive response to a dispatch call, from characteristics of the event (Incident) or volunteer Insignificant Factors Population of the Incident Zip Code Income of the Incident Zip Code Distance from the Rauner Center Time of Day – Late Night Significant Factors Volunteer on Schedule Role – Trainee vs. Full Responder Location – Downtown vs. Suburbs Time of Day – Afternoon Time of Day – Evening Time of Week – Weekend 300%* (65%) 35% 15% (15%) (25%) Dispatch Impact

41 How can this model be used?We can predict a volunteer’s response to a dispatch 75% of the time based on certain factors Use these factors to inform scheduling Times when volunteers are less likely to respond should have more volunteer slots to minimize risk When a high-reputation volunteer is on the schedule, consider calling less experienced volunteers first to get them engaged Engage newer volunteers on dispatches where they are more likely to respond, e.g. afternoon Incidents occurring downtown Emphasize the importance of a full schedule – 3x more effective

42 Analytics for Social GoodA humanitarian and non-profit logistics course that reaches across Northwestern to bring together students from a multitude of disciplines and engages students in challenging case studies, presented by leading experts Students will work in interdisciplinary teams on case studies related to humanitarian and non-profit logistics The course will feature weekly lectures, led by the professor, complemented by experts in the field Students will complete four 2-week case studies with weekly presentations to smaller breakout groups, led by the professor and teaching assistants The course will culminate with a day-long hackathon to design solutions to a disaster response volunteer problem presented by the American Red Cross of Greater Chicago Region

43 Questions? More information aboutHumanitarian Logistics at Northwestern at: