Abstract: This paper seeks to establish a machine learning driven method by which a military veteran with Post-Traumatic Stress Disorder (PTSD) is classified as being in a crisis situation or not, based upon a given set of criteria. Optimizing alerting decision rules is critical to ensure that veterans at highest risk for mental health crisis rapidly receive additional attention. Subject matter experts in our team (a psychologist, a medical anthropologist, and an expert veteran), defined acute crisis, early warning signs and long-term crisis from this dataset. First, we used a decision tree to find an early time point when the peer mentors (who are also veterans) need to observe the behavior of veterans to make a decision about conducting an intervention. Three different machine learning algorithms were used to predict long term crisis using acute crisis and early warning signs within the determined time point.

Abstract: Several attempts have been made at creating mobile solutions for patients with mental disorders. A preemptive approach would definitely outdo a reactive one. This project seeks to ensure better crisis detection, by assigning patients (veterans) to caregivers (mentors). This is called the mentor-mentee approach. Enhanced with the use of mobile technology, veterans can stay connected in their daily lives to mentors, who have gone through the same traumatic experiences and have overcome them. A mobile application for communication between veterans and their mentors has been developed, which helps mentors get constant feedback from their mentees about their state of well-being. However, being able to make good deductions from the data given as feedback is of great importance. Under-represent ing or over-representing the data could be dangerously misleading. This paper presents the design process in this project and the key things to note when designing a data visualization for
timely crisis detection and decision-making.

Abstract: Researchers and end users generally demand more trust and transparency from Machine learning (ML) models due to the complexity of their learned rule spaces. The field of eXplainable Artificial Intelligence (XAI) seeks to rectify this problem by developing methods of explaining ML models and the attributes used in making inferences. In the area of structural health monitoring of bridges, machine learning can offer insight into the relation between a bridge’s conditions and its environment over time. In this paper, we describe three visualization techniques that explain decision tree (DT) ML models that identify which features of a bridge make it more likely to receive repairs. Each of these visualizations enable interpretation, exploration, and clarification of complex DT models. We outline the development of these visualizations, along with their validity by experts in AI and in bridge design and engineering. This work has inherent benefits in the field of XAI as a direction for future research and as a tool for interactive visual explanation of ML models.

Abstract: In disasters, several national, international, and non-governmental organisations such as police, health, ambulance, fire and civil defence services are usually involved in the response process. Therefore, it is crucial that responding agencies communicate effectively to avoid fragmentation and duplication in services, to harmonise separate activities, and to clarify roles and responsibilities. Central to communication is information exchange. Effective information exchange enhances not only the appropriateness and success of disaster response, it also ensures timeliness. However, cross-agency communication is extremely challenging especially at times when there are high stress levels, incomplete data, and minimum time to make critical decisions. This paper attempts to specify a 'best-practice' model for cross-agency communication built around the specific information requirements of disaster management and disaster medicine agencies, with the aim of improving the overall quality of healthcare services provided to disaster victims.

Abstract: The COVID-19 pandemic has generated unprecedented, devastating impacts across the United States. However, some communities have disproportionately endured adverse health outcomes and socioeconomic injuries. Ascertaining the factors driving these inequities is crucial to determining how policy could mitigate the impacts of future public health crises. We have established research-driven metrics, aggregated as the Community Vulnerability Index (CVI), that quantify vulnerability to public health and economic impacts of COVID-19. We performed two analyses to better understand similarities between communities in terms of the vulnerabilities represented by the metrics. We performed an unsupervised k-means clustering analysis to understand whether communities can be grouped together based on their levels of negative social and health indicators. Our goal for this analysis is to determine whether attributes of the constructed clusters reveal areas of opportunity for potential policy impacts and future disaster response efforts. We also analyzed similarities between communities across time using time-sensitive clustering analysis to discover whether historical community vulnerabilities were persistent in the years preceding the pandemic and to better understand the historical factors associated with disparate COVID-19 impacts. In particular, we highlight where communities should invest based on their historical health and socioeconomic patterns and related COVID impacts. Through extensive interpretation of our findings, we uncover how health policy can advance equity and improve community resilience.

Abstract: This research uses multiple methods to investigate the use of an enterprise mobile multimedia information system aimed at improving handover of patient and emergency incident information from pre-hospital Emergency Medical Services (EMS) to hospital emergency department providers. A field study was conducted across EMS and hospital organizations in the Boise, Idaho region of the United States for three months to examine use of the system and to assess practitioner perspectives. Findings include perceived benefits and challenges to using digital audio recordings and digital pictures, captured using a smartphone application, for improving the timeliness, completeness, accuracy, convenience, and security of patient information for handover in EMS; limitations on how much data can be collected in the field due to a wide variety of contextual constraints; and a need to better understand the value of video within the EMS handover context.

Abstract: Low socioeconomic status (SES) and inadequate nutrition during pregnancy are linked to health disparities and adverse outcomes, including an increased risk of preterm birth, low birth weight, and intrauterine growth restriction. AI-powered computational agents have enormous potential to address this challenge by providing nutrition guidelines or advice to patients with different health literacy and demographics. This paper presents our preliminary exploration of creating a GPT-powered AI chatbot called NutritionBot and investigates the implications for pregnancy nutrition recommendations. We used a user-centered design approach to define the target user persona and collaborated with medical professionals to co-design the chatbot. We integrated our proposed chatbot with ChatGPT to generate pregnancy nutrition recommendations tailored to patients’ lifestyles. Our contributions include introducing a design persona of a pregnant woman from an underserved population, co-designing a nutrition advice chatbot with healthcare experts, and sharing design implications for future GPT-based nutrition chatbots based on our preliminary findings.

Abstract: Crisis and disaster management in the Netherlands has made huge leaps forward in recent years with regard to different organisations trying to manage one or more aspects of the safety chain. This research focuses on the information structure of health care during disasters with an aim to improve disaster management and tries to answer the following question: How can location based services improve information services within health care during disasters? Through the use of literature and interviews this thesis describes how disaster management can be improved through the use of Location Based Services (LBS). The scope of this research is aimed at better understanding the organisational processes during somatic health care. By defining a case and on the basis of literature and interviews in the Province of Utrecht, it was possible to develop a three layer graph model (3LGM). This model shows an overview of information processes performed by the health care organisation during the first hour after an accident. In this context, the 3LGM model is used to obtain an overview of the quality of information processing in such a problem area. The organisational structure, which deals with disaster management, consists of a strong co-operation between the police, fire departments, the local government and the 'Medical Aid during Accidents and Disasters' (GHOR). The size of the organisation depends largely upon the scale of the disaster, however the current information structure is not suitable for storing and processing the information in an efficient and effective manner. The same applies when displaying information related to casualties and safety within an area. With the help of location based services consisting of, geographical information systems (GIS), global positioning systems (GPS) and second or third generation telecommunication technologies, the existing information structure can be optimised. Expected advantages are higher accessibility to health care, a safer environment for rescuers, more time for managing the healthcare processes and an improved interdisciplinary co-operation between the police, fire departments, the local government and the GHOR. © Proceedings ISCRAM 2004.

Abstract: During disasters public health organizations increasingly face challenges in acquiring and transforming real-time data into knowledge about the dynamic public health needs. Resources on the internet can provide valuable information for extracting knowledge that can help improve decisions which will ultimately result in targeted and efficient health services. Digital content such as online articles, blogs, and social media are some of such information sources that could be leveraged to improve the health care systems during disasters. To efficiently and accurately identify relevant disaster health information, extraction tools require a common vocabulary that is aligned to the health domain so that the knowledge from these unstructured digital sources can be accurately structured and organized. In this paper, we study the degree to which the Unified Medical Language System (UMLS) contains relevant disaster, public health, and medical concepts for which public health information in disaster domain could be extracted from digital sources.

Abstract: Designing mHealth applications for mental health interventions has largely focused on education and patient self-management. Next generation applications must take on more complex tasks, including sensor-based detection of crisis events, search for individualized early warning signs, and support for crisis intervention. This project examines approaches to integrating multiple worn sensors to detect mental health crisis events in US military veterans. Our work has highlighted several practical and theoretical problems with applying technology to evaluation crises in human system, which are often subtle and difficult to detect, as compared to technological or natural crisis events. Humans often do not recognize when they are in crisis and under-report crises to prevent reputational damage. The current project explores preliminary use of the E4 Empatica wristband to characterize acute aggression using a combination of veteran self-report data on anger, professional actors simulating aggressive events, and preliminary efforts to discriminate between crisis data and early warning sign data.