AI-Driven Volunteer Selection

By Jesko Elsner

In case of medical emergency situations, a Volunteer Notification System aims to alarm potential helpers who can arrive at the victim fast enough to provide cardiopulmonary resuscitation until the professional EMS arrive on scene. A simplistic solution for selecting the corresponding volunteers is described by a so called notification radius, i.e., alarming any volunteer with a geographic location that is within a maximum distance from the victim. Whilst the actual geographical distance is an important parameter to be considered, this work will illustrate that various additional decision parameters are of importance, and elaborate an AI-driven volunteer selection system to increase the reliability and efficiency of occurring notifications.

• Language: English
• Publisher: Books on Demand
• Release date: Mar 18, 2015
• ISBN: 9783738678451

Jesko Elsner

Since 2011, Jesko Elsner is a research scientist and postgraduate at the Institute of Information Management in Mechanical Engineering (IMA) of the RWTH Aachen University. Mr. Elsner has more than 8 years of practical experience in the field of computer science as well as Business Informatics. His research interest focuses on applied Machine Learning, Profiling and Software Engineering.

Book preview

Bibliography

1 Introduction

1.1 Motivation

Within the western industrialized world, about every fifth occurring death is the direct result of the medical emergency condition referred to as Sudden Cardiac Arrest (SCA]. With an annual rate of incident of nearly 0.2% within the adult population, SCA is by far the most common manifestation of cardiac disease and claims – solely in Germany – about 150,000 lives per year (Roskamm et al. 2004). Within the state of a SCA, blood circulation is stopped and oxygen is no longer being delivered to the brain and other vital organs (Kasper et al. 2004), reducing the chances of survival with every passing minute. While most emergencies do not involve an immediate life danger, in case of a SCA the first minutes are of utter importance. Jan Bahr (Bahr 2007) states that as little as three minutes without treatment is usually enough for victims of SCA to suffer permanent brain damage and Karin Grassl (Grassl 2009) describes, that every minute without treatment drastically decreases the chance of survival, indicating that a time interval of more than five minutes without treatment will most likely result in the death of the patient.

Whereas professional emergency medical services (EMS) are working on shortening the time that is needed for arriving on scene, there are various parameters that limit potential performance increases regarding this topic; e.g., the current traffic situation, the travel distance and the delay between an incoming emergency call and the march out of the professional helpers. In Bavaria (Germany) for example – a region with good infrastructure and a highly advanced medical system – reoccurring studies are made in order to analyze the effective time interval that local EMS need until arriving at the place of incident. The institute for emergency medicine (INM) in Munich states in a recent study, that professional EMS in the area of Bavaria require approximately nine minutes for arriving on scee (INM Munich 2010). Furthermore, the study underlines an ongoing increase in this deficit, due to a diversity of reasons and illustrates that the severity of the time deficit generally correlates with the infrastructure that a country can provide; resulting in intensification for less advanced countries and regions. The consequential gap between the arrival time of the professional EMS on scene and the survival rates of SCA incidents (if no medical measures are applied in the meantime) is illustrated in Figure 1 based on (Larsen et al. 1993).

Figure 1: Sudden cardiac arrest survival chances

Various publications, e.g., (Valenzuela et al. 1997) (McNally et al. 2011), have analyzed cardiac arrest interventions and the effect of bystanders delivering cardiopulmonary resuscitation (CPR) to a victim suffering SCA, compared to no medical measures being applied until the arriving of the EMS on scene. The results indicate, that CPR applied in an early stage in out-of-hospital emergency scenarios can increase survival rates by up to 2-3 times (Larsen et al. 1993) (Holmberg et al. 2000). Thus, a patient suffering SCA is in need of most urgent medical care that professional EMS alone cannot always sufficiently provide; alternative solutions need to be found, utilizing the positive effect of early applied CPR, even if no bystanders are directly located at the place of incident.

One suitable approach is the integration of medically trained volunteers into the EMS workflow as discussed in recent publications, such as (Elsner et al. 2013a; Elsner et al. 2013b). The approach is built upon the idea that most medical emergencies occur in areas with other people nearby. It is rational to assume, that with an increasing number of people in the closer vicinity of a SCA emergency, some of these people will have at least basic medical expertise and are generally willing to provide CPR to the victim. The actual problem is the unawareness of the ongoing emergency situation. To integrate those nearby helpers into the EMS workflow, a system is required that is able to identify potential volunteers in order to notify them of the emergency; afterwards guiding them to the victim so that they can initiate first aid measures and therefore bridge the time until the professional EMS arrive on scene. This type of system will hereafter be referenced as a Volunteer Notification System (VNS) as firstly introduced in (Elsner et al. 2013c). This work will introduce the decision criteria on which volunteers are to be notified in an ongoing case scenario and discuss suitable implementation details on an intelligent volunteer selection algorithm.

1.2 Medical Background

The human heart has an electrical conduction system that controls the rate and rhythm of the heartbeat. Problems with this electrical system can cause irregular heartbeats called arrhythmias, which can lead to SCA – a condition in which the heart suddenly and unexpectedly stops beating. The hereby resulting loss of blood flow will prevent the brain and any other vital organ from getting oxygen and without immediate treatment, the victim will die within minutes. It’s a common misconception that SCA is the same as a heart attack, while in reality they are quite different; SCA is an electrical problem that prevents the heart in its whole from functioning, whereas a heart attack occurs when part of the heart’s blood supply is reduced or blocked, causing the heart muscle to get injured or die; a heart attack can therefore lead to a cardiac arrest (Kasper et al. 2004).

The term SCA describes the medical emergency condition that a patient is suffering and is usually used by medical personal referring to an unexpected (i.e., sudden) cardiac arrest. An occurring death that results from SCA is described by the term Sudden Cardiac Death (SCD). SCD generally refers to an unexpected death from cardiovascular cause in a person with or without preexisting heart disease (Deo, Albert 2012).

Cardiopulmonary Resuscitation

Cardiopulmonary resuscitation (CPR) describes a medical emergency procedure which utilizes manual chest compressions to partly restore blood flow for victims in a cardiac arrest. While CPR alone is unlikely to restart the heart function, it is indicated for any person who is unresponsive with no breathing in order to manually circulate blood through the victim’s body; thereby supplying the brain and other vital organs with oxygen and preventing the cells from dying (Ornato, Peberdy 2005). Various studies indicate that the application of CPR before the arrival of professional EMS on scene can greatly increase the survival rate of victims suffering SCA (Valenzuela et al. 1997) (Holmberg et al. 2000).

1.3 Methodology

With Section 1 introducing the motivation and general context of this work, the upcoming section will evince and define the actual research focus within the motivated domain. That is, a suitable solution for the volunteer selection problem, as will be introduced and discussed in Section 2. Section 3 will analyze and argue the fundamental requirements followed by an introduction to the contextual terminology in Section 4; concluding machine learning, as major branch of artificial intelligence (AI), a suitable approach for generating sophisticated results within uncertain problem scopes. After discussing currently available notifications systems and their underlying selection approaches, Section 5 will furthermore argue the State of the Art in respect to the introduced requirements by illustrating the possibilities and restrictions of available technologies in the scope of implementing an intelligent volunteer selection system. Section 6 will afterwards elaborate the component based model (i.e., a first system draft) and introduce the individual components of an AI driven selection algorithm. Section 7 will define the process based model that will further be formalized into a generic process model, enabling the derivation of a suitable implementation context and architectural features. In correlation to the elaborated system draft, pseudo-code based implementation details on various components will be discussed in Section 8, whereas Section 9 will evaluate the accomplished functionalities in respect to different scenarios. After a short synopsis, Section 10 will close this work by an outlook and a discussion on the transferability of the research results.

2 Definition of the Research Topic

With EMS having limited resources and not always being able to assure early enough first aid treatment to victims suffering time critical emergencies, reported out-of-hospital SCA incidents currently only have a survival rate of about eight percent (Eisenberg et al. 1990) (McNally et al. 2011). Whereas the positive impact of bystander CPR has been motivated in Section 1, most emergency situations are occurring in the absence of medically skilled observers. Whereas potentially valuable helpers might be geographically nearby, they are unaware of the ongoing medical emergency and unfortunately, the common EMS workflow has no suitable possibility to find or notify them. One approach to fully integrate volunteers with at least basic medical expertise into the EMS protocol is the implementation of a VNS, as described in the upcoming section.

2.1 Basic Concept of a Volunteer Notification System

Introduced by (Elsner et al. 2013c), a VNS is basically described as an IT system that tracks the location of registered users by utilizing state of the art location services, which are available for smartphones and other mobile devices with internet access. By having access to the current whereabouts of the registered volunteers, the system is able to identify those potential helpers, who are at the time of the incoming emergency call, close enough to the reported victim to arrive on scene faster than any professional EMS; thus, being able to provide CPR to the victim and thereby gap the time without treatment for the patient before the professional EMS arrive at the place of incident.

It is important to stress that this general concept does not interfere with any corresponding emergency standard procedures, but can rather be described as an optional add-on to existing EMS; i.e., the responsible dispatcher will decide if to involve this optional feature. Therefore, the potential volunteers must not be a replacement for emergency physicians or any professional helper that is usually involved in a medical emergency workflow, but instead, their main purpose is to arrive at the victim fast enough to provide CPR until the professional helpers are able to initiate further treatment. The effectiveness of a VNS approach will intuitively increase with the number of registered users. Whereas the quality of applied first aid measures generally varies with the medical skill level of an individual volunteer (Bahr et al. 1997), the chances of finding potentially suitable volunteer candidates for an emergency situation will - by bare statistic - increase in correlation with the number of registered users within the system.

Figure 2 illustrates the integration of volunteers into EMS as follows: In case of an incoming emergency call, the responsible dispatcher will follow standard protocol and alert the professional EMS; initiating the march-out of paramedics and emergency physicians. Additionally (i.e., optionally), if a cardiac arrest or any type of emergency requiring immediate first aid treatment is suspected, the dispatcher can forward the assessed information into the VNS infrastructure; triggering the selection and notification of nearby volunteers.

Figure 2: Integration of volunteers into EMS

2.2 Selection Imperative

Once the VNS is provided with information on an ongoing medical emergency, the system will determine which volunteers are in the closer vicinity of the victim and send notifications on the ongoing case to the most promising candidates. Arguing, that alarming any registered user available within the system will for sure include the most promising available helpers is of trivial matter, but with an increasing number of participants, such an approach will result in an immense amount of notifications with mostly no chance whatsoever to arrive on scene in time. Implying a scenario of a VNS covering a region way bigger than a single city and taking into account the therein frequent occurrence of medical emergencies, each registered user would furthermore be flooded with notifications that are most likely out of concern; hiding the ones that might be of actual importance. Regarding this matter, expert interviews with experienced emergency volunteers in the Netherlands have been organized as part of various workshops held between the IMA and the HartslagNu project (IMA 2013/2014). During these interviews, more than 100 active volunteers of the HartslagNu project, as well as the involved project coordinators have been interviewed regarding requirements for occurring notifications and the general framework for integrating a pilot VNS into existing EMS within the province of Limburg. The interviews dictated the requirement of minimizing unnecessary notifications; i.e., notifications that will immediately appear irrelevant to its recipient and those notifications without any plausible chance of reaching the victim in time.

Whereas general software usability standards (Law et al. 2008) (Wan et al. 2013) are not being discussed within the scope of this work, they will be of high importance to any VNS approach since an increased user acceptance will boost the number of participant and thereby also the efficiency of the implemented VNS. Furthermore, technical limitations on both the server side as well as the mobile client side will demand the efficient use of the available resources, once more underlining the requirement of a selection algorithm; as any incoming notification on the client for example will drain the mobile device’s battery and thus quickly result in full discharge. Details on smartphones and corresponding technologies will be discussed in Section 5.2. Financial aspects regarding the total cost of the VNS implementation have to be considered as well, since the absence of a selection algorithm will result in immense message flow, as the number of mobile clients will increase.

Finally, as legal advices indicate (Fehn 2009, 2012), country specific aspects have to be considered before implementing a VNS into a regional system; i.e., volunteers have to fulfill different legal criteria (e.g., specific medical certificates) based on country specific regulations in order to be integrated into the EMS workflow. While this demands the consideration of regional differences within the selection process, especially in cross border emergency scenarios, it again underlines the requirement of a sophisticated selection algorithm.

2.3 Volunteer Selection Problem

As stated in the basic concept of a VNS, location services (as part of modern mobile devices with internet access) are utilized to track the current whereabouts of the volunteers. Whilst available location data is generally not fully accurate and GPS data currently only being available outdoors (as discussed in Section 4.2.1), it is the theoretical knowledge of the current whereabouts of the registered volunteers that will enable different selection algorithms based on the geographical location. However, this section will furthermore illustrate, that even implying perfectly reliable location data, various additional information is needed on an individual volunteer in order to make a dependable decision whether a specific notification is sensible or not.

Notification Radius

With location data available, a simple solution for selecting volunteers is the utilization of a notification radius, defining a maximum distance around the place of incident and alarming those volunteers who are within this radius. This approach will provide a set of helpers who are geographically close to the victim, but will they also arrive on scene faster than potential helpers outside the notification radius?

Whilst the actual distance is an important parameter to be considered when deciding if a volunteer should be notified or not, this data is influenced by technical limitations and subject to fluctuations, especially inside buildings. Moreover, due to impassable environmental obstacles – e.g., rivers, mountains, highways or railways – linear distance calculation does not offer an accurate solution for estimating the arrival time of a potential helper. Whereas the current state of the art regarding the availability of (accurate) map material will follow in Section 5, even the hypothetical scenario of having access to flawless and complete map data for any place at any time – thereby enabling a perfect shortest way calculation, maybe even considering real time traffic information – will still require additional information on an individual volunteer in order to reliably forecast the arrival time; thus, the type of movement, the individual physical performance, the knowledge of the area or the current situation a volunteer is involved in at the moment (e.g., in the company of small children or in the middle of an important appointment).

To summarize the volunteer selection problem at this point; the expression close is definitely suitable for describing the general concept of a VNS, but determining which volunteers are actually able to arrive on scene in time, turns out to be a non-trivial decision problem. As illustrated in later section, an upper limit of hypothetically valuable candidates will be generated by assuming perfect traveling conditions in relation to the current geographical distance. Implying that every registered volunteer is furthermore able to provide at least basic CPR – and therefor leaving individual medical expertise a secondary parameter – as well as ignoring ethical arguments that might prevent volunteers from helping in specific circumstances, the most important information that is to be acquired on an individual volunteer during the selection process, is thereby the expected time of arrival at the victim. This information is not directly assessable but instead has to