17.2.1 Point-of-Care Environment

The effect of the ambient environment at the Point-of-Care (PoC) can have a significant impact on the patient being treated. This involves the study of physical condition as well as chemical and biological contaminants on patient's health while being treated. Advances in IoT for health care allow environmental sensing systems to be incorporated as part of the smart ambulance (Kelly et al., 2013). Through integrating ambient sensing in the smart ambulance as illustrated in Figure 17.2, it is possible to assess the impact of varying environments on changes in the physical condition of patient's state of health as well as the treatment being provided at the PoC in order to determine the environmental factors that are most critical to enable better patient treatment in the ambulance setting. Ambient environmental effects could impair the recovery of patients that have certain respiratory conditions (Roche et al., 2008). Monitoring the environment the patient is in and ensuring it is optimal for specific patient conditions would improve recovery. In the center of the IoT–PoC environment is the smart ambulance that serves as a connected support console. Additionally, it can initiate an automatic request for a rescue helicopter in one of the two conditions: difficult access to patient, road condition either not allowing rapid return or number of patients exceed available road resources; or the patient exhibits time sensitive condition that entails clinically important time saving in reaching hospital. Clinically significant information about the patient can also be sent to the helicopter support staff prior to dispatch.

Other forms of basic paramedic support include the following:

• Vital signs if a patient that may indicate the mortality of a patient such that no treatment remains meaningful.
• Oxygen administration with SpO2 measurement below 94% for identification of high risk patients.
• Respiratory determination of type and severity.
• Cardiac conditions with ECG in the diagnosis of anterior ST elevation myocardial infarction (STEMI) as well as other conditions such as cardiac arrest and atrial fibrillation or flutter.
• Shock and trauma entails image processing engines and determining causes of hypovolemia. The former allows fast and accurate diagnosis of burns severity, spinal cord injury, joint dislocation, and fracture misalignment; whereas the latter may involve mechanism for fluid loss detection or infrared sensing for hyperthermia patients.
• Metabolic analysis involves protection of frontline paramedics' personal safety; this concerns agitated delirium in case a patient exhibits sign of anxiety with physical aggression. In addition, biosensors play an important role in the measurement of altered consciousness.

Perhaps the most important aspect of safety concern is commutable infectious disease. Similar to minimizing the risk of hospital acquired infection (HAI) (Dancer, 2009), keeping the ambulance clean is also a great challenge in the avoidance of infection. An infection occurring in a patient prior to arriving at the hospital as well as occupational infections among supporting staff of the ambulance requires close monitoring of the ambulance environment itself. However, the current medical settings and guideline-based standard operating procedures (SOP) in providing on-scene treatment does not specify a technology or technique to reduce HAI (Liu et al., 2011). The existing practices are more related to discipline in paramedicine without any associated assistive technology. In this regard, technological advancements are beginning to change the way infection prevention scientists tackle HAIs by adopting IoT and assistive technologies in a smart ambulance. Earlier work has described various technologies and the way it is facilitating measurement and compliance from hand hygiene to environmental cleaning and instrument decontamination (Pyrek, 2014). Any effort in tracking would entail the deployment of wireless system to instrument hygiene events. IoT enables improvement in the analysis of on-scene treatment and prognostics due to the increase of collected data about environmental effects that could affect patient health. Such IoT system would enable determining environmental factors that are most critical to enable better treatment.

Patient monitoring and managing remote rescue operations can deliver significant value in reducing recovery time and operating costs. Before the introduction of connected ambulance when paramedics were inadequately networked, they were effectively serving in a closed system unable to communicate with the hospital network. Capturing, aggregating, filtering, and sharing data from health systems seamlessly into emergency medicine is both a challenge and promise of the Medical IoT, where IoT serves as the point where patient data are directly linked to the national health care system via telemedicine. IoT allows monitoring in remote and demanding locations through the smart ambulance as a hub. Data filtering and aggregation in the ambulance gateway facilitates full duplex data transfer and storage to the hospital.

Central to the PoC environment featured in Figure 17.2 are intelligent wireless networks that consists of both short-range local coverage and long-range links, wearable devices, and low-power sensing arrays coupled with medical data analytics combined to form the basic building block of medical IoT in emergency medicine. In the IoT context, this combination of connected technologies enables a multitude of biomedical and environmental sensors to be temporarily placed in any location not restricted to where network coverage and power source are available. All information paramedics need about the patient and details on treatment options is assembled according to the condition of the patient. The concept of instrumenting “things” such as medical devices, health monitors, consumer wearables with biosensors are all proliferated in the ambulance setting. Even before IoT became popular in recent years, emergency medicine systems have operated as separate networks like one between paramedics and the hospital via traditional cellular phone networks; this is linked to another completely isolated wireless local area network (WLAN) within or surrounding the ambulance. Maintaining adequate network reliability and security for emergency medicine cannot be simply met with traditional network management strategies (Ansari et al., 2006). Moreover, the way these networks are implemented determines whether the patient data manually collected at the scene can be reliably and securely transmitted to the hospital under harsh environments typical of causing accidents under heavy rain (Fong et al., 2003b).

IoT sensing networks facilitate the study of development of hazard response and any methodology can be derived for a sensor network within an ambulance setting. This will complement the biosensing network for monitoring the patient's physiological and vital signs that is already in place. With current focus of biosensor research on acquiring patient data, integrating environmental sensing into the ambulance setting toward human prognostics based on environmental factors would improve on-scene treatment, which expands on the field of biosensing networks to also monitor environmental changes in order to better accommodate the recovery of patients. This would complement the use of traditional wireless sensor networks for patient monitoring and increasing efficiency of the emergency support system itself.

17.2.2 Biosensing Network

Biosensors form the basic building blocks of IoT in emergency medicine that drives the development of biocompatible composite materials and the design of reliable biosensing networks using novel composite materials (Guiseppi-Elie, 2010). In order to maximize biosensors' reliability, the sensor head should be encapsulated in a cavity to avoid interference from external sources as shown in Figure 17.3. The primary consideration here is optimizing cavity creation and the method of actuating the sensor head inside the cavity from the interaction between encapsulation and the biological parameters to be measured. In the IoT context, the wireless biosensing system consists of a network of environmental biosensors that have minimal space requirements. The embedded microelectromechanical system (MEMS) is mounted with an adequate degree of freedom to oscillate during use since it may be subject to surface stress and analyte interaction.

Some biosensors employ a biocatalyst that maps a substrate into a corresponding electrical signal for subsequent data processing. This output signal is often weak and degraded by additive noise. Wearable biosensors can effectively double in size when another adjacent sensor, with biocatalytic membrane omitted, taking a reference baseline reading for noise compensation is added to the sensor head. Given that the ambient environment can have a significant impact on biosensors' reliability (Kress-Rogers, 1996), the biosensors in a PoC environment, where many sensors operate under exothermic enzyme-catalyzed reactions, are often deployed for both patient and environmental sensing (Yakovleva, 2013). Heat sensing biosensors commonly found in emergency medicine are summarized in Table 17.1. For improved reliability, thermistors can be used to monitor any changes in ambient temperatures while measurement is taken. It is worth noting that these types of heat sensitive biosensors as a worst case scenario as other types are less prone to variations in the ambient environment.

In the ambulance setting, it is virtually impossible to control the environment where a rescue takes place, attending to an accident scene under snow will be very different from attending to a patient indoor. One major advantage of using IoT in the ambulance is the ability of supporting self-calibration (Song et al., 2014). Environmental awareness is extremely important in reliability assurance through condition-based monitoring (CBM) (Fong and Li, 2012a).

17.2.3 Hierarchical Cloud Architecture

Integrating the biosensing network into an ambulance setting, a hierarchical cloud platform model for context-aware emergency support services would be useful to control both resources and scheduling by supporting a range of context-aware IoT services in the cloud control layer (CCL) (Carvalho et al., 2017). The main advantage is to control each context-aware service for the patient and the paramedic in the user control layer (UCL). This supports interoperability irrespective of communication protocols and operating systems, the important issue of interoperability will be discussed in the next section.

Controlling access in an IoT environment is an important part in providing context-aware services (Perera et al., 2014). When a paramedic accesses a context-aware service upon arriving at the scene, an association between service and the paramedic needs to be generated, this is referred to as a service binding (Chen et al., 2016). The context-aware service binding is controlled by the control process shown in Figure 17.4 that manages access, service, resource, context, and performing binding adaptation management. The main modules of this control system are a cloud control layer (CCL) and a user control layer (UCL).

CCL is responsible for resource and service management, which consists of three main modules, namely, a cloud optimization controller COC) that controls cloud resources and service scheduling, a cloud service controller (CSC) that controls service configuration, and a cloud network controller (CNC) that controls cloud-related network information. UCL is responsible for context management and real-time service binding adaptation, which is made up of four main modules, namely, patient context controller (PCC), binding controller (BC), user network controller (UNC), and environmental device controller (ECC). The PCC controls various services for patient health context whereas the BC controls service binding adaptation according to the change of the patient health context that links directly to the hospital via the telemedicine network.

Service binding entails both application and transmission binding (Curbera et al., 2002). Application binding module represents an association among application-related objects, these cover the user interface (UI), programming language, and application protocols; transmission binding module represents an association among data transmission related objects, this is related to interconnection of devices that cover compressor, filter, modulator, queuing, caching, and transmission protocols. Service binding operates upon a binding identifier (BID) such that each object in the binding is also assigned with a system-object identifier (SID). These identifiers are used for the platform-level control as shown in Figure 17.5. Upon detecting a change of service context such as in the case of accessing EPR medical history, BC controls the service binding in conjunction with CCM, ECC, and CNC. The BC needs to initiate real-time service binding adaptation through its control agents through the application binding adaptor (A-adaptor) and transmission binding adaptor (T-adaptor). In this system, both the A and T adaptors update the system objects or protocols inside the application and transmission bindings.

In situations where high demands of network resources are needed, such as sending video images showing the extent of injury suffered by the patient and information about an accident scene, the paramedic uses video service context such that video streaming is served with adequate resolution. While it is not necessary to support near real-time video playback, the video resolution must be adequately high to show key features needed by hospital staff so that advance preparation can be made prior to the patient's arrival. The A-adaptor and T-adaptor configure image processing objects in the application binding and high-throughput transmission objects in the transmission binding. The network status has to be continuously monitored so that in the event that the network conditions degrade, the T-adaptor will attempt to maintain the connection while performing data compression through local-binding adaptation. In case the network status further deteriorates, the T-adaptor will attempt to maintain the quality of service (QoS) corresponding to the video context by searching for alternative routing path (Zhang and Ansari, 2011). Greatest challenge is posed if the T-adaptor is unable to establish an alternative route and, therefore, unable to maintain the necessary QoS for context delivery, the T-adaptor should initiate a QoS violation to the BCM, so that the CCM determines its new service context, for example, through a store-and-forward approach. The A-adaptor will store the video captured by the paramedic continuously as transmission objects in the transmission binding.

17.2.4 Weather Observation for Remote Rescue

While California battles significant widespread fires, emergency rescue operations also face challenges associated with frequent earthquakes across California and Nevada as reported by the California Earthquake Authority (CEA). Many parts of the Central America face tornadoes that often occur with rescue resource virtually having no advance warning. On the Eastern side, hurricane season in the Atlantic can stretch emergency rescue resources to their limits with millions of peoples affected across multiple states. Throughout most of the United States, ambulances must be able to obtain accurate real-time meteorological information both locally and across adjacent states for efficient and safe rescue operations.

Weather observation plays an important role in remote emergency rescue where accidents are often directly caused by adverse and abrupt change in weather conditions. IoT allows analysis of meteorological information surrounding the ambulance while attending to a remote accident scene (Mendonca et al., 2001). In the context of IoT for meteorological parameters acquisition, IEEE 802.15.4 ZigBee or Bluetooth are commonly used for low-power short-range wireless data communication between connected devices (Tung et al., 2013). Earlier work in wireless sensor networks (WSN) deployment that measure primitive parameters, such as wind speed and direction, can be measured by using a system on chip (SoC) implementation (Du, 2011) such that a microwide velocity sensor and a digital compass are embedded. IoT combines different communication networks from stationary meteorological sensors mounted on the ambulance to wearable biosensors that can be moved around by paramedics or patients deployed around an accident scene. While high-precision instruments are needed for the accurate observation of any sudden changes in weather condition that may affect the rescue operation, specific communication interfaces may be necessary for the avoidance of any interference to medical devices (Tung et al., 2014). The basic schematic is shown in Figure 17.6.

In this implementation, a pattern recognition algorithm is implemented inside the universal asynchronous receiver/transmitter (UART) module. The UART is driven by one of the internal timers, which analyzes different types of samples from either internal (for diagnostics and power management) or external (meteorological parameters such as temperature, humidity, and rainfall) sensors. All these sensors are simultaneously interconnected so that any risk that could impact a rescue operation can be detected.

17.3.1 Operational Consistency and Reliability Assurance

Connected devices can be made self-cognizant by maintaining their own database of certain system health parameters. An individual device can apply statistical modeling to predict the failure of its electronic components and subsystems so that preventive maintenance can be scheduled prior to an anticipated failure. Given that the smart ambulance environment involves many types of devices and sensors with varying life expectancies, accurate prediction of their behavior under different operating conditions would ensure continuing operability.

Operational reliability is perhaps the most important aspect of any service system as its major objective is to safeguard both paramedics and patients by providing timely and accurate information at all times. While calibration guarantees measurement accuracy for a certain period thereafter, further assessments are necessary to deduce the deviation from expected precision. In addition, any impact on measurement due to changes in environmental parameters, such as ambient temperature, humidity, shock, skin condition where sensors are placed on a patient, have to be continually assessed. Calibration is a vital process to ensure long-term operational worthiness. Various aspects of prognostics and system health management will provide solutions for determining an optimal service interval for calibration. It will also help to determine a statistical model that can be used for self-calibration, thereby eliminating the annoyance of having to send the device in for periodic calibration.

Calibration regression analysis with linear interpolation of a reference measurement with the field measurement can therefore provide a system for customized calibration (Maruo et al., 2006). Linear interpolation is used to assign laboratory values to the field measurement, between adjacent pair of the reference measurements performed on a device. Compensation for calibration errors resulting from changes in parameters, such as the user's skin temperature and time of reading taken, should also be taken into consideration. The former can be accomplished by temperature regulation in the confined area during the calibration process, while the latter can be dealt with by statistically modeling the variation pattern caused by use conditions. The known characteristics of an individual device can therefore be deduced. The accuracy will primarily be a function of the number of reference samples taken such that the number measurements is much more than the number of regression terms in the regression algorithm.

While calibration plays an important role in ensuring the trustworthiness of a medical device, continuing reliability assurance entails the development of a comprehensive set of reliability engineering modeling and analysis techniques (Fong et al., 2013). The framework is to facilitate data management and reporting for reliability centered maintenance (RCM) analysis. It will provide support for the different industry sectors and extensive customization options to fit a particular analysis approach.

An important aspect of lifetime analysis is to find a lifetime distribution that can adequately describe the ageing behavior of the device concerned. Most of the lifetimes are continuous in nature and hence many continuous life distributions have been proposed in the literature. IoT allows discrete failure data to be analyzed for the following:

• Reports on field failures are collected continually, and the observations are the number of failures, without specification of the failure times.
• A device operates in cycles and the experimenter observes the number of cycles completed successfully prior to failure. A frequently referred example is a copier whose life length would be the total number of copies it produces. Another example is the number of on/off cycles of a switch before failure occurs.
• An experimenter often discretizes or groups continuous data while attending to patient to construct a model over time.

IoT facilitates analyzing reliability or survival data set to determine which ageing class an individual device belongs to. Thus, tests of stochastic ageing play an important role in such reliability study. Connected devices continually capture data to generalize these concepts to multivariate lifetimes because a complex system usually consists of several components that are working under same environment and hence their lifetimes are generally dependent. Indeed, many such bivariate and multivariate aging concepts have already appeared in the literature for a long time. The concept of dependence permeates throughout our daily life. There are many examples of interdependence in the medicine, economic structures, and reliability engineering, to name just a few. A typical example in engineering is that all outputs from a device will depend on the inputs in a broader sense that include material, equipment, environment, and others. Moreover, the dependence is not deterministic but of stochastic nature.

Prognostics is the process of predicting the future reliability of a product by assessing the extent of deviation or degradation of a product from its expected normal operating conditions (Lau and Fong, 2011). Device or system health monitoring, just like health monitoring of a patient, is a process of measuring and recording the extent of deviation and degradation from a normal operating condition. On-going operational reliability will be assessed by computational algorithms and data collection techniques, condition-based maintenance, prognostics, and system health management for the application of in situ diagnostics and prognostics.

At the hospital end, a computer will be running the host-side software, which not only maintains an electronic patient record of each patient but also maintains record of medical devices. When the computer has connection established with the ambulance, the software will also update the data on a central medical device database similar to the case of saving individual patient's medical records for any actions to be taken such as replacing batteries or other consumables. Under normal conditions, the response center does not need to be alerted as data archival and responsive alert generations are all performed automatically. Manual intervention is only necessary for unscheduled system maintenance and update. Prognostics technology will allow the administrator to be warned in advance of any system problem prior to a complete failure so that corrective actions can be taken as appropriate.

The IoT environment enables each device to obtain information about the ambient environment that in turn facilitates the establishment of a maintenance database that is constructed using data on actual use condition of each individual device. One of the important aspects of IoT is the ability to connect devices of different types. Their operational reliability can therefore be closely monitored and at the same time the operating environment can also be monitored. A wide range of information needs to be assessed for ensuring operational reliability; these include disease surveillance-related data, community archive information in social demographic structures, Red Cross blood indicators, Internet search queries, health care facilities operation, and flow data. Data gathered from these sources need to be studied by close linkages for the purpose of compiling available census and research resources. The accuracy, reliability, and robustness of the surveillance algorithms for efficient disease monitoring will be verified by employing principled meta-analysis methods to handle multiple data sources. Another important aspect is to minimize false alarms and missing signals in these surveillance systems.

In order to support real-time device monitoring irrespective of platform, it is necessary to deploy a meta-object-based binding adaptation scheme to detect any change to a device's internal system health (Fong et al., 2012b). Meta-class defines a set of generic functions for bindings control in object-oriented (OO) classes (Costanza and Hirschfeld, 2007). A meta-object refers to the instance of the OO classes of which the methods are defined. Referring to Figures 17.5 and 17.6 that illustrate a meta-object space of both A-adaptor and T-adaptor that are utilized for their respective application and transmission binding management, the binding meta-object space can be shown as illustrated in Figure 17.7. The meta-object space contains the main features of configuration and functional meta-objects. The former detects and processes the configurations of the methods, system objects, and the connection topology inside the bindings. It manages connection from point-to-point connection to a point-to-multipoint connection of a specific binding adaptively based on the current network status. The latter detects and manages the binding's protocols and QoS of the bindings that adaptively changes the queuing policy from first input, first output (FIFO) to class-based low latency queue (CB-LLQ) based on current network status.

Before rolling out in an ambulance, it is necessary to develop a series of design study methods specific to a specific country's regulations and validate the patterns of disease models in both controlled laboratory and field experiments inside an ambulance cabin. The laboratory experiment will mainly be conducted in the controlled test chamber facility in a hospital setting that entails a class-100 clean room chamber, a controlled environmental test chamber, and a twin-room test chamber. This should then be verified in an ambulance before being taken out for rescue operations.

The disease prevalence parameters will then be aggregated into the stochastic simulation models to define a set of baseline. The calibration and validation of these models will need to be carried out prior to developing the statistical models.

17.3.2 Electronic Patient Record Retrieval in Multihop Communication

Multihop (ad hoc) communication makes use of multiple wireless hops to relay patient information from the hospital to on-scene paramedics or vice versa. Generally, multihop makes use of either multihop cellular networks or mobile ad hoc networks (MANETs) (Wang, 2008). The former is an important configuration for an ambulance traveling on rural roads. As illustrated in Figure 17.8, vehicle-to-vehicle (V2V) communication can convey data from the nearest base station to other nearby vehicles so that data communication is still possible even when the ambulance is outside cell coverage range. In this example, only Vehicle A is within the boundary of cellular coverage, data packets are relayed from one vehicle to the next with short range V2V direct communication so that the ambulance can still maintain network connection even while traveling outside cellular coverage range.

MANET is also an important topic in supporting smart ambulance in an IoT environment where a MANET on site consists of a group of mobile nodes that communicate without the presence of a wireless link to the hospital. In this arrangement, each node within the proximity of the ambulance is interconnected in a peer-to-peer (P2P) mode instead of the kind of master–slave relationship between nodes. The key feature is that communication between nodes is accomplished by either direct connection or through multiple hop relays.

EPR has been progressively replacing paper scribbled by physicians and other health professionals. In the case of retrieving EPR resources at the ambulance, it usually involves a multihop communication environment. At the same time, it should also allow simultaneous updating of the patient's health while being treated by paramedics. With the interconnection of different sensors and devices discussed earlier in this chapter, the proliferation of these IoT devices form a collection of nodes with unique IDs across different platforms and support data sharing between patient and paramedic devices. In an IoT configuration where different devices are interconnected with cross-platform compatibility, power optimization is necessary to receive and transmit data packets (Han and Ansari, 2014). Nodes in the network are often required to relaying packets destined for other nodes.

Position identification and localization of all IoT devices surrounding the ambulance can provide an efficient method of relaying information to paramedics and becomes possible to extend the coverage range through multihop communication (Abdulla et al., 2012). When a paramedic leaves the ambulance, it can be difficult to continuously track the position through both the distance and angle of the ambulance when using a GPS-enabled device alone. The paramedic cannot be detected if entered a sheltered location or carrying out underwater rescue. While acoustic positioning can provide location tracking (Mohl et al., 2001), there is still a potential issue with the tracking signal having a lower signal-to-noise ratio (SNR) than the ambient noise of the rescue area. Given that offshore rescue is beyond the scope of this text, we shall concentrate our discussion on the technology aspects of providing a reliable link for paramedics to access patient information or any additional information needed for on-scene diagnosis.

17.4.1 On-scene Diagnosis and Prognosis

When a paramedic arrives at the scene of an accident, it is likely that an initial judgment will be made from observed syndrome since most incidents are reported without accurate medical description suffered by the patient. Commonly observable syndrome such as cough and respiration anomaly can be linked to COPD or infectious diseases such as influenza (Benfield et al., 2008). As observed in the outbreaks of SARS and swine flu, a patient can exhibit multiple syndromes due to a combination of chronic and infectious diseases during pandemic outbreaks (Fong, 2011). Currently, on-scene diagnosis methodologies lack the ability to interrogate disparate data with diverse conditioned datasets from sources such as the national health system and hospital databases. In this respect, an approach that facilitates on-scene disease data collection to enable fast and reliable data-oriented disease prognosis, disease propagation analysis, and risk analysis is needed. By considering a case study involving the on-scene treatment of COPD patient, we have the following developments:

• Syndromic surveillance algorithms for analyzing public health data together with correlated indicators such as patient record data to provide accurate prognosis of respiratory diseases.
• Quantitative methods for modeling disease transmission behavior in highly infectious risk areas and stochastic simulation methods for mimicking infectious disease spreading under complex community contact structures and settings. This is a vitally important topic in ensuring the health and safety of paramedics when arriving at the scene with little knowledge of the patient.
• Aggregate the community-based disease transmission simulation models with physical disease transmission patterns to enable risk modeling, health economic analysis, and performance evaluation of disease surveillance methods and mitigation strategies under a variety of outbreak scenarios.
• Validate infectious disease models through epidemiological knowledge and experience, carefully designed clinical and field experiments, and medical records and data collected during previous pandemic and infectious disease periods.

Advances in IoT enhance algorithms and methodologies that provide quantitative disease modeling solutions to enable the preemptive detection, identification, and comprehension of respiratory diseases as well as scientific justifications for mitigation strategies. Currently, most disease outbreak surveillance methods apply standard monitoring methods used in industrial applications, such as Shewhart, CUSUM, and EWMA charts (Sparks et al., 2010), and disease propagation studies are direct extensions of over-simplified SIR models that were developed over 20 years ago and are often poor indicators of developing disease trends. Part of the challenge is the complexity of disease incident data, which is likely to be heterogeneous, multivariate, intercorrelated, of multiple data types, and often exhibits seasonal patterns over time. There are clearly substantial algorithmic obstacles to interrogating disparate datasets with such complexity or diverse conditioned datasets. Based on the current understanding of surveillance algorithms and modeling methods, an effective, reliable, and credible public health and health care disease monitoring and mitigation framework has not yet been achieved. IoT allows data acquisition from multiple sources such that analysis and modeling methods can enable reliable and data-oriented disease forecasting, disease propagation analysis, and risk analysis. This entails data collection and surveillance algorithms for monitoring multiple streams of data, including disease symptoms, correlated indicators, and incidents under various contagious disease assumptions.

17.4.2 Data Acquisition and Analytics

The smart ambulance entails a wide range of data types from multiple sources as shown in the overall IoT smart ambulance architecture summarized in Figure 17.9.

Automated time series modeling and forecasting methodologies, such as regressions, autoregressive integrated moving average (ARIMA) (Williamson and Hudson, 1999), and Holt–Winter exponential smoothing methods (Burkom et al., 2007), are common syndromic surveillance algorithms used to predict the occurrence of future health events for prognosis. For the purpose of COPD diagnosis and prognosis, a weighted CUSUM chart for detecting patterned mean shifts resulting from forecasting or feedback control can be developed (Shu et al., 2008). Optimizing the effectiveness of a multivariate control chart may relate to the correlation of variables as well as the direction and magnitude of the process shift (Han et al., 2010). Spatial scan statistics have also become popular for the evaluation of geographical disease clusters in a wide range of application areas making it particularly suited for respiratory diseases analysis (Kulldorff, 1997), thereby distinguishing from infectious diseases (Washington et al., 2004). Scan statistic-based prospective spatiotemporal surveillance methods are useful in detecting increase in incident rates in clusters of regions (Woodall et al., 2008). The main objective is to carry out a generalized likelihood ratio test that uses estimated parameters through the maximum likelihood principle. A generic framework based on likelihood ratio statistics for both spatial surveillance and spatiotemporal surveillance under independent or correlated regions is developed for this purpose, as shown in Figure 17.10.

17.4.3 Decision and Selection Process

Most basic methods such as statistical process control (SPC), regression, time series, and forecast-based methods were originally developed as temporal approaches. On the other hand, health surveillance methods such as scan statistics were originally developed as spatial approaches and later extended as temporal and spatiotemporal processes. Most spatial surveillance techniques rely on statistical clustering methods (Sonesson and Bock, 2003). The objectives in temporal surveillance and spatial clustering methods are generally different such that the former involves detection delay whereas the latter is for correct identification. When deployed in the selection process, this can be conflicting in some situations. It is, therefore, challenging to develop effective methods for timely detection with high identification rates even for standard problems with homogeneous populations and independent distributions of occurrences among individual patients. The disease incident rates are heterogeneous, correlated, and often exhibit seasonal patterns over time particularly with COPD patients. Effective selection methods under these situations thus require further discussion.

Another concern in syndromic surveillance is that it encounters a multiple testing problem. There are multiple data sources from the hospital, devices on paramedics as well as on the patient, and within each data source there are usually multiple series. Many of these series are further divided into subseries. Currently, most series are monitored in a univariate manner, after which multiple detection algorithms are applied to each series. Each of the common methods for handling multiple data streams has its limitations. Commonly used methods like Bonferroni is considered overconservative (Reiner et al., 2003), false discovery rate (FDR) corrections exhibit inherent deficiency with too few hypotheses (Benjamini and Yekutieli, 2001), and Bayesian methods are sensitive to the choice of a prior and it is unclear how to choose a prior (Robert, 2007). In helping with selection process on emergency treatment, faster detection by monitoring disease symptoms and correlated indicators in addition to monitoring observed symptoms is needed.

To eliminate symptoms due to infection, the selection process needs to quantify infection probabilities for microscale or society-scale models of infectious disease spread. Previous studies on pathogen-laden expiratory aerosols do not investigate infectious source strengths under the effects of particle size-dependent dynamics on the removal and dispersion of expiratory aerosols in an actual clinical environment. By integrating fluid dynamic modeling into infection exposure assessment, the selection process entail an accurate prediction model for infection probabilities in contained environments within the PoC to better understand the transmission mechanisms of respiratory diseases.

Most retrospective analyses of health surveillance focus on studies of clusters based on static covariates and exponential smoothing forecasting methods. In order to model discrete count data in health surveillance, it is necessary to construct a dynamic generalized linear model (DGLM) that can incorporate covariates such as subregions, days of the week, holidays, and seasonality. The DGLM accommodates hierarchical modeling of spatial data so that spatiotemporal data can be appropriately handled in a systematic framework (Banerjee et al., 2014). When modeling the relationships among heterogeneous health-related data streams, sampling frequencies are often different while frequently sampled streams may have leading power to predict the key performance indicators. The DGLM allows accurate disaggregation of space–time data so that different levels of disease-related information can be integrated and synchronized in combination with the Bayesian network model (Moauro and Savio, 2005). DGLM can include dynamic covariates for identifying dynamic clusters and their relationships during disease prognosis using data mining such as FDR and variable selection to identify clusters based on dynamic covariates. An algorithm based on a Bayesian framework to calibrate and update model parameters can be used. Predicted values will be compared to observed values, and any significant differences will be indicated in the application for detecting health anomalies.

17.4.4 Patient and the Ambient Environment

The first piece of information to gather upon arriving to the scene to carry out syndromic surveillance is to detect disease outbreaks (Chen et al., 2010) in addition to prior received information from conventional methods through reporting of the case. This can be accomplished by monitoring data that are related to the symptoms, such as influenza-like illness (ILI) symptoms or rapid analysis of the cough sound (Shin et al., 2009), recent over-the-counter (OTC) drug sales, or hospital telephone hotline calls made by the patient. By monitoring various disease-related indicators, the type of disease suffered by the patient can be diagnosed so that countermeasures will be implemented effectively and proactively, possibly even saving a trip to hospital admission given proper treatment at the scene. This entails collection of health care registration data in real time from participating hospitals. To diagnose a COPD patient through elimination of other causes, data acquisition and analytics play a vital role in supporting on-scene diagnosis and prognosis.

The sensing network that consists of paramedics and patient devices will be created as a collection of sensor nodes that sense changes in the environment. In the context of IoT, the physical parameters of the biosensor are environment dependent, as described earlier in this chapter. Deploying wireless sensor networks with specific biomaterial at the sensor head require a paramedic to identify locations that are not suitable for a particular patient affected with a common type of disorder such as in the case of asthma. The biosensor, when networked to a central server, will be vital for PoC environment assessment such that the paramedic can retrieve patient medical history as well as any allergic information in relation to the type of biomaterial to be used. The biosensor module comprises of biomaterial, a processor with on-board memory, analog-to-digital converter (ADC), control circuits, and a wireless transceiver section. IoT will enable real-time monitoring of PoC sites through the architecture in Figure 17.11.

This PoC sensing system will capture, store, and transmit the reading of the biomaterial sensor. For this system, the greatest concern in this implementation is the cost of custom-made components that may end up being expensive as they are likely to be disposable. It is, therefore, more economically desirable to use less powerful hardware on the biosensor module and have more processing done at the receiver station. Recent advances of IoT would support high-speed data transfer and computational processes to be carried out at the server end.

Simulation studies play a unique and significant role in supporting pandemic scenario prediction and enabling understanding of disease spread, which are paramount for mitigation and containment of pandemics. Some disease-spread simulation models aim to understand the effects of changes in citizen behavior or government policies. Others study disease outbreak parameters such as community, demographic, physiological, behavioral, epidemiological, and mitigation-strategy-related features (Cauchemez et al., 2008). PoC sensing provides paramedics with vital information to assess the effectiveness of a treatment plan and nonpharmaceutical interventions to protect themselves from infection while reducing the risk of infection in a hazardous environment (Nuno et al., 2008). This can be accomplished using a Susceptible–Infected–Removed (SIR) model to quantify states of human infection. The Susceptible–Exposed–Infected–Removed (SEIR) model considers the exposed state between the susceptible and infected states, and the compartmental model separates an asymptomatic state from an infectious state by a latent state.

17.5.1 Reliability

Current methods for reliability assessment of electronics-rich systems within the smart ambulance have fundamental flaws due to their inability to keep pace with new technologies. This is due to the fact that any test conducted in a controlled environment during the design phase is unable to account for complex and unpredictable usage profiles. Furthermore, it is equally important to address soft and intermittent faults that are common failure modes in many smart systems. This is especially problematic given the fact that these systems can fail at any time. Vehicle crashes caused by failed electronics, LED cabin lighting systems that cannot survive several months, a host of automotive engine controller failures, life supporting apparatus shutdowns caused by failed sensors, and patients' medical history cannot be retrieved due to failed servers. The impact on safety, availability, and cost pose tremendous challenges to the reliability of smart ambulance.

IoT utilizes connected sensors and monitors that continuously track system performance. These include system degradation assessment, fault diagnostics, and the real-time prediction of reliability through prognostics of subsystems within the smart ambulance. System health information can be gathered while the ambulance is in operation. Such information can be sent to a backend server for analysis and automated scheduling of preventive maintenance based on actual state of health (SoH) of the system.

17.5.2 Standards

The fact that there is currently no standards to ensure operational compatibility across different connected devices mean that data exchange between the paramedics and the patients may not go as smoothly as what standardized devices can offer. It would be virtually impossible to define one single set of standard for all connected devices in a smart ambulance setting given the diverse range of devices involved. There are medical devices that comply with FDA regulatory requirements, legacy instrument, consumer wearables, wireless nodes, and integrated smart home consumer electronics such as assistive elderly care systems and smart drug dispensary systems. Standardizing all these devices and systems can be extremely challenging.

17.5.3 Staff Training and Operating Procedures

Over decades, paramedics are professionally trained to strictly adhere to a set of standard operating procedures optimized for a specific country. When the ambulance goes smart, paramedics too need to be trained to adopt new technologies. It is natural to assume that these health care professionals are not trained with in-depth technical skills.

User interface (UI) must therefore be made to be suitable for the unique environment of the ambulance setting. For example, special gloves may be needed for interaction with touchscreens while these gloves must be suitable for paramedics carrying out their regular duties. Furthermore, user interaction must be minimized in order to ensure that the paramedic can concentrate on the emergency task when supported by smart technologies.

During the process of revolutionizing the way paramedics are supported, paramedics need to be trained with the necessary technical skills without affecting their regular roster for carrying out emergency support. Furthermore, it will also be necessary to aggregate the standard operating procedures with actual operating patterns to enable risk modeling, health economic analysis, and performance evaluation of surveillance methods and mitigation strategies under a variety of rescue scenarios.

17.5.4 Security and Privacy

The successful deployment of smart ambulance relies on sharing of information about the patient such as recent activities, drug consumption, and health indicators. Given that it is highly unlikely that all the necessary information for diagnosis and prognosis are stored in a stand-alone dedicated device, that is, such information is almost certainly stored in an ordinary consumer electronic device that is multipurpose by nature. There is an inherent challenge with having all the necessary health information being made available to on-scene paramedics without any consent being explicitly given by the patient (this is extremely important since the patient may not even be conscious when being attended by the paramedic). At the same time, absolutely no other kind of information can be made accessible to the paramedic or anyone at the scene.

The fact that the smart ambulance environment has so many different types of connected devices with varying connection methods pose a security challenge in ensuring that many possible points of attack within the entire network are all safeguarded.