Perspective Chapter: Edge Computing in Digital Epidemiology and Global Health

Abstract

Edge computation (EC) will be explored from the viewpoint of complex systems. An evolutionary and ecological context will be described in detail, including the subjects of epigenetics, self-domestication, attachment theory, scientific cosmology, deep learning, and other artificial intelligence issues and the role of wireless data acquisition analysis and feedback. A technical exemplar will be described and examples of potential integration with various systems such as public health and epidemiology, clinical medicine, operations, and fitness will be proposed. Also, various system vulnerabilities and failures will be discussed and policy implications in the global and clinical health and wellness domains will be identified.

1. Introduction

As with any rapidly developing technology, Edge Computation (EC) seems to offer significant benefits to humans and society but also will likely have unanticipated consequences, some of which may harm or immiserate humans and even affect environmental functionality. Within this context, this chapter will attempt to define and describe EC from a variety of perspectives including germane theoretical frameworks and issues, methodological principles, and operational/technological applications. The major substantive issues covered will include defining EC, EC and global health, epidemiological aspects of EC, EC from an ecological perspective, EC and modern evolutionary neurobiology, EC and artificial intelligence, integrated sensor/software/hardware/network development, heart rate variability(HRV), and advanced technology for psychobiological assessment and intervention, including nanomaterials, implantable technology, and bioelectric power generation. The chapter will conclude with an identification of high-impact, high-risk applications and policy directions to address such applications.

2. Toward a definition of EC

The most immediate context for the emergence of edge computation is more general computation and the information sciences. Far from the roots of the abacus, the human “computers” in use during the Babbage age, and the Bletchley mechanical computers of Turing’s time, the exponential growth of electronic information processing has been dramatic from the initial solitary transistor to more than 42 million transistors in current integrated circuits, following the prophetic Moore’s Law. This has, of course, required an equally significant development of network systems designed by electronic engineers to take advantage of this increase in potential computing power. In the 1950s, electronic engineers began using a fuzzy circular icon in their otherwise very specific circuit design schematics to indicate an important but not specified part of the system, which came to be called the “cloud.” This term has been applied as a metaphor in current usage and actually refers to the plethora of data centers using many sophisticated information processing schemes and interconnected by extensive arrays of communication links, including cable and wireless transmission modes. This has been a great opportunity for proprietary interests to market “cloud services,” while many individual and corporate users have only a vague and metaphorical awareness of the real physical constituents of the “cloud,” which otherwise may seem amorphous and impenetrable.

In truth, the cloud is a market-segmented information processing arrangement that uses the same general technology as more local computation but with a much-enlarged scope and scale. Like its less complex origins in “personal computing,” the cloud of course is subject to issues of available computing power and efficient system design. In an effort to establish a specific boundary condition in information processing, the term “edge computing” has enjoyed a similar metaphorical usage functioning as the use of any type of computer program that delivers low latency nearer to requests. Of course, an air-gapped laptop carries out this function without the benefit of network availability. MIT’s MTL Seminar, in 2015, defined edge computing broadly as all computing outside the cloud happening at the edge of the network, and more specifically in applications where real-time processing of data is required. In their definition, cloud computing operates on big data while edge computing operates on “instant data,” which is real-time data generated by sensors or users. This, of course, begs the question, where is the edge of the cloud? According to The State of the Edge report, edge computing concentrates on servers “in proximity to the last mile network.” Alex Reznik, Chair of the ETSI MEC ISG standards committee, loosely defines the term as “anything that is not a traditional data center could be the ‘edge’ to somebody.”

For our purposes, edge computation will be the use of electronic data acquisition, processing, analysis, and actionable feedback with little and intermittent assistance from larger data processing resources. This definition demetaphorizes the use of “cloud” to indicate the use of literally nonlocal system intervention, although contact with larger systems is also a potential use of EC. Later in this chapter, a number of examples of integrated data acquisition, algorithmic analysis, and feedback systems will be described, most of which do not need “cloud services” to function.

The current scale and scope of global overall computing are massive with almost 200 zettabytes predicted by 2025, less than two years hence. With the rapid growth of IoT and particularly the internet of healthy things (IoHT), limitations of centralized data center nodes will become increasingly cumbersome and even prohibitive. The growing use of personal devices including smartphones and wearable devices, as well as smart objects and secure network gateways, will catalyze more autonomous EC. Such development will require a high degree of effective privacy, security and personal data retention, and ownership practices and privileges.

3. Theoretical and conceptual issues

The radical theoretical framework adopted here is complex systems theory as explicated by Wilson [1], Wilson [2], and Capra and Luisi [3]. Examining any system from this perspective includes identifying a delimited set of agents acting within operating rules and consistent with well-studied ecological principles, which constitute a scientific cosmology and resultant worldview. The importance of viewing any system as nested and situated within a larger context is essential and forces identification of boundary conditions, such as “cloud” versus “edge” as well as operative rules followed by system agents. This aids in the clarification of interface requirements such as rules of engagement for any edge computation process. It seems likely that EC will follow the common finding of fractal self-similarity, which can aid in stable and efficient system design. Systems should be designed and maintained with sensitivity to emergent and self-organizing phenomena, as well as the frequent finding of sensitivity to initial conditions (the “butterfly flapping its wings in Rio may cause a deluge in Boston” issue) which may greatly affect output variables.

The significant role of evolutionary processes known as the completed Darwinian revolution includes not only genetic technologies such as CRISPR-CAS 9 but the role of epigenetics, attachment theory, and self-domestication, known popularly as “survival of the friendliest.” The importance and real necessity of acquiring large-scale longitudinal psychobiosocial data sets make EC ideally situated to better understand and manage these important processes since the pace of sociocultural and psychological evolution is orders of magnitude faster than classical genetic natural selection. This broadened understanding of the evolutionary process to include group selection and epigenetic modulation of methylation processes has massive implications for a variety of uses including computation-based interventions. Aside from Wilson’s revered biophilia, appreciation of both ecological and evolutionary processes needs to include ecocognosy, the term meaning the acute observation of and learning from nature. Many invaluable lessons and facts have been drawn from acute observation of nature outside the built environment. Similarly, many important scientific discoveries have occurred after “accidental” events observed by perspicacious scientists that have been integrated into canonical science. A related principle is a biomimicry, which is the imitation of natural processes. This approach operationalizes the important Hippocratic Oath, second only to “Primum Non Nocere,” which is to “follow the healing path of nature.” A timely example of this may be the functional abilities of the octopus, which has not one “central nervous system,” but a distributed distal system with each arm possessing an autonomous nervous system capable of many adaptive tasks that are only occasionally surveilled and supervised by the nervous plexus located in the octopus’ head. It is perhaps only a slight exaggeration to say the octopus has nine brains, with each tentacle included “at the edge.”

Another discipline that plays a central role in the systematic approach advocated here is epidemiology and its related health profession, global public health. The primary concern of epidemiology is the study of morbidity and mortality in specific populations, and the knowledge developed is invaluable in both disease prevention and management of disease manifestation progressing from outbreaks to epidemics to pandemics and transdemics (multiple interacting pandemics). A critical point is that epidemiology does not focus on physiological disease pathology alone but includes the psychosocial realm of dysfunctions as well, so biomedical problems such as obesity are appropriate for study, as is the occurrence of gun violence and traffic fatalities. The study of life expectancy and excessive mortality are also highly relevant areas of inquiry. As the Lancet Commission on lessons learned from the COVID-19 pandemic has noted, the development and widespread adoption of sensitive sentinel surveillance systems make effective use of epidemiological data on outbreaks usable to the global public health community. The chronic underfunding and low prioritization of both epidemiological research and public health planning and preparation may well turn out to represent existential threats, as it is common knowledge within the scientifically literate that future pandemics are certain to arise, and experience with COVID-19 demonstrates that as mutation leads to variants of concern, the next pandemic may have both greater virulence and transmissibility, requiring novel approaches to containment of outbreaks and disease management. It is of some comfort that historically, public health has shown major benefits from use of nonmedical interventions (NMI) such as improved sanitation, nutrition, air quality management, and self-management behaviors such as masking, distancing, and avoidance of crowds. Almost equally distressing is the ease with which reasonable conservative scientific pronouncements have been distorted into misinformation by politically conservative and reactionary interests, as exemplified by a recent Cochrane review of mask-wearing. The conclusions were that the studies reviewed had high risk of bias, which hampers drawing firm conclusions regarding the efficacy of mask-wearing. This was miscast by the less scientifically literate but politically astute to conclude that mask-wearing was ineffective at controlling aerosolized infectious agents. Further, a politically expedient tendency to declare “the pandemic is over” overrides scientific public health practices at great risk to society and tends to delegitimize and discredit scientific knowledge. This has led to attacks on prominent scientists and health professionals. As the technology of data acquisition, analysis, and data-based intervention continues to mature, digital epidemiology will become increasingly valuable, especially regarding wireless sensors, deep learning algorithmic analysis, and last-mile EC, which provided that the distortions caused by misinformation and disinformation are identified and discredited.

Related to these developments, science itself has recently suffered damage by failure to replicate key findings and the withdrawal of peer-reviewed studies. The epidemic of drug-related morbidity and mortality, especially related to opioids like fentanyl, has been poorly understood and framed by the scientifically illiterate as needing a renewed “war on drugs” aimed either at limiting supply, either illicit or professionally prescribed and commercially marketed, or criminalization and punishment of users. This approach has repeatedly failed since it does not address the demand side of the issue, and efforts at mandated “treatment” have shown equivocal results at best. Health literacy is often neglected and research has shown that the more negative attitudes toward science and medicine are not justified. Recent attention to terms and concepts such as polycrisis, traveler surveillance, food wastage, aridification, gender food gap, climate-inspired resilience, poverty, and zero-dose children by the World Economic Forum has been poorly understood or misunderstood. For example, zero-dose children, those that have received none of the generally recommended childhood vaccinations are commended by some ill-informed parents. Another troubling development facilitated by the prominence of social media, powered by internet availability is stochastic violence and terrorism, whereby provocative public pronouncements increase the level of perceived fear, threat, and danger and lead to incidents of aggression, while the instigators claim innocence, in that they “never directly” advocated the aggressive act. Such pronouncements have been issued even at the highest level of government responsibility—the President of the US.

The polyvagal theory proposed by Porges [4] and the neurovisceral integration model described by Thayer [5] highlight the role of the autonomic nervous system in mediating and modulating a wide variety of health-related systems including the central nervous system, cardiovascular system, the respiratory system, the digestive system, and the immune system as well as the sensory and motor components that embody these systems. They both focus particularly on heart rate variability as a key biomarker of health and various disease states. The great number of pathological states and functional indicators have been reviewed by Laborde et al. [6] and Drury et al. [7] and various metrics of HRV are described as well, including time-domain, frequency-domain, and nonlinear analyses [8]. A key conceptual component of these theories is the social engagement system, which is the basis for all attachment phenomena and sociality. The neurological CNS substrates for this system have been identified to include the orbitofrontal cortex, the fusiform gyrus, and the cingulate cortex. This system appears to be very similar to the default mode network, which is active when a person is not focused on external events. Together with the central executive network, they are perhaps the brain’s dominant control networks, crucially involved in social competence and interactional skill. Since HRV is based on easily obtainable heart rate interbeat intervals, it is an ideal candidate for wireless sensor longitudinal data acquisition and local algorithmic data processing, given the considerable power of current smart devices. This will be discussed in detail in Section 5.

4. Information science including artificial intelligence (AI)

While innovations and advancements in electronically mediated information processing have led to countless valuable applications, it has been said that there has only been one breakthrough since the term artificial intelligence was introduced by John MccCarthy in 1955: the startling arrival of deep learning. In particular, the victory of an AI-mediated deep learning program over human players of the complex board game Go in 2016 produced a shocked reaction globally. This has been deemed a “Sputnik moment” regarding its impact since this technological achievement threatens the putative superiority of humans, and technophobes fear a singularity where the aggregate of computers, robots, and nanoparticles overpower, enslave, or even eliminate humanity. Technophiles, on the other hand, foresee a future of plenitude and security with humans only engaged in work that they deem worthy and non-repetitive, boring, or dangerous. A more moderate position recognizes both the achievements in information science and the often over-hyped promises of some investigators in the algorithmic artificial intelligence field. A very promising approach advocated by Topol [9] is deep medicine, which allocates routine tasks that deep learning excels at, while focusing the provider on the more difficult and nuanced process variables such as empathic engagement. The practice of digital twinning also shows promise for efficient use of both edge and cloud resources. Notorious and highly visible debacles, such as IBM’s Watson effort at cancer intervention with MD Anderson Cancer Center are cautionary but illustrate the importance of scientific persistence and diligence in discriminating science from public relations. Indeed, this is only one illustration of the need to bolster the scientific literacy and transparency of current practice for both journalists, “media influencers,” and the general public. One of the most important critiques of AI is the anthropocentric and narcissistic identification of human intelligence as the paragon and peak of all possible types of intelligence. The search for artificial general intelligence (AGI) needs to be informed by the reality that there are many forms of intelligence that are not premised on human problem-solving ability and that human intelligence is frequently very flawed and biased. Of course, the recent emergence of both focused and active disinformation campaigns targeting not only COVID-19 issues but science in general, and issues of journal retractions and non-replicability of findings have damaged the healthy formation and dissemination of public health information, weakening the essential role of public health advocacy. A very substantive critique of AI is that if it succeeds in replacing many repetitive, boring, or dangerous jobs, there will be many displaced workers who may become part of an “unnecessary class” similar to the role of many of our elderly population. Anticipating such conflicts, there have been calls for an AI code of ethics and regulation of professionals, which extend the excellent but mainly ignored Asimov’s three laws of robots. To add to the anxiety and uncertainty of the general public, much AI research and development has been initiated by the military, often with minimal transparency and justified weakly by claims of national security and document classification. Recent developments have shown the dysfunctional nature and negative outcomes associated with overclassification of documents, often based on political expediency not national security.

5. Heart rate variability (HRV) and networked wireless sensor parameters and applications

As discussed in section three, HRV is an outstanding candidate for remote patient monitoring since increasingly unobtrusive, noninvasive, and efficient wireless sensor systems have been developed. Briefly, HRV is derived from the interbeat interval, which is defined as the period between successive R waves in the ECG signal and can be reliably and noninvasively obtained by photoplethysmography (PPG), which is included in many fitness training belts and smartwatches, as well as recently developed rings which capture interbeat data as well as three-axis accelerometer, temperature, and blood oxygen saturation data in longitudinal time series form. Further, this data can be transferred to a smartphone application that can store, analyze, and display those data to derive various HRV metrics, the most common being the root mean square of successive RR interval differences (RMSSD). Grounded in the theoretical and conceptual issues noted in section three, particularly polyvagal and neurovisceral integration theory, we have proposed deployment of the canary system, a geocoded networked wireless sensor system [10], based on existing proof-of-concept research [11] using the RMSSD HRV metric, which has been shown the ability to detect the onset of COVID-19 up to 9 days before the development of symptoms in symptomatic individuals and laboratory signs such as positive PCR results in both symptomatic and asymptomatic subjects [12]. Notably, Hirten et al. used a gradient-boosting machine-learning algorithm to detect circadian HRV variation in making the most accurate predictions. This targeted technology in response to the massive costs in mortality, morbidity, and socio-economic costs engendered by the COVID-19 pandemic, which is still producing variants of concern that may have both high transmissibility and virulence. Notably, the US DARPA has identified the important role of implantable aptamer-based biosensors to track ongoing health status of military personnel, especially in mission-sensitive settings, and has funded such development. As is typical, massive spending on military applications is rationalized as a national security priority, while the huge social and economic costs of the poorly managed COVID-19 pandemic are not identified as significant national and global security issues. In fact, concern regarding epidemics, pandemics, and transdemics is usually forgotten soon after the disturbance is deemed “over” by national governments, and funding and planning are cut or discontinued completely.

The role of the Canary System has been described above in application to the COVID-19 pandemic for several reasons. Most importantly, it is based on a rapidly scalable commercial technology of sensor devices and smartphones. Thus, its role in the detection of outbreak prevalence and spread is critical, with medical laboratory testing both expensive, time-consuming, and frequently inaccessible, while less expensive antigen tests are less reliable and subject to uneven application and reporting. The system is also well suited to the important but often neglected sentinel surveillance, which can massively improve response to outbreaks that otherwise can go undetected for weeks or months and, in fact, may facilitate original identification of new variants of concern. A recent Lancet Planetary Health recommendation [13] notes the urgency to identify “salient symptoms which need documentation of early routine evaluation of data validity, sentinel site designs and data collection methods to enable rapid implementation and analysis.” Such sentinel site designs are applicable not only to high-risk populations but specific individuals.

An example of use by individuals or small operational groups is in military settings where infectious disease is not the only risk. The Canary System can also track mobility and operational behavioral status, which can be categorized in the typical simplified military jargon of green—“good to go” or fully functional; yellow—impaired capability; and red—nonfunctional or deceased. As explored by Thayer [5] and many others, HRV is not only a health biomarker but also an indicator of positive and adaptive psychosocial functioning. HRV has been used in tracking executive CNS function and in improving stress management and resilience enhancement [14]. When combined with longitudinal temperature, blood oxygen saturation, and activity level, HRV could also constitute a routine vital sign monitoring system useful in clinical medicine for both prevention activities and evaluation of clinical status of existing patients.

6. Conclusion

While the Canary System as currently conceptualized is premised on a wearable sensor system, the rapid development of microelectronics and materials science makes other enhancements feasible. Recently, an innovative use of the popular Raspberry Pi technology for ongoing EEG monitoring was described [15] and a wearable device has been shown to be able to accomplish single-neuron CNS recording [16]. The role of mosaic RBD nanoparticles in assessment and intervention in SARS CoV-2 virology has also been explored [17]. The use of wireless data transfer and battery charging has already been accomplished and will further the development of the Canary System. Developments in molecular biology may make it possible for implanted devices to be powered by internal body chemistry, as well. A related area of significant development We have described the positive uses of wireless networked devices here, which we advocate as a dynamic element of iP4 healthcare, which is an integrative approach to health that is personalized, preventive, prescriptive, and participatory [18]. A related highly significant development is the use of genetic and epigenetic interventions in regenerative medicine, which may allow regrowing of damaged or dysfunctional organs such as teeth using native DNA [19].

It is also important, however, to be vigilant regarding misuses of such approaches. In particular, the maintenance of stringent personal data privacy and confidentiality is an issue that has been identified in other applications but would be acute in this type of application. EC must assure that local data is as secure as other settings such as “the cloud.” Also, concern with data ownership is salient, since HIPAA provides for health data accessibility but not strict ownership. In the era of increasing data monopolization and commodification by huge commercial ventures intent on profiting from ownership of the data of individuals, it would be pathological for individuals to lose their own biomedical data to commercial interests such as proprietary concerns, healthcare systems, or professional providers. Only with such protections can the potential of such applications flourish. It is also essential to recognize the crucial role played by scientific psychology since all efforts to implement sound scientific and technological innovations and quality improvements are premised on skillful use of the principles of behavior underlying human nature.