How Big Data Analytics and AI can improve Public Health Policies

Public health policies can be defined as a set of laws, actions, and decisions, implemented by authorities at an international, national or local level, that aim to the satisfaction of specific health goals and the promotion of common wellness.[1] Owing to the spread of COVID-19 pandemic, our everyday routine is bombarded by guidelines and instructions, deriving from policies, that are directed towards the containment of the coronavirus. In this context, the term ‘health policy’ has gained its deserving glory. Public health policies have been guiding our wellness and healthcare decisions for many years. Since the public health act of 1848,[2] which is considered as the first organized attempt to commit to specified actions for the promotion of proactive public health in England and Wales, several documents have been issued with common objectives. Initially driven by the eminence of well-respected professionals in the domain of healthcare, isolated opinions, and sporadic reports, the public health policies lacked in the establishment of a standardized formulation procedure.[3] In modern years, all stakeholders agree that this standardized procedure should be based on the evidence that is generated from policy-related research and the application of evidence-based medicine (EBM). Evidence-based medicine has a direct connection to evidence-based health policymaking (EBHPM) since it is utilized as a pattern to formulate the latter.[4] Medicine practitioners and scientists have learned to leverage scientific information, own experiences, and values to create evidence for their clinical decisions, a fact that adds transparency and trustworthiness to the act of providing quality healthcare services. 

In parallel with the evolution of health policymaking and its transition from the eminence to the evidence-based era, a different transformation is occurring in the information and communication technology (ICT) domain. Numerous technological advancements regarding hardware, storage capacity, computational, and communication speed have affected all aspects of our everyday life.[5] The same holds for healthcare, since technologies such 4G communications, 3D printing, Robotics, Augmented and Virtual Reality, the Internet of Things, Blockchain, AI, and Machine Learning are changing traditional healthcare procedures to their digital analog with overwhelming results to the provided services. 

Even though gaps between health research evidence and public health policies are still documented in various countries of the world irrespectively of their financial status,[6] the belief that policymaking should be based on scientific evidence is universal and guides the whole process of its issuance. In ‘grosso modo’, the procedure of designing a health policy consists of four phases. Initially, the planning phase is responsible for the definition of specific health goals within a physical environment of financial, geophysical, social, and political constraints. Consequently, the assessment phase takes place which accounts for the evaluation of the effects of the policy on the health of the targeted population. After the assessment phase, the implementation phase follows where the definition and creation of appropriate tools for the realization of the designated objectives occurs. Finally, the monitoring phase takes place to control and verify whether the assumed laws, actions and decisions achieve the health goals.[7]

Evidence Based Public Health Policies 

Among the basic principles for this standardized procedure, the access, analysis, and management of the scientific information to create a well-targeted policy are highly appreciated. Policymakers should have the ability to review the current data, understand and analyze them, to balance competing values fairly in order to draw objective conclusions. To a certain degree, policy-making refers to decision-making based on a-priori information, which is an ideal scenario for artificial intelligence (AI) to kick in. Able to harness the massive amount of data in order to approximate a non-linear imaginary function that can efficiently map the input (a-priori information) to the output (specified conclusion), machine learning (ML) and deep learning (DL) algorithms have made their appearance to provide a helping hand to the respective stakeholders. 

One does not have to be a computer expert to know that we are living in a period of vast data generation on a daily basis. Indicative of this immense production is the fact that, approximately, 90 percent of the world's data has been generated in the last two years (according to Forbes, calculated in 2019). Big data is the main vehicle that drove the change from the information revolution to the AI revolution. Following the same trend, health-related data are created in large volume, uncertain veracity, and high velocity and variety. The existence of these four dimensions in expanding growth provides a very difficult operational scene for the human perception to analyze and make a profit of. On the other hand, Machine Learning (ML) and Deep Learning (DL) approaches are data-driven and data-hungry mechanisms that have proven their ability to discover interconnections and patterns that are hidden in multiple dimensions. By providing quantitate assessment of these patterns, they are relieved from the subjectiveness of human nature and therefore can be efficiently utilized for the extraction of knowledge to achieve specific health goals. When these approaches are enhanced with interpretability capabilities, meaning that they can describe their inner mechanisms in a way that is understandable to humans,[8] the requirement of evidence-based AI is met and can lead to EBM[9] and EBHPM. In this context, AI can be utilized for the detection of events that affect the individual or public health, the prognosis or diagnosis of harmful health conditions, drug development, population health, organization, and patient-facing applications. Moreover, interpretability by design in healthcare decision-making systems can assist in their embrace by the related professionals and their smoother integration in the clinical workflow, since the understanding of their inner workings can make them more transparent, trustworthy and explain their failures towards further improvements. An interesting example of the application of explainable AI in the medical imaging field is the work presented in the article.[10] The authors present a convolutional neural network (CNN) methodology for the classification of 2D mammographs in a binary task (benign/malignant), while providing useful interpretations of the internal representations of the CNN that agree with the experts’ experience. An important contribution of this work is that it keeps the experts in the loop by making use of seamless interaction between the deep learning system and the humans involved. Concerning the application of DL models for public health policies, a Long Short-Term Memory (LSTM) model is utilized to predict the epidemic peaks in China and enforce health policy for large-scale quarantines before the peaks to prevent the eventual COVID19 epidemic spread.[11] The advantageous utilization of AI-based systems in public health policymaking is highlighted by the European Union as well, with the funding of the research projects that aim to integrate heterogenous health-related data from different sources for the support of policy-making decisions. An example of such an R&D project is CrowdHealth,[12] which was technically coordinated by University of Piraeus. This project developed a web-based environment (named Policy Development Toolkit - PDT) for the creation of public health policies by means of a well-defined model. The basic elements of this model are the Public Health Policies (PHPs), the Key Performance Indicators (KPIs), which arecountable and measurable indicators linked with the related PHP, and the Health Analytics Tools (HATs) that may be used to calculate or predict the policy KPIs. The latter is the element that incorporates AI in the policy-making mechanism by connecting data analytics such as risk stratification, causal analysis, and multimodal forecasting to health policies via KPIs. The overall architecture of the developed PDT is illustrated in Fig. 1. PDT enables the processing of health data through the exploitation of collective knowledge that emerges from multiple information sources. The platform explores mechanisms that can be clustered across two main areas: analytics on holistic health data at population and individual level and predefined policy models exploiting community knowledge across the complete health ecosystem. PDT may be considered as the point of integration and interaction of the platform with the policy makers. Through the PDT, the policy makers will be able to question the platform data and exploit the health analytics tools to perform policy creation and evaluation.[13]

Figure 1: The conceptual architecture of the Policy Development Toolkit (PDT) developed in the Crowdhealth project

Simulation of Health Policies Before Actual Application (i.e Digital Twins)

Another important requirement that needs to be addressed for the formulation of a public health policy is the evaluation of its effectiveness. The design of a health policy is a very complex procedure that takes place in a continuously evolving environment, characterized by great uncertainty and noise. Since the release of a policy is meant to have a profound effect on public health, it is important to make sure that the assumed actions are directed towards the tackling of the real health challenges and the satisfaction of the corresponding goals. Therefore, it is considered a best practice approach to utilize protected ‘playgrounds’ for the simulation of real-world ‘what-if’ scenarios to verify the results and avoid side effects. The creation of simulated environments for predictive solving problems in dynamic systems is not a newly breed idea. Utilized as a test-bench in various engineering for the prevention of potential failures, its application on the evaluation of potential effects of a health-related policy can be beneficial. Towards this notion, the important enabling technologies are IoT and 4G/5G connectivity. The burgeoning advancements in the fields of smart biosensors, wearables, inhalers, dispensers, and injectable devices along with the digitization of health and administrative services provide all the necessary information for the construction of a digital twin of a system, environment, or procedure. The value of this data is multiplied by the fact that all these devices and structures are interconnected, thus, forming, networks of health entities that can interact and exchange important information. Once the digital twin is formed, it can be utilized along with AI predictive models for the prediction of potential outcomes for a public health policy release. The combination of forming digital twins and applying AI algorithms on them can have various applications in the healthcare domain such as developing improved drugs,[14] providing real-time analysis of the health condition of a patient[15] or population. The integration of health-related historical data and real-time information in a simulated environment and the application AI algorithms to produce a potential outcome of clinical interventions and health policies provide the basis for evidence-based medicine and policymaking. 

Monitoring of Individuals and Provision of Personalized Recommendations for Public Health

The road to public health policies passes inevitably from its core element, the human, and the promotion of individual health and well-being. The improvements in targeted against ‘one size fits all’ clinical decisions, treatment plans, and predictions of risk stratification have been pursued through the development of AI technologies that integrate seamlessly heterogeneous health data like medical images, omics, lab tests, socio-demographic particularities to better identify and treat an individual’s disease. Smart wearables, mobiles, and biosensors is a consistent ally to this effort since personal health-related data produced by such devices can further enhance the electronic health record and provide the source of meaningful patterns concerning the real-time health status, adherence to medical advice, the existence of health emergencies and the delivery of personalized consultation that is tailored made on the person’s needs. Towards this purpose, the European Commissioner for the Digital Agenda and Commission Vice-President shared her vision for the personal health navigator, which is a mobile application that acts as a coach for the engagement and support of healthy living of the individuals, while enabling data sharing and communication in a cross-border context.[16] Due to the advances in edge computing, the extraction of knowledge from these patterns can be conducted in-site (where the patient is located) by smart AI agents to reduce undesired latencies or a centralized server for analysis on a larger scale.[17] Being able to capture the variety of evidence for an individual’s health status will lead to the creation of a computation avatar that can be exploited as a simulation test bench for the prediction of pathophysiology progression and the effects of received therapeutics.[18] All the aforementioned technologies can form a collaborative ecosystem that promotes the idea of personalized medicine that, in the eye of an inattentive reader, might seem the very opposite to public health. However, this hypothesis is far from reality. From one point of view, focusing on the individual traits hidden in the connections of various omics can unveil new knowledge for a more precise treatment plan against a disease of a more effective prevention program.[19] Moreover, the application of personalized medicine requires intensified data sourcing in the sense that health professionals and, even patients to a certain extent, bear increased responsibilities as they are dynamically involved in the data gathering. The accountability in the process of data gathering contributes to public health policies with more data of better quality.[20] The opposite contribution is observed in the process of aggregating extracted knowledge from personalized medicine into larger health-related information groups by gaining insights from population-level data. This reciprocal communication highlights that personalized medicine and public health cannot be regarded as contradicting notions.

Conclusion

Artificial intelligence is drastically affecting all aspects of human life and public health bears no exception to this rule. Although the procedure of health policymaking is influenced by numerous financial, social, environmental, political, national, and foreign factors, the principal motivation should derive from the evidence that is provided from the systematic review and critical appraisal of scientific information. To support this idea, policymakers can rely on the extraction knowledge capabilities of AI algorithms, while keeping experts in the loop. The strict and blind adoption of analytics tools in the procedure of policymaking should not be deemed as a panacea since the plethora of scientific information that the models are based on containing noise and discrepancies, and uninformed ‘a-priori’ knowledge. The experts’ experience and insights are an asset that cannot be excluded from the process. Taking into consideration that ‘all models are wrong, but some can be useful’, an aphorism attributed to the British statistician George E. P. Box, the necessity of a meta-model that keeps human experts’ experience in the loop and updates the knowledge of the models in concern with all the social, environmental, epidemiological, and health-related changes is evident. Other concerns such as data biases, security and privacy, and the preservation of related ethics will continue to pose significant obstacles towards the decrease of gap between evidence-based medicine and health policymaking.