Public health prioritizes community medical conditions and population health factors. Promoting population health and preventing disease outbreaks and epidemics are the main goals. Targeting populations based on territorial factors, socio-economic and environmental determinants, and phenotypic profiles is essential for developing precise preventive or health promotion measures. Digital Twins (DTs) technology enables data acquisition, hypothesis generation, and in-silico experiments and comparisons. Thanks to Internet of Things and Artificial Intelligence, digital twins can collect a wider range of real-time data from various sources in addition to traditional data sources like Electronic Health Records. Thus, comprehensive simulations of physical entities, their functionality, and their evolution can be created and maintained. This position paper proposes using DT technology, Public Health instruments, knowledge graphs, and AI to enable Precision and Predictive Public Health for population health. In particular, it introduces Neuro-symbolic DTs, which combine semantic reasoning supported by a knowledge graph, deep-learning's predictive power, and a DT's agility to simulate public health interventions in a virtual environment.
analytical techniques in order to expedite the resolution of pressing public health dilemmas
worldwide and solve global public health issues eficiently. While predictive modelling is to
identify the likelihood of future events, in public health, advanced predictive models are now
utilized to anticipate health events and screen high-risk individuals [
We hypothesize that it is possible to go far beyond the current practices and achieve better
results in terms of public health studies impact thanks to the exploitation of Digital Twins (DTs)
technology that will be used to ingest and process near real-time heterogeneous health related
data. More importantly, the use of prior knowledge encoded into knowledge graphs [
The ultimate goal of relying on DTs enabled by KGs will be to be able to target populations
according to territorial factors, socio-economic and environmental determinants. It will also be
phenotypic profiling in order to develop specific (or precision) preventive or health promotion
measures. Establishing a virtual system for managing disease outbreaks is a vital research
opportunity. For instance, Deren et al. have suggested an integrated system as a smart city
component, drawing from their experience with COVID-19 in China. The model utilizes
multiple elements, such as a spatio-temporal patient database, cloud computing, and AI location
technology, to efectively respond to outbreaks [
The overall framework of the neuro-symbolic DTs, which enables precision and predictive public health, is depicted in figure 1. It comprises three main components: a data and knowledge layer, the DTs core engine and a services layer.
Thanks to the capability of ingesting and processing in a timely manner various health
related data and non health per-se, such as socio-economic determinants of health (SDoH) [
In the context of health, DTs are virtual copies of human organs, tissues, cells, population
dynamics, or micro-environments that adapt to online data and predict the future of the
corresponding physical entity [
Due to the ability to understand complex structures and engage in contextual reasoning using specific knowledge, Neuro-symbolic AI [ 9, 10] paradigm could be a valuable technology for
DTs. This capability facilitates comprehension of human language in DT contexts. The merging of neural networks and symbolic reasoning in neuro-symbolic AI provides an advantageous combination, empowering complex systems to incorporate domain knowledge and execute complex reasoning and decision-making based on said knowledge.
The three main components of the framework are depicted in figure 1.
The core symbolic DTs engine. The Symbolic DTs engine, represented in the central panel of the figure, constitutes the core component of the framework. It includes a regularly updated repository of DTs, a set of tools for On-demand Twins designing, components for multimodal interaction (voice, text) with the DTs, etc. The engine is aware of the available DTs and is able to identify and suggest a suitable one to reuse or extend for a particular use case or research question.
The data and knowledge source layer. It is represented on the left panel of figure 1. In that part, there are the following data or knowledge sources: (i) the so-called non-traditional health data per-se are depicted at the top (Multimodal Data), such as that generated by social networks and online activities, wearables and sensors and data from satellites or mobile geolocalized devices (e.g. Call Data Records [11]); (ii) traditional health data that are used for public health activities, such as EHRs and clinical data warehouses that could be made available through platforms such as i2b2 [12], or, at a European level for instance, the European Health Data Space (EHDS) infrastructure for the secondary use of health data [13] ; (iii) knowledge generated either by Semantic Web technology [14] and the Linked Open Data cloud, or by standard health related coding systems (e.g., the International Classification of Diseases from the World Health Organization) and ad-hocs Knowledge Graphs developed for a specific purpose. The services layer. The third main component of the framework, the services layer, is represented on the right part of figure 1. It is dedicated to providing a set of services towards diferent end-users of the framework. Four groups of end-users are identified: (i) DTs experts, (ii) patients, (iii) citizen or public and (iv) clinical experts or research scientists.
Digital Twins expert will be able to design and instantiate new DTs. Patients will be able to interact with their twins, either for giving (treatment) feedbacks or user profiling [ 15] and personalized coaching. Citizens will be subject to health promotion campaigns, for instance. Finally, a clinical expert or researcher in public health will be able to test hypotheses or simulate some efects of interventions [16].
Neuro-symbolic DTs ofer a promising avenue for the advancement of public health research, allowing for a more efective implementation of digital solutions to various crucial concerns. These DTs have the potential to substantially improve pharmacovigilance and pharmaco-epidemiology including post-market drug monitoring, assessment of a treatment eficacy, but also diseases phenotyping, and other areas of public health research. In addition, the incorporation of neuro-symbolic DTs can improve health literacy among citizens [17]and promote data-driven decision-making through real-time reporting.
However, the adoption of neuro-symbolic DTs presents significant concerns and obstacles that require cautious considerations. An important aspect is the interaction between end-users and DTs, taking into account their diverse characteristics, which include expert designers, laypersons, and researchers. It is essential to investigate how natural language processing technologies, such as prompt engineering, text, and speech, can be utilized to facilitate seamless interaction with DTs for various user groups.
In addition, when utilizing AI-based prediction models, it is crucial to address the issue of supplying customized explanations for the outcomes and results generated by DTs. In public health applications, ensuring that users comprehend the reasoning behind the DT’s predictions can increase transparency, trust, and acceptance of the technology.
It would also be interesting to investigate methods for empowering patients to actively contribute additional data to the DT system. Patients’ participation in supplying the system with pertinent data has the potential to improve the accuracy and applicability of diagnostic tests in public health settings.
Ethical considerations also play a significant role in the public health planning application of digital twins technology. Identifying and addressing crucial ethical issues, such as equal access and averting bias, is of the utmost importance. It is essential to ensure that neuro-symbolic DTs are accessible securely to all individuals and communities, while mitigating both the risk of perpetuating inequalities and privacy violation.
In conclusion, neuro-symbolic digital twins hold a lot of promise for advancing public health research by allowing for a more eficient evaluation of a variety of crucial questions. To fully harness the potential of these technologies for the advancement of public health, it is essential to address challenges associated with user interaction, explanation of results, patient empowerment, and ethical considerations. Future research should concentrate on creating userfriendly interfaces, refining explanation techniques, examining patient engagement strategies, and establishing comprehensive ethical frameworks for the responsible use of neuro-symbolic DTs in public health planning.
The authors would like to thank the Agence Universitaire de la Francophonie (AUF) which partly funded this study, as part of the SAMIA project, agreement OPE 2023-0038.