1 Introduction
Epidemiological surveillance is an essential tool for detecting, monitoring and controlling the spread of infectious diseases [1]. It plays a fundamental role in public health by enabling the collection, analysis and dissemination of information on the occurrence and trends of diseases.
The increase in human mobility has considerably amplified epidemiological risks [2]. The contemporary world is highly interconnected, as evidenced by extensive and diversified transportation networks (land, air and sea), which continue to expand in reach, speed and volume of passengers and goods [2]. Recent research indicates that human movement plays an important role in the dissemination of infectious diseases [3–7]. A notable example is the recent COVID-19 pandemic, marked by the emergence and rapid global spread of successive SARS-CoV-2 variants. In Brazil, the emerging virus spread rapidly throughout the country, reaching remote regions far from the major urban centres in a few weeks [8]. This scenario highlights the need for accurate and timely epidemiological surveillance to guide effective public health responses. Thus, the development of systems focused on early warning tools is a fundamental requirement to improve pandemic preparedness and response [9,10].
In recent years, data science has played a crucial role in improving surveillance methods and systems [11,12]. There has been a rapid increase in the availability of large health-related datasets that include, for example, clinical diagnostics, molecular characterization of pathogens, healthcare attendance data and social media publications [11]. Consequently, methods for analysing diverse data types have evolved, shifting from traditional techniques based on control charts to advanced artificial intelligence algorithms [13,14]. Various methods can be applied to collect and analyse real-world temporal data and alert authorities about diseases as early as possible, enabling evidence-based decision-making and risk assessment [15]. However, some classical methods designed to detect temporal outbreak do not adequately integrate spatial data [16], limiting their ability to assess disease spread. For instance, partial differential equation (PDE) models often rely solely on diffusive terms to represent mobility, which oversimplifies spatial dynamics [17].
Public health authorities routinely monitor trends in disease incidence and public health events linked to emerging diseases. This is typically achieved through sentinel surveillance networks, which consist of selected healthcare units that monitor or test specific diseases more intensively [18]. By prioritizing locations where diseases are particularly relevant, these networks provide precise data and enable early interventions. The task of choosing specific regions and healthcare units is a key step to set up a sentinel surveillance network [18–22]. In Europe, North America, Brazil and in many other countries [23–27], sentinel surveillance networks have been essential in detecting emerging viruses, guiding the adaptation of seasonal influenza vaccines and monitoring endemic respiratory diseases.
Despite these efforts and successes, ongoing research is necessary to address persistent challenges in current systems, such as sparse geographical coverage of sentinel units, suboptimal locations, low-quality datasets and inadequate infrastructure [7,18,28]. Methodologically, the study of infectious disease spread, primarily driven by human interactions, has increasingly incorporated ideas from network science, metapopulation models [7,21,29,30], and the inclusion of long displacement terms in PDEs [31]. In particular, this work aligns with the objectives of the Alert-Early System of Outbreaks with Pandemic Potential (ÆSOP) initiative [10,32,33], which has already made relevant contributions to the mobility influence on disease spread and sentinel network design [7].
Most existing methods for selecting sentinel units rely on modelling infectious disease spread through human interactions [7,21,29,30]. A natural way to modelling these interactions is through mobility networks, where nodes represent regions and edges represent interactions, usually weighted by the number of people moving between regions [34]. Topological properties and key network metrics can be extracted and used to inform sentinel site selection. Additionally, mobility networks can structure metapopulation models to estimate each region’s contribution to new infections or healthcare visits in other regions [6], aiding sentinel unit selection by assessing disease exportation risk.
