Malaria is an infectious disease with significant impact on developing countries. In 2016 alone, it caused 445,000 deaths worldwide, and globally around 216 million cases of malaria have been reported in 91 countries [1]. Populations in sub–Saharan African countries are most susceptible, with 80% of observed cases and recorded deaths worldwide [2,3]. Whereas a decrease in case incidence has been observed [1] since 2010, the rate of decline appears to have stalled, in part due to lack of adequate surveillance and intervention programs. An essential prerequisite for accelerating the decline of the disease and optimally targeting resources is an efficient surveillance infrastructure that can reliably deliver robust datasets. Poor data quality and sparseness and the coordination between different surveillance systems are ongoing challenges for the malaria surveillance community [4]. The increasing number of stakeholders, including international organizations, governments, nongovernmental organizations, and private sectors [5], that contribute to gathering the data can lead to siloed heterogeneous information systems and data sources that need to be integrated [6]. Overall, infrastructures for malaria surveillance are brittle [7], and stakeholders do not have sufficient confidence in the aggregated datasets to adopt any conclusions that could be meaningfully derived [8].

A comprehensive study [9] showed that malaria information systems that can collect, store, and analyze data and provide feedback to developers based on real-time information are few. Moreover, the most effective systems are still limited by the absence of real-time data aggregation, inconsistent decision support, and low levels of resolution, such as no mapping to households and no extrapolation across geographic borders. In specific cases where an information system [10-12] has been upgraded with improved visualization and reporting tools, other challenges remain unresolved. Data entered from field stations are managed centrally, and updates occurring at the central data source are not reflected in the field level immediately. Also, the options to generate reports are generally predetermined; hence, ad hoc queries cannot be answered without significantly overhauling the systems. Consequently, the types of surveillance queries that can be run to derive actionable knowledge in a timely manner are relatively few.

Furthermore, 11 widely used Web platforms were studied to assess how internet and Web technologies are used in the fight against malaria [13]. The elements of this study were focused on data, metadata, Web services, and categories of users. The results revealed that, although heterogeneous spatiotemporal malaria data came from multiple disciplines, they were rarely updated dynamically, no metadata were used to standardize them, Web services were inflexible for reuse and nonstandardized, and the platforms primarily served the scientific communities. The authors identified that, to improve these systems, interoperability through standardization is necessary. The World Health Organization’s global technical strategy [14] also identified that surveillance challenges in sub–Saharan African countries exist partly because mechanisms for facilitating data integration from distributed data silos and ensuring system interoperability are lacking.

Indeed, the malaria surveillance community is not alone in facing these challenges, and many other communities are investigating how to bring distributed datasets together in real time to support decision making. Researchers in other domains have sought to introduce guidelines for ensuring that source data are published in ways that ensure they are findable, accessible, interoperable, and reusable [15], albeit no specific technical solutions are proposed or mandated. In addition, standards and software prototypes have emerged specifically from the Semantic Web community [16] but, given the nature of these challenges, it is the combined benefits of applying guidelines and technical solutions that would provide a transformational development for surveillance practitioners.

The specific challenge of interoperability has two dimensions, namely structural and semantic interoperability. Structural or syntactic interoperability can be achieved by defining common syntax and formats for data exchange. For example, if two systems such as the Malaria Atlas Project [17] and Africa Development Indicators [18] are using demographic data from a census stored in another database, having interoperability between the two systems means that users of these databases can mutually access and reuse census data without having to store them locally or reformat the data on import. Semantic interoperability, on the other hand, is much harder to achieve. The goal is to ensure that the integrity and meaning of the data is preserved throughout the integration process. This can be achieved by mapping data to standardized vocabularies or terms in ontologies [19], such as the Vector Surveillance and Management Ontology (VSMO) [20], Infectious Disease Ontology-Malaria (IDOMAL) [21], or Mosquito Insecticide Resistance Ontology (MIRO) [22], which provide means to formally model a domain of interest using concepts in the domain, relations among the instances of concepts, and complex logical axioms. Two sets of data mapped to the same ontology terms or vocabularies can be regarded as having the same meanings. Ontologies facilitate semantic interoperability in integrated biomedical and health systems and can also be used to make malaria data sources comparable. Semantic interoperability achieved through mapping to community standard ontological terms is therefore an essential property for surveillance routines and enables more integrated access to data.

Overcoming the challenge of distributed data access has relied on established technologies such as Web services, but interoperability is still lacking in many implementations. In recent work, Web service-based data access and interoperability challenges have been tackled together using Semantic Web service infrastructures [23]. The benefits of this approach include enhanced findability of distributed online resources, easy construction of workflows to fetch data from multiple Web services, and access to source data in interoperable formats. Existing deployments of this approach [23,24] as part of the Semantics, Interoperability, and Evolution for Malaria Analytics (SIEMA) [23] platform have aimed to address the interoperability between multiple dynamic and evolving malaria data sources. Early trials have reported advanced query options [24] for end users without advanced technical skills. In addition, SIEMA seeks to manage infrastructure changes, where change management is defined as the process of preserving the integrity and consistency of an integrated system while keeping the information up-to-date. Here the primary activities during change management are detection, representation, validation, traceability, and rollback, as well as reproduction of the changes [25]. SIEMA has used authentic scenarios (use case) provided by the Uganda Ministry of Health to illustrate effectiveness in providing seamless access to distributed data and preservation of interoperability between online resources.

We introduce a prototype surveillance and change management platform, known as SIEMA, built from a combination of third-party tools, community-developed terminologies, and custom algorithms. We illustrate the methodology and core infrastructure used to facilitate interoperable access to distributed data sources using Semantic Automated Discovery and Integration (SADI) [26] Semantic Web services. We show a dashboard that reports on terminology changes that can render SADI services inactive, jeopardizing system interoperability, allowing end users to control and reactively rebuild services to preserve interoperability and minimize service downtime.

SADI is a representational state transfer (RESTful) Web service framework that provides a set of conventions for creating Semantic Web services. The framework uses resource description framework schema (RDF[S]) [27] and Web Ontology Language (OWL) [28] standards for data representation and modeling, and HTTP-based recommendations (GET, POST) for interacting with the services. SADI services consume and produce RDF instances of OWL classes where the input data are decorated until they become an instance of the output OWL class. The services are deployed in a registry and can be automatically discovered, orchestrated, and invoked to return data in RDF on the query clients SHARE [29] and HYDRA [30,31].

Our research focuses on middleware for enabling discovery of datasets and tools for agile query composition, but deployments of these methodologies beyond prototypes will require full access to malaria data. Existing data repositories such as the Scalable Data Integration for Disease Surveillance [32] or Global Malaria Mapper [33] would be target resources for building service descriptions and including them in a generic registry. However, a recent study [13] suggested that in order to facilitate change management and semantic interoperability these data sources must be represented by standard terminologies related to their domain. Rather than defining new terminologies, it is a best practice to use community-adopted domain terminologies, if available, for defining the services. For this purpose, several ontologies are leveraged, including VSMO [20], IDOMAL [21], MIRO [22], and the Public Health Ontology [34]. The Clinical Data Interchange Standards Consortium [35] also offers ontologies.

SHARE is a specialized open source query client that enables end users to discover, plan, and orchestrate SADI services in a registry and invoke them automatically from SPARQL [36] queries. The services are discovered by finding a match to the predicates attached to the input URIs after comparing both the semantic descriptions of the services’ input and the output OWL classes. Each triple pattern in the SPARQL query is resolved by checking the predicates in the service registry. The order of invocation of services is strategized during the planning stage. The services are then orchestrated and invoked to execute the desired tasks. The output triples are produced through the binding of subject-object values to the corresponding variables in the SPARQL query. To generate the complete output, every triple in the query is resolved. HYDRA is a commercial query client for SADI services, developed by IPSNP Computing Inc. HYDRA can be used as a Java application programming interface, as a simple command line application, or through an intelligent graphical user interface (GUI) supporting ad hoc query composition by nontechnical users by combining Google-style keyword-based querying with query graph editing. Instead of queries being written in SPARQL, the GUI allows using a graph representation that is easier to understand.

Detecting changes in the source data schema, as well as the domain and service ontologies, is a prerequisite step for change management. Studies on the evolution of large domain ontologies [37] have shown that the vocabularies are subject to change [38] because they have shared authorship, and they evolve to represent new knowledge. Because of the high degree of dependency between system components in a relational data schema, even a small change in a data schema could result in cascading impacts across an application stack. In previous work, we presented features of a Web-based analytics dashboard for detecting and reporting changes [39].

Underpinning the dashboard, software agents enable two key actions: (1) detecting changes and identifying their types, and (2) restoring the modified component to an operational state through repair and rebuilding. The types of changes we accommodate are addition (ie, extension), deletion (ie, obsoleting), and renaming (ie, refining) of components in domain ontologies and service ontologies [40,41], such as concepts, property restrictions, individuals, and axioms. Changes in the source data, such as tables and their attributes, data types, and indices, can also be reported [42]. Likewise, changes to saved queries and the corresponding service dependencies can be detected. Brenas et al [23] report further details of the agents designed to capture changes in various components of the infrastructure and their interaction with other software agents in the system.

Valet SADI [43] generates SADI services automatically based on the formal definition of a service and rules for mapping a data source to ontology terms. The knowledge required to interpret the data is expressed using rules scripted in positional-slotted object-applicative (PSOA) RuleML [44]. These rules define how each element coming from a data source has to be matched to the vocabulary defined in the ontologies. Valet SADI, with the help of the domain ontologies and the semantic mapping rules, rewrites the declarative input and output definitions of a service into a query that is specific to the source data schema, such as Structured Query Language (SQL), and creates the complete Java code implementing the service functionality. It facilitates the generation and deployment of a service in a matter of seconds, which is essential in the context of change management, where uptime is a priority.

The implementation presented in this paper is inspired by the objectives stated in the National Malaria Control Program by the Ugandan Ministry of Health [45]. The questions, shown in Textbox 1, based on these objectives resemble authentic queries for a surveillance practitioner in the malaria domain. In this research, we converted these questions into formal queries by semantically presenting them as one or more SADI services. This section illustrates how the queries can be represented using a combination of SADI services on a graphical query canvas of the HYDRA query engine.

The previous section outlined how, using Semantic Web services, it is possible to answer complex questions relevant to malaria surveillance. Special attention is needed before considering the introduction of a new methodology in a dynamic context where data and middleware are not static. Several possible changes that could occur in a malaria surveillance framework have been described and classified according to the degree to which they affect data access and their likelihood to affect interoperability of the system [23]. In the following, we illustrate an approach to address the notion of change management through the introduction of change detection tools and triggers for the reactive rebuilding of service to ensure service uptime. Specifically, we discuss the addition of terms that are used in service ontologies and the use of a reporting tool for detecting and displaying these types of changes.

The SIEMA framework comprises several technologies and standards and is further customized to address the proposed targeted needs and interests of surveillance practitioners. The contribution takes three main directions. First, using SADI Web services allows for easy access to distributed data. This task is simplified further using Valet SADI, which enables a programmer to create services in an efficient and straightforward way. Second, due to the user interface features of the HYDRA query engine, SIEMA offers end users a more appealing way to build surveillance queries. HYDRA’s ability to discover and call the services that are needed for a query permits the user to simply use the data as an abstract construct without having to look at its actual structure. Third, to make the system more robust and flexible, a dashboard has been introduced. The dashboard informs users when changes have occurred that render the services or queries inactive. This enables users to know which queries may no longer be reliable and to identify which parts of the service infrastructure must be rebuilt to restore it to its fully interoperable state. Deployed together, this combination of technologies offered by the SIEMA framework exhibits key functionalities that are of great value to the community.

Our initial studies in malaria surveillance [23,24] and other domains [30,48,49] showed us that this approach is a viable solution for enhancing interoperability. The successful extension of this methodology to the malaria use case appears promising based on the results from the initial implementation. The addition of a dashboard and Valet SADI extends the capabilities of the SADI framework to make it suitable for change detection, service restoration, and preservation of interoperability.

Whereas a systematic evaluation of the many components of this framework, individually and together, is beyond the scope of this initial study, we are aware that other malaria surveillance systems in sub–Saharan African countries have been reported and evaluated in part [12,50] according to attributes recommended by US Centers for Disease Control and Prevention [51]. Indeed, it would be valuable to make some direct system comparisons. Regrettably, few technical details of these systems have been reported, and their key features—such as their architecture, supported data types and data quality, use of data representation standards, capacity for change management, and degree of interoperability supported—have not been disclosed. Without such information, it is not possible to compare them with SIEMA. It is apparent, however, that the systems were designed for centralized data warehousing rather than seamless access to distributed data, and this is a significant distinction.

A brief assessment of SIEMA according to attributes recommended by Centers for Disease Control and Prevention [52], namely simplicity, data quality, flexibility, stability, and timeliness, can be made based on our initial experiments. To address the attribute of simplicity, the activity of creating surveillance queries can be assessed. Whereas writing syntactic queries (in SPARQL) for submission to a query client manually requires that a user have expertise with the query language, the HYDRA GUI can save users time and allow more complex surveillance queries to be composed by persons with less technical skill. Figure 5 shows a graph representation of a query built with the GUI, which is in turn translated to SPARQL automatically. To address the attribute of data quality, SIEMA’s adoption of SADI ensures that the World Wide Web Consortium standard OWL and RDF[S] are used to describe and format every piece of data accessed through the framework. The attribute of flexibility is also met, since access to distributed data can be provisioned by using SADI services that are easily deployed in a registry, making them discoverable and readily usable for a variety of ad hoc queries. Likewise, the attributes of stability and timeliness of the system are met as the implemented dashboard tracks service uptime and reports failures immediately after detecting the faults. System rebuilds with Valet SADI ensure stability so that data can continue to be provided and the system stays operational if changes occur. In this way, interoperability is preserved and any service downtime is kept to a minimum.

Overall, we anticipate that the ongoing trials with the SIEMA framework will give the research and development team further insight into real-world requirements for interoperability and change management in malaria surveillance, leading to further improvements in adaptability and performance. Given the critical need for timely integration of distributed data from multiple heterogeneous sources in an efficient way, we hope to build cooperative partnerships between multiple disciplines, organizations, and sectors. In addition, insights gained from this research are likely transferable to a range of global surveillance projects.

We have demonstrated that authentic questions asked in malaria surveillance can be formalized as queries and mapped to a combination of Semantic Web services designed to deliver target data from distributed data sources. We have shown that using SIEMA and leveraging terminologies from community-developed ontologies offer flexibility both for integrating data and for easily composing queries. The developed infrastructure also offers a solution to the problem of change management, an important process for maintaining interoperability and integrity of an integrated surveillance system. Given that changes in the form of addition, renaming, and deletion of terminologies can frequently occur in the face of evolving system requirements, we introduced a change management dashboard. This makes it possible to identify important changes, report on the status of services as a consequence of changes, and offer users the option to rebuild inactive services. The dashboard and service reauthoring routines serve as an important vehicle to maintain system interoperability of mission-critical global surveillance programs. The infrastructure has been implemented and its relevance has been demonstrated with an authentic use case, with the goal of soliciting further requirements from the malaria analytics community. In future work, we will deploy SIEMA on live dynamic data sources.