We designed an ontology to enhance the collaboration between IoT-driven patient health monitoring, clinical management of individual patients, and infectious disease surveillance. The overall goals of the ontology are to enable the sharing and integration of data collected from disparate heterogeneous surveillance systems and to support risk analysis and early warning for better patient management and triage as well as for early response to infectious disease epidemics. As requirements for the ontology to achieve the intended goals, we formulated the competency questions (CQs) described in Table 1. CQs specify functional requirements for an ontology and are used to evaluate whether the ontology fulfills the elicited requirements [38]. The use of CQs has been proposed in several ontology engineering methodologies such as the Tropos methodology [39] and the NeOn Methodology framework [40].

The IoT-MIDO was developed using the BFO as the top-level ontology. BFO is a domain-independent upper-level ontology created to provide a common top-level structure for enhancing semantic interoperability across different domain ontologies [41]. This facilitates information sharing with multiple ontologies, built upon the BFO by making data smarter by adding both computer and human-interpretable semantics to the raw data. Moreover, when developing the IoT-MIDO ontology, we reused the classes from existing BFO-based ontologies as much as possible to maximize interoperability with other BFO-based ontologies and databases that rely on them. Instead of introducing all of the classes and their definitions and properties, we demonstrate 5 use cases in the results section to show the potential usage of IoT-MIDO, using the simplified ontological models in a similar way to that done by the authors of IDODEN [34]. The complete ontology model and the definitions of all the classes are shown in ﷟Multimedia Appendices 1 and 2.

Table 2 shows the lists of ontology prefix classes that we reused in the use cases. When describing the following use cases, classes are written in italics starting with an upper case (eg, IoTStream), and associations are written in italics starting with a lower case (eg, generatedBy).

This study did not collect any primary or secondary data. Therefore, it was not necessary to obtain approval from the institutional review board. The authors, however, recognize the importance of responsible data handling. Furthermore, informed consent was not applicable, as the study did not directly involve human subjects.

We used generative artificial intelligence only for checking and improving formulations. We did not use it for creating the content of the study, including the ontology and use cases.

The first use case is remote monitoring of patients’ vital signs and other health information using IoT devices to detect their health risks proactively (Figure 1). We reused our IoT for patient health monitoring (IoT4PHM) ontology with modifications to create IoT-MIDO [61]. The IoT4PHM was built on the IoT-Stream ontology, consisting of 4 original concepts: IoTStream, StreamObservation, Analytics, and Event. In addition, the ontology is linked with 6 concepts imported from external ontologies: qoi:Quality, iot-lite:Service, sosa:Sensor, qu:QuanityKind, qu:Unit, and geo:Point. We developed the IoT4PHM by adding 3 classes to the IoT-Stream ontology: Patient, UnderlyingHealthcondition, and PatientManagement. The goal of the ontology was to facilitate data integration and sharing, knowledge representation, reasoning, and computer-assisted data analysis to enable IoT-based patient health monitoring and management for the prevention, early detection, and mitigation of patient deterioration. We first modified the IoT4PHM ontology to better conform to BFO by importing classes from BFO-based ontologies (eg, replacing the qu:QuantityKind class with the Measurement class from Experimental Factor Ontology).

The second use case is clinical patient management of infectious diseases (Figure 2). When a Person who has acquired an infection seeks health care services, they start to play the role of a Patient. The Patient receives a Diagnosis of InfectiousDisease and is classified into CaseClassification imported from the Ontology for General Medical Science [54]. The subclasses of CaseClassification are ConfirmedCase, ProbableCase, SuspectedCase, and Negative. The ConfirmedCase further has a subclass of LaboratoryConfirmedCase, a case confirmed by 1 or more of the laboratory methods that conform to the laboratory criteria included in the case definition. The Diagnosis includes information on CaseClassification. The latter only specifies whether a person has a particular infectious disease, while Diagnosis is a more comprehensive summary of patients’ medical conditions. The Diagnosis is based on the LaboratoryTest imported from the clinical LABoratory Ontology [53]. The class includes information on the type of laboratory test the Patient has undergone.

An example of a laboratory test is a nucleic acid amplification test, such as a reverse transcription-polymerase chain reaction, an antigen test, and an antibody test in the case of testing for SARS-CoV-2. When a Person is identified as a case of a particular infectious disease, they start to play the role of InfectiousAgentHost, which has 2 child classes: AsymptomaticHostRoleOfInfectiousAgent and SymptomaticHostRoleOfInfectiousAgent. In some infections such as SARS-CoV-2, a patient who never has symptoms associated with an infection (ie, an asymptomatic patient) can still transmit an infectious agent to others. The Person who plays the role of SymptomaticHostRoleOfInfectiousAgent has the developes association with the Symptom.

The Patient experiences the DiseaseProcess, which realizes the InfectiousDisease. The Vaccine administered to the Person may mitigate the DiseaseProcess. The Treatment may either provideProphylaxis to the Person or is usedForCuring the DiseaseProcess that the Person experiences. The DiseaseProcess has the resultsIn association with the DiseaseOutcome such as death, convalescence, or long-term sequelae. The Person receives Vaccination using the Vaccine, which results in having VaccinationStatus.

The third use case is epidemic risk assessment at the public health level (Figure 3). Risk assessment is essential for informing evidence-based public health decision-making about preparedness for and response to an infectious disease epidemic [62]. We thus added the RiskAssessment to handle information on risk assessment for the Epidemic regarding the InfectiousDisease. The information may include geographical scope (ie, national, subnational, and local or community level), the time of the risk assessment performed, population group (eg, general population and vulnerable population), and the person or the organization that has performed the risk assessment. Various risk-scoring criteria can be used for risk assessment. It is informative to understand which risk-scoring criteria are implemented for risk assessment. The RiskScoringCriteria handles information on the criteria’s source, version, and developer. The RiskScore represents the overall risk score for an outbreak of an infectious disease in a specific geographical area. Since risk scores may change over time, we also add the RiskScoreChange class to track the risk trajectories.

In addition, we included the RiskLevel class because risk level (eg, high, moderate, low) is often determined based on the risk score and is frequently used for risk communications (eg, risk assessment reports). Because risk scores can change over time, the risk level also changes accordingly. The RiskLevelChange is thus also included. An infectious disease epidemic’s risk depends on the infectious agent’s transmission mode. In our ontology, the InfectiousAgent is transmitted by the TransmissionProcess. For vector-borne infectious diseases, the InfectiousAgent has the InfectiousVector. Based on the risk assessment outputs, the Country may implement InfectiousDiseaseControlStrategy, which has a response level according to the ResponseMeasureIndex. The ResponseMeasureIndexChange is included to assess the changes in implemented infectious control strategies over time.

There are various ways of operationalizing risk assessment. For example, one possible way is to use risk score criteria that are proposed by the World Health Organization or the European Centre for Disease Prevention and Control [62,63] (Figure 4 [63]). Both scoring criteria output the overall risk score based on impact score and likelihood or probability scores. For the World Health Organization criteria, the impact is determined by 3 factors: vulnerability assessment, severity assessment, and coping capacity assessment. Another possible way to operationalize is to use the criteria suggested by Lesmanawati et al [64] (Figure 5 [64]). The scoring criteria of their risk analysis framework, called “EpiRisk,” provide rapid risk prediction based on country-specific risk scores computed by summing disease and country risk scores. The disease risk score has 7 parameters (ie, the type of pathogen, basic reproductive number, mode of transmission, the occurrence of asymptomatic transmission, case fatality rate, therapy or drug availability, and vaccine availability). On the other hand, the country risk score has 7 parameters, including the World Bank’s income classification, the proportion of health expenditure to the country’s GDP, the state of peace (assessed by the peace index proposed by the Peace Institute), the type of country border, population density, physician density, and hospital beds per 1000 individuals.

The fourth use case is modeling infectious disease surveillance (Figure 6). The InfectiousDiseaseSurveillance is a subclass of DiseaseSurveillance, which is a subclass of HealthSurveillance. The HealthSurveillance is performed for an InfectiousDiseaseAgent. The InfectiousDiseaseSurveillance, performed in a particular Country, collects DiseaseSurveillanceData that are reported to the DiseaseSurveillanceSystem and are used for computing the PopulationBasedStatistic based on the Population data of the Country. The DiseaseSurveillanceData has 2 subclasses: CaseBasedDiseaseSurveillanceData and AggregatedDiseaseSurveillanceData. We also included DiseaseSurveillanceDataChange for handling longitudinal surveillance data.

HealthSurveillance refers to the EpidemicThreshold, which defines the Epidemic regarding InfectiousDisease. The CaseDefinition is a set of standard criteria for identifying cases to monitor the trend of the InfectiousDisease under investigation [65]. Using uniform case definitions is essential for public health surveillance. By ensuring that every case is equivalent, the number of cases and disease incidence across different time points and geographical areas can be meaningfully compared [66]. Case definitions can vary across countries, especially at an early stage of newly emerged infectious disease outbreaks [67]. Furthermore, the definitions can be modified as new evidence on infectious diseases becomes available. When analyzing and interpreting surveillance data, it is essential to know on which case definition the diagnosis of an infectious disease is based. The number of cases needs to be interpreted with caution when a new version becomes available. The CaseDefinition may include information on the source of the case definition and its version and has 4 subclasses: laboratory criteria, clinical criteria, epidemiological criteria, and diagnostic imaging criteria.

HealthSurveillance can use the ProxyDiseaseActivityDataItem, which can include data items that may provide an earlier indication of an epidemic spread than traditional epidemiological metrics such as confirmed cases or deaths [68]. Kogan et al [68] evaluated 6 digital data sources as proxies of COVID-19 activity to detect COVID-19 outbreaks as early as possible [68]:

They found that increased digital data stream activity anticipates the increase in confirmed cases and deaths 2-3 weeks earlier than traditional surveillance methods. Although all metrics discussed in their study have limitations, the authors proposed using the combination of disparate health and behavioral data or early warning of increased COVID-19 activity [68]. We believe that those digital proxy data become an important asset for future infectious disease surveillance; thus, they are included in our ontology.

The last use case is transforming patient information into surveillance information (Figure 7). Individual patient information such as VaccinationStatus, CaseClassification, and DiseaseOutcome is included in the CaseBasedSurveillanceDataChange. The information on ContactTracing may also be included in the DiseaseSurveillanceDataChange. This contributes to transforming individual patient information into data sets that can be used for infectious disease surveillance. The rest of the parts in Figure 5 [64] have already been described in the use cases 2 and 4.

The aim of our work was to design an infectious disease ontology that can support semantic data integration from disparate heterogeneous sources, including IoT-driven patient monitoring, clinical management, and infectious disease surveillance systems, and interlinking them. For this end, we developed the IoT-MIDO ontology, and by addressing CQs and presenting 5 use cases, we demonstrated the potential of this ontology to enhance data interoperability, integration, and analysis across health care and public health domains, and to assist health care practitioners and public health decision makers in interpreting data available and their semantic relationships. The IoT-MIDO ontology can also serve as a starting point for enabling automated reasoning to drive actionable insights and informed decision-making to improve the health of individual patients as well as populations at large.

Similar to our work, the OMOP CDM can be used to support data integration from disparate sources in the context of infectious diseases. The model was developed by the Observational Health Data Sciences and Informatics as an open-source common data standard to store observational health data [73]. The OMOP CDM effectively leverages a relational database schema to represent structured tabular data. Through integration with standardized vocabularies, it facilitates data analysis across multiple data sources and meaningfully compares and reproduces results from various observational studies [73].

However, in contrast to OMOP CDM, we have chosen an ontology modeling approach for knowledge representation because of the following capabilities. Ontologies offer more comprehensive and explicit knowledge representation, enriching data with richer semantic contexts and meanings, and modeling relationships between concepts in both human- and machine-interpretable formats [74]. Their expressiveness provides capabilities for automated reasoning, inferencing, and more precise querying over data.

Furthermore, ontologies provide greater flexibility in modeling complex relationships beyond mere hierarchies, accommodating any data formats, including structured, unstructured, and semistructured data [75]. While their flexibility enables more seamless data integration, their ease of extensibility allows for adaptation to the dynamic growth of data.

Ontologies are compatible with the World Wide Web Consortium standards for the Semantic Web. The World Wide Web Consortium endorses the use of International Resource Identifiers [76] and Uniform Resource Identifiers [77] and Uniform Resource Identifiers allows data from distributed and heterogeneous systems and databases to be linked together on the Web of Linked Data. This enables the establishment of semantic links between IoT-MIDO and existing specific infectious disease ontologies such as IDODEN [34] and IDO-COVID-19 [33], addressing the major issue of siloed information across disconnected systems, which prevents a comprehensive understanding of public health data.

There are many well-established BFO-based ontologies existing, such as VIDO [35], CIDO [36] and IDO-COVID-19 [33], which are all ultimately extended from IDO [51]. IoT-MIDO adopts terms from those established ontologies, such as Infectious disease and Infectious agent, and infectious disease surveillance from IDO. This provides the possibility of easily extending IoT-MIDO further, according to the needs of ontology users to create semantic relationships with other BFO-based ontology concepts, for example, to characterize virus species and to overlapping or concurrently occurring multiple infectious disease epidemics.

The novelty and uniqueness characterizing our ontology lie in adopting concepts from IoT-Stream [44], such as IoT stream and stream observation. Incorporating these concepts, together with concepts related to infectious disease surveillance, control strategy, and response measure, which are both newly created and adopted from existing ontologies, builds a bridge to link IoT-based individual patient monitoring and early warning based on patient risk assessment to infectious disease epidemic surveillance at the public health level.

This study introduced IoT-MIDO, making a crucial initial step in addressing the long-standing issue of information silos caused by the historical segmentation between clinical medicine and public health, as well as the lack of interoperability across disparate systems. We made efforts to minimize the creation of new ontology concepts, opting instead to reuse existing domain-specific ontology concepts, particularly those based on BFO, to enhance semantic interoperability with external ontologies. Furthermore, digital data collected through IoT-driven systems provide information, including patients’ real-time physical and mental health statuses, lifestyles, and sociodemographics [78]. Such data are underutilized in the context of infectious disease surveillance, yet they can enhance existing surveillance systems by integrating with data obtained using conventional surveillance approaches. This integration can contribute to the development of more comprehensive epidemiological profiles at public health level.

However, 2 primary limitations must be acknowledged. First, the ontology has yet to undergo formalization using dedicated ontology development tools, such as Protégé. We recognize this shortfall and intend to rectify it by formalizing the ontology in the Web Ontology Language using Protégé in future iterations. Second, our evaluation of the ontology has been primarily focused on its ability to answer CQs. We acknowledge the need for iterative empirical evaluation using live data sets to assess the capabilities of the ontology to integrate individual patient and surveillance data. Such evaluations are crucial not only for validating its practical use but also for facilitating continuous enhancements to its quality over time.

These limitations highlight areas for future research and development. Moreover, several studies have proposed ontology-driven approach to ensure data compliance [79-81]. For instance, Debruyne et al [79] proposed an ontology, an extension of the provenance ontology called “PROVO,” to represent collected informed consent and its changes over time [79]. While many of existing studies predominantly focus on verifying compliance with policies regulations during the data-processing stage, their ontology model is used to assess potential compliance issues at data set creation stage before the data undergo processing.

Inspired by these studies, we plan to expand our ontology to include entities representing health data security, privacy, informed consent, and compliance with the legal requirements such as General Data Protection Regulation in future versions. Incorporating these entities is essential for safeguarding patients’ sensitive information and ensuring compliance with legal requirements and contractual agreements. Enabling compliance verification at the data set creation stage saves time and resources that would otherwise be spent on post hoc compliance analysis during data processing. In addition, it strengthens data security and privacy measures by identifying and addressing vulnerabilities or compliance gaps early in the data life cycle.

The primary aim of IoT-MIDO is to address the issue of information silos between clinical medicine and public health and enhance interoperability across different systems. Demonstrating IoT-MIDO’s capability to answer diverse CQs highlights its potential as a tool for achieving this goal. The seamless integration and sharing of clinical and public health data are particularly important during the rise of emerging infectious disease outbreaks when knowledge and evidence about infectious disease agents, transmission routes, and disease profiles are still scarce. This integration can accelerate the understanding of comprehensive and consistent epidemiological profiles and facilitate the effective and timely planning and execution of preventive and control measures.

As health data volume and complexity grow with increasing IoT applications in health care, standardizing such disparate heterogeneous data is essential for the effective and efficient use and integration of such data in infectious disease surveillance. Ontologies such as IoT-MIDO help systematically represent knowledge to make disparate heterogeneous data ready for integration and comparable for further analyses, enabling informed and proactive interventions at both individual and population levels. IoT-MIDO’s interoperability with existing ontologies, especially BFO-based ones like IDO, also facilitates seamless data exchange and the creation of comprehensive surveillance ecosystems.

Moreover, while the ontology standardizes the way of data sharing, it also provides flexibility to meet local needs, for example, by adding country-specific concepts and reference data such as codes while semantically linking them to global concepts or reference data.

Looking ahead, IoT-MIDO could underpin advanced decision support systems and predictive analytics tools in health care. These systems, leveraging computational semantic and logical reasoning, can assist in accurate diagnoses and optimal treatment decisions while enabling proactive surveillance and early detection of public health threats.

IoT-MIDO contributes to using IoT technologies to enhance health care delivery and modernize infectious disease surveillance by providing means to bridge individual patient data with epidemiological data. By enabling the integration of these data from multiple domains, the ontology enhances interoperability, thereby advancing the use of IoT technologies in providing personalized preventive measures and care for individual patients as well as improving preparedness and response to infectious disease epidemics.