The FRMO aims to provide a model representing knowledge in the domain of fall risk management. The scope of this ontology includes the identification of individuals at risk of falling and fall prevention strategies, as well as information about the consequences of a fall.

The step (1) was divided into two stages. First, the ontology’s domain was defined by answering questions such as “What is the ontology’s domain?,” “What is the purpose of the FRMO?,” and “Who will use it and maintain it?” Second, its scope was defined by defining a set of competency questions that an ontology-based knowledge base is expected to be able to answer. Competence questions can be used to express the requirements (expressiveness) for an ontology and to informally justify the need to develop the ontology [17]. Furthermore, the FRMO should provide answers to questions related to fall risk, such as “What factors cause falls?” Later, these questions were used as tests to determine whether the FRMO provides adequate information that can assist in fall management.

Upon step (2) searching the ontology digital search engines, The search did not yield any ontology related to fall risk management. However, some ontologies that have defined terms related to fall risk management have been integrated with the FRMO, such as the Human Disease Ontology (DOID), Symptom Ontology (SYMP), and Human Developmental Anatomy Ontology (TEMP). The search of the electronic libraries revealed several keywords related to the topic, including “fall”, “fall risk factors,” “fall risk assessment tool,” “fall prevention,” “fall risk management,” “fall risk scale,” “fall risk controlling,” “fall injuries”, “fall consequences,” “fall risk factors,” “fall and older people,” and “fall risk predicting.”

In step (3), the literature was reviewed to extract terms and concepts to achieve two aims. Our primary aim was to identify the key terms and concepts related to fall risk management. Our secondary aim was to provide a better understanding of those concepts. Significant concepts and terms related to fall risk management knowledge were extracted from the literature and coded in the ontology editor Protégé. The extracted nouns were divided into attributes, instances, and concepts. The main objective of the FRMO is to manage fall risk through concepts related to fall risk management, including fall risk factors for predicting falls before their occurrence; concepts and entities related to preventing falls; and the consequences of falls. After collecting all the relevant literature, we selected the most appropriate articles. Of the 1,146 articles collected, 112 articles remained after applying the inclusion and exclusion criteria (Fig. 2).

Flowchart of the modified PRISMA [15]

In step (4), the classes of the ontology were derived from the extracted nouns, while the hierarchy of the FRMO classes was based on the literature, with additional input from biomedical websites and services to provide more accuracy for the hierarchy of medicines and medical classes. To classify the medicines, first, the medicines leading to falls were identified. Then, the Drug Information Portal [18] was searched for their classifications and synonyms. To learn more about each drug, the search engine on the Therapeutic Goods Administration (TGA) page was used [19]; this search engine for medicines in Australia helped us to provide more information about the medicine and to classify it. The hierarchy of subclasses for the Medical class (Diagnoses) was determined through the Unified Medical Language System (UMLS) Metathesaurus Browser [20]. Each class contained annotations when applicable, such as a label, definition, synonyms, SNOMED_CT, and resources. A definition was added to classes whose purpose was unclear.

The FRMO consists of 890 classes arranged in a hierarchical structure, including six top-level classes. The six top-level classes are Consequences of Falls (an individual’s injuries and complications following a fall, including death), Fall Risk Assessment (fall risk assessments are used to determine if an individual is at low, moderate, or high risk of falling), Patients (any person who may suffer falls, be prone to falls, or be at risk of falling), Prevention (a fall prevention strategy is anything that can help prevent or reduce falls among vulnerable individuals), Risk Factors (risk factors for falls are parameters that increase an individual’s risk of falling), and Affected Body Parts (parts of the body that are injured when an individual falls).

A class hierarchy framework is not sufficient to address competency-related questions. Such questions require the definition of the relationships between the classes and between individuals. Step (5) The use of object properties enables classes to be related to other classes, and individuals to other individuals. For instance, the object property “has Risk Factors” connects the class “Patients” to the class “Risk Factors.” Ontologies also include data type properties, which describe the relationship between individuals and data types. As an illustration, “First Name” has the string data type. Determining the value type of the data type properties is known as defining the facets, which is the sixth step in the Stanford approach.

Consistency among the classes and rules governing the ontology allows software to reason about the ontology [21]. The ontology’s completeness reflects how well a particular domain is covered [22]. In addition, an ontology is accurate when it describes classes, properties, and individuals correctly [22]. There are several subclasses for each top-level class, and each of these subclasses may further have its own subclasses. Furthermore, we developed a total of 43 object properties to link classes with domains and ranges. Some of these object properties are inverses of others. One example of a resultant object property is “hasFallRisk” and “hasPatients.” In this case, the property “hasFallRisk” links the individual “Patients” to the individual “RiskFactor.” For example, patient A has the risk factor “muscle weakness”. On the other hand, the category “muscle weakness” includes patient A. Moreover, “hasFallRisk” and “hasPatients” are inversely related (Fig. 3). In addition, we added 28 data properties to the FRMO to link values to certain classes, such as the data property ‘PatientFirstName,’ which is linked to the class ‘Patients’ (Fig. 3).

Screenshots of the FRMO within the Protégé ontology editor showing the hierarchy of a object properties and b data properties as defined in the ontology

Clarity in an ontology means that it should convey the meaning of its defined classes and entities clearly [23]. As part of step (6), we added classes within the FRMO that correspond to BMI classes (High BMI, Normal BMI, Low BMI) and age groups (e.g., 100 age group). To clarify, the High BMI class is equivalent to the Patient class and the score for the High BMI class (Fig. 4) [24, 25].

Screenshot from the Protégé ontology editor illustrating the logical conditions defining the ‘High BMI’ class within the FRMO

Pellet, the first OWL 2 reasoner that comes with the Protégé editor, was used during the process of implementation to verify and evaluate the consistency of the relationships between classes, subclasses, and instances, as well as the class–instance relationships. Pellet allows ontology engineers to debug ontologies, evaluate instances, and define data types. In addition, it provides a range of features such as query response, rule reasoning support, axiom formulation, and model validation [26]. Pellet’s final results showed that the FRMO is consistent, indicating that every class, object property, and data property in the ontology has been assigned to its correct class and follows the correct rules. The reasoner also displayed the inferred classes, properties, and data properties, if any, following the rules of the FRMO. For example, Pellet inferred patients’ BMI subclasses and age groups when they were entered into the ontology as individuals.

For further evaluation, we used OntOlogy Pitfall Scanner! (OOPS! ), which is a web application that assists in identifying ontology development errors [27]. OOPS! detects common pitfalls in ontology development, such as wrong inferences and issues related to ontology clarity and real-world modeling. OOPS! checks a total of 41 ontology pitfalls in two categories: dimensions and evaluation criteria. Dimensions include structural dimensions (wrong inference, no inference, ontology language, and modeling decisions), functional dimensions (real-world modeling or common sense, completeness of requirements, and application context), and usability-profiling dimensions (ontology understanding, ontology clarity, and ontology metadata). The evaluation criteria category includes consistency, completeness, and conciseness [27].

The OOPS! framework classifies pitfalls into three categories: minor, important, and critical. There were three minor pitfalls and one important pitfall in the FRMO. The first minor pitfall was that 41 object properties did not have inverses. In ontologies, however, not all object properties must have inverse functionality; therefore, this pitfall is not a problem (Fig. 5). The second minor pitfall occurred when the word “other” was used (e.g., OtherLocation, Other_Shoe_Style, and Other_Direction_of_Fall). The word “other” was used in case any related information was not included in the ontology. The third minor pitfall was that the name of the ontology elements differed from the naming convention as some classes’ names did not include “_”; however, this will not affect the ontology (Fig. 5). The important pitfall was that the ontology metadata failed to include the license information for the FRMO. To prevent license conflicts, ontologies should be made public, meaning that they should be released without restrictions. This issue was solved by adding a Creative Commons license (http://creativecommons.org/license/zero) in the ontology header [28]. We created a new annotation property, “dc: rights,” and then used it to include the license in an ontology header.

OOPS! evaluation for FRMO

We conducted a manual evaluation of the FRMO to assess its completeness, accuracy, and clarity. Following feedback on completeness from an expert nurse, we expanded the ontology to include two new terms. We added the “Vertigo” class as a subclass of “Dizziness” and the “Unexpected_Death” class as a subclass of “Consequences of Falls.” The expert nurse, with profound knowledge of fall risk factors, prevention strategies, and consequences within the field of fall risk management, provided crucial insights in affirming the completeness and applicability of the ontology’s terminology. As she was not a member of the development team, her assessment offered an unbiased perspective on the ontology’s accuracy and clarity.

As part of the FRMO, 869 terms related to fall risk management are used, including 39 terms related to affected body parts, 81 terms related to fall consequences, 68 terms describing fall characteristics, 4 terms defining fall risk assessment, 8 terms related to patients, 46 terms referring to prevention, and 623 terms related to risk factors. The study [29] provides general information about risk factors such as muscle weakness, whereas the FRMO provides more specific terms related to muscle weakness, such as leg weakness [30]. In addition, [29] targeted inpatients older than 64 years of age, while the FRMO targets patients older than 44 years regardless of whether they are inpatients or not.

Thereafter, we confirmed the accuracy of the FRMO. Finally, the FRMO was assessed by evaluating its ability to answer the competence questions. The competence questions were represented as queries using SPARQL to extract data from the FRMO. Several SPARQL requests were applied to the FRMO, such as “List all environmental risk factors among the extrinsic risk factors”, “List all risk factors for falls affecting the head in the 70 age group”, and “List the history of falls of patients associated with Rheumatoid Arthritis”. The SPARQL codes can be found online (https://github.com/ftuhaifa/SPARQL-for-FARMO). SPARQL for Competence question 1 “What are the factors that cause falls?” was as follows in Algorithm 1:

We found that the FRMO produced the correct answer for each competence question based on the correlation between the original data and extracted data. The final version of the developed FRMO can be found online (https://bioportal.bioontology.org/ontologies/FRMO).