Chapter 4 A CONCEPTUAL MODEL FOR TB ADHERENCE FACTORS
4.1 Definition of Purpose
4.1.1
Determination of Modelling Scope
The disease of focus is TB and the geographical region to be considered is SSA. The motivation of selecting TB adherence in SSA has already been discussed in section 1.4.1 and section 2.1.2.
4.1.2
Design Goals for TB Adherence Factors Ontology
The main goal of designing the TB Adherence Factors Ontology is to capture explicit knowledge about factors that influence TB adherence behaviour in a manner that can support construction of a predictive model for predicting adherence risk for TB communities. This goal includes the following objectives:
To capture, consolidate and structure explicit knowledge about adherence factors as an evidence-base for decision-making in public health
To present a computational representation of adherence behaviour that can be queried, navigated and shared among experts and is understandable by machines
To develop a knowledge base for determining the community-specific adherence factors and to support the predictive model’s construction
To capture findings from the current and future scientific publications on adherence behaviour.
4.1.3
Evaluation Criteria for TB Adherence Factors Ontology
The ontology should have the following characteristics in order to validate that it achieves the above set design objectives:
It should be comprehensive, consistent, clear and unambiguous
It should allow categorisation of factors and represent their effect on adherence behaviour
It should permit structuring, curating and exposing scientific knowledge emanating from clinical studies about factors that influence adherence
It should be a representation that is understandable by machines and can form the basis of a shared, computer-based knowledge repository for treatment adherence behaviour
It should be able to be extended, navigated and queried, and be useful for computer-based prediction
Lastly, it should enable the linking of factors to clinical studies that provide evidence for their predictive value
4.1.3.1 Competence Questions
Three sets of CQs are proposed to evaluate the TB adherence factors ontology. The first set is to determine the possibility of generating output from the ontology. This is to evaluate the capability of the ontology to produce results about adherence behaviour when queried. The second set is to identify the parameters that can be used to query the ontology successfully. This is to evaluate the usefulness of the adherence factor categories for querying and navigating the ontology. The last set is to evaluate the usefulness of the outputs from the ontology for constructing a predictive model. The CQs are:
CQ1: What are the possible outputs that can be derived from the TB adherence factor ontology?
o CQ1a: Is it possible to search the ontology for factors that influence specific TB communities in sub-Saharan Africa?
o CQ1b: Is it possible to search the ontology for evidence that asserts specified factors that influence the adherence behaviour of TB patients?
o CQ1c: Can the ontology provide location information about the influence type, influence period and interrelationship between two or more factors?
CQ2: What are the categorisation dimensions that can be used as parameters to query the ontology?
o CQ2a: Can the ontology be queried using a combination of some or all of the dimensions of the influencing factors?
o CQ2b: Can the ontology be queried using the community characteristics as the only query parameter?
o CQ2c: Can the ontology be queried using any of the influencing factors’ categories and properties as the only query parameter?
o CQ2d: Can the ontology be queried using the evidence characteristics as the only query parameter?
o CQ2e: Can the ontology be queried using publication characteristics as the only query parameter?
CQ3: Is the ontology useful for predictive model construction?
o CQ3a: Can the ontology be used to generate the variables and states for a BN model?
o CQ3b: Can the ontology be used to generate the probability tables for a BN model?
o CQ3c: Can the ontology be used to generate the BN model structure?
4.1.4
Use Case Description
Two types of user groups, community and global users, are identified to test the ontology with the CQs stated above. The main concern of both groups of users is to identify community-specific influencing factors. Additionally, global users are concerned with knowing broader factors pertaining to multiple communities in a region or country. Typical examples of such users are:
A TB programme officer (community user) is planning a new intervention plan for her community. The data he/she collected at the point of care show that there is a high number of defaulters in the communities but she does not understand the reason for this high rate. Thus, she wishes to identify a list of potential factors that are influencing TB patients in her
community. He/she requires this knowledge for the development of a proper community intervention plan that will reduce the rate of treatment defaulting in the community.
Additionally, he/she wishes to develop a predictive model that can help in predicting which community or individual is as risk of poor adherence.
A TB researcher (global user) is saddled with the task of understanding the most common factors for certain countries in SSA. This information will help in her proposal for an alternative treatment plan for TB endemic communities in the region. Hence, she wishes to identify factors that have been established as risk determinants in specific communities of interest and the type of influence they have on patients. She also wishes to identify the existing scientific studies that have identified these factors.
The users described above represent groups of users who can use the ontology to support their tasks.
Based on the level of their knowledge, these users can make requests to the ontology and the result will be the required output that is needed to further carry out expected tasks. In the case where a user is interested in establishing influencing factors, the result from the ontology will contain influencing factor classes and instances as the information required. Other influencing factor-related information can also be obtained from the ontology. Such information includes evidence that shows a factor and the location where the studies were carried out.