Background
Term | Definition | Source |
---|---|---|
Algorithm | Sequence of actions to perform calculation, data processing and automated reasoning tasks. | |
Annotation | Process of identifying selections of content from BCI evaluation reports describing features of BCI evaluations, together with specification of the features described using the BCIO. | |
Artificial Intelligence (AI) | The theory and practice of building computer programs to perform tasks that a human would reasonably regard as requiring intelligence. | [31] |
Attribute | A quality or disposition of an object, collection of objects, process or collection of processes. | [10] |
Automated annotation | Annotation that is undertaken by a computer program. | |
Basic Formal Ontology | An upper level ontology consisting of continuants and occurrents developed to support integration especially of data obtained through scientific research. | [10] |
BCI database | The database containing all information about BCI evaluation reports and inferences from these, organised according to the BCIO. | |
BCI evaluation | A comparison between two or more BCI scenarios focusing particularly on estimating the differences in outcomes between these scenarios. | |
BCI evaluation report | Description of a BCI evaluation, usually in the form of a published research report. | |
BCI Knowledge System | An automated system delivering comprehensive, high quality, timely and accessible syntheses and interpretations of evidence in the domain of behaviour change. | |
BCI ontology (BCIO) | An ontology that represents entities and relationships related to BCI evaluations and their reports. | |
BCI scenario | A scenario (a sequence or development of events) consisting of a BCI, its target behaviours, and factors that influence the outcome of the BCI in relation to the target behaviour. | |
Behaviour | Anything a person does in response to internal or external events. Actions may be overt (motor or verbal) and directly measurable or, covert (activities not viewable but involving voluntary muscles) and indirectly measurable; behaviours are physical events that occur in the body and are controlled by the brain. | [32] |
Behaviour Change Intervention (BCI) | A product, service, activity or structural change, intended to achieve behaviour change. It can be specified in terms of the content of the intervention and the way this is delivered. | [33] |
Behaviour Change Technique (BCT) | The smallest component of an intervention compatible with retaining the postulated active ingredients, and can be used alone or in combination with other BCTs. | [34] |
Behaviour Change Techniques Taxonomy version 1 (BCTTv1) | A hierarchical classification system (taxonomy) for reliably specifying intervention components in terms of 93 well-defined behaviour change techniques (BCTs), organised into 16 groupings. | [35] |
Cochrane Collaboration | A global independent network of researchers, professionals, patients, carers and people interested in health. It is a not-for-profit organisation with contributors from more than 120 countries working together to produce credible, accessible health information that is free from commercial sponsorship and other conflicts of interest. They work to produce reviews that summarise the best available evidence generated through research to inform decisions about health. | [36] |
Context | Features of a BCI scenario, independent of the BCI itself that may influence the outcome. | [33] |
Delivery | Features of a BCI related to the manner in which the intervention is enacted. | [33] |
Effect | The estimated effect size for the combination of intervention, usage (exposure and engagement), context, mechanism of action and behaviour, always specified in relation to a comparator. | [33] |
Engagement | The amount and manner of use of, or interaction with, an intervention among people who use it at least to some degree. | [33] |
Entity | Anything that exists, that can be a continuant or an occurrent as defined in the BFO. | |
Exposure | Factors relating to the interaction between the intervention and the target population (the extent and nature of the target population’s access to and engagement with the intervention) that may influence the intervention’s effect. Consists of reach and engagement. | |
Extensible Markup Language (XML) | A markup language that defines a set of rules for encoding documents in a format that is both human-readable and machine-readable. | |
Feature | An instance of a BCIO entity that forms part of a BCI evaluation or BCI evaluation report. | |
Human Computer Interaction (HCI) | An area of study that focuses on ways in which humans and computers interact. | |
Knowledge base | A repository of information from the domain of interest linking classes into the ontology to instances. | [11] |
Machine learning (ML) | Computer algorithms that learn from sample inputs and apply that learning to make predictions on data or classify data into categories. | [37] |
Mechanism of Action | Process that mediate the effect of the intervention on the behavioural outcome. These can be specified in terms of changes to capability, opportunity, motivation or other behaviours. | [28] |
Method | The set of features of methods used in BCI evaluations, containing features relating to study design (e.g., controlled trial), measures, sample identification and recruitment, sample size, and risk of bias. | |
Natural Language Processing (NLP) | Algorithms that extract meaning from passages of text in a form that can be used for inference by computers. | [38] |
Object | A material entity that is 1) spatially extended in three dimensions, 2) causally unified, and 3) maximally self connected. | [10] |
Ontology | A standardised representational framework providing a set of terms for the consistent description (or “annotation” or “tagging”) of data and information across disciplinary and research community boundaries. | [10] |
Outcome | Absolute numerical value of target behaviour associated with a BCI scenario. | |
OWL | A formal language for describing ontologies. It provides methods to model classes of “things”, how they relate to each other and the properties they have. OWL is designed to be interpreted by computer programs and is extensively used in the Semantic Web where rich knowledge about web documents and the relationships between them are represented using OWL syntax. In the HBC project elements of OWL are used to express ontologies relevant to behaviour change in a way that can be processed by reasoning and machine learning systems. | [39] |
PICO | An ontology used by Cochrane that represents important entities in medical and population science, focusing on evaluations of clinical and public health inteventions. The acronym stands for: Patient, Population or Problem, Intervention, Comparison (group intervention is compared to) and Outcome. | [40] |
Population | Characteristics of the individuals, groups, sub-populations or populations whose behaviour one is seeking to change, including their other behaviours, mental health status etc. | [33] |
Process | An entity that exists in time by occurring or happening, has temporal parts, and always depends on at least one object as participant. | [10] |
Reach | Uptake of intervention. | [33] |
Reasoning algorithms | Computer programs that can generate conclusions from available knowledge. In the HBCP project, reasoning algorithms may derive conclusions through combinations of logic based reasoning (where basic axioms about the behaviour of the environment are provided as a basis for reasoning) and statistical learning (where patterns are used to construct new facts). | |
Risk of bias feature | Features of BCI evaluation method and reporting that may lead to the reported effect size of the evaluation not being accurate. | [41] |
Setting | Features of the social and physical environment that may influence the outcome of a BCI. | [33] |
Target Behaviour | Behaviour that a BCI seeks to influence. | |
Taxonomy | A classification system in which classes are uniquely assigned to a higher level class. | [42] |
User Interface (HBCP) | The means by which the user and a computer system interact, in particular the use of input devices (keyboards, screens etc) and software. The HBCP interface will consist of a machine interface and a human user interface. The machine interface will provide application programming interfaces that will allow other programs to query and provide information to the BCI Knowledge System (e.g. results of searches for BCI evaluation reports). The human user interface will be a website and associated supporting programs to allow users to query and inform the BCI Knowledge System. | [43] |
The challenges of a rapidly expanding, complex evidence base
Challenge | Solution |
---|---|
Research methods: Diversity of research methods and topics, and inconsistency and incompleteness in reporting of study methods and findings | Development and application of an ontology of behaviour change interventions |
Human limitations: Insufficient human resources to undertake reviews and syntheses in a timely manner given the volume of findings and increasing rate of evidence accumulation | Use of automated literature searching and study feature extraction |
Research findings: Equivocal or contradictory findings, sparseness of findings relative to the number and variety of behaviours, interventions, populations and settings about which information is required, complexity of interactions between intervention components, populations, settings and behavioural outcomes | Use of machine learning and reasoning algorithms for evidence synthesis and interpretation. Focus will be on methods providing a confidence level associated with the prediction so as to be able to rigorously incorporate conflicting and missing information |
The Human Behaviour-Change Project (HBCP)
Organising and classifying research, and generating inferences: The role of ontologies
-
BCI evaluation report is a written description of a BCI study, which provides information about one or more BCI evaluations (see below), including the intervention(s) being evaluated, study methods and findings. It will typically involve a published paper but may include information from more than one paper, for example if important features of the methods are described in a protocol paper.
-
BCI study is an empirical data-gathering activity consisting of one or more BCI evaluations.
-
BCI evaluation is a comparison between two or more BCI scenariosg.
-
Method g defined as the set of attributes of BCI evaluation methods. These include study design (e.g., controlled trial), measures, sample identification and recruitment, sample size, and ‘quality’
-
Effect g defined as the result of a comparison between outcomes of each pair of intervention and comparator scenarios. It is specified in terms of an effect descriptor (e.g., odds ratio, risk difference), effect size and confidence intervals.
-
Risk of bias features g are features of the BCI evaluation report and method that may have an impact on the observed effect of a BCI evaluation. These include study design, blinding, method of randomisation etc.
-
BCI scenario g is a scenario (a sequence or development of events) consisting of a BCI, its target behaviours, and factors that influence the outcome of the BCI in relation to the target behaviour (Fig. 2). A BCI scenario may be hypothetical (if it is one that is being considered for modelling purposes), planned (if it is one that is or has been intended), or realised (if it has been enacted, for example in a BCI evaluation). When annotating BCI evaluation reports (see below) the aim is to capture the realised BCI scenarios based on information from the reports. When querying the knowledge base (see below) the aim will be to present features of a planned or hypothetical BCI scenario with a view to obtaining a prediction of the likely outcome.
-
Outcome (behaviour) g defined as the type(s) of behaviour that the BCI seeks to change (e.g., tobacco smoking) together with a collection of attributes (e.g., duration, frequency or incidence) that together make specific types of outcome measure (e.g., self-report of not smoking for 6 months supported by a salivary cotinine concentration of less than 15 ng/ml measured at the final follow up point) [17].
-
Intervention g defined as a set of types of policies, activities, services or products that are intended to result in a specified outcome in relation to the target behaviour. The intervention is specified in terms of summary descriptors (e.g., ‘brief opportunistic advice from a GP on smoking’) together with detailed descriptions of ‘content’g such as the techniques used (e.g., pharmacological support, verbal persuasion about capability etc.), and ‘delivery’g (e.g., 5 min, single session, verbal, face-to-face, during a routine consultation, by GP, trained with UK National Centre for Smoking Cessation Very Brief Advice online course). The term ‘intervention’ is also used to refer to any comparator in a BCI evaluation (e.g., usual care).
-
Context g defined as factors (consisting of characteristics of the population and setting) not directly connected with the intervention that may influence the intervention’s effect.
-
Exposure g defined as factors relating to the interaction between the intervention and the target population (the extent and nature of the target population’s access to and engagement with the intervention) that may influence the intervention’s effect. Consists of reachg (e.g., the proportion of the target population that has access to, or is exposed to, the intervention) and engagementg (e.g., the extent and nature of the target population’s interaction with intervention components).
-
Mechanism of action g defined as the type(s) of process by which interventions influence the target behaviour (e.g., through increasing strength and frequency of feelings of concern about the risks of an unhealthy behaviour; providing a physical or social cue to action).
-
Outcome (behaviour) value g defined as the value attaching to the target behaviour for a given BCI scenario (e.g., the outcome would be 15% of the population where the target behaviour was six months of continuous abstinence from smoking).
The role of computer science in the HBCP
Aim and research questions
1. Creating the BCI knowledge system
2. Evaluating the output
Methods
Overview
Development of the HBCP interface
Type of user | Requirement | Query |
---|---|---|
Directors of Public Health for a consortium of Local Authorities in a deprived region of England | To identify effective messaging in a mass media campaign to promote smoking cessation in their region | What is the optimal content, timing and patterning of delivery of a mass media campaign that aims to increase attempts to stop smoking in economically deprived smokers in the North East of England? |
A research team developing a research proposal to evaluate a mindfulness smartphone application to promote healthier eating | To find out what evidence there is on whether mindfulness interventions can help people to achieve lasting change in eating patterns | What is the knowledge base on the effectiveness of mindfulness interventions in achieving long-term behaviour change, and are there any general conclusions that can be drawn from this about who it works for, for what behaviours and delivered in what ways? |
Highways England (responsible for safety on major roads) | To re-evaluate policies on speed cameras as a way of reducing excessive speed | What effect if any have speed cameras had on reducing the incidence of driving above the speed limit on major roads in England? Are there any factors that influence the effect in terms of geography, type of road, and type of road user? |
NHS England | To develop a national campaign to reduce hospital acquired infections through improved hand-hygiene | What interventions have been found to be effective in improving hand hygiene in hospitals? How effective are they? Are some more effective than others? Are there contextual factors (population and setting) that influence the effect of these interventions, and if so what are they? |
A cancer charity | To identify the most effective strategies for increasing ultra-violet protection behaviours | What is the relative effectiveness of different interventions aimed at increasing the ultra-violet protection behaviours? How far is this influenced by contextual factors (population and setting)? |