Developing general models and theories of addiction - Nfap

[Pages:14]Developing general models and theories of addiction

Robert West Department of Behavioural Science and Health, University College London

robertwest100@

Simon Christmas Independent consultant simon@

Janna Hastings Babraham Institute, University of Cambridge

janna.hastings@babraham.ac.uk

Susan Michie Centre for Behaviour Change, University College London

s.michie@ucl.ac.uk

Introduction

The science of addiction is being hampered by confusion in concepts and terms, and a multiplicity of models and theoretical approaches that make little reference to each other. In this respect it has much in common with other areas of social, clinical and behavioural sciences. Technologies now exist and are being rapidly advanced that can address this problem, and other sciences are already making use of them. In particular, what are known as `ontologies' (as used in computer science) and the `Semantic Web' could revolutionise our ability to formulate models and theories in addiction which can then provide much needed direction to the scientific endeavour.

The field of biology suffered from a similar problem until the development of what is known as the Gene Ontology (Ashburner et al., 2000). The gene ontology is not just about genes, but is a representational system for the whole of biology, unifying terms, definitions and models across species and research groups in a way that has revolutionised the field (Lewis, 2017).

This chapter introduces readers to ontologies and the semantic web, and explores their potential use in developing and expressing models and theories of addiction in ways that allow relationships to be examined between them, and between these and more general models and theories in clinical,

General theory of addiction

Page 1

population and behavioural sciences. These technologies also allow investigation of construct relationships within models, necessary for testing and hence refining and advancing them.

We begin by describing a central challenge facing the study of addiction: the need to achieve clarity of constructs and develop consensus while at the same time recognising that divergent views have utility. We then move on to describe some key characteristics of the Semantic Web, and the ways in which these provide a pragmatic way of responding to this challenge. We conclude by looking at the potential value of existing ontologies in developing a general theory of addiction.

The challenge: clarity and diversity

Models of addiction are necessary for building addiction science and developing effective interventions to combat this problem. If they go beyond describing observed relationships (descriptive models such as tobacco price elasticity (Gallus, Schiaffino, La Vecchia, Townsend, & Fernandez, 2006)) and attempt to explain phenomena we refer to them as `theories' (e.g. the dopamine theory of drug reward (Blum et al., 2015)). There are a plethora of models and theories of addiction differing in scope, emphasis, constructs and propositions but they have not been expressed in ways that allow them to be compared, tested or integrated.

The term `addiction' is itself an example of this lack of clarity in relation to key constructs. Many cases of psychoactive drug use have features that lead to the users being labelled as suffering from a condition called `addiction'. Sets of such features have been listed in `diagnostic criteria', such as DSM-5 (American Psychiatric Association, 2013) and ICD-10 (World Health Organisation, 2016). These include continued use despite harmful consequences, experience of adverse withdrawal symptoms during periods of non-use, difficulties controlling use, high levels of use, and repeated strong motivation to use. Features can be present to different degrees, which means that thresholds are required for deciding whether or not a pattern of drug use is addictive. These thresholds are to a large extent arbitrary and context dependent. An alternative, rather than considering addiction to be present or absent depending on whether some threshold is exceeded, is to assess the degree to which the features are evident to specify a `degree' of addictedness or `severity' of addiction (Gossop et al., 1995).

With multiple features potentially being involved, different ways of characterising these features, and different thresholds potentially being applicable, there can be substantial differences of view in whether an individual's pattern of drug use is considered addictive, or the degree of addictedness. These differences mean that issues such as the prevalence of addictions and theories concerning the causes of addiction are subject to differing viewpoints that cannot be reconciled solely by reference to objectively determined facts.

The picture is further complicated by the fact that different psychoactive drugs have different patterns of use and show different degrees and patterns of addictive features, as do different ways of using the same drug. An additional complication is that behaviours that do not involve psychoactive drugs, such as gambling, can show similar characteristics to drug addiction and so the concept of addiction appears to apply to those behaviours as well.

A similar lack of clarity exists for many other constructs which are important to a scientific understanding of addiction. Any general model or theory of addiction would need to capture processes that have been identified as important in its development and maintenance (Orford, 2001; West & Brown, 2013) (Table 1). Although conceptually different, many of these processes have features in common, or else the distinction between them is nuanced. For example, incentive sensitisation and drive theory both involve craving. In the case of incentive sensitisation, cravings are

General theory of addiction

Page 2

generated in response to cues through repeated exposure enhancing a direct link between those cues and the experience of `wanting' whereas drive theory proposes that cravings are generated through exposure to a drug leading to a state that is relieved by taking the drug. Some models and theories are fundamentally neurophysiological whereas others focus on social processes.

Table 1: Processes commonly included in models of addiction (West, 2017)

Processes

Description

Propositions in existing models

Cost-benefit

The benefits of the addictive

At least some people addicted to

analysis

behaviour are judged by the addict to alcohol believe that the benefits

outweigh the costs

in terms of anxiety relief and

mental escape are worth the

financial, social and health costs

Incentive

Repeated exposure to addictive drugs Smoking crack cocaine leads to

sensitisation

leads to sensitisation of brain

feelings of craving in situations

pathways that generate feelings of

similar to those where this has

`wanting' in response to drug cues

occurred, independent of feelings

independent of feelings of `liking'

of euphoria produced by the drug

Reward seeking

Addicts learn that addictive behaviours Methamphetamine users seek

provide positive feelings of enjoyment the `rush' provided by the drug

and euphoria

Attachment

Addicts become emotionally attached Smokers often report feeling a

to drugs or addictive behaviours

sense of bereavement during the

because these have been reliable

early stages of stopping smoking:

sources of comfort or gratification

like they have lost a cherished

friend or family member

Drive reduction

Repeated engagement in an addictive Repeated use of nicotine alters

behaviour results in development of brain physiology so that

an acquired drive, which is

abstinence results in an acquired

experienced as craving, after a period drive state, experienced as

of abstinence.

craving

Distress avoidance Addicts learn that addictive behaviours Repeated use of heroin results in

relieve mental and physical distress changes in brain chemistry

caused by mental health problems, life leading to adverse mood and

circumstances, and/or withdrawal

physical symptoms when

symptoms

concentrations of the drug fall

below certain concentrations in

the brain

Social influence

Cultural, sub-cultural, peer group

Family and peer group are

and/or family norms promote or are important factors influencing the

permissive of addictive behaviour

development of smoking and

alcohol consumption

Impaired control

Addictive behaviours develop, and are Use of stimulant drugs leads to

maintained by, pre-existing or

impairment in frontal lobe

acquired inefficiencies in brain

functioning required to inhibit

systems required for impulse control impulses to continually repeat

the use leading to bingeing

Classical

Repeated pairing of stimuli (cues)

Lights, images and sounds are

conditioning

associated with effects of addictive

used by gaming machine

behaviours leads those stimuli to

General theory of addiction

Page 3

Operant conditioning

generate anticipatory reactions to those effects Addictive behaviour are followed by powerful positive or negative reinforcers (rewards or offset of aversive stimuli) in the presence of discriminative stimuli (cues), so that those stimuli come to provoke a strong impulse to engage in the behaviour

manufacturers to promote high rates of use of those machines Use of psycho-stimulants is maintained by the positive reinforcing properties of these drugs

None of the processes believed to underlie addiction are limited to addictive behaviours; they are all involved in the development of other motivations (West & Brown, 2013). What makes a behaviour pattern addictive is the strength of the motivational forces generated and/or the way that particular drugs or behaviours interact with the motivational system to create a positive feedback cycle rather than the self-correcting systems that normally operate to ensure that no one behaviour receives an unwarranted priority at the expense of others.

For example, with palatable food, our natural processes of satiation and habituation reduce its rewarding value as we eat more of it during a meal. By contrast, with cocaine and amphetamine, reward mechanisms become sensitised to the impulse generating effects of the drug (Berridge & Robinson, 2016). In the case of alcohol dependence, addicts develop craving, rebound anxiety and adverse physical symptoms which provide a very powerful motivation to resume drinking (Seo & Sinha, 2014). In the case of tobacco smoking, rapid ingestion of nicotine leads to strong cravings through multiple mechanisms, including creation of an acquired drive state, similar to hunger, when CNS nicotine concentrations become depleted (West, 2009). So development of general models and theories of addiction will involve bringing together a wide range of constructs and processes relating to canonical motivational theory (i.e. a theory of what normally happens) and how abnormalities in motivation occur. These processes will be both internal and external to individuals, groups and populations.

Diversity in the study of addiction

While there is a need for greater clarity of constructs in the study of addiction, no investigator or organisation has the authority, or expertise, to propose a single unifying conceptual framework.

Figure 1: du Plessis' classification of addiction theories (du Plessis, 2014)

a) Eight major methodologies

b) Types of addiction theory classified according to the eight methodologies

General theory of addiction

Page 4

A general theory of addiction has yet to be developed, but a key requirement for such a theory is that it should recognise and accommodate multiple viewpoints on addiction, and not be limited to a single viewpoint such as the `medical model' (construing addiction in term of a mental disorder, disease or disease process). Figure 1 shows du Plessis' classification of the types of theory that would need to be recognised and accommodated by a general theory of addiction according to eight major methodologies (du Plessis, 2014). Some specific theories and several theoretical approaches have attempted to span a number of the zones in this classification. Of the specific theories, the theory of Excessive Appetites and PRIME Theory have been elaborated in some detail (Orford, 2001; West & Brown, 2013), respectively).

It is important not to overstate this diversity. Despite the complexities discussed earlier, the observed phenomena that are captured by a term like `addiction' are important, and it is useful to give a label to a construct that captures these. Moreover, there are shared understandings of many of the features that characterise addictive drug use, even where there may be disagreement around details and where emphasis should be placed.

On the other hand, it should go without saying that a general theory which achieved greater clarity by simply ignoring this diversity of viewpoints would not in fact be a general theory at all. To move forward, we need a way to achieve clarity of constructs and develop consensus, while at the same time recognising that divergent views have utility.

Addressing the challenge in the Semantic Web

The technology associated with the Semantic Web provides the basis for a response to the challenge we have outlined: a way for the science of addiction to move from its current state of confused, confusing and imprecise terminology to a way of representing addiction and models of addiction that will promote collaborative working, respect and preserve different viewpoints on addiction, but nevertheless allow all research results to be integrated, and theoretical predictions to be tested, and so advance the field.

What is the Semantic Web? One way to answer this question is to chart the development of the Semantic Web from the Worldwide Web (www). The Worldwide Web has revolutionised our lives by making information available from a vast range of sources. It defines a technological framework for locating and exchanging diverse content types ? for example, text, images, and films. At the heart of this framework is the Uniform Resource Locator (URL), a unique `address' for each of the billions of different web pages. These URLs are stored as a code that means nothing to human readers, but a

General theory of addiction

Page 5

system of `domain names' has been developed to link them to a name that humans can read and understand. These also have to be unique. Thus is a domain name for the journal, Addiction, that uniquely points to the journal's home page. Web pages use URLs to link to other pages to create the Worldwide Web.

The www has a huge limitation, however. The information contained in web pages is designed for humans to consume, and is difficult for machines to interpret, which hinders the ability of software to help in the discovery, synthesis, and summary of online content. Google, Yahoo and other search engines make their developers billions of dollars by trying to address this limitation, but for the most part all that search engines are able to do is to point to web pages that might contain the information you want; finding the right pages can often require multiple attempts to formulate a search in the right way. For example, a search for "Addiction conferences" will not return results from pages entitled "Addiction meetings"; a search for "addiction" will not return results from pages entitled "dependence". There is no representation of the underlying meaning of content which is sensitive to synonyms or closely related constructs, only the superficial textual representation of the content encoded directly in the web pages.

The Semantic Web is a radical advance on this. Instead of only making content available in text on web pages, the Semantic Web allows individual pieces of information within those pages to be assigned a meaning and a specific location. Rather than only having a string of text to interpret ("addiction" / "dependence"), within the Semantic Web units of information are explicitly defined in shared vocabularies or "ontologies" ? computational representations of knowledge in a particular domain. Ontologies link together the words, expressions and language which humans use to refer to things, with computable formal definitions of those things that allow computers to distinguish one type of thing from another.

The Semantic Web is implemented in the most general sense as a collection of statements which take the form subject-predicate-object, known as `triples'. This very generic form of representation is then populated with entities (subjects and objects) and relationships drawn from shared, welldefined vocabularies in order to be clear on their interpretation and meaning.

Units of information are assigned Uniform Resource Identifiers (URIs), each of which points to a unique entity, concept or resource that has an address on the web, and each URI is then used in triples represented in what is known as the `Resource Description Framework' (RDF). Computers are thus given the ability to `understand' terms such as `psychological condition' by how such a term is related to other terms set out in a network of relationships in RDF triples and definitions within ontologies.

This forces the definitions of all terms used within the Semantic Web to be explicit and distinguishable. Humans, by contrast, understand terms in ways that combine formal definitions with natural language meanings, which creates ambiguity. Although nuances of natural language provide a richness that is helpful in certain circumstances, ambiguities impede the scientific testing and advance of ideas. To make content computationally interpretable within the Semantic Web forces the use of clear and distinguishable representations for each different sort of thing, which explicitly addresses the sorts of confusion that are hindering progress in developing the science of addiction.

For example, Table 2 shows examples of propositions about addiction expressed in the form of triples. These provide a way of explicitly expressing models of addiction that allow different models

General theory of addiction

Page 6

to be compared and related to more general models of behaviour. Setting out each proposition in a model of addiction in this way allows that proposition to be examined and tested.

Table 2: Examples of propositions about addiction in the form of `triples' expressed as an approximation to natural language

addiction

is a subclass of has subclass has subclass inheres in has manifestation has attribute has attribute has attribute has attribute has part has part has manifestation

mental disorder substance addiction behavioural addiction people addictive behaviour addiction duration addiction strength addiction harmfulness addiction severity strong motivation significant harm potential repeated motivation episodes

A considerable amount of unproductive debate in the field of addiction, as in many other areas of social, clinical and behavioural science, arises because the individual propositions within theories cannot be separated out for scrutiny. Thus different researchers may take differing views on any or all of the propositions in Table 2, but the fact that they are set out in this way allows for discussion about the merits and demerits of each one to be analysed systematically, whether by reference to evidence or inference.

Diversity of perspectives in the Semantic Web

The way in which the Semantic Web works, and in particular the fact that the elements from which RDF triples are composed are explicitly assigned meaning in ontologies and by reference to their links to other elements as stated in other triples, ensures clarity of constructs. At the same time, the Semantic Web also provides a way of accommodating diversity. Indeed, a founding principle of the Semantic Web, just like the Worldwide Web, is that anyone can say anything about anything.

For example, different people can continue to use terms such as `addiction' differently, since these terms will be assigned separate URIs and definitions despite sharing the same superficial label. In fact, there is a built in way to represent different "perspectives" on the Semantic Web through use of different `namespaces'. Stated simply, if I want to define addiction in a new way I can simply create my new meaning of addiction in my own unique namespace, and explicitly refer to that version of addiction in my content. This makes it completely clear what the term means as I am using it. Importantly it also allows for systematic comparisons between my usage and someone else's.

In principle the use of namespaces could result in anarchy if no-one agreed about the definitions of anything. But this is no different from what happens with natural language. The difference in the case of the Semantic Web and ontologies is that disagreements in how terms are used are explicit and defined. They can then be subject to scrutiny, and differences potentially resolved where such resolution is useful. In practice, many researchers do agree on key definitions and can therefore use a single shared namespace to reflect this consensus. This is directly analogous to referencing the

General theory of addiction

Page 7

definition of a construct in another paper, with the important practical difference being that such links become computable.

It is also possible that namespaces could be subject to arbitrary (in the scientific sense) bias. For example, a powerful research group might insist on use of a particular namespace for anyone wanting to receive research funding or publish their work in a prestigious journal. Again, this is no different from what already pertains with models and theories that are expressed in natural language. The key difference is that everything is made explicit so there is a greater opportunity to expose this tendency through automated information searches and to challenge it with analysis and evidence.

The concept of a `namespace' allows for a diversity of perspectives on the Semantic Web, and ensures that anyone can say anything about anything. There are, however, practical limits to diversity. In order to have debates about the construct of `addiction', for example, we do need to agree on the meaning of a set of more fundamental terms and predicates we use in that debate: terms such as `is a subclass of' and `has attribute' (see Table 2 above).

The Semantic Web as a whole is underpinned by the existence of shared namespaces in which fundamental terms such as these are defined. The namespaces associated with RDF and OWL, the Web Ontology Language, provide a suite of general-purpose computable vocabulary elements that are needed to say anything at all. In the next section, we will explore the potential to use other preexisting ontologies in the development of a general theory of addiction.

Building on existing ontologies

Ontologies have been developed covering a range of topics in science, humanities and computing, and there is a very strong case for using existing, broader ontologies such as these wherever possible. Of most direct relevance to the study of addiction are the Mental Functioning Ontology (MF) (Hastings, Smith, Ceusters, Jensen, & Mulligan, 2012) and the WHO's International Classification of Functioning Disability and Health (?st?n, Chatterji, Bickenbach, Kostanjsek, & Schneider, 2003). Concepts in the field of addiction can be represented by linking constructs from relevant ontologies (Hastings, Le Nov?re, Ceusters, Mulligan, & Smith, 2012); and addiction ontologies can be developed using principles established by the Basic Formal Ontology (BFO) (Ceusters & Smith, 2010) and implemented by the Open Biological and Biomedical Ontologies (OBO) Foundry (Smith et al., 2007).

We now take a closer look at the potential value of two existing ontologies in developing a general theory of addiction: the Basic Formal Ontology (BFO) and the Ontology for General Medical Science (OGMS). It should be noted that BFO represents only one of a number of top level ontologies. It is not static, and operates under the philosophy that if changes are required in order to be able to create better models, these can be implemented.

Basic Formal Ontology

BFO has been developed to provide a top level ontology for use in domain-specific ontologies. The purpose of BFO is to provide a categorisation of the sorts of things that exist that is generic enough to be agreed on across multiple domains, thereby providing a common vocabulary which ensures computers will be able to interpret relevant general distinctions in the same way across the different domain implementations. The Mental Functioning Ontology has been developed as one such domain-specific ontology and has been implemented in OWL. These make a suitable starting point for addiction ontologies.

General theory of addiction

Page 8

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download