Search

Abstract Background It is challenging to predict long-term outcomes of interventions without understanding how they work. Health economic models of public health interventions often do not incorporate the many determinants of individual and population behaviours that influence long term effectiveness. The aim of this paper is to draw on psychology, sociology, behavioural economics, complexity science and health economics to: (a) develop a toolbox of methods for incorporating the influences on behaviour into public health economic models (PHEM-B); and (b) set out a research agenda for health economic modellers and behavioural/ social scientists to further advance methods to better inform public health policy decisions. Methods A core multidisciplinary group developed a preliminary toolbox from a published review of the literature and tested this conceptually using a case study of a diabetes prevention simulation. The core group was augmented by a much wider group that covered a broader range of multidisciplinary expertise. We used a consensus method to gain agreement of the PHEM-B toolbox. This included a one-day workshop and subsequent reviews of the toolbox. Results The PHEM-B toolbox sets out 12 methods which can be used in different combinations to incorporate influences on behaviours into public health economic models: collaborations between modellers and behavioural scientists, literature reviewing, application of the Behaviour Change Intervention Ontology, systems mapping, agent-based modelling, differential equation modelling, social network analysis, geographical information systems, discrete event simulation, theory-informed statistical and econometric analyses, expert elicitation, and qualitative research/process tracing. For each method, we provide a description with key references, an expert consensus on the circumstances when they could be used, and the resources required. Conclusions This is the first attempt to rigorously and coherently propose methods to incorporate the influences on behaviour into health economic models of public health interventions. It may not always be feasible or necessary to model the influences on behaviour explicitly, but it is essential to develop an understanding of the key influences. Changing behaviour and maintaining that behaviour change could have different influences; thus, there could be benefits in modelling these separately. Future research is needed to develop, collaboratively with behavioural scientists, a suite of more robust health economic models of health-related behaviours, reported transparently, including coding, which would allow model reuse and adaptation.

Abstract Background Problem gambling can lead to health-related harms, such as poor mental health and suicide. In the UK there is interest in introducing guidance around effective and cost-effective interventions to prevent harm from gambling. There are no estimates of the health state utilities associated with problem gambling severity from the general population in the UK. These are required to determine the cost-effectiveness of interventions. This study aims to use an indirect elicitation method to estimate health state utilities, using the EQ-5D, for various levels of problem gambling and gambling-related harm. Methods We used the Health Survey for England to estimate EQ-5D-derived health state utilities associated with the different categories of the Problem Gambling Severity Index (PGSI), PGSI score and a 7-item PGSI-derived harms variable. Propensity score matching was used to create a matched dataset with respect to risk factors for problem gambling and regression models were used to estimate the EQ-5D-derived utility score and the EQ-5D domain score whilst controlling for key comorbidities. Further exploratory analysis was performed to look at the relationship between problem gambling and the individual domains of the EQ-5D. Results We did not find any significant attributable decrements to health state utility for any of the PGSI variables (categories, score and 7-item PGSI derived harms variable) when key comorbidities were controlled for. However, we did find a significant association between the 7-item PGSI derived harms variable and having a higher score (worse health) in the anxiety/depression domain of the EQ-5D, when comorbidities were controlled for. Conclusions This study found no significant association between problem gambling severity and HRQoL measured by the EQ-5D when controlling for comorbidities. There might be several reasons for this including that this might reflect the true relationship between problem gambling and HRQoL, the sample size in this study was insufficient to detect a significant association, the PGSI is insufficient for measuring gambling harm, or the EQ-5D is not sensitive enough to detect the changes in HRQoL caused by gambling. Further research into each of these possibilities is needed to understand more about the relationship between problem gambling severity and HRQoL.