Economics of Public Health Blog 10

We know from our last blog that models can help us evaluate a whole range of different interventions and provide useful information to policy makers.  They bring together the available data and allow us to project the long term outcomes of an intervention which can be especially useful given the generally short term nature of many evaluations collecting primary data.

While the concept may sound straightforward, it can be more difficult to picture what type of information you really need for a model and in what ways it can be used to assess different policy options. In this blog we try to demonstrate a bit more of this using the case of cardiovascular disease (CVD). There are two main approaches to primary prevention. The first is a targeted approach which screens individuals, while the second is a population-wide approach where interventions are intended to impact on everyone without engaging specific individuals. We now turn to briefly describing models which have been developed to estimate the impact of both approaches.

Targeted CVD interventions:  Keep Well – Scotland’s CVD primary prevention programme

Keep Well is Scotland’s primary prevention programme with the aim to prevent the premature onset of CVD, associated health service costs and the reduction in health inequalities. Keep Well is (at present) focussed on the most deprived 15% of areas of Scotland. Individuals aged 45-65 are screened using the ASSIGN risk score to assess the risk of an event over the forthcoming ten years. This is composed of nine variables, a mixture of “modifiable variables”, such as systolic blood pressure and cholesterol and “non-modifiable” variables such as age, sex and a measure of socioeconomic deprivation to detect the underlying social gradient in the incidence of CVD. High risk individuals, defined as those with a risk score ≥20%, are referred onto tailored multi-factorial programmes, including pharmaceutical and behavioural interventions. No existing economic models are equipped to evaluate the impact of this programme.

This background formed the rationale behind the recently created Scottish CVD Policy Model, funded by the Chief Scientist Office. The model was built using the same longitudinal dataset used to create the ASSIGN ten year risk score. But rather than estimate the relationship between the ASSIGN risk factors and CVD over ten years, we instead estimated how risk factors predict quality adjusted life expectancy (QALY), and lifetime health service costs. The model can then be used quite easily to estimate the impacts of interventions (single or multiple) on changing QALYs and costs. Further, given the inclusion of a measure of deprivation, the model also estimates the impact on health inequalities.

To date, there has not been a comprehensive evaluation to assess whether the programme is reducing CVD. The Scottish CVD model was used by the Scottish Government and NHS Greater Glasgow and Clyde to project the likely cost effectiveness of Keep Well. In the absence of evidence, the model used secondary data sources on the efficacy of interventions (e.g. statins and smoking cessation) and expert opinion. Through a sensitivity analysis the model illustrated why the programme may not actually be cost effective and highlighted key areas of uncertainty (e.g. adherence to medication or behaviour change) and made the unequivocal case that a robust economic evaluation is needed, if scarce health service resources are to be used efficiently. The good intentions of the programme may not be matched by the reality of effectiveness and cost effectiveness. 

Population CVD interventions – salt reduction

Eating a diet high in salt increases your risk of high blood pressure which in turn increases your risk of having a cardiovascular event such as a heart attack or stroke. Reducing the amount of salt the population consumes has the potential to reduce the overall level of CVD across a country.  One study designed to evaluate policies to reduce dietary salt intake was the MedCHAMPS project which covered four Eastern Mediterranean countries. The starting point was an epidemiological model which explains CVD trends across a population based on a number of risk factors including obesity, smoking, cholesterol and blood pressure (the IMPACT CHD model).  To this we needed to add in information on what our policies to reduce salt intake would be and the effect these policies would have on blood pressure. Two polices selected were; a health promotion campaign highlighting the health effects of a diet high in salt and reformulation of food products.  As is the case in most models, we took estimates on the effectiveness of each policy, in this case reduction in blood pressure, from the literature. We then estimated the expected reduction of current salt consumption for each policy. The expected change in salt intake was then translated into a change in blood pressure and the resulting change in blood pressure was used to estimate the number of deaths prevented or postponed. We then estimated the number of life-years gained from this information on the number of deaths prevented or postponed. Costing these different policy options involved bringing together data from a number of sources; existing comparable policies, data from ministries of health and, for the cost of reformulation, from the food manufacturers. Health care costs of CVD events were also incorporated and data was taken from standard hospital tariffs.

The data on the costs of introducing each policy was combined with the effectiveness information generated by the epidemiological model and we estimated that in Palestine, for example, the health promotion policy would save PPP$4 million with 97 life years gained across the population, while the reformulation campaign would result in an incremental cost of PPP$2 million and 479 life years gained.  

What have we gained from modelling?

The construction of a model may seem like a daunting task so we hope that these case studies highlight that models are constructed from the same type of information that we would need to collect for any type of evaluation. Our previous blogs on costs and health outcome measurement provide lots of detail on where those data could come from. Although not the solution for every evaluation of an intervention or policy, in both of the case studies the model provided us with information which would have been difficult to obtain otherwise. This can help us to identify future research needs, potentially eliminate waste and identify the relative value of different policies. Modelling allows us to do all of this using the best available data at the point in time when a decision has to be made, providing information with which to make an informed decision.