Cost studies

Many studies went through the costs of several components of the project, but not as a whole economic evaluation study. Cost-effectiveness analysis of other aspects of CVDs treatments was applied in Finnish studies, such as secondary prevention, and the effects of small changes in diets (Martikainen et al. 2011, Soini et al. 2010). However, the lack of some information concerning costs and operation were described briefly for North Karelia to indicate some difficulty in such evaluation of the project (Puska et al. 2009). On the other hand, since evaluation studies do not need to have complete lists of requirements in order to get conclusions about the economic value and consequences of such projects, therefore, some assumptions in these cases are commonly used in economic evaluation studies as long as they do not harm the internal and external validity of the research.

Some information regarding the cost of illness due to North Karelia Project has been elaborated earlier (Puska et al. 2009). The book shares some results from a separate publication about the cost of CVDs in Finland (Kiiskinen et al. 1997). Both sources relied on prevalence approach in estimating costs, where both the direct and indirect costs resulting from disease are assigned to the years in which they are borne. The value of expected future production losses caused by premature mortality is assigned to the year of death. The methods recognized but couldn’t take into account the intangible costs like non-economic consequences of illness. Presenting the results followed 5 years gap pattern from 1972 till 1992, with full information regarding the type of direct costs was only in the last year due to the lack of information for previous years.

Rest of years have more generalized data regarding the cost of drugs, hospitalization, working hours lost due to CVDs. And assumptions were made in some district or year data.

The results indicated increase of directs costs of CVDs healthcare like treatment, and medical expenditure due to enlargement of quantity and quality of resources dedicated for CVDs both in North Karelia and all Finland. However, there was a decrease in the indirect costs associated with time lost from work due to illness in the working-age group. The total proportional decrease in CVD-related total costs ranged from about 5 to 50 % in the overall period, with greater results cited in North Karelia than the rest of Finland (Puska et al. 2009).

Other studies that discussed the costs around North Karelia Project involved its components of risk reduction, like cost-effectiveness of hypertension program of North Karelia (Nissinen et al. 1986). In this study, cost-effectiveness model was used. It depended on a formula that has the relevant cost calculations along with gains of QALYs, and the period of interest was the first five years of the project: 1972-1977. Results showed that the cost per QALY gained is

$3,612 at zero discount and $5,830 at 10% discount. While drugs costs represented 86% of the total cost of reducing hypertension, the study suggested that it would be much more cost-effective if hypertension could be treated as cost-effectively without medications or if the costs of medications could be reduced.

A more recent study in the early 2000s focused on the costs of smoking in North Karelia and combined sample with Kuopio (Kiiskinen et al. 2002). The population group was middle-aged men only and were followed up for 19 years. Results indicated the mean difference between a current smoker and a never-smoker in health service costs was € 2,900, and the difference in mean total costs was € 69,300. While no difference in mean health care costs between current smokers and former-smokers was found but the difference in mean total cost was € 44,000.

This study was based on the reduction of the percentage of current smokers from the same population group in the same period from approximately 52% to 33% (Puska et al. 2009).

Another cost-effectiveness study of population-wide educational approaches to reduce serum cholesterol levels based in the USA, analysed the work of North Karelia Project (Tosteson et al. 1997). It estimated that a national program as such, with costs of $16.55 per person in year 1 and $8.28 for each year after (including media campaigns), as well as the 3% reduction of

cholesterol similar to those found in the North Karelia study, would cost ≈$6100 per year of life saved. Another estimation if the program cost continued at $16.55 per year, with a decline in cholesterol of ≥2%, the cost-effectiveness measure would still be relatively acceptable compared with most accepted medical interventions, with estimated costs per year of life saved of ≤$38 500. Unfortunately, the source of which this study relied on to get cost information was a report to the Henry J. Kaiser Family Foundation following Rand Corporation that according to contacts with the author was never published later.

3 AIMS

This paper aims to define what an economic evaluation would mean for North Karelia Project using Markov model. Building a Markov model will be explained in steps in the frame of North Karelia Project as a public health community-based intervention to limit CVDs burden. Also, the paper will illustrate the information needed for such model, the available inputs, possibly missing parts from literature, and how decisions can be made accordingly.

Study questions

1- How to build a Markov model for North Karelia Project?

2- What information is needed for a Markov model of North Karelia Project? Which parts are available from literature and what is missing?

4 MATERIALS AND METHODS

The methodology of this paper relied on 2 guidebooks in the field of economic evaluation and decision analytic modelling in healthcare. The first book (Drummond et al. 2015) titled Methods for the Economic Evaluation of Health Care Programmes was used initially to understand the rationale of using economic evaluations in public health prevention program like North Karelia, the connection between these two domains conceptually and theoretically, and general guidance in setting steps, scope, and limitations in using economic evaluations in healthcare.

The other sourcebook used was Decision Modelling for Health Economic Evaluation, part of the Oxford series of handbooks in health economic evaluation (Briggs et al. 2006). It supported building the structure for modelling, particularly, Markov model due to its important utilization in healthcare topics. The book provided the recommendation materials to find suitable model inputs and guided the process of taking decisions in conducting economic evaluation using Markov model. Some exercises, formulas, and templates were used as well in the process of learning how to involve the model in the study.

All the information required by the model and its entries concerning North Karelia Project were derived from literature, by viewing secondary data sources. In this study, I will apply the principles of economic evaluation for the North Karelia Project. Relying on Markov model framework, I will study the possibilities and challenges in such evaluation of community-based CVD prevention program. The model framework will be utilized also for assessing information needs to conduct an economic evaluation modelling study and to search for other questions and matters relevant to my study questions that may arise.

To answer the research questions, the stages of developing a decision model (Briggs et al. 2006, Drummond et al. 2015) were used. It is a set of guidelines that describes what the author should think about and search for when aiming to conduct an economic evaluation in healthcare, to have a reliable structure of study. They were employed in the results section to evaluate the steps and possibilities to conduct an economic evaluation for North Karelia Project and to reflect my own choices. The stages used in this paper are summarized as follows:

1- Specifying the decision problem 2- Defining boundaries of the model

3- Structuring a decision model

4- Identifying and synthesizing evidence 5- Dealing with uncertainty and heterogeneity 6- Assessing the value of additional research

In order to locate information regarding North Karelia Project and the literature needed to accommodate its model, triangulated search approach was used. Starting from the most recent in-depth publication of the project team (Puska et al. 2009), the book titled The North Karelia Project: From North Karelia to National Action was initially used as a reference book for the development of the project, its stages, structure, health outcomes, and costs. Then backward and forward reference searching was conducted to determine usable findings of the model inputs. First, I used the search engine Primo provided by the UEF library to scan for relevant literature to find the right keyword. This was followed by keyword searching with suitable search modifiers using Medline database for “North Karelia Project”, “economic”,

“evaluation”, “costs”, “analysis”, “CEA”, “CBA”, “CUA”, “CVDs”, “CHD”, “Finland”,

“incidence”, “mortality”, “Markov model”, “CVDs health states”, “death rate”, “causes of death”, “CHD risk ratio”, “recurrent CHD”, “QALYs”, and “CVDs unit prices” and

“cardiovascular”. All publications having Professor Pekka Puska as an author were scanned as well since some of his articles were found relevant without having the mentioned keywords directly stated in the title or abstract. The searches above took place between December 2019 and May 2020 for English publications since the 1960s till the time of writing this paper.

Markov model, the literature available and the gaps in research will be described in detail in the results section.

5 RESULTS

5.1 Markov model to evaluate the cost-effectiveness of North Karelia Project

As stated earlier, Markov model is one form of decision-analytic modelling used to evaluate the cost-effectiveness of health interventions, or other cost/consequences measures. In the case of North Karelia Project, it has the capability to combine the complicated map of factors associated with the project’s targeted illness and the interventions against it, in this case, CVDs.

Many factors are associated with the incidence of CVDs and its complications. Its chronic nature, and inability to assign limited paths for follow-up and treatment that patients will go through, both resulted in the difficulty to use simpler forms of decision-analytic models like the decision tree. Especially since the project is under the public health domain where no single medical intervention was applied, rather than the combination of various efforts on a multi-sectoral level. These forms of disease occurrence, causes, and public approach of prevention following long time horizon would make the decision tree too unwieldy, hence Markov model has better ability in handling this function (Drummond et al. 2015).

In a review of economic evaluations for lifestyle interventions to prevent diabetes and CVDs (Saha et al. 2010), Markov model has been the most frequently used tool to describe the evaluation among decision-analytic modelling designs, to make up 20 studies out of total 31.

This result has backed up the decision of preparing this paper based on a Markov model design.

Nevertheless, many papers in another review that focused on medication intervention have combined Markov model with decision tree design, as well as other approaches (Ward et al.

2007). The model is used as a tool to draw inputs and outputs for CVDs prevention, therefore, the time and data available for such studies would basically answer better which model to use and how results can be interpreted.

The Markov model to be used for North Karelia Project involves the different stages of person’s lifetime, where a certain hypothetical population will develop CHD and continue with its consequences, or simply live healthy and die due to other reasons. The stages of a lifetime, or health states person experiences are set separately. Each year or discrete duration of time, described as Markov cycle, a person is expected to move between states or remain in the same state. So that, in every cycle (a healthy person for example) has a chance of having CHD event, fatal CHD death, other reasons for death or to remain healthy. Figure 5, found in section 5.3,

is an illustration of these Markov health states, where a person will definitely be at one stage at a time. An important consideration in states’ design decision is to limit the probability that a given patient could experience more than one event in the period of the cycle so that they have to be mutually exclusive (a given patient can only follow one of the pathways) and exhaustive (a given patient must follow one of the pathways) (Drummond et al. 2015). While the length of these cycles usually varies on the disease and interventions.

The chances a person has to move between states are called event probabilities, which are usually affected by the risk of disease incidence, risk of surviving an event, risk of death due to CHD or any other cause. These probabilities describe who’s healthy and who’s not in each year and by the end of years modelled. Therefore, choosing the evidence for event probabilities is very critical and can highly alter the results of the study. The health states incur different costs and health consequences, where they’re entered also as values separately associated with the states like QALYs utilities and unit prices or cost of illness. So that, by the end of model duration and after applying the event probabilities, we can calculate the health outcomes and its associated costs.

To measure cost-effectiveness, the results of the first model are compared to other scenario models where, for example, we assume no intervention of North Karelia Project was applied, or another approach is used. If the project wasn’t spread nationally, and evaluation is based on the design of first 5 years of the implementation, Kuopio statistics might have been suitable for the other scenario since it was the reference area for North Karelia. But, in this case, a hypothetical design would work. Either only 2 or multi-options can be compared in economic evaluations involving steps or stages. They can be mutually exclusive or not. In this paper, assuming “no intervention” scenario was the comparison model of North Karelia; these 2 models would differ slightly based on the preventive factors added by the project.

Usually, prevention would be visible in the model as event probabilities alterations of developing CHD or its consequences like what would the prevention project result in. Then, the sum of QALYs, if used as the outcome measure, and costs value would differ because a different number of people reached to the disease stage in the model time. Subtracting the values of costs and QALYs for the 2 models, the North Karelia Project and “do nothing”

scenarios, would equal to the incremental cost-effectiveness ratio (ICER) of the project, which

means the added economic value of effectiveness (or the cost of 1 added QALY) due to the project compared with doing nothing in this case. The choice of the other scenario or alternative of treatment or prevention would directly affect this incremental value, that’s why the other alternative might cause drawing hasty conclusions if it’s not realistic or rather expensive. All the previous inputs entered to model (event probabilities, costs, QALYs utilities) are described as parameters of the Markov model, which will be described in details how can be assigned following the steps stated in the methods section.

5.2 Stages of developing the Markov model of North Karelia Project.

Step 1: To specify the decision problem

Following the guidance in the book, in order to specify the problem to be answered by the economic evaluation, we should start by defining the population of interest in the study. In the case of North Karelia, as a population-level intervention, it targets healthy people and patients in the first place, through primary and secondary prevention. So they’re the group of interest, whether this includes high-risk individuals, patients suffering from developed complications or average people.

On the other hand, we should understand the perspective of the study and its timeline. From one side, the QALYs, humanitarian suffering and economic costs of this lost life are what causes the CVDs burden on people’s level. While on the other side, the funding organization, which is the taxpayer or public sector in the case of Finland has the other important view on the issue, as well as the health provider’s view with its limited capabilities. These 3 dimensions of interest would affect what parameters will be associated with each health state.

To answer what possibly is chosen to be the perspective of this model, I would like to highlight the public sector view on costs and QALYs. Studying what the health sector invested in North Karelia Project would provide a holistic approach that gives weight for both government and people’s concerns. Time-wise, the model setting to be in present time would reflect both its results from an ex post perspective, and also help in the assessment of future consequences of similar projects.

Moreover, since the project had wide applications on a national level, and involving in different sectors, we ought to specify the settings of interest. According to North Karelia publications

(Puska et al. 2009, Puska 2010), even after the widespread of the project work in the country, North Karelia due to its initial start and different CVDs prevalence than other Finland’s counties, it remained a model or demonstration for the national work. Therefore, if the economic evaluation of the whole project is considered more feasible when applied in limited areas, North Karelia has the potential for such choice. On the other hand, the variations of prevention work on a multi-sectorial level can cause complexity of what to be included in the model, however, since all the different approaches had aimed for reduction of CVDs risk factors, the combined result can be described by limited event probabilities that caused the reduction of incidence rates of CVDs. So one’s don’t need to be distracted by the number of interventions in the project, rather than focus on its unified result, as it is what would actually affect in terms of model reflection of reality.

Another concern in defining the problem would the demographics of interest. The project started with middle-aged men and women, concentrating more on men who had a higher risk in the first place. Later it involved other age groups in the society. Inclusion of both men and women in the study would better describe the differences they had in costs and QALYs since the topic of variation of health results by sex was discussed earlier (Puska et al. 2009). And mostly all resources found had available data of both men and women. Despite the same need to include the different age groups studied, no enough information seems to be available on all ages in the publications reviewed by this paper. Most studies that included long term follow-up periods, focused on 25-64 or 30-59 years old age grofollow-up (Laatikainen et al. 2005, Borodulin et al. 2015, Jousilahti et al. 2016, Jousilahti et al. 1999, Jousilahti et al. 1998). Although modelling usually helps in predicting the scenarios based on what’s available, the missing data might affect the validity of the research. However, this doesn’t mean data of other ages isn’t available but mainly, the current literature review couldn’t find clear indications of it. Also, since the initial focus of the project was on this middle-aged group, the economic model of it doesn’t need to involve other groups in the meantime.

Another concern in defining the problem would the demographics of interest. The project started with middle-aged men and women, concentrating more on men who had a higher risk in the first place. Later it involved other age groups in the society. Inclusion of both men and women in the study would better describe the differences they had in costs and QALYs since the topic of variation of health results by sex was discussed earlier (Puska et al. 2009). And mostly all resources found had available data of both men and women. Despite the same need to include the different age groups studied, no enough information seems to be available on all ages in the publications reviewed by this paper. Most studies that included long term follow-up periods, focused on 25-64 or 30-59 years old age grofollow-up (Laatikainen et al. 2005, Borodulin et al. 2015, Jousilahti et al. 2016, Jousilahti et al. 1999, Jousilahti et al. 1998). Although modelling usually helps in predicting the scenarios based on what’s available, the missing data might affect the validity of the research. However, this doesn’t mean data of other ages isn’t available but mainly, the current literature review couldn’t find clear indications of it. Also, since the initial focus of the project was on this middle-aged group, the economic model of it doesn’t need to involve other groups in the meantime.