Content aspects
The observed distribution of studies over time supports the impression that the evaluation of cost-effective strategies to combat cardiovascular disease in developing countries has been a neglected topic for decades. However, around the time of the release of the second edition of the report on Disease Control Priorities (DCP2, April 2006), which among other issues covered interventions to address chronic diseases in developing countries, the number of publications on cost-effective strategies to reduce the burden caused by CVD in low- and middle income countries increased significantly. This could mean that the work on DCP2 has directly led to the publication of relevant work and/or that it has spurred the research interest in this area.
Nevertheless, large research gaps do remain in the area of economic evaluations. While certain countries in Latin America, Africa, Europe and South Asia have been subject to some formal assessment, there are regions in the world that have only been studied from an aggregate perspective. These countries are typically located in the regions of North Africa and the Middle East as well as Central Asia, South Asia and East Asia. This may reflect deficits in our research strategy (e.g. due to non-coverage of relevant languages), or--perhaps more likely--it may indeed mean a paucity of research efforts.
Ideally, a systematic review of this kind should provide an answer to the question: which are the most cost-effective ways of addressing CVD in developing countries? While we cannot provide a satisfactory answer to this question, simply because the evidence base is too limited, not enough transparent, and incomplete, we are in a position to describe for which strategies there is arguably strong evidence and where it is that research is missing.
There is significant evidence for pharmaceutical strategies to tackle risk factors as part of secondary prevention and--in some cases--also for primary prevention. While there appears to be a consensus on the utility (up to a point) of certain pharmaceutical strategies in general and the need for some form of their scaling up in developing countries [
12], the debate continues to revolve around specific implementation and organisational issues [
13]. This includes the discussion between those advocating the targeting of patients with a single but high risk factor (e.g. high blood pressure) on one hand and those arguing for an overall absolute risk approach (e.g. on the basis of 10-year risk of CVD), independent of the particular risk factor, on the other hand. There are also diverging views around the introduction of a 'poly-pill', a medication consisting of multiple pharmacological agents at a fixed dose, as a means to provide more generic treatment options (compared to treating each individual risk factor with a specific drug and dose). Some argue that the poly-pill would allow a broader population to access and use pharmacological care, due to lower requirements in risk factor assessment and monitoring. Even though large trials in developing countries have been undertaken to prove the effectiveness of the poly-pill approach [
14], its overall consequences that would capture potential adverse effects, the impact on health inequalities, the consequences of mass medicalization for healthcare budgets in developing countries, with a lot of resources allocated to a few major medications, as well as the role of patient compliance, still awaits a thorough assessment.
By contrast, there has been remarkably little research coverage and discussion of non-clinical, population based approaches, e.g. health promotion through social marketing, or legislative actions as a way to tackle CVDs in developing countries.
Apart from the methodological issues in evaluating these interventions, at least two reasons may help explain this bias in the research. First, research on population based, non-clinical interventions is likely to be subject to a market failure: private actors do not have the incentive to engage in such research, because (a large share of) the resulting knowledge would become a public good that everybody could use, without having to pay the often substantial research costs to arrive at that knowledge. From a sheer economic efficiency (and not even from a moral or public health) perspective, this type of knowledge will be undersupplied compared to the social optimum [
15]. An analysis of the funding sources of the articles included in our review shows that out of the three studies reporting industrial support, two evaluated pharmaceutical interventions and one a medical technology. (Given the few studies reporting a funding source, this observation may of course be of limited generalizability.)
Second, primary data on the effectiveness of specific population interventions typically does not exist for certain countries or regions. Since the results of interventions targeted to change defined health behaviours or implementing social marketing are highly dependent on cultural, infrastructural and other system-related aspects, scientists often argue that it is less feasible than in clinical evidence to transfer such results from developed regions to developing regions [
16]. It is a widely held assumption in pharmaceutical research that a drug affecting biomedical processes would have approximately identical effects, irrespective of the ethnic context in which it is applied. We will scrutinize this hypothesis below, when considering the transferability of results among countries and regions.
Despite our general endorsement of some form of scaled up pharmaceutical support, it is also important to be mindful of the limitations of such a strategy. This is to do with the observation that any approach that defines the benchmark risk level (e.g. on blood pressure) as high as most current approaches recommend, inevitably misses out the typically large amount of people that is below that threshold but nevertheless shows ailments that are related to their (less than nominally "too high") risk factor levels (e.g. blood pressure) [
17]. Even though an approach assessing the total absolute risk of individuals would address some of the issues, it would imply the allocation of (possibly disproportionately) large resources to healthcare for the elderly population due to the high contribution of age to these risk calculations. Clinically managed chronic care often is expensive and may be required for the remaining lifetime. An extension of the target group for treatment, though clinically justified, would cause higher pressures on the already constraint budgets of LIMCs. A population-based approach, such as reducing salt intake, would at least in principle also effect change in the entire population in the long term and not only in the highest risk group. Therefore including the larger group of beneficiaries in this outcome calculation might in some cases render such approaches attractive because it could be more cost-effective. This could be the case even when the overall population risk reduction is limited [
18]. However, strong local level evidence of such a shift to proof these approaches cost-effective, accounting for obstacles in large-scale implementation and financing is still missing and requires further analysis.
In our search we found only a small number of studies assessing strategies to combat tobacco use. This is surprising in light of the otherwise well-established evidence on cost-effective strategies to address smoking-related health loss. In particular, taxation and legislation options have been rather well evaluated, certainly for developed regions and countries but also in developing countries [
19]. The achieved reduction of smoking rates is shown to have lowered the burden of disease caused by CVD by about 36% in the UK [
20]. One explanation for the few studies we identified might lie in our search strategy, which focused on studies concerning primarily CVD. Smoking interventions, by contrast, might be more often labeled in connection with lung diseases or as an independent disease. Indeed, upon closer scrutiny, more evidence for efficient strategies to reduce smoking in developing countries does exist. Those include other review articles, as for instance the 2003 study by Shibuya et al. [
21] or the Chapter in the DCP2 publication by Jha P et al. [
22], both of which describe an increase in tobacco tax as the most cost-effective strategy to reduce smoking prevalence, followed by comprehensive advertisement campaigns and bans on smoking in public places.
In addition, we were surprised to realize that contrary to an earlier review of ours on economic evaluations of primary prevention of CVD in developed countries [
5], in the present review there were nearly no studies evaluating the effects of statins--alone or in combination with other drugs--on dyslipidemia. Neither for primary, nor for secondary prevention did our search reveal any such evidence for developing countries. Even though the first statin, Lovastatin, went off-patent in the US and Europe in 2001, it took until the patent expiration of the popular drug Simvastatin in 2006, for a statin (Simvastatin for high risk patients) to be added to the
World Health Organization Model List of Essential Medicines in 2007 [
23]. Until then, evaluations of strategies targeting dyslipidemia through statins might not have appeared useful for developing countries, since broad access to the drug had not been feasible. In addition, the diagnostic costs for determining blood lipid levels are relatively high, when compared for instance to measuring blood pressure. Therefore targeting specifically dyslipidemia is less feasible in developing countries, since the direct costs (and infrastructure costs) for diagnosis and monitoring of patients would require a considerable share of the scarce resources.
It may have come as a surprise that we found more studies in our review on primary prevention than on secondary prevention of CVD. In interpreting these results, however, it has to be acknowledged that we used the definition of secondary prevention as stated by the American Heart Association [
10], i.e. meaning treating risk factors in patients with established cardiovascular disease (e.g. ischemic heart disease). Some other disciplines, e.g. public health, tend to follow a broader definition of secondary prevention that includes any treatment of evident hypertension or dyslipidemia [
24]. Another influential institution, i.e. the European Medicines Agency [
25], regards these definitions of prevention in CVD as artificial and outdated and prefers to discuss overall CVD risk on a continuum which needs to be tackled by suitable measures.
Methodological aspects
In general, few of the studies adopted a comprehensive perspective in their analysis. The more comprehensive a study is, the easier it is for decision makers to compare the intervention to other alternatives available for funding. This applies to the computation of health benefits as well as to the economic perspective adopted. Only 12 out of the 33 studies included in our review used comprehensive units such as "life years gained" or the surrogate measure of "QALYs" or "DALYs" in their analysis. The remaining studies preferred biomarkers or CVD related incidents, which are easier to measure, but harder to compare to other interventions within or outside the health care sector. No article applied a cost-benefit approach.
Moreover, it is surprising to find that no study explicitly applied a societal perspective to the evaluation. Both the cost-benefit method and the societal perspective would in principle be helpful approaches for decision makers, in particular in developing countries. In these regions budgets are even more constraint and investments in healthcare compete heavily with those in other budgetary sectors, such as education or public infrastructure. To support the decision maker in her task of allocating resources across and within sectors, more comprehensive and hence more comparable evaluations might have been desirable.
Modeling is a useful or indeed often necessary method to produce economic evaluations in particular when certain data is missing or when long-term results represent a core interest of the analysis. Eighteen out of 33 studies included in our review used some form of modeling. In modeling, one main decision concerns what effect measure to apply to compute intervention outcomes. In general, in our review, both efficacy or effectiveness studies are used. Revicki and Frank analyzed the importance of both types of studies for pharmacoeconomic evaluations [
26]. While efficacy studies or RCTs demonstrate the performance of an intervention under ideal and controlled conditions, effectiveness studies show the impact of treatments under regular clinical conditions or "real world" circumstances. Efficacy studies focus on the internal validity of results and therefore accuracy of conclusions--however, their practical use is limited due to potential lack in the generalisability of their results. Effectiveness studies have a more real-life set-up and lead to results of more practical value, increasing external validity--at the potential cost of internal validity. Revicki and Frank conclude that cost-effectiveness studies with RCTs "may provide a very precise answer to the wrong question" [
26]. In general, efficacy rates are higher than effectiveness rates, therefore giving the decision maker a biased impression on success within his population of concern. Due to the lower adherence to treatment guidelines (by doctors and patients), co-morbidities and limited patient monitoring, efficacy rates usually drop in effectiveness trials [
27,
28]. Goldenberg and Glueck [
29] reviewed retrospective studies concerning the goal attainment for statin therapy in managing CVD and found that ~20% of patients did not receive the necessary medication by their doctors when compared to guidelines. Furthermore, only 50% of treated patients achieved lipid-lowering goals with significant consequences for CVD mortality and morbidity. Another study focused on the patient side of adherence in an Italian population of 10,890 patients [
30]. Only half of the patients who started on statins, continued to take the medication after 1 year. In patients for primary prevention of CVD, only 19% adhered to the regimen prescribed by the doctor. Predictors for non-adherence were younger age, total number of daily drug doses and having multiple prescribing physicians. Similar adherence rates can be found in the context of developing countries. Bowry et al. [
31] systematically reviewed studies on the adherence to cardiovascular medication in resource-limited settings and found an average adherence rate of 58% according to pill-count and self-reporting. Common predictors of non-adherence to medication were poor knowledge, negative perceptions about the medication, occurrence of side-effects, high medication costs, and lack of family support. Factors such as age, gender, lifestyle, complex treatment regimens, and lack of access to health care services were not consistently associated with non-adherence.
Revicki and Frank conclude that RCTs are a precondition for conducting effectiveness studies. However, for providing the decision maker with relevant information about the pharmacoeconomic outcomes of an intervention, evaluations based on RCTs are of limited use, particular in a community setting [
26].
In our sample of studies applying a modeling approach, 17 out of 18 studies incorporated large RCTs to calculate health benefit outcomes, either as single source or in metaanalyses. Moreover, these studies were all conducted in developed countries instead of the country under analysis or a country with similar conditions. This problem of transferring results will be discussed in detail below.
Transferability of results between regions--opportunities and limitations
Conducting original economic evaluations for every intervention in every LMIC is well beyond the means of most developing countries' monetary and human resources. This general lack of capacity has to be considered when analyzing the studies and drawing conclusions. For example, a third of studies included in our review were conducted by authors at institutions in solely developed countries (n = 11).
Hence the idea of transferring results from one country to another, in particular from developed to developing countries, has always been a potentially attractive and fairly widely accepted alternative for researchers and decision-makers. However, this approach also bears several challenges, especially including differences in health system costs across countries, differential effectiveness of the same intervention, differential disease prevalence, differential valuation of outcomes, and differential efficiencies in the implementation of interventions. In what follows we discuss these issues in light of the existing cost-effectiveness evidence for CVD interventions.
Use of external data
Disease modeling is widely applied in research on developing countries, as is shown extensively in our review. Modeling approaches transfer data on disease epidemiology, risk factor associations, relative clinical efficacy, resource utilization, and unit cost from the country where the original study took place to a target country of interest. Half of the studies included in our review used this type of information in a model-based evaluation of interventions. A disease model is expected to incorporate as much data from the examined country [
32] as possible, e.g. information on disease and strength and prevalence of risk factors, effectiveness of interventions within the population, the resource use needed, as well as prices for goods and services. Obviously, not all data is always available for all developing countries for reasons mentioned already above. This forces the scientist to transfer data from countries where this information is available to the country under analysis. Among those 18 studies that did use modeling in our review, none used external data on prices for services and goods, three used epidemiological and risk factor data from other countries, five used data on resource use from other settings, and all 18 studies used aggregated efficacy data from other healthcare settings to model the cost-effectiveness of interventions, without accounting for the presence of other risk factors. This shows that in particular data on effectiveness (and efficacy) of multiple and combined risk factor interventions is scarce for developing countries. In particular, large RCTs or meta-analyses of effectiveness studies are missing. Most estimates for effectiveness in those cases were based on data from developed countries. A study by Goeree et al. [
33] confirms that this is a general obstacle when transferring data among regions. Goeree et al. analyzed 40 economic evaluations, which had tried to transfer results of studies to other geographic areas (not necessarily developed to developing countries). They developed a scoring system for the comprehensiveness of transferability (see Table
2 Modeling approaches based on the three most commonly advocated transferability factors).
Table 2
Modeling approaches based on the three most commonly advocated transferability factors (adopted from Goeree et al.2007 [
33])
Least to most country-specific analysis ↓ | 1 | Studied country only | Studied country only | Mixture of studied and target country |
| 2 | Studied country only | Studied country only | Target country only |
| 3 | Studied country only | Mixture of studied and target country | Target country only |
| 4 | Studied country only | Target country only | Target country only |
| 5 | Target country only | Target country only | Target country only |
While costs (in 39 out of 40 studies) and (some) resource use (28/40) was described as being most often adapted to the local setting, clinical efficacy was directly transferred from one geographic area to another. Only two out of 40 studies used efficacy data from the target country in their analysis. We applied the scoring system to the articles included in our sub-selection of modeled studies. Out of the 18 studies, none were category 1, six would be classified as category 2, two as category 3, seven as category 4, and none could fall into category 5. Three studies did not supply enough information for an explicit classification.
It can be expected that the effectiveness of risk factor and disease interventions will differ between developing and developed countries, given the often large differences in cultural, economic, infrastructural, and health care aspects and differences in risk factor epidemiology. In addition, biological differences may exist between ethnic groups [
34], possibly based on pharmacogenetics, which might in some cases contribute to differences in the efficacy of certain drugs.
There are no specific guidelines on how to handle this type of uncertainty in modeling interventions for developing countries. While the guidelines for conducting a CEA in the methods section of the DCP2 acknowledges the lack of sufficient effectiveness data, no recommendations for transferring data from developed countries to the context of low resource settings is provided. One explanation suggests, however, that:
"Besides the quality of the evidence at its source, how the results will apply to other settings matters, particularly when the data are limited to high-income countries. The more that outcomes depend on underlying biology, the more the findings will apply to low- and middle-income countries. Outcomes depending more on cultural or environmental factors are less readily transferred and require judgment and evidence as to their applicability elsewhere." [
16]
This may reflect the fact that many researchers believe or assume that clinical effects (in particular biological effects) of the interventions are transferable across health care systems but resource use and unit costs are more location specific. This is often common practice in HIC settings, while others tend to be more cautious in this respect: the WHO-CHOICE project developed in their guidelines a method on how to deal with differences in effectiveness. Effectiveness for developing countries is obtained by adjusting efficacy of clinical studies in developed countries by a factor between 0 and 1, based on the literature--or expert opinion as a last resort to account for these uncertainties [
35]. Even though this method seems basic and its validity is not proven, it emphasizes the need to consider these influences in modeling [
28].
As in all quantitative research, the use of valid and/or appropriate mathematical models is key when evaluating interventions in a generic way and for long-term [
36]. Its critical assumptions may affect outcomes directly. The validity of the used results from randomized controlled trials for a particular study population may be limited as there can be a bias [
37]. The impact of both the model and trial assumptions may be equally large. A thorough assessment of the models used in the reviewed papers, while desirable in principle, has on the whole not been possible due to the lack of detailed information on the precise model used in the studies. To allow more transparent analysis and appreciation of the modeling, we recommend, however, that future modeling analyses employ check-lists on good-modeling practices [
27]. Unal et al. [
38] already systematically assessed the quality of 42 different models on cardiovascular disease and stated that only 5 (12%) of them were comprehensive enough, considering both all relevant risk factors and types of treatments. The authors describe a vast variety in the quality and utility of mathematical models, in particular limited validation and calibration against observed data. Quality standards for modeling studies should be part of review processes (such as proposed by ISPOR [
39]).