Cost effective interventions for the prevention of cardiovascular disease in low and middle income countries: a systematic review

Background While there is good evidence to show that behavioural and lifestyle interventions can reduce cardiovascular disease risk factors in affluent settings, less evidence exists in lower income settings. This study systematically assesses the evidence on cost-effectiveness for preventive cardiovascular interventions in low and middle-income settings. Methods Design: Systematic review of economic evaluations on interventions for prevention of cardiovascular disease. Data sources: PubMed, Web of Knowledge, Scopus and Embase, Opensigle, the Cochrane database, Business Source Complete, the NHS Economic Evaluations Database, reference lists and email contact with experts. Eligibility criteria for selecting studies: we included economic evaluations conducted in adults, reporting the effect of interventions to prevent cardiovascular disease in low and middle income countries as defined by the World Bank. The primary outcome was a change in cardiovascular disease occurrence including coronary heart disease, heart failure and stroke. Data extraction: After selection of the studies, data were extracted by two independent investigators using a previously constructed tool and quality was evaluated using Drummond’s quality assessment score. Results From 9731 search results we found 16 studies, which presented economic outcomes for interventions to prevent cardiovascular disease in low and middle income settings, with most of these reporting positive cost effectiveness results. When the same interventions were evaluated across settings, within and between papers, the likelihood of an intervention being judged cost effective was generally lower in regions with lowest gross national income. While population based interventions were in most cases more cost effective, cost effectiveness estimates for individual pharmacological interventions were overall based upon a stronger evidence base. Conclusions While more studies of cardiovascular preventive interventions are needed in low and mid income settings, the available high-level of evidence supports a wide range of interventions for the prevention of cardiovascular disease as being cost effective across all world regions.


Background
Chronic diseases were estimated to account for approximately 50% of the total disease burden in low and middleincome countries in 2005 with further marked increases expected in the coming years [1].
It has been shown that the concomitant modification of multiple known risk factors (principally blood pressure and serum cholesterol concentration) could reduce cardiovascular disease to a large extent [2].
Both pharmacological and non-pharmacological strategies are likely to have a key role in tackling Cardiovascular Disease (CVD) in low and middle income countries [3]; non pharmacological strategies because of their potential for wide dissemination as well as their ability to be delivered more cheaply than pharmacological strategies to low and middle income populations, [4][5][6] pharmacological strategies because of the large absolute benefits conferred to those treated and the greater certainty in attribution of benefits [7].
While there is evidence to show that population-based and lifestyle interventions can reduce cardiovascular disease risk factors in affluent settings [8], as well as some evidence supportive of longer-term benefits in disease reduction, [6] less evidence exists in lower income settings. To generalise results from high income setting is not entirely satisfactory because reasonable thresholds for cost effectiveness will vary markedlyas will affordability [9]. Additionally setting specific information is important because population-based and preventive interventions are often, to some extent, context specific.
In this paper we evaluate and summarise the existing evidence on the cost-effectiveness of interventions for the prevention (primary and secondary) of cardiovascular disease in low and middle income countries. Furthermore, we describe how the level of cost-effectiveness differs by setting and intervention type.

Eligibility criteria
Studies were included if they were [i] randomised controlled trials assessing any cardio-protective intervention to prevent fatal or non-fatal CVD events (including myocardial infarction, coronary heart disease, stroke and heart failure); [ii] cohort, case-control, cross sectional studies or controlled trials reporting economic outcomes, or studies utilising the results of such studies to model economic outcomes; [iii] reported economic outcomes in terms of costs per YLG (years of life gained)/events averted, or cost-utility ratios, (ie, cost per QALY (quality adjusted life year) or DALY (disability adjusted life year)) of interventions aimed to prevent CVD; [iv] included adult participants (≥ 18 years old); and [v] published in any language. We excluded studies if they [vi] were letters, abstracts, case reports, editorials, descriptive studies, ecological studies or conference proceedings; [vii] involved nonhuman subjects; [viii] were conducted in affluent settings/ rich countries (see Additional file 1: Appendix 1); [

Study selection
Title and abstract for all studies identified by our search were screened by two independent reviewers (AS, RA, SS, TH, RC, PB, OhF) against our eligibility criteria to determine inclusion for full text review. AS reviewed all abstracts and RA, SS, TH, RC, PB and OhF equally shared the task of reviewing a duplicate list of all abstracts. Eligibility criteria were systematically applied to each abstract to derive a list for full text review, where an abstract was rejected the criteria barring eligibility were noted. In this way 2 lists of articles for further review were produced. These were then compared and disagreements were resolved by discussion; where disagreement persisted a third investigator was consulted (RA, OhF). In this way we arrived at an agreed list of articles for full text review.
All full text manuscripts were then assessed using a standardised checklist to ascertain definitively whether they met all eligibility criteria for this review. This was done independently by two reviewers (AS, OhF). Each reviewer then compared their selection with that of the other, reassessing against eligibility criteria in all cases of disagreement. Disagreements which persisted were resolved through discussion. Where agreement was not reached the opinion of a third party was sought (RA). The remaining studies were included within this review. For definitions used see Additional file 1: Appendix 1. For income groupings see Additional file 2: Appendix 2.

Data collection process
From each study selected for inclusion we extracted a prespecified set of data items using a data extraction form which was piloted before use. Data extraction was carried out by two independent reviewers (AS, OhF). The two reviewers compared data extraction results, resolving disagreements by discussion, before producing a final data extraction form which was entered into Epi Info.

Data items
We extracted data on year of publication, study setting, geographic origin of publication, publication date, target population, intervention type, whether embedded within a trial, the nature analysis undertaken, modelling techniques used, main economic findings and funding source. We also extracted data on the analytic parameters used. To aid synthesis of results intervention effect estimates were categorised by metric used, the setting they related to and by intervention type.

Risk of bias in individual studies (Quality review)
The quality of included articles was rated independently by 2 reviewers (AS, SS) according to the checklist for economic evaluations produced by Drummond [10]. We chose to present findings of quality review as either ++, +, oras has been used by The UK National Institute for Health and Clinical Excellence (NICE) [11].

Synthesis of results
We used broad categories of cost effectiveness to compare results between studies which we considered useful to facilitate comparison between studies while allowing for setting specific variation in costs and effects. Categories used are those suggested by WHO, whereby if cost/DALY ≤ (Gross national income) GNI per capita the intervention of interest would be considered: i) very cost effective, ii) with a cost/DALY of 1-3 times GNI per capita it would be categorised as cost effective and iii) with a cost/DALY more than 3 times GNI per capita classified as "not cost effective" [12]. We added a further category iv) of "extremely cost effective" in order to further differentiate cost effectiveness results. We arbitrarily defined this category as ¼ GNI per capita per DALY gained.

Risk of bias across studies
We examined whether there was a systematic difference in the quality of evidence underpinning effectiveness estimates according to the modality of intervention. (pharmacological vs lifestyle interventions).
Study conducted and reported in line with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses PRISMA statement guidelines [13].

Study selection
Our initial search yielded 9729 results from all databases with two further studies obtained based upon responses of experts who were contacted. From these 9731 search results 93 studies were retrieved in full text after review of title and abstract against eligibility criteria. Review of the reference lists of the retrieved studies did not yield any additional studies.
After review of the full text of these 93 studies a further 77 were excluded leaving 16 articles, which met our search criteria. In each case the criteria by which a study was deemed ineligible was recorded. Most studies were excluded at this stage because they were not of the study type required [14], or they did not provide results in terms of QALYs/DALYS/LYG. Eight studies were excluded due to being conducted in affluent settings ( Figure 1).

Study characteristics
The majority (12/16) of studies retrieved were published within 5 years of this review. Eight papers evaluated pharmacological interventions only, while a further 8 papers evaluated a mixture of both pharmacological and lifestyle interventions. None of the included papers evaluated only lifestyle interventions (Table 1).
In 6/16 cases the study originated from the setting of interest, with the other 10 studies originating from the USA or Europe. Within included papers over 20 different preventive interventions were evaluated in total. Geographical categories in which interventions were evaluated included World Bank and World Health Organisation (WHO) regions as well as individual country level.
Most studies (15/16) were based on stochastic simulation and we found only one economic evaluation embedded within an intervention study [45]. In 14/16 cases some form of sensitivity analysis was undertaken. In most cases (9/16) where this was carried out the impact of changes in the most consequential variable had a large (>1 order of magnitude) impact upon results.

Results of individual studies
All of the papers retrieved presented positive results, supporting some or all of the interventions considered as cost effective in the setting/s of interest. Where study authors categorised cost effectiveness using GNI the same thresholds as applied here were used. In all but one case [46] we arrived at the same categorisation as study authors.

Tobacco control
We found 6 studies that evaluated tobacco control interventions in one or more low and middle income country [5,15,17,32,34,36]. Although personal interventions such as nicotine replacement therapy (NRT) were generally found to be cost effective, population-based interventions were much more cost effective (by an order of 10-100 fold) (see Figure 2).
Pharmacological primary prevention using an absolute risk based approach 7 studies evaluated the cost effectiveness of providing preventive medication on the basis of absolute risk. [15,17,18,20,23,24,30] 5 of these found this approach to be "very cost effective" or "cost effective" in all settings for which they evaluated this intervention, [17,18,23,24,30] with cost effectiveness generally increasing at higher risk thresholds for treatment (see Figure 2).
Less favourably, studies in Tanzania [20] and Kyrgyzstan [15] reported borderline and negative results respectively for the cost effectiveness of an absolute risk based approach.

Individual risk factor reduction approach
Of the 16 studies included, 5 evaluated the use of pharmaceuticals for individual risk factor lowering [15,[23][24][25]30]. Drugs to lower "high" blood pressure were found to be in the "very cost effective" or "cost effective" range in all studies. In the case of statins, 3 studies reported them to be cost effective [23][24][25] while 2 studies found them to be not cost effective [15,17].

Pharmacological secondary prevention
All studies evaluating pharmacological secondary prevention of CVD found this approach to be in the "cost effective" or "very cost effective" range [15,46,47].

Use of mass media
We found 5 studies, which evaluated the use of mass media to reduce salt consumption, to stop smoking, to improve dietary characteristics as well as joint campaigns [12,15,17,24,32].
The source of effect estimates underpinning cost effectiveness calculations were derived from a mixture of trial data, observational data and expert opinion. The use of mass media was generally found to be "very cost effective" (Figure 2) [12,15,17,24,32].

Other legislative interventions
Four of the studies included reported evaluations of legislative interventions such as the compulsory reduction of salt in food [12,17,23,32] (Figure 2). Interventions of this type were generally found to be very cost effective, or extremely cost effective, with the exception of salt lowering legislation in Sub Saharan Africa which was found not to be cost effective [32].

Provenance of studies and study estimates of costs and effects
In all but one case [45] the population-based interventions estimates of effect size were derived from settings other than the setting of interest. Effect estimates were obtained utilising the results from a mixture of observational and experimental studies. We found variation in  Meta analysis [16] Reduced relative risk of CVD RR=0.12 [17] for cardiovascular disease Estimate based on RCT evidence [2] RR=0.29 for IHD and 0.4 for stroke (primary prevention) [18] Overview of RCTs [19] RR=0.12 for CHD and RR=0.2 for stroke [20] Multiplicative effects [21,22] Reduction in BP and cholesterol + reduced absolute risk (to account for effects of aspirin) 20% reduction in cholesterol+33% reduction in difference in BP between 115** and current + 20% reduction absolute risk CVD (to account for benefits aspirin) [23] Product of estimates from RCT estimates used for Cholesterol and BP. For Aspirin [16] 20% reduction in cholesterol+28% reduction in difference in BP between 115** and current + 18% reduction absolute risk CVD (to account for benefits aspirin) [24] Product of estimates from RCT estimates used for Cholesterol and BP. For Aspirin [16] Treatment of "high"cholesterol Reduction in total serum cholesterol concentration −20% [15]) RCT [21] −20% [23] RCT [21] −22% [25]) RCT [26] Reduction in relative risk of cardiovascular disease RR=0.84 [20] Heart Protection Study Group [21].
−15% [15] No reference for impact on salt intake, impact of salt reduction on BP from trial data [41] Reduced CVD prevalence −4% [32] Review [14] and expert opinion Combined mass media Relative risk of CVD RR=0.98 [17] Meta analysis [42] Legislative Interventions Salt in bread-voluntary/other Reduced CVD relative risk RR=0.99 [17] No reference for impact of legislation, review of trials supports impact of salt on CVD [43] Legislation on salt in food Reduction in total dietary salt intake 30% reduction [23] No reference for impact of legislation. Impact of salt on BP from observational data [44] Reduced salt intake via legislation + education Reduced systolic BP −2 mmHg (1-4) mmHg [32] Review [14] and expert opinion Reduction in the difference between actual SBP & 115 mmHg 33% reduction [15] *Where multiple source citations provided the highest in hierarchy of evidence is shown. **115 mmHg suggested as theoretical minimum risk level for systolic blood pressure by WHO. the nature of effects modelled for the same intervention as well as in the magnitude of change that intervention was assumed to produce (see Table 2).
In the case of studies evaluating policy interventions, all effect estimates were based on the observed experience of other locations implementing such policies and/ or expert opinion. Effects on risk factor levels were used to model expected changes in mortality.
Costs for population-based interventions were generally based upon a theoretical estimation of likely costs derived from summing individual strategy components, rather than by measuring the cost of delivering the intervention as a whole in a real life setting.

Parameters used in economic models
Most studies (11/16) used a 3% discount rate for costs and effects and all but one [46] used the same discount rate for costs and effects.
In the majority of papers retrieved (10/16) adherence was not incorporated directly in modelling, although by using trial data adherence was in effect incorporated at high levels in others. Where modelled, levels of adherence ranging from 50% [17] to 95% [15,17,24,25,47] were used.
Six studies used a lifetime time horizon, five used a 10 year horizon, one used 20 years and one 5 years with the remaining three studies not reporting the time horizon considered. Where altered in sensitivity analysis the potential impact of the chosen time horizon upon overall cost effectiveness was of at least one order of magnitude [45,46].

Discussion
Economic evaluations of cardiovascular prevention in low and middle income countries have found a wide range of interventions to be cost effective across all world regions. Given the limited evidence base, findings should be interpreted with caution; yet can aid rational resource allocation and implementation.
Cost effectiveness estimates for pharmacological interventions were generally supported by stronger evidence than those for other interventions. Additionally agreement on effect sizes between studies was generally greater for pharmacological interventions.
Virtually none of the evaluations are fully based on data derived from LMICs. We found a consistent difference between the sources of the effect estimates for population-based interventions compared to personal interventions, with the latter generally based upon studies lower in the hierarchy of evidence ( Table 2). The effectiveness estimates on personal, pharmacologic interventions for the most part are based on studies with reliable effect sizes in high income countries, supplemented with LMIC cost estimates. The evaluations of population-based interventions lack any RCT level of evidence which leads   to greater uncertainties. Whereas individual interventions may be attractive, although based on HIC evidence, this is not the case in the area of tobacco control. Personal interventions such as Nicotine Replacement Therapy (NRT) appear to be far less cost effective than population-based interventions in low-resource settings. There are few trials of population-based interventions in low and middle income countries, leading to a lack of effectiveness information (Table 2). Our findings are relevant for policy makers at the international level e.g. UN agencies with responsibilities in health across societal sectors, national government agencies and ministries, local private, non-profit and for-profit organizations in health care as well as professional medical societies and other health professional bodies. Given the uncertainties in the study findings, local relevance should be assessed, given disease epidemiology and available resources, and, next, taken into account, while making decisions and formulating country policies and guidelines.
Several methodological observations can be made. Most importantly, study methodologies across the identified articles are highly heterogeneous, in terms of analytic methods, input parameters and data used and the baseline against which the intervention of interest was evaluated. We utilise a null baseline to aid comparison, real world costs of implementing a given intervention may be higher if on going activities need to be wound down. Choice of time horizon was of particular consequence and some interventions were found on sensitivity analysis to move from cost ineffective as this parameter was altered [46,47]. The adoption of standard parameters for discounting rates, time horizons and study perspectives would help address this as would widespread adoption of a standard (counterfactual or null) baseline as proposed by WHO [12]. The published articles and background documentations do not allow for a detailed analysis of how this diversity would affect our comparison of the study findings nor do they allow adaptations of the calculations to facilitate a better comparison.
Next, most studies use known or predicted changes in risk factors, associated with each intervention of interest, within a stochastic or deterministic model, to estimate the anticipated changes in disease occurrence that would result. The Framingham equations, employed in most studies, also limit the reliability of results as they under predict risk in high risk populations while over predicting risk in low risk populations [50][51][52][53] Resulting cost effectiveness estimates may therefore be unduly favourable in low risk populations and vice versa. Lastly, most studies do not distinguish trial efficacy results from real-life effectiveness of implemented interventions. Especially, in many rural and urban areas, provider compliance, system compliance, and patient compliance, may lower the impact of the intervention and may raise the health care costs and broader societal costs for patients or the existing systems (see Additional file 3: Appendix 3).
Finally, the limitations of the cost effectiveness categorisation we have used should be acknowledged. Specifically, presently labelling an intervention as "cost effective" at less than 3 x GNI/capita does not necessarily imply that it should be adopted. Shifting resources from this to another intervention, even a very cost effective one, could lead to unacceptable transaction costs and be unwise if there are other compelling societal reasons to allocate resources in a different way.

Conclusions
In sum, there is evidence supportive of a wide range of interventions to prevent cardiovascular disease in most parts of the world; nevertheless further setting specific research of preventive interventions in this field is needed and should include economic evaluation. Lifestyle interventions appear to be of generally greater cost effectiveness, while pharmacological interventions offer an impact of greater certainty and magnitude. These modalities of interventions can thus be seen as complementary, offsetting potential gain against certainty of outcome. Policymakers should aim to balance distribution of relevant resources between these areas, favouring the most cost effective in each class, while accounting for other criteria such as, affordability, access, and equity. Healthcare infrastructures concerned differ markedly [54], both among countries and within countries.
The economic evidence on both pharmacological and lifestyle interventions supports large-scale implementation strategies and efforts in all settings confronted with the growing NCD epidemic.

Additional files
Additional file 1: Appendix 1. Definitions used in this review.
Additional file 3: Appendix 3. Findings from studies reporting costs per treated individual. Per capita costs for selected interventions to prevent cardiovascular disease in LMIC settings compared to per capita expenditure on health for the year considered in that study.