FormalPara Key Points for Decision Makers

Current evidence suggests that early CVD detection strategies are predominantly cost effective compared with no early detection.

Direct comparison between study outcomes is complicated due to lack of standardisation, but this review is capable of guiding future research towards the most promising early detection strategies.

1 Introduction

The global cardiovascular disease (CVD) burden has been steadily increasing over time with prevalence almost doubling between 1990 and 2019 [1]. Consequently, CVD has become the most prominent cause of death and led to 17.9 million deaths and 365.8 million disability-adjusted life years (DALYs) worldwide in 2017 [2, 3].

The main challenge in reducing the CVD burden is that progression is often unnoticed as CVD is typically asymptomatic in its early stages. Moreover, when symptoms become apparent in later stages, this is often in the form of life-threatening events, such as acute myocardial infarction and ischaemic stroke. Despite recent advances in CVD treatment, acute care remains very expensive and cannot always prevent premature death or (permanent) disability reducing quality of life. With its high prevalence and high treatment costs following events, CVD puts major pressure on constrained healthcare budgets [4]. In 2017 alone, the cost of CVD in the European Union was 210 billion Euros, of which 111 billion Euros were attributed to direct healthcare costs, such as diagnostic tests and treatment, 54 billion Euros were attributed to productivity losses and 45 billion Euros to informal care [5].

Previous studies have argued that the CVD burden may be reduced more efficiently through preventive strategies than curative strategies [2, 6,7,8]. Preventive strategies may rely on screening for CVD risk factors or early detection of CVD in asymptomatic individuals to identify individuals who could benefit from preventive medication, such as anti-hypertensive drugs and statins, or lifestyle changes [9]. Early treatment of these individuals at high risk of CVD may prevent the occurrence of life-threatening cardiovascular events and hospital admissions, resulting in potential health benefits and reduced CVD costs.

To assess the balance between the health benefits of early detection strategies and their costs, health economic evaluations have become increasingly important [10]. It is essential for such analyses to estimate both short-term and long-term health effects and costs, as the time between preventive interventions, initiated after early detection, and the resulting future health benefits may be considerable. Clinical trials are suitable to determine short-term outcomes (e.g. occurring within 1–5 years), but only some can be used to determine long-term outcomes, due to time and budget constraints. Consequently, the long-term health and economic impact of early detection strategies are increasingly evaluated by using simulation models. Simulation models are particularly valuable in this context since they allow the estimation of unobserved long-term health outcomes and costs by extrapolating observed intermediate outcomes, such as the yield of early detection strategies. However, choices and assumptions made during modelling may influence health economic outcomes [11]. Therefore, reported outcomes can only be interpreted in light of the choices and assumptions underlying the model and analysis.

Currently, it is unknown which early detection strategies for (risk factors of) CVD are (most) efficient, as systematic reviews concerning the cost effectiveness of such strategies are scarce. To our knowledge, only one systematic review including publications from 2005 to 2015 reported on both the health and economic impact of screening strategies for cardiometabolic diseases [12]. This review showed large heterogeneity in study objectives, country setting, comparators, methodology, outcomes, and screening programmes between studies and between healthcare systems in different countries. Consequently, the authors were unable to make uniform policy recommendations. Moreover, evidence on screening for CVD was limited, as only three out of the 17 included studies focussed explicitly on CVD. Even though specific (cost) outcomes cannot be directly generalised to other countries, a review on different early detection strategies can be valuable to guide future country-specific research towards most the promising strategies, as (new) early detection strategies are identified and health outcomes on different populations are reported. This study aims to systematically review recent health economic evaluations assessing the cost effectiveness of recent early detection strategies targeting CVD in adult populations without prior CVD diagnosis and at risk of developing CVD.

2 Methods

2.1 Literature Search

A literature search was performed in the online databases PubMed and Scopus. The search strategy included a broad range of early detection strategies for CVD (search queries can be found in Electronic Supplementary Material [ESM] 1). Early detection strategies were defined as strategies aimed at screening for risk factors of CVD or the early detection of CVD in asymptomatic individuals without a previous cardiovascular diagnosis. Given the recent increase in interest in early CVD detection and continuing where the review mentioned in the introduction [12] stopped, only articles published from 1 January 2016 until 30 April 2022 were included. This systematic review (CRD42022321585 in the International Prospective Registry of Ongoing Systematic Reviews) was structured according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (ESM 2) [13].

Covidence software was used for screening articles identified through the search and for data extraction (version 2819 47a0a5d6, Veritas Health Innovation, Australia) [14].

The following inclusion criteria were applied:

  1. 1.

    the study was healthcare-related;

  2. 2.

    the targeted disease was a CVD, defined as diseases within the vascular bed or cardiac area;

  3. 3.

    a full health economic evaluation was performed (i.e. comparing costs and health outcomes of multiple strategies);

  4. 4.

    an early detection strategy was evaluated, defined as a strategy aimed at screening for risk factors for CVD or early detection of CVD in asymptomatic individuals with subsequent appropriate treatment in identified individuals.

The following exclusion criteria were applied:

  1. 1.

    the study focused on animal testing;

  2. 2.

    the target individuals for the early detection strategy were < 18 years old, as populations < 18 years are defined as paediatric care and the majority of CVD events occur in adults;

  3. 3.

    the study focused on secondary prevention, defined as early detection strategies focusing on patients with a prior cardiovascular diagnosis;

  4. 4.

    the study used a time horizon < 1 year.

Assessment of eligibility was performed by one author (MOW), with a second author (CB) screening a random sample of 10% of the title and abstracts and another 10% of the full texts for validation purposes. Discrepancies regarding the inclusion of studies between MOW and CB were discussed and resolved through discussion with a third author (HK).

2.2 Data Extraction

Data extraction was performed by one author (MOW). The extracted data were categorised into three sections: general study information and PICOTS (Patient Population, Intervention, Comparator, Outcome, Time, Setting), methodology and outcomes.

2.2.1 General Information and PICOTS

The general study information included first author, country in which the study was conducted (if not explicitly mentioned, the country affiliation of the first author was used as proxy), and year of publication. The PICOTS section consisted of mean age and standard deviation (SD) of target population, sex (% female), intervention, comparator, outcome, time horizon, (clinical) setting where the early detection strategy was initiated, and perspective (in case of multiple applied perspectives, the broadest perspective was reported).

2.2.2 Methodology

The methodology included the type of early detection strategy (screening for risk factors or early diagnosis of CVD), subsequent management of high-risk individuals or asymptomatic patients, type of included costs, currency, type of health economic analysis, discount rate(s), willingness-to-pay (WTP) threshold, description of different variations on initial strategy assessed (if any), subgroup analyses (if any) and type of study (trial-based or model-based). For trial-based health economic evaluations, additional information regarding study design, inclusion start and end date, duration of follow-up, and method for uncertainty analysis was extracted. For model-based health economic evaluations, information on the type of model, cycle length (for discrete-time models), (time-dependent) estimation of CVD risk based on population characteristics, whether a deterministic sensitivity analysis was performed and whether a probabilistic analysis was performed was extracted. Finally, any methods for (model) validation described were extracted.

2.2.3 Outcomes

Regarding results, the following items were extracted: total and incremental costs of the early detection strategy and comparator per person, type of reported health outcomes, total and incremental health outcomes for the early detection strategy and comparator, the incremental cost-effectiveness ratio (ICER) of the base-case analysis, the probability of the (optimal) early detection strategy being cost effective at the applied WTP threshold, and the reported conclusion on cost effectiveness. The following items could also be estimated based on information in the text, tables or figures: total and incremental costs of the early detection strategy and comparator per person, total and incremental health outcomes for the early detection strategy and comparator, the ICER of the base-case analysis, and the probability of the (optimal) early detection strategy being cost effective at the applied WTP threshold. All costs were, if necessary, first converted to Euros using historical exchange rates using OECD data [15] and subsequently indexed to 2021 Euros using Dutch consumer price indices [16]. In case multiple early detection strategies were compared within the same study, the best strategy was described, that is, the strategy providing the highest health benefits with an ICER still below the reported WTP threshold. Converted incremental costs and quality-adjusted life-years were plotted in an incremental cost-effectiveness plane for study comparison. As different WTP thresholds were applied in different countries over the world, all outcomes were also subsequently compared with the Dutch WTP threshold. All results were presented per targeted disease.

2.3 Reporting Quality

The 2022 CHEERS checklist was applied to assess the reporting quality of all included articles according to 28 items [17]. An item could receive a score of 0 (insufficient) or 1 (sufficient) for each item. Subsequently, all points were aggregated and divided by the maximum points that could be received (28) to determine the quality score. A score of 85% or higher was considered high quality, between 60% and 85% medium quality, and < 60% as low quality. The reporting quality was assessed over time and per item.

3 Results

3.1 Screening

Of the 4994 unique articles that were identified in PubMed and Scopus after deduplication, 50 articles were considered for data extraction, as shown in the PRISMA flowchart (Fig. 1). As two included articles reported on the same study in different forms (i.e. one as a journal article [18] and one as a report [19]), the journal article was excluded from data extraction. Finally, two articles studied the impact of the same early detection strategy in different contexts, presumably using the same health economic simulation model [20, 21], resulting in 48 unique early detection strategies included for data extraction. Detailed information on screening can be found in ESM 3.

Fig. 1
figure 1

2020 PRISMA flow chart showing the reviewing process

Full size image

3.2 General Characteristics

The general characteristics of the included articles are shown in Table 1 per targeted disease. The country perspectives most often used were the United States in nine articles (18.4%), Sweden in six articles (12.2%), United Kingdom in six articles (12.2%), and the Netherlands in five articles (10.2%). Only three articles (6.1%) focussed on low- and middle-income countries, namely Nigeria [22, 23] and Vietnam [24]. One study [20] compared the cost effectiveness of an early detection strategy in multiple European countries including Serbia, which is considered a middle-income country. More than half (55.1%) of the studies were funded by non-profit organisations. Ten out of 49 articles (20.4%) reported funding from industry. A quarter of all articles (24.5%) did not report any funding (ESM 4). The target populations ranged from samples of the general population to specific patient groups, such as patients with diabetes mellitus type 2 and autosomal dominant polycystic kidney disease. Most studies compared an early detection strategy with no early detection (n = 41, 83.7%). The type of early detection strategies varied substantially between studies from CVD risk prediction tools to identify high-risk individuals to ultrasound scans and computed tomography-based calcium scoring to identify aneurysms and coronary artery disease in asymptomatic individuals. Time horizons ranged from 10 years to lifetime. Out of 49, 34 articles (69.4%) focussed on early detection of CVD in asymptomatic patients, whereas the other 15 articles (30.6%) described screening for CVD risk factors. Early detection of atrial fibrillation in asymptomatic patients was most frequently assessed (30.6%), followed by abdominal aortic aneurysm (18.4%), hypertension (14.3%), and 10-year CVD risk based on Framingham or SCORE risk prediction models (12.2%).

Table 1 General characteristics of included articles
Full size table

3.3 Health Economic Methodology

Methodological characteristics of the health economic evaluations are described in Table 2 per targeted disease. Costs were discounted between 1.5% and 5%, whereas health effects were discounted between 0 and 5%. All studies used simulation modelling to perform the health economic analysis. Markov cohort state transition models (STMs) were used most often, in 27 articles (55.1%), followed by the combination of a decision tree and Markov cohort STM (n = 11, 22.4%), microsimulation (n = 5, 10.2%), discrete event simulation (n = 3, 6.1%), and a decision tree (n = 1, 2.0%). Two studies did not report which model type they used [25, 26]. Real-world data from observational studies, registries and national databases were included to inform disease incidence in 18 articles (36.7%), mortality in 30 articles (61.2%), treatment effectiveness in seven articles (14.3%), utility in five articles (10.2%), and resource use and costs in 24 articles (49%) (ESM 5). Overall, 38 studies (77.6%) included a deterministic sensitivity analysis, 33 studies (67.3%) included a probabilistic sensitivity analysis and 26 (53.1%) included both. Finally, only 13 articles (26.5%) reported on (types of) model validation applied (ESM 5).

Table 2 Methodology of health economic evaluation
Full size table

3.4 Outcomes

The outcomes of all health economic evaluations are summarised in Table 3 per targeted disease. The majority of studies (n = 47, 95.9%) reported quality-adjusted life-years (QALYs) as the primary health outcome. One article (2.0%) [23] reported disability-adjusted life-years (DALYs) as the main health outcome and one article (2.0%) [25] reported CVD events as the main health outcome. All studies considered direct healthcare costs (i.e. costs of the early detection strategy plus potential events) during the time horizon. Additionally, four studies included costs related to productivity loss and four articles included costs for patients and family. Finally, two studies reported costs per life-year gained. Total costs per person for both the intervention and usual care strategy were reported in 34 articles (69.4%), ranging from €69 to €373,884 for the intervention strategy and €42 to €317,074 for the control strategy. Incremental costs were reported or could be calculated based on total costs in 43 articles (87.8%) and ranged from − €127,266 to €2810 per person. Similarly, average QALYs per person for the intervention and usual care strategy were reported or could be calculated in 30 studies (61.2%) and ranged from 5.96 to 27.41 in the intervention strategy and from 5.95 to 27.35 in the usual care strategy. Incremental QALYs were reported or could be calculated in 40 articles (83.7%) and ranged from 0.00 to 2.87. All outcomes of articles of both incremental costs and incremental health effects were reported or could be calculated are shown in Fig. 2. In total, 47 studies reported ICERs (i.e. costs per QALY gained), ranging from dominance to about €340,000 per QALY gained. Converted WTP thresholds ranged from €1808 to €99,364. In total, 43 out of 49 unique early detection strategies (87.8%) were reported to be cost effective (considering only the best early detection strategy reported per article in case multiple strategies were reported) and 11 articles (22.5%) reported their early detection strategy to be dominant over the comparator considering the applied time horizon. Considering the four most targeted diseases, four out of 15 early detection strategies focussing on atrial fibrillation were dominant, none out of eight early detection strategies focussing on abdominal aortic aneurysm were dominant, one out of seven early detection strategies focussing on hypertension were dominant, and four out of six early detection strategies focussing on 10-year CVD risk were dominant. When comparing the base-case ICERs of all early detection strategies with the Dutch WTP threshold, which is €20,000 when considering preventive strategies, 36 out of 49 (66.7%) were still deemed cost effective.

Table 3 Outcomes of health economic evaluations
Full size table
Fig. 2
figure 2

Incremental health economic outcomes of all papers reporting both costs and quality-adjusted life-years (QALYs). The colour represents the targeted disease of the early detection strategy and the shape reflects the authors’ conclusion. The dashed line represents the Dutch willingness-to-pay threshold of €20,000/QALY. One publication [39] reported both incremental costs and QALYs, but had such large QALY gain (2.87) that it was removed from this figure for visualisation purposes

Full size image

3.5 Reporting Quality

The reporting quality ranged from 25 to 86% (median = 57%) and scores per article are shown in ESM 5. In total, 26 articles (53.1%) were labelled as low quality, 22 (44.9%) as medium quality, and one (2%) as high quality. To assess the reporting quality of articles over time, the reporting quality of all articles is plotted per year of publication in Fig. 3. No improvement in reporting quality could be seen over the years. Most articles reported on how costs were measured and valued (n = 46, 93.9%) and on the effect of uncertainty on outcomes (n = 46, 93.9%), as can be seen in Fig. 4. Distributional effects and effects of engagement with patients and stakeholders on the design and outcomes, for example, were only reported in one article (2.0%). On the contrary, the valuation and measurement of costs and the effect of uncertainty were reported in most studies (n = 46, 93.9%).

Fig. 3
figure 3

Reporting quality of health economic articles over the included time period, i.e. from January 2016 until May 2022. The colours represent the target disease and the horizontal lines separate low, medium, and high quality articles. For visability, some noise was added to the x-value

Full size image

4 Discussion

This systematic review identified and assessed 49 unique health economic evaluations that focussed on 48 unique early detection strategies for CVD. Almost all included health economic evaluations were performed from a high-income country perspective. Most evaluations compared early detection strategies with no early detection and simulation modelling was used in all studies to estimate (long-term) health and economic impact. Early detection strategies were predominantly cost effective with approximately a quarter also claiming cost reductions. This suggests that early detection of CVD is likely to be cost effective given the respective WTP thresholds applied. However, this could also be (in part) explained by the fact that studies with a negative outcome may be published less often [27]. No disease-specific early detection strategy appeared to be much more cost effective than others. Moreover, of the four most targeted diseases, 10-year CVD risk prediction showed the most promising results being dominant in 67% of all studies. When comparing the ICERs with the Dutch WTP threshold, two-thirds of the reported ICERs fall below this threshold, indicating that the majority of strategies would be cost effective when consistently applying this WTP threshold. Compared with the previous systematic review mentioned in the introduction [12], substantially more studies focussing on early detection of CVD were identified (49 vs 5). Both reviews showed that reported ICERs varied substantially. However, 11 studies (22.4%) in this review reported the early detection strategy to be dominant over the comparator, whereas the earlier review did not mention any early detection strategy to be dominant.

Fig. 4
figure 4

Overview of how many articles reported on each item of the 2022 CHEERS checklist

Full size image

The median reporting quality of studies according to the CHEERS 2022 was 57% (ranging from 25 to 86%) and quality was considered high in only one health economic evaluation. Whereas reporting quality varied per year, it seems to remain consistent over time. No conclusions on the studies from 2022 can be made yet, as only two relevant articles were published in the included months (January 1–April 30). When investigating the items individually, it was apparent that several specific items consistently scored poorly. For example, engagement with patients and stakeholders and the effect thereof were rarely reported. Only one study created a patient committee that was involved in the design of the study [19], whereas two studies involved a multidisciplinary team with clinical experts in the development of the simulation model [28, 29]. Furthermore, distributional effects and health (in)equality were only mentioned in one study [30]. Only two studies referred to a health economic evaluation plan [31, 32]. All the above-mentioned items were added in the latest version of CHEERS in 2022. This may explain why few studies reported these items, as all included studies were published before or at the beginning of 2022. Surprisingly, the items mentioning perspective, time horizon and to a lesser extent discount rate were also scored poorly. Although these choices were mentioned in most articles, the reasons for choosing a certain perspective, time horizon and discount rates were not reported, leading to low scores.

Several findings were striking. Regarding the model type used, the Markov cohort STM was by far the most used simulation model. However, such models may not be most suitable to simulate the long-term impact of early detection strategies for CVD. Limitations of Markov cohort STMs are that they are rather unsuitable to include heterogeneity and have limited flexibility to consider the history of patients [33]. When estimating the occurrence of CVD events, the risk of developing CVD may depend on many different (risk) factors and medical history which may be harder to adequately incorporate in a cohort Markov STM compared with a patient-level simulation model. While a Markov cohort STM may require fewer inputs than patient-level simulation models, a Markov cohort STM may yield comparable results when inputs are carefully assessed and patient-level parameters are appropriately considered and reflected [34]. Justification for the choice of a Markov cohort STM was, however, lacking in most included studies. It can be argued that patient-level models may be more suitable for early detection strategies in particlular, as they are well suited to include heterogeneity, long-term memory, and account for the (long-term) clinical and treatment history of simulated individuals. In addition, current computing power, detailed tutorials and supporting programming code [35] have made the use of patient-level models for standard health economic analyses quite feasible. Still, only eight studies (i.e. 5 microsimulations and 3 discrete event simulations) used a patient-level simulation model.

The aggregated costs per individual deviated substantially between studies, ranging from below 100 Euros to hundreds of thousands of Euros. Multiple causes could (partly) explain the differences in costs. Firstly, the target population ranged from general population samples to samples of specific patient populations. For example, one study only included patients on the waiting list for kidney transplantation [36]. These patients already incur high costs for dialysis regardless of potentially developing CVD, contributing to very high costs per individual. Secondly, the type(s) of costs included varied per study. Whereas some studies only included direct healthcare costs, other studies included productivity losses and costs per life-year gained, that is, the additional healthcare costs an individual makes for living longer, in addition to direct healthcare costs. Thirdly, the time horizon and age of inclusion varied greatly between studies, which may influence total costs and QALYs per individual, as younger simulated individuals typically consume more healthcare resources and collect more QALYs due to longer survival. Fourthly, the costs of early detection strategies varied greatly between studies, depending on the screening method used and the management of individuals after a positive screening result. Finally, medical guidelines and care pathways may differ per country, particularly between low-, middle-, and high-income countries, leading to different healthcare resource use and costs of treatment. However, despite the large and expected variations in (cost) outcomes, our results allow clinicians to identify promising strategies based on conclusions of health economic assessments performed in countries with largely similar health systems. This identification can be based on, for example, combined (health) outcomes, on targeted population, or on the type of early detection strategy.

Generally, age is considered an important risk factor for CVD [37]. However, of the included studies, only 12 reportedly implemented age-dependent risks for developing CVD or developing CVD events, while the remainder used constant age-independent risks (ESM 4). The use of a constant, average CVD risk regardless of age could easily lead to overestimation of disease incidence early in the simulation or underestimation of disease incidence later in the simulation, affecting incremental health effects and the resulting ICER. Furthermore, some studies mentioned that early detection strategies for CVD will lead to cost savings, because future cardiovascular events and associated costs may be avoided, thus improving survival and quality of life. However, living longer due to avoided CVD events also leads to additional healthcare costs, which need to be considered as well [38]. Only two included articles mentioned they included these additional costs in their health economic evaluation. Most reported conclusions that an early detection strategy is cost saving from a health care or societal perspective should therefore be considered with caution and interpreted only in terms of an expected reduction in CVD-related costs. Finally, it was striking that only about a quarter of all studies reported on validation methods of the simulation model (shown in ESM 5), as proper validation is essential for good outcome interpretation.

This review has several strengths and limitations. This is one of the first systematic reviews focussing on both health and economic outcomes of early detection strategies for CVD. Health economic model developers could benefit from learning about existing models and their structure, when developing their own. This may render model development more efficient. CVD is a very complex disease including all diseases to the cardiac area and vascular bed. While likely introducing large heterogeneity to our findings, broad search terms were used to ensure all types of CVD and all known risk factors that increase the risk of developing CVD were included. We chose to exclude studies in which other diseases known to be a risk factor for CVD, such as diabetes mellitus and chronic kidney disease, were targeted, as this complicates determining the specific impact of those strategies on the CVD burden. Therefore, only studies remain that clearly focus on CVD and described the outcomes within that context. One limitation is the exclusion of grey literature. Therefore, policy-related documents discussing health economic evaluations could have been missed. However, such documents are unlikely to (independently) report on a full health economic evaluation. Moreover, data extraction was performed by a single reviewer, which may have led to some inconsistencies. Additionally, this study attempted to compare the health and economic outcomes of different health economic evaluations with many different country perspectives and with varying underlying assumptions and choices. However, health economic outcomes will be influenced by methodological choices and country perspectives. No correction was applied to address these issues, as no widely accepted correction method is currently available. Converting all costs to 2021 Euros using Dutch consumer price indexes likely affects outcomes, but it is yet unknown how large differences are when using consumer price indexes of other (Euro) countries. However, other consumer price indexes do not impact the cost effectiveness of dominant interventions. Finally, there were no exclusion criteria based on language, but articles without an English abstract could not be found due to the English search strategy.

Many health economic evaluations focussing on the impact of early detection strategies regarding CVD compare early detection strategy with no early detection. While many early detection strategies may be cost effective when compared with no early detection, some strategies will certainly outperform others. Given different methodological choices and country perspectives, including differences in (cost of) usual care, it is challenging to directly compare early detection strategies described in different health economic evaluations. A way to perform such comparison would be to assess the transferability of results from different evaluations to a specific jurisdiction (e.g. Dutch setting). However, performing such systematic comparison was beyond the scope of the current study. Moreover, comparison of different early detection strategies for CVD with consistent methodological choices, in a single validated and accepted simulation model (i.e. a ‘reference model’) would be valuable to support policymakers with identifying and implementing the most efficient early detection strategy. For instance, uniformising methodological choices concerning the type of model to use, the methods for extrapolating CVD risk and cost categories and health outcomes to consider may contribute to such an endeavour.

5 Conclusion

Current evidence suggests that early detection strategies for CVD are predominantly cost effective and may reduce CVD costs compared with no early detection. However, the lack of standardisation complicates the comparison of cost-effectiveness outcomes between studies. Real-world cost effectiveness of early CVD detection strategies will depend on the target country and local context.