Introduction
Most current guidelines recommend a single threshold for initiating statins for primary prevention of cardiovascular disease (CVD) [
1‐
3]. Irrespective of age, sex, or other factors that influence the benefit-harm balance of statins, common thresholds are 7.5% or 10% 10-year risks for CVD, with or without consideration of additional risk factors [
1‐
3]. Many regions, especially the low- and middle-income countries (LMICs), usually simply adopt recommendations from the authoritative guidelines [
4], such as from the US or Europe, without a formal adaptation process [
5]. The World Health Organization (WHO) recommends 10-year risk thresholds of 20% in high-income, 30% in medium-income, and 40% in low-income settings [
6]; but this seems to rely entirely on resource availability and not on the clinical benefit-harm balance, which would suggest different thresholds. Recently, the introduction of generic statins would reduce the cost of statins [
7]; yet even if costs are too high, subsidized access to medication programs in resource-limited settings for patients with high risk populations– as in the example of antiretroviral therapy for HIV–, can be devised if statins have relevant net benefit in reducing CVD burden.
The use of a single threshold globally to initiate statins is very likely to lead to an over- and underuse of statins in different target groups. We found in our recent quantitative benefit-harm balance modelling study for a few specific countries that age and sex had a large influence on appropriate risk thresholds since these factors are associated with differences in baseline incidences of the health outcomes, both of which impact have significant impact on the benefit-harm balance of statins [
8]. Thus, even within a population or country, recommendations for statin use should take into account age- and sex of target subgroups or patients. In addition, the analyses in the same study performed for the populations in Switzerland, the UK, and the US showed somewhat different thresholds, as the outcome baseline risks differ between the countries [
8]. Without considering such variations between different environments and countries, there is a great risk that a single threshold leads to an over- or underuse of statins. It would be important to take local population factors into account and inform and guide national policy makers to develop recommendations contextualized to national or even local settings. Against this expected significance, this study aimed to determine age- and sex-specific CVD risk thresholds for 186 countries, particularly focusing on baseline risk variation such as type 2 diabetes, hemorrhagic stroke, and competing risk for non-CVD death.
Discussion
We performed the first worldwide benefit-harm balance modelling study and found that 5th and 95th percentiles of the 10-year CVD risk-thresholds in 186 countries varied from 14 to 20% in men and from 19 to 24% in women and with median varying from 14 to 18.5% and 19–22% respectively, depending on age. While most countries had 10-year risk thresholds below 21%, some countries had higher thresholds. The differences in thresholds across populations should be interpreted with respect to baseline risks of harm outcomes and the competing risk varying across countries while we kept, according to current evidence, treatment effects and preference values similar. The small differences in thresholds between men and women were due to variation in baseline risks of the outcomes and effect estimates.
Our study showed variation in risk thresholds across countries and, as in our earlier study [
8], differences according to age and sex. As reported previously, the higher thresholds in the older age groups were due to the fact that the excess harms increased with age, offsetting the benefits of statins. The small islands countries in the western pacific region had higher risk thresholds in our study. This was due to higher risk rates for diabetes, which would be further increased with the use of statins that counterbalance the possible benefit of statins in reducing CVD events. This imply that use of a one-size-fits-all threshold or recommendation for all populations would lead to gross over- or underuse of statins in different populations, depending on the distribution of the benefit and harm risks. The threshold variation could have even been higher if country-specific rates for all outcomes were available as well as other factors (e.g., absolute proportion of people above the calculated risk threshold) were considered. The findings suggest the need of contextualized thresholds that reflect national or subnational distributions of the harm risks and CVD risks, rather than importing recommendations from authoritative guidelines without proper adaptation to the local settings.
There are no well-documented approaches of how countries determine thresholds to initiate statins. Some countries adapted guidelines to their settings or developed their own, including Japan and Fiji (threshold ≥30% 10-year risk), but it is not clear what evidence was considered in the customization of the guidelines. Nonetheless, many countries globally, especially LMICs, do not have clear guidelines. Those that have some sort of guidelines, for example, Kenya, India, Fiji, South Africa, and Brazil, are not clear about how they established their thresholds, which seem to be adopted from guidelines developed for other populations or settings [
26‐
29]. For example, the South African guideline was adapted from the European one (probably with some changes) and thus recommends statins for people with at least a 10-year CVD risk of 30% (3–15% risk with additional consideration of lipid target) [
29]. However, our thresholds for this country were 14–23%, depending on age and sex. Similarly in Kenya where statins are recommended starting at a 20% 10-year risk, our findings showed lower thresholds, especially for men [
26]. This implies that there could be missed opportunities of preventing CVD events with statins in these populations. On the contrary, an overuse of statins in some settings is also possible such as in China, Brazil (for women), the UK, and the US where a 10% or 7.5% risk is used to start therapy (e.g., NICE or USPSTF use 10% whereas ACC/AHA uses 7.5%), which is lower than ours [
1,
2,
6,
30]. This generally suggest the need of systematic and quantitative approaches to establish recommendations based on the distribution of risks, health system, infrastructure and resources.
The idea of contextualizing recommendations to different populations may not be new. There are initiatives by WHO and World Heart Federation (WHF) to develop roadmaps to promote development of national policy and health systems, especially to support LMICs in detection, treatment, and management of CVD [
6]. For example, the WHO/International Society of Hypertension (ISH) as well as the INTERHEART developed risk scores for different countries and sub-regions to detect people at high-risk [
31,
32]. While absolute risk scores assist in risk stratification or detection of people at high-risk, they do not provide direct information on who should take statins, and how possible harm risks may influence the treatment outcome. Harms of statins need to be weighed against the benefit to determine even if a group of people have higher risk for CVD. Thereby, our country-specific thresholds could serve as a base-case–or insight– for guideline developers to extend the thresholds to more precise and tailor to country-specific circumstances in order to harness the benefits of statins for primary prevention and minimize related risks and costs. It shoulld therefore be noted that the thresholds might not be ready-to-use for treatment decisions, because it was difficult to find all detailed information in each country that may have significant impact on the precision.
Our thresholds show the lowest possible CVD risk levels across countries at which statins would be more likely to provide more benefits than harms. For example, the absolute prevented MI events in 10 years were 22–41 among hypothetical 10,000 people with a baseline CVD risk equivalent to the calculated thresholds. However, these events do not show the total prevented CVD events in the general population of the different countries, which national guidelines need to take into account. That is, the thresholds could be increased or decreased, thus can be updated to specific settings, depending on the distribution of CVD risk factors in populations, country’s target to reduce morbidity or deaths due to CVD with statin use, and resource availability. For example, lower (or similar) threshold in the Sub-Saharan Africa does not mean that statin therapy would lead to higher proportion of prevented CVD events in the region than high-income countries with higher thresholds. Such impact estimation is beyond this paper, as data on the distribution of CVD risk in the respective populations would be necessary. Our study was directed to assisting decision-making for clinical guideline developers. The findings were not intended to dictate clinicians when to prescribe statins by using only the calculated thresholds. In fact, the findings also convey important message for individual patient decision-making that clinicians need to evaluate the patient risks for the potential harm outcomes (besides risk for CVD) and involve patients to discuss the tradeoff between the benefits and harms of statins.
Our findings target the general primary prevention populations and their applicability to the diabetes population should be considered with caution. People may have different levels of glycemic impairment, although it may not reach the threshold for diabetes diagnosis. Thus, the use of statin could still raise glycaemia in diabetic patients, but the extent of the excessive effect is not clear in this population. In addition, it would be difficult to understand whether the excessive risk of glycaemia due to statins would be negligible if patients are under well-controlled antidiabetic treatment. On the other hand, it is clear that diabetic patients would be more likely to be eligible for statin treatment anyway (with a higher 10-year baseline risk) than non-diabetic patients, as diabetes is a strong predictor of CVD. In general, the application of the thresholds to diabetic patients might lead to under-treatment with statins, especially in patients with well-controlled blood glucose treatment. This can be left to the discretion of doctors and patients, to decide whether to initiate statins, taking into account the patients' glycemic control.
Our study is the first global benefit-harm balance modeling study that quantitatively considers important parameters required to determine thresholds that have practical relevance for patients and clinical decision-making. However, it comes with limitations such as that we obtained country-specific baseline risks only for part of the harm outcomes. We believe that this limitation of not having baseline incidences for some harm outcomes, including myopathy, renal and hepatic dysfunctions, and cataract, for each country may have led to an underestimation of the threshold variation between countries. Specific countries could update the thresholds according to harm outcome risks available for their specific context. The debate on what evidence to consider for harms of statins is still ongoing. The effect estimates on harms range from almost none to large effects, depending the sources but all data sources have their own strengths and limitations [
12,
21,
33‐
36]. Hence, we considered convergent estimates, specifically for the harms, from both RCTs and observational data that were likely to be valid, have less bias, applicable and do not represent the extremes. We excluded observational data and post-market surveillance sources that reported extreme and contentious estimates [
21]. Indeed, the RCTs contributed substantially more to the convergent estimates because of their higher precision; thus, the effect estimates used for harms were rather small. The estimates from observational studies were less precise likely due to adjustment for several covariates, which may have increased the standard error. In addition, the different outcomes considered in the analysis were defined a priori and thus considered regardless of their effect size or baseline rate. For example, hemorrhagic stroke and cancer were included in the model, but their influence on the results was not considerable due to the low rates or effect size of the outcomes.
In terms of patient preferences, we took average estimates from Switzerland and Ethiopia. While we elicited preferences of the different outcomes related to statins in socio-demographically and economically disparate countries, Ethiopia and Switzerland [
13], there may be other factors that could have influenced the estimates that remain unaccounted for (e.g., cost). In fact, the GBD study also reported consistent disability weights across countries [
23], which supports our assumption of taking the average estimates. However, it is worthwhile to test how the preferences are affected if economic factors are taken into account. Of note, our risk thresholds cannot be directly applied within the European Society of Cardiology and European Atherosclerosis Society guideline, because this clinical guideline addresses fatal CVD, while our risk thresholds considered fatal and non-fatal CVD events combined [
25].
In summary, our worldwide map of 10-year CVD risk thresholds above which statins for primary prevention of CVD is likely to provide net benefit show high variation across populations due to differences in the baseline incidence of some harm outcomes and the competing risk for non-CVD death. The thresholds were also higher in the older people than in younger adults. The findings provide an insight for policy makers and guideline developers to consider contextual data to revise their guidelines on the use of statins, and adapt eligibility criteria for their specific country to avoid an over- and underuse of statins for primary prevention.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.