Statins in Healthy Adults: A Meta-Analysis

Background and Objectives: In this paper, we investigated the efficacy of statin therapy on cardiovascular disease (CVD) reduction in adults with no known underlying health conditions by undertaking a meta-analysis and systematic review of the current evidence. Materials and Methods: We performed a systematic search to identify Primary Prevention Randomized Controlled Trials (RCTs) that compared statins with a control group where CVD events or mortality were the primary end point. Identified RCTs were evaluated and classified into categories depending on relevance in order to determine which type of meta-analysis would be feasible. Results: No differences were observed between categories with the exception of relative risk for all CVD events combined which showed a 12% statistically significant difference favouring studies which were known to include participants without underlying health conditions. Strong negative correlations between number-need-to-treat (NNT) and LDL-C reduction were observed for all Coronary Heart Disease (CHD) outcomes combined and all CVD outcomes combined. Conclusions: This project highlights the need for further research on the effects of statins on participants who do not suffer from underlying health conditions, given that no such studies have been conducted.


Introduction
People may be healthy while presenting hypercholesterolemia, and more specifically, increased Low Density Lipoprotein Cholesterol (LDL-C) levels. We do not know what strategy to adopt to prevent cardiovascular disease (CVD) in these cases. The critical issue for such people is that the standard of care for the management of high LDL-C is statin treatment; a practice which has become increasingly controversial due to side effects [1], claims that they have failed to substantially reduce CVD outcomes [2], statistical misrepresentation of the benefits [2] and financial conflicts of interest as a source of bias [2].
In terms of the evidence, there are hundreds of published randomised-controlled trials (RCTs) and several meta-analyses favouring statins [3][4][5][6]. Despite this, there is scepticism within the scientific community on the efficacy of statins, backed by a large number of scientific studies and publications [7][8][9][10][11]. The question of whether cholesterol plays a causal role in CVD is essential for proposing optimal strategies, not only pharmaceutical, but also dietetic.
The primary focus of the present work will be an investigation into the evidence on the benefits of statins on "otherwise healthy" participants. The justification for this is that the current standard of care for the management of high LDL-C is the ubiquitous prescription of statins, including participants with no underlying health conditions and where high LDL-C is the only exhibited CVD risk marker.
The key outcome of the study will be a review of the evidence on the efficacy of statins in reducing CVD risk in people with no underlying health conditions. We hypothesise that the CVD risk reduction effect size of statins is lower in participants without underlying health conditions compared to those with underlying health conditions.

Search Protocol
MEDLINE, PubMed (1946-January 2020) and the Cochrane Database of Systematic Reviews were searched. Relevant studies were identified by using a combination of MeSH terms, text words and search equations focussing on the following terms: statins, cardiovascular disease and mortality, cholesterol, primary prevention and randomised controlled trials. All RCTs, reviews and references were examined to identify studies potentially eligible for inclusion.

Study Selection
We shortlisted studies for further investigation if they met the following criteria: We designed a structured data abstraction form which included a comprehensive list of data points, and each study was categorised according to relevance (see Table 1). Titles and abstracts of all potentially eligible studies were then read in order to classify studies and where required papers were read in full. We also extracted information and assessed study quality using the Cochrane risk-of-bias tool.

Study Categorisation
All studies were evaluated to determine the type of outcome and population under investigation. Secondary prevention studies were not evaluated and categorised as they were not relevant to the research question. Primary prevention studies were classified into three categories as described in Table 1 with the objective of determining which type of meta-analysis would be feasible.

Meta-Analysis
No Category 1 studies were found, meaning a standard aggregate meta-analysis was not feasible. Several Category 2 studies were found; however, although all studies reported sub-group data, none were sufficiently reported to enable a meta-analysis on the sub-group relevant to the present study. For example, studies may report sub-group data for "no hypertension", however, within those sub-group data, participants may be included with other underlying health conditions, such as T2DM. This finding meant that a standard aggregate meta-analysis was not feasible.

Secondary Statistical Analysis and Meta-Regression Analysis
Given that meta-analyses were not possible, we decided to undertake a statistical analysis on two further levels. First, we compared the pooled results from the Category 2 and 3 studies to determine whether there was any difference between the two groups for the three outcomes below. Second, we conducted a meta-regression analysis to examine the relationships between number-needed-to-treat (NNT) and various confounding factors.

Outcomes
We performed a statistical analysis on the following three outcomes:

•
Primary outcome: all-cause mortality; • Secondary outcome: all CHD events combined; • Tertiary outcome: all CVD events combined.
Some aggregation was required to determine the secondary and tertiary outcomes:

Search Results
The search protocol identified a total of 143 relevant studies, with 97 RCTs and 46 reviews found. From the RCTs identified, 74 were disregarded as ineligible following a review of titles and abstracts, and 12 were duplicates. As such, 11 RCTs were deemed eligible for further, more comprehensive evaluation. A further 202 studies were found through a manual readthrough of study references, reviews and meta-analyses. See Tables 2-5 for further details. Figure 1 shows the logical flow used to classify RCTs identified through the search protocol.    Table 6 lists the RCTs included in the meta-analysis and further statistical analysis. 140 This table includes relevant information pertaining to each study, including the type of 141 statins studied, year of publication, mean follow-up and the number of participants in-142 cluded. In addition, this table includes data on potential confounding variables, such as 143 average age, BMI, LDL-C at baseline, % LDL-C change, and smoking status as the per-144 centage of participants with underlying health conditions, namely, hypertension and 145 T2DM.  Table 6 lists the RCTs included in the meta-analysis and further statistical analysis. This table includes relevant information pertaining to each study, including the type of statins studied, year of publication, mean follow-up and the number of participants included. In addition, this table includes data on potential confounding variables, such as average age, BMI, LDL-C at baseline, % LDL-C change, and smoking status as the percentage of participants with underlying health conditions, namely, hypertension and T2DM.

Meta-Analyses Comparing Category 2 and 3 Studies
Figures 2-7 show relative risk ratios and absolute risk differences across Category 2 and 3 studies for the primary, secondary and tertiary outcomes. Stain therapy favoured control in both categories of study in terms of relative risk and risk difference for all outcomes (See Table 7 for effect sizes and statistical significance). No differences were observed between Category 2 and 3 studies with the exception of relative risk for all CVD events combined, which showed a 12% statistically significant difference favouring Category 2 studies, X 2 (1, N = 88,877) = 4.06, p < 0.05 (Table 7, Figure 6).       Table 7. Relative Risk Reduction (RRD), Absolute Risk Difference (ARD) and Number Needed to Treat (NNT)/year in the trials for all of the present study's outcomes. * Significant to p ≤ 0.05; ** significant to p ≥ 0.01.

Meta-Regression Analysis
Analysis was undertaken to examine the relationship between NNT and various confounding factors (Table 8). In order to achieve this, Spearman's rank-order correlation was used, the results of which can be seen in Table 8 and in Figures 8-11.     Figure 11. The relationship between LDL-C reduction % and number-needed-to-treat (NNT)/year.

Primary Outcome
In terms of the primary outcome of all-cause mortality, no statistically significant correlations were observed between NNT and any LDL-C metrics (Figures 8-11). There was a strong positive correlation between the percentage of smokers and NNT per year with regard to all-cause mortality (r s (13) = 0.703, p < 0.01) [ Table 8].

Secondary Outcome
There was a strong negative correlation between the NNT per year for the secondary outcome of all CHD events combined and mean LDL-C reduction (mg/dL: r s (13) = 0.764, p < 0.01; mmol/L: r s (13) = 0.757, p < 0.01) [Figures 9 and 10, respectively].

Discussion
Through a meta-analysis of primary prevention studies, this study is not able to confirm any difference in statins efficacy between RCTs which included participants with and without underlying health conditions. This was with the exception of a 12% relative risk difference in the tertiary outcome of all CVD events combined favouring studies with the highest number of "otherwise healthy" participants. Should our hypothesis be true, we may have expected to see reduced efficacy in Category 2 studies, given that they did not primarily focus on participants with underlying health conditions. However, it is extremely difficult to draw conclusions from this result given that the percentage of "otherwise healthy" participants in each study is unknown and all studies included a significant percentage of participants exhibiting known heart disease risk factors. For example, 44% of participants in the WOSCOPS trial were smokers, and only 22% were reported as never having been smokers [12]. Similarly, 38% of HOPE-3 participants were reported as being hypertensive [15], and 21% of participants in MEGA were reported as having T2DM [14].
Meta-regression analysis showed an absence of an association between LDL-C reduction and NNT for the primary outcome of all-cause mortality. However, statistically significant negative associations between LDL-C reduction and NNT in the secondary and tertiary outcomes support a number of other meta-analyses that suggest that lowering LDL-C has a positive effect on reducing the risk of CHD and CVD events [3][4][5][25][26][27]. Correlation, however, does not prove causation, so such associations should be interpreted with significant caution. This is particularly pertinent to the results of statins trials, given the complexity associated with heart disease and the constantly evolving science. Central to this evolution is the emerging theory that LDL-C in its own right may not be as significant in the heart disease story as previously thought. For example, a wide range of potential heart disease risk markers are now considered to be more important than LDL-C, such as inflammation [28][29][30][31], LDL particle size [32][33][34], Apolipoprotein B [35] and LDL particle number [36], Lipoprotein(a) [37], lipoprotein ratios [38] and the ratio between triglycerides and HDL [39]. Inflammation specifically needs a special mention, as it has been the cause of much controversy in statins trials. For example, JUPITER reported statins as deriving a 20% RRD in all-cause mortality and almost 50% RRD in heart attacks and strokes. However, although an association with CRP reduction was shown, an important inflammation marker, no association was observed for LDL-C [31].
Conversely, it cannot be neglected that both naturally randomized genetic studies and randomized intervention trials have consistently shown that lowering plasma LDL particle concentration should reduce the risk of CVD events. This suggests that, through different approaches, LDL reduction lowers CVD risk, even though the effect of LDL increase has not been formally tested [25][26][27].
Our research contributes to the evidence that the results from statins trials may have been misrepresented. Most trials report either relative risk or hazard ratios and conclude that statins have significantly positive effects on CVD outcomes. However, although relative risk is one of a number of appropriate ways to present results, it can result in an overestimation of effect size, given that the underlying absolute risk is concealed [40]. JUPITER, for instance, reported that "rosuvastatin significantly reduced the incidence of major cardiovascular events". As a specific example, the hazard ratio reported for myocardial infarction (MI) was 0.46, which could be interpreted as rosuvastatin reducing heart attacks by 54% compared to the control. This was based on 31 and 68 MIs in the intervention and control groups, respectively, unquestionably showing more than double the number of MIs in the control group.
However, in terms of absolute risk, the difference between groups was only 0.42%, given the large sample size of almost 9000 in each study arm. Thus, whilst the study authors are technically accurate in reporting a significant difference between groups, the hazard ratio is potentially misleading without the context of absolute risk. In real-world terms, only 0.35% and 0.76% of participants suffered myocardial infarctions in the rosuvastatin and control groups, respectively, which is a miniscule effect size compared to that presented through a hazard ratio. In another example, HOPE-3 reported "a significant reduction in the risk of cardiovascular events with the use of rosuvastatin". This was on the back of a hazard ratio of 0.76 for the coprimary outcome of a composite of death from cardiovascular causes, nonfatal myocardial infarction, or nonfatal stroke. The absolute risk difference for the same endpoint was 1.10%, deriving an NNT of 91. This common method of reporting has been highlighted in many critiques of statins trials, and is one of the factors which compounds statin scepticism [2,[41][42][43][44].
This type of selective reporting may, in fact, be at least partially attributed to bias. Although all studies were found to be at low risk of bias after assessment using the Cochrane risk-of-bias tool, it must be acknowledged that there may be a risk of bias given that the majority of studies were funded by the pharmaceutical companies manufacturing the drugs under examination [45,46].
In conclusion, the efficacy of statins in reducing CVD risk in people with no underlying health conditions cannot be directly quantified. Our study has highlighted a significant gap in the literature regarding the efficacy of statins therapy on healthy adults with no underlying health conditions. It appears that the majority of RCTs conducted to date have focussed on participants exhibiting multiple heart disease risk factors, such as smoking, hypertension, obesity and diabetes. Although the majority of these studies paved the way for the current guidelines on statins therapy, none seem to have adequately isolated LDL-C as a standalone risk factor. From the studies that we identified as potentially having included a significant percentage of participants without underlying health conditions, none adequately reported this cohort in sub-group analyses to enable the present study to execute the intended aggregate meta-analysis. Given the evolving research on the relationship between LDL-C and heart disease risk, a review of the current guidelines is highly recommended. Statins clearly demonstrate benefits in terms of reducing CVD risk, that is, when the cardiovascular risk is high as a whole, but whether this is due to LDL reduction, or whether the statins are acting on other heart disease risk factors is not sufficiently understood.
Author Contributions: L.S. designed the study under the supervision of P.F., collected the data, did the analyses and interpreted these with P.F., drafted the text which was reviewed by P.F. All authors have read and agreed to the published version of the manuscript.