In this post hoc analysis of the REAL-CAD study, using a novel method named the “bottoming-out model,” we found a “threshold” value of LDL-C, below which further reduction did not affect the onset of cardiovascular events in patients with CAD given statins for secondary prevention of cardiovascular disease. Our analysis model suggests a threshold value to be 70 mg/dl for primary composite outcomes, 80 mg/dl for cardiovascular death, 70 mg/dl for myocardial infarction, and 60 mg/dl for ischemic stroke. From the results of our analysis, we can envision that the “The lower, the better” hypothesis does not always apply to Japanese CAD patients.

In the subanalysis as a preliminary analysis, to estimate the existence of the “threshold” value of LDL-C, we divided the patients into 6 categories by LDL-C level at 6 months and calculated event rates and multivariable-adjusted hazard ratios within each category. Regarding primary composite outcomes, the event rates and adjusted hazard ratios were the lowest in the 50 ≤ LDL-C < 75 mg/dl category in overall patients as well as in pitavastatin 4 mg/day group patients. The result suggested that there might be the threshold value of LDL-C around 50–75 mg/dl. Also, for other endpoints, we estimated the threshold values in the same way. Next, we performed a “bottoming-out model,” i.e., we calculated multivariable-adjusted hazard ratios for each of the threshold LDL-C values set by every 10 mg/dl. We tried several models with different threshold values of LDL-C. In the fitting of multiple models, the fitting was certainly improved in the model with the threshold set. Regarding primary composite outcomes, the hazard ratio was significant and the model fitting was best when the threshold LDL-C was assumed to be 70 mg/dl. Similarly, regarding secondary endpoints, the best-fit threshold LDL-C values were 80 mg/dl for cardiovascular death, 70 mg/dl for non-fatal myocardial infarction, and 60 mg/dl for non-fatal ischemic stroke. Our analysis model is based on the establishment of a monotonic relationship [21]. Figure 4 is a simple illustration of the bottoming-out model on the absolute risk scale. From the results of this analysis, it appears that the bottom of LDL-C is at about 70 mg/dl, as demonstrated by the black solid line in the figure. The line indicates that when LDL-C is below 70 mg/dl, the risk of cardiovascular events is constant, and when it is above 70 mg/dl, the risk increases as LDL-C rises. The uniqueness of our model is that when LDL-C is lowered below 70 mg/dl, the risk remains independent of the LDL-C level despite further reduction. These analyses were performed based on LDL-C levels at 6 months but not last LDL-C levels during the follow-up period. Since the LDL-C control continued well at the final visit as well as at 6 months in overall patients (Fig. 2C, D) and did not deteriorate before the events even in patients who had developed cardiovascular events (Fig. 3), we believe it would be rational that our analyses were conducted based on the LDL-C levels at 6 months.

The “The lower, the better” concept for LDL-C has been proposed by many statin trials [15, 22,23,24,25] and was strengthened by the recent non-statin lipid-lowering therapy trials [16,17,18]. The 2019 European Society of Cardiology and European Atherosclerosis Society guidelines for the management of dyslipidemias recommend aggressive goals for LDL-C lowering, such as <1.8 mmol/l (<70 mg/dl) for patients at high risk of atherosclerotic cardiovascular disease, <1.4 mmol/l (<55 mg/dl) for patients at very high risk or with clinically evident atherosclerotic cardiovascular disease, and <1.0 mmol/l (<40 mg/dl) for very high-risk patients who experienced a secondary vascular event within 2 years [26]. Some believers in the “The lower, the better” concept, especially the followers of PCSK9 inhibitors, state that no level of LDL-C below which benefit ceases or harm occurs has been defined. On the other hand, the 2018 American College of Cardiology/American Heart Association guidelines identified a threshold LDL-C value of 70 mg/dl, at which addition of a non-statin to high-intensity statin therapy could be considered, but recommends high-intensity statin therapy for the secondary prevention and for patients with high risk of atherosclerotic cardiovascular disease regardless of the baseline LDL-C without setting a specific target LDL-C, i.e., “Fire and forget,” which is the opposite concept of the “Treat to target” [27]. Therefore, regarding statin treatment, no consensus has yet reached as to “The lower, the better,” “Fire and forget,” or “Treat to target,” which has been in debate for a long time. Moreover, the universal target LDL-C for lipid-lowering treatments has not been established. In the present subanalysis, we assumed a threshold value of LDL-C every 10 mg/dl, and 70 mg/dl LDL-C showed the best-fit as the threshold value for the primary composite outcomes. The novelty of the present subanalysis lies in an aggressive search for the “bottom” of LDL-C value, i.e., the value below which events may not decrease even if it is reduced further. We believe that our “bottoming-out model” would have a certain impact, in which the shape of the assumed model was changed including the threshold value, and the model fit was compared.

Very recently, a meta-analysis result of 21 randomized clinical trials that examined the efficacy of statins on primary and secondary prevention for death and cardiovascular outcomes demonstrated was published [28]. This article demonstrated that the absolute risk reduction of treatment with statins was 0.8% (95% CI, 0.4–1.2%) for all-cause mortality, 1.3% (95% CI, 0.9–1.7%) for myocardial infarction, and 0.4% (95% CI, 0.2–0.6%) for stroke, while the relative risk reductions was 9% (95% CI, 5–14%), 29% (95% CI, 22–34%), and 14% (95% CI, 5–22%) respectively. The authors conclude that the absolute risk reductions of treatment with statins are modest compared with the relative risk reductions. From this article, the association between LDL-C levels and cardiovascular events seems modest as far as the view of absolute risk reduction is concerned, and the authors state that a conclusive association between absolute reductions in LDL-C levels and individual clinical outcomes seems pending and that discussing absolute risk reductions would be important when making informed clinical decisions with individual patients. In addition, this article seems to warn us regarding the overtreatment of patients with low baseline LDL-C levels. For that as well, we believe it is valuable that we found the threshold value of LDL-C in the present subanalysis.

The primary mechanism of statins for the prevention of cardiovascular events depends on lowering LDL-C. On the other hand, it has been proposed that statins also exert cardiovascular protective effects that are independent of LDL-C called pleiotropic effects [29, 30]. Among various pleiotropic effects, the anti-inflammatory properties of statins have been the focus of a number of clinical trials. The Myocardial Ischemia Reduction With Aggressive Cholesterol Lowering (MIRACL) trial was an acute coronary syndrome trial comparing atorvastatin 80 mg/day with placebo. Atorvastatin lowered the primary endpoint in patients with both high and normal LDL-C and also lowered the level of hsCRP by 83% [31]. JUPITER was a primary prevention trial of rosuvastatin in patients with LDL-C levels < 130 mg/dl and CRP ≥ 0.2 mg/dl. The main analysis of JUPITER found that rosuvastatin reduced LDL-C by 50%, hsCRP level by 37%, and the primary endpoint by 44% [32]. Plotting the expected benefit from JUPITER based on LDL-C lowering on the Cholesterol Treatment Trialists’ (CTT) collaboration regression line suggests that the achieved benefit may be greater than the expected benefit based on LDL-C reduction alone [25]. These observations suggest rationales for statin therapy targeting inflammation. The Canakinumab Antiinflammatory Thrombosis Outcome Study (CANTOS), which was a non-statin intervention trial for CAD, demonstrated that anti-inflammatory therapy targeting the interleukin-1β innate immunity pathway with canakinumab at a dose of 150 mg every 3 months led to a significantly lower rate of recurrent cardiovascular events than placebo, independent of LDL-C lowering [33]. Therefore, anti-inflammatory interventions as well as LDL-C lowering might be beneficial in reducing cardiovascular events in patients with CAD. In the present subanalysis, we demonstrated that a threshold value of LDL-C was present and that it was 70 mg/dl for the primary composite outcome. From our results, we can envision that therapy targeting residual risk factors beyond LDL-C, including inflammation, would be promising, for secondary prevention in CAD patients with LDL-C < 70 mg/dl. Also, in the REAL-CAD study, another subanalysis focused on hsCRP levels is currently in progress.

In this subanalysis, we used a novel analysis procedure called the “bottoming-out model” to assess the monotonic between LDL-C levels and cardiovascular event onset. There are some points to keep in mind in this method. The thresholds of LDL-C relationship around 70–90 mg/dl obtained from our analysis also rely on the modeling assumptions of proportional hazards, that is, time-constant hazard ratios and the linear association with hazard for each adjustment variable. Although we checked these assumptions in primary models via the Schoenfeld (for proportional hazards assumption) and martingale (for linearity assumption) residuals and did not find evidence for gross deviation (Additional file 4: Fig. S1 and Additional file 3: Table S3), partly linear relationship with a single threshold characterized by our “bottoming-out” models is itself a strong assumption. Hence, the results should be taken as a simplified approximation of the possibly non-monotonic associations between LDL-C and cardiovascular risks. In the first place, the “bottoming-out model” did not attempt to clarify the “threshold” with a statistical significance. Although we found from the shape of model fitting that the “threshold” might exist, the value (negative double of the log-likelihood or −2LL) of the model fitting does not have criteria to interpret how much the difference is significant. Therefore, it might be impossible to make a strong claim that there is the definitive “threshold,” only from our results. The bottoming-out model (Table 3, Fig. 4) directly incorporates a hypothesized piecewise log-linear association between LDL-C levels and cardiovascular events with specific thresholds, where the lower LDL-C level is not associated with an increased risk of cardiovascular events. The advantage of the model would be the simple, parsimonious generalization of the strictly increasing log-linear association typically assumed in the standard Cox models. However, our model also relies on the abovementioned restricted piecewise log-linear relationship. If the model severely misspecifies the true hazard changes by LDL-C levels, the results from the bottoming-out model would be misleading as with other statistical modeling techniques.

The REAL-CAD trial is the largest scale trial in the world. However, it is a study of Japanese CAD patients alone and does not take into account racial differences. Primarily, cardiovascular risk is lower, and cardiovascular event rates are lower in Japanese CAD patients, compared with Western patients. In also this subanalysis, the number of cardiovascular events was too small, especially in patients with LDL-C <70 mg/dl, and was still smaller in patients with LDL-C <50 mg/dl. It is undeniable that these facts affected the conclusions obtained from this subanalysis. Although we have discussed the inflammatory response as a residual risk beyond LDL-C, hsCRP levels in Japanese CAD patients are lower than those in Western patients. Therefore, detailed investigations for racial differences are needed. We hope a similar analysis to the present one will be applied also to the Western CAD patients. Finally, in the REAL-CAD trial, the overall population was only followed for 5 years and some cardiovascular events may not have been fully exposed. In most of the large clinical trials, long-term results have been observed for 2, 3, or 5 years at most. Still longer-term follow-up would be necessary when considering the life span of a person, although it may be very difficult.

PCSK9 inhibitors, when added to a statin, can achieve an LDL-C level as near to zero as possible [17, 18]. Estimating from the CTT collaboration regression line, theoretically, the incidence of cardiovascular events would be zero, if the LDL-C level were under 30 mg/dl in patients with CAD [25]. Actually, however, cardiovascular events may occur even if the LDL-C level is near to zero. We reported a case of a 76-year-old woman with CAD who experienced several repeated cardiovascular events after PCI, despite receiving a PCSK9 inhibitor, evolocumab (420 mg every month), along with a moderate dose of a statin (5.0 mg/day rosuvastatin) and that her LDL-C was as low as 10 mg/dl. However, finally by increasing the dose of rosuvastatin to 20 mg/day (the maximum in Japan), her hsCRP level was reduced from 2.4 to 0.09 mg/l, after which she no longer developed cardiovascular events [19]. In the main results of the REAL-CAD trial, the pitavastatin 4 mg/day group achieved a significant reduction of hsCRP at 6 months from baseline at randomization, while pitavastatin 1 mg/day showed no change in hsCRP, possibly explaining in part the greater reduction of cardiovascular events in the pitavastatin 4 mg/day (maximum dose in Japan) group. In the present subanalysis, we first estimated that a “threshold” value of LDL-C might exist around 50–75 mg/dl for primary composite outcomes, especially in patients receiving high-dose statin that suppressed inflammatory reaction more strongly. Next, following “bottoming-out model” exhibited that the “threshold” value of LDL-C might be 70 mg/dl. These results suggest that the “The lower, the better” concept applies until LDL-C is reduced to 70 mg/dl, but not when the level is below 70 mg/dl. Therefore, our analysis model provided us a certain proposal that we should aim “Treat to target” until lowering LDL-C to less than 70 mg/dl. When the LDL-C level reaches less than 70 mg/dl, we should then use high-dose statins with the concept “Fire and forget.” Taken together, we can envision a lipid-lowering strategy for secondary prevention of CAD as follows: First, we aim to reduce LDL-C to less than 70 mg/dl and also to reduce inflammatory status, by increasing the dose of statins to the maximum. Next, if the LDL-C level does not reach 70 mg/dl even after using the maximum dose of the statins, a non-statin drug such as ezetimibe or PCSK9 inhibitors is used in combination.