In an analysis comprising a uniquely large number of individuals with very low LDL-C levels, our data show that estimation of even lower LDL-C levels is even more inaccurate. This study expands upon our prior work in the Very Large Database of Lipids [
24] by closely examining the accuracy of LDL-C within the very low range. More often than not, LDL-C levels estimated by the Friedewald equation are classified falsely low if < 40 mg/dL, when accurate safety monitoring is most needed, and this proportion exceeds 80% at TG ≥ 150 mg/dL. Although estimation of LDL-C remains imperfect by the novel method, it provides a substantially more accurate estimation than the Friedewald equation at very low LDL-C levels, halving the proportion of falsely low classifications.
Implications for efficacy and safety assessment
The IMPROVE-IT trial [
11] testing the addition of ezetimibe to statin therapy helped confirm the “the lower the better” hypothesis and supports aiming for lower LDL-C levels if it can preserve a favorable risk-benefit ratio [
34]. Moreover, recently FDA approved monoclonal antibodies to PCSK9 appear safe through 1 year and robustly lower LDL-C [
35]. The addition of alirocumab and evolocumab to standard of care in the ODYSSEY LONG-TERM and Open Label Study of Long Term Evaluation Against LDL-C Trial (OSLER) studies, respectively, yielded mean LDL-C reductions of approximately 60% to levels of approximately 50 mg/dL and preliminary short-term outcome data show an incremental 50% relative reduction in cardiovascular events compared to standard of care [
12,
13]. A meta-analysis found that PCSK9 inhibitors reduced all-cause mortality (OR, 0.45; CI, 0.23–0.86) [
35], and long-term outcome trials are eagerly awaited [
36‐
39].
An eligibility criterion for those long-term trials is an on-treatment LDL-C ≥ 70 mg/dL. Based on our data, 29% of persons with Friedewald LDL-C levels of 50–69 mg/dL actually have a directly-measured LDL-C ≥ 70 mg/dL. Therefore, individuals may be excluded from the long-term trials because of an underestimated LDL-C level. These trials are focused on high-risk patients, one feature of which is a concurrently high TG level, a setting wherein LDL-C underestimation is more likely to occur by Friedewald estimation.
For those patients who do qualify for trial participation or receive therapies in routine practice to treat their LDL-C down to very low levels, extra concern over ensuring appropriate risk-benefit ratio is warranted. This issue was raised in recent FDA advisory deliberations on PCSK9 inhibitors and some PCSK9 inhibitor trial protocols included active safety monitoring for LDL-C levels < 25 mg/dL and drug discontinuation when LDL-C was < 15 mg/dL [
14,
18]. Since these LDL-C cut-off points were not derived from Friedewald LDL-C values, our findings raise the question of potential misinformed decision making due to LDL-C underestimation when relying on the Friedewald equation. This might translate into undue anxiety, increased resource utilization (e.g., clinic visits, additional lab work), and inappropriate therapeutic adjustment.
The association between very low LDL-C levels and adverse events has been controversial. Although the Pravastatin or Atorvastatin Evaluation and Infection Therapy (PROVE-IT) study demonstrated augmented risk reduction without safety concerns among participants with LDL-C < 40 mg/dL [
40], a post-hoc analysis of the Justification for the Use of Statins in Primary Prevention: An Intervention Trial Evaluating Rosuvastatin (JUPITER) trial showed a significantly higher incidence of new-onset diabetes, hematuria, and hepatobiliary disorders in rosuvastatin-treated participants with LDL-C ≤ 30 mg/dL as compared to rosuvastatin-treated participants with LDL-C > 30 mg/dL and placebo-allocated participants [
41]. Of note, LDL-C was estimated by Friedewald equation, and therefore more than 50% of values were discordantly low based on our results. In a recent meta-analysis, PCSK9 therapy was associated with a significant increase in neurocognitive adverse events compared with placebo, although this number was yet small [
16]. However, this preliminary finding is still undergoing further investigation.
Several PCSK9 trials directly measured LDL-C by ultracentrifugation at different time points to support treatment effect data obtained using the Friedewald equation [
14]. In trials of alirocumab [
13] and evolocumab [
42], the placebo-subtracted percentage change in LDLf-C was 2–4% greater than that of direct LDL-C, consistent with modest group averaged LDL-C underestimation at lower LDL-C levels in follow-up (in patients with generally well controlled TG levels). However, actual difference between LDLf-C and direct LDL-C in the patient was not evaluated as we have done in this study. While accurate measurement of LDL-C is important in accurately assessing group averaged treatment effects in clinical trials, it is also important for accurate patient-level monitoring of adverse events at very low LDL-C levels (< 25 mg/dL) as suggested by the FDA [
14]. For example, approximately 40% and 26% of trial participants using alirocumab [
18] and evolocumab [
42], respectively, had Friedewald-estimated LDL-C levels < 25 mg/dL. Applying the findings from this study, we can estimate that approximately four in five patients actually could have had LDL-C levels ≥ 25 mg/dL.
As clinical practice moves towards lower LDL-C levels than ever before with the availability of new cholesterol lowering drugs, our findings will tend to have greater clinical relevance. The size of the present study and methodological approach, detailing accuracy of estimation across multiple clinically relevant reference categories, adds to prior work. We suggest that an update to the current de facto LDL-C assessment will likely be crucial for personalized clinical decision-making, clinical trial design, and adverse event monitoring and prevention.
Potential alternatives to Friedewald LDL-C
If an accurate method for directly measuring LDL-C was widely available, or at least widely scalable, that could be a simple solution, assuming reasonable cost. However, no such method exists. Since the introduction of the Friedewald equation, multiple chemical based assays for direct LDL-C measurement have been introduced, but do not appear to provide an improvement [
19‐
22]. These assays show non-specificity toward abnormal lipoproteins and fail to meet accuracy standards in diseased individuals. While ultracentrifugation was used to assess some participant samples in PCSK9 inhibitor trials, this was for research purposes only and cannot be practically implemented in clinical practice, as noted in the FDA proceedings [
40].
In this context, a more accurate estimate of LDL-C is desirable from both a cost and accuracy perspective. Although multiple other groups have proposed alternative methods for LDL-C estimation, our novel LDL-C estimation appears most accurate [
27] and is best validated. The performance is consistent across TG levels, the main component of the lipid profile that varies with fasting, and novel LDL-C requires no additional testing. The method can be implemented by incorporation into laboratory information technology systems for automated reporting, via Excel and Stata software available for free download at ldlcalculator.com, or via the Johns Hopkins LDL-C Calculator smartphone app that is freely available for iOS and Android.
While LDL-C is the focus of most clinical practice guidelines and clinical trials, non-HDL-C and apolipoprotein B are also included in some guidelines and warrant consideration for guiding treatment. However, clinicians are not as familiar with these [
39] and clinically relevant reference values for efficacy and safety assessment at very low levels have not been established. Moreover, their responsiveness to more intensive lipid-lowering agents like PCSK9 inhibitors differs from LDL-C.
Limitations
Limitations of our database have been discussed previously in detail [
27], the main limitation being the lack of clinical, medication, or demographic information other than age and sex. However, lipid distributions closely match a nationally representative US survey (NHANES) [
27]. Individuals with TG < 400 mg/dL were excluded in this study as it is well-known that LDL-C estimation is highly inaccurate in this setting and the clinical priority is managing hypertriglyceridemia. Moreover, the samples in this study were obtained for clinical purposes and thereby include a mix of fasting and non-fasting samples. Friedewald estimation may have performed better if only fasting samples had been included; however, inclusion of non-fasting assessments is representative of current clinical practice in Europe [
43] and the US [
44]. Given that novel LDL-C estimation showed more stable performance across TG levels, it may be more suitable for both fasting and non-fasting lipid assessment. Finally, external validation of our results are required; in particular, similar analyses would be of interest in patients treated with high-intensity statin therapy with or without PCSK9 inhibitors given the high proportion of individuals with very low LDL-C levels and the availability of direct ultracentrifugation measurement in subsamples of trial participants.