The Lipid Paradox is present in ST-elevation but not in non-ST-elevation myocardial infarction patients: Insights from the Singapore Myocardial Infarction Registry

Lowering low-density lipoprotein (LDL-C) and triglyceride (TG) levels form the cornerstone approach of cardiovascular risk reduction, and a higher high-density lipoprotein (HDL-C) is thought to be protective. However, in acute myocardial infarction (AMI) patients, higher admission LDL-C and TG levels have been shown to be associated with better clinical outcomes - termed the ‘lipid paradox’. We studied the relationship between lipid profile obtained within 72 hours of presentation, and all-cause mortality (during hospitalization, at 30-days and 12-months), and rehospitalization for heart failure and non-fatal AMI at 12-months in ST-segment elevation myocardial infarction (STEMI) and non-ST-segment elevation myocardial infarction (NSTEMI) patients treated by percutaneous coronary intervention (PCI). We included 11543 STEMI and 8470 NSTEMI patients who underwent PCI in the Singapore Myocardial Infarction Registry between 2008–2015. NSTEMI patients were older (60.3 years vs 57.7 years, p < 0.001) and more likely to be female (22.4% vs 15.0%, p < 0.001). In NSTEMI, a lower LDL-C was paradoxically associated with worse outcomes for death during hospitalization, within 30-days and within 12-months (all p < 0.001), but adjustment eliminated this paradox. In contrast, the paradox for LDL-C persisted for all primary outcomes after adjustment in STEMI. For NSTEMI patients, a lower HDL-C was associated with a higher risk of death during hospitalization but in STEMI patients a lower HDL-C was paradoxically associated with a lower risk of death during hospitalization. For this endpoint, the interaction term for HDL-C and type of MI was significant even after adjustment. An elevated TG level was not protective after adjustment. These observations may be due to differing characteristics and underlying pathophysiological mechanisms in NSTEMI and STEMI.

on oral medication for hyperlipidemia, age, sex, race, body mass index, history of diabetes, history of hypertension, smoking status, history of AMI/coronary artery bypass grafting (CABG)/PCI, Killip class on admission, presence of cardiopulmonary resuscitation (CPR) in ambulance, random blood glucose levels within 72 hours of onset of AMI, admission creatinine, admission haemoglobin, presence of elevated first troponin within 72 hours from AMI onset, left ventricular ejection fraction (LVEF) of <50% during hospitalization, presence of anterior myocardial infarction (for STEMI only), and symptom-to-balloon time (for STEMI only). These variables were selected to be included in the multivariable models as they were statistically different across the categories of lipids and they were clinically associated with the outcomes of interest. We performed further analyses to study the interaction between the type of MI (STEMI/NSTEMI) and lipids in relation to the primary and secondary outcomes. Supplementary Fig. 1 shows the patient selection criteria. A sensitivity analysis for missing data was also performed. This was done using multiple imputation with 20 imputed datasets and no auxiliary variables based on the Markov Chain Monte Carlo procedure assuming all variables in the model having a joint multivariate normal distribution. Sensitivity analysis showed that the results were in the similar direction albeit with differing magnitudes and statistical significance. As such, we opted to maintain the data in its original form and missing data were dropped from analyses through case deletion without any imputation.
The institutional review board granted an exemption for conducting this study without need for informed consent (SingHealth Centralised Institutional Review Board Reference No: 2016/2480) as this study involved analysis of a dataset without identifiers. The research was conducted in accordance with the Declaration of Helsinki. The statistician had access to anonymizedindividual data points while the other co-authors had access to analyzed, aggregated data. Statistical analysis was performed using Stata SE Version 13 (StataCorp. 2013. Stata Statistical Software: Release 13. College Station, TX: StataCorp LP). All reported p-values were 2-sided and p-values < 0.05 were considered to be statistically significant.

Results
The final patient population comprised 20013 patients. There were 11543 STEMI patients and 8470 NSTEMI patients available for analysis. Patient characteristics are shown in Table 1. STEMI patients were about 2 years younger than NSTEMI patients and more likely to be male. Fewer STEMI patients had a prior history of diabetes mellitus, hypertension, history of AMI/CABG/PCI but they were more likely to be smokers. In terms of cholesterol levels, STEMI patients had a higher LDL-C (3.4 mmol/l vs 3.2 mmol/l, p < 0.001), lower TG levels (1.4 mmol/l vs 1.6 mmol/l, p < 0.001) but there was no difference in HDL-C levels. There were a higher proportion of STEMI patients than NSTEMI patients with higher TC. STEMI patients were less likely to be on oral medications for hyperlipidemia prior to presentation (62.6% vs 75.1%, p < 0.001). STEMI patients were also more likely to have a depressed ejection fraction (61.2% vs 39.6%, p < 0.001).
AMI Population. Firstly, we analyzed the entire AMI population. Higher LDL-C levels were associated with a lower risk of death during hospitalization, at 30 days and at 1 year on both unadjusted and adjusted analysis ( Table 2) but not for rehospitalization for heart failure or myocardial infarction within 1 year from discharge. While there was a similar association for TG levels for the risk of death, this did not persist after adjustment (Supplementary Table 1). There was an association between higher TC levels with worse primary outcomes on unadjusted and adjusted analysis for the primary outcomes but not for the secondary outcomes of interest (Supplementary Table 2). After adjustment, there was actually a higher risk of death at 30-days and at 1-year with HDL-C levels of <1.0 mmol/L which suggest that there was no paradox present for HDL-C levels for the entire AMI population (Supplementary Table 3). STEMI Patients. We next examined the correlations of lipid levels with the primary and secondary outcomes, for both STEMI and NSTEMI patients. Of note, in terms of primary outcomes for STEMI patients, a higher LDL-C was inversely correlated with a lower risk of death during hospitalization, at 30 days and at 1 year on both unadjusted and adjusted analyses (Figs. 1 to 6 and Supplementary Table 4). In terms of secondary outcomes for STEMI patients on unadjusted analysis, a higher LDL-C was similarly inversely correlated with a lower risk of rehospitalization for HF and AMI within 1 year from hospitalization discharge. After adjustment, the observed lipid paradox for the secondary outcomes of HF and AMI hospitalizations within 1 year from AMI discharge and higher LDL-C levels was no longer present. The significant variables after adjustment for HF hospitalizations were use of oral medications for hyperlipidemia, older age, Malay and Indian ethnicities, history of diabetes mellitus, history of hypertension, history of AMI/CABG/PCI, higher Killip class on admission, elevated troponin levels, longer symptom-to-balloon time and depressed LVEF. The significant variables after adjustment for AMI hospitalizations were driven by female sex, Malay, Indian and other ethnicities, smoking status and history of AMI/CABG/PCI. There was a similar relationship between a lower LDL-C and worse secondary outcomes for hospitalization for HF and AMI on unadjusted analysis but these correlations did not exist after adjustment (Supplementary Table 4).
There was an association between higher TC levels with worse primary outcomes on unadjusted and adjusted analysis for the primary outcomes but not for the secondary outcomes of interest (Supplementary Table 5).
For HDL-C, there was an association between lower HDL-C levels and better primary outcomes on unadjusted analysis but only the primary outcome of death during hospitalization persisted after adjustment. There was no such correlation for the secondary outcomes (Table 3). For TG, there was no strong association between TG levels and primary or secondary outcomes (Supplementary Table 6).

NSTEMI Patients.
There was an inverse correlation between LDL-C levels and the primary outcomes of risk of death during hospitalization, at 30 days and at 1 year as well as the secondary outcomes of rehospitalization (2020) 10 www.nature.com/scientificreports www.nature.com/scientificreports/ for heart failure and myocardial infarction within 1 year on unadjusted analysis for NSTEMI patients but was not present after adjustment (Supplementary Table 7).
There was an inverse relationship between the primary outcomes and TC levels and the secondary outcomes of rehospitalization for HF and AMI, but these correlations did not exist after adjustment (Supplementary Table 8).
Lower HDL-C levels appeared to increase the risk of death during hospitalization after adjustment in contrast to the trend demonstrated in STEMI patients (Table 4). There was no correlation between TG levels and primary and secondary outcomes (Supplementary Table 9).

Interaction between type of myocardial infarction and lipids. Further analyses for interaction
between the type of MI and lipids demonstrated a significant interaction term after adjustment for HDL-C only for the outcome for death during hospitalization (p = 0.015) ( Table 5). There was no significant interaction after adjustment for all other outcomes and lipid levels.

Discussion
Our main study findings were as follows: 1. The lipid paradox for LDL-C exists for STEMI patients undergoing PCI for the primary outcomes of death during hospitalization, at 30 days and at 1 year, but not for NSTEMI patients i.e. a pseudo-paradox was present for NSTEMI patients; 2. The lipid paradox for TG levels for STEMI patients undergoing PCI did not exist in our study after adjustment i.e. a pseudo-paradox is present; 3. HDL-C levels trended towards a paradox for STEMI patients but not for NSTEMI patients, and there was significant interaction between the type of MI and HDL-C levels for the outcome of death during hospitalization.
A number of studies have investigated the lipid paradox in patients with acute coronary syndromes. These studies were done in acute coronary syndrome populations that involved STEMI and NSTEMI populations as a whole, but did not specifically compare between these 2 groups 11,13,26-28 . Cho et al. studied a population of AMI patients post-PCI in relation to 30-day and 1-year outcomes, but did not stratify between the STEMI and NSTEMI groups. In their study, they found that patients with higher LDL-C levels, except for patients with LDL-C > 160 mg/dL (>4.1 mmol/L), were related to better outcomes. However, they reported independent predictive factors of 12-month mortality being age, systolic blood pressure, acute myocardial infarction, LVEF, renal function, Killip class, N-terminal-pro-B-type natriuretic level and use of renin-angiotensin receptor blockers (RAB) use, and concluded that their observation was an apparent paradox due to confounding factors. In our study, we accounted for the above variables (except for biomarker and RAB use) and demonstrated that the lipid paradox persisted in the STEMI but not the NSTEMI population. Interestingly, while we did not account for RAB use, the duration required for RAB use to effect positive myocardial remodeling in post-AMI patients would require time 29 . We observed the lipid paradox being present even for LDL-C in STEMI patients during the index STEMI (n = 11543) NSTEMI (n = 8470) p www.nature.com/scientificreports www.nature.com/scientificreports/ hospitalization for myocardial infarction, which would not have been a sufficient duration of time for RABs to exert their myocardial remodeling effects. As such, we believe that RAB use, while prognostically useful in the long-term, would not explain our short-term observation.
For triglyceride levels in ACS patients, Cheng et al. studied a cohort of STEMI patients who received primary PCI in a single tertiary referral hospital and found that the serum triglyceride level had an inverse relationship with in-hospital death and late outcomes 30 . They postulate that higher TG levels may have a role in infarct size stabilization, reducing the risk of arrhythmias. An alternative postulated explanation is that TG actually reflects nutritional status and a lower TG means that the body's nutritional state is poorer and hence may halt the patient's    www.nature.com/scientificreports www.nature.com/scientificreports/ recovery from STEMI. We did not find the same results in our study cohort after adjustment, nor was there any major differences between STEMI or NSTEMI groups for TG levels. A possible explanation is that our study adjusted for more variables compared to the study by Cheng et al., and there might be an apparent paradox for TG in that study due to residual confounding.   www.nature.com/scientificreports www.nature.com/scientificreports/ STEMI patients have been described to have an increased pro-inflammatory state compared to NSTEMI patients 15,31 . Our study also supports the role of inflammation as the underlying factor in the lipid paradox, as we demonstrated a lipid paradox in STEMI patients but not NSTEMI patients. Furthermore, this may be contributed by the fact that the STEMI and NSTEMI patients have different clinical characteristics. STEMI patients in our population were more likely to be smokers -the latter contributes to a pro-inflammatory state 32 . A counterpoint to this argument would be that the subjects in our STEMI population were more likely to be on oral medications for hyperlipidemia, and it is known that statins exert a pleiotropic anti-inflammatory effect 33 . Statins result in lower TC, LDL-C and TG levels 34 . A study demonstrated the effect of statins on outcome modification in patients with low LDL-C levels. Oduncu et al. demonstratedthat patients with statin-induced low LDL-C on admission had better outcomes in STEMI and predict lower mortality, but patients with spontaneously low LDL-C without statin treatment predict higher mortality 35 . Similarly, they postulate that statin exert an anticoagulant, anti-platelet and anti-inflammatory effect. Those with spontaneously low LDL-C in their study were associated with increased inflammation as reflected by higher inflammatory markers (leukocyte count, neutrophil/lymphocyte ratio and C-reactive protein levels) 35 .
Building on this, with regards to HDL-C, patients with a lower HDL-C trended towards better outcomes for STEMI patients (HDL-C lipid paradox). On the contrary, a lower HDL-C trended towards worse outcomes for NSTEMI patients, although this was only statistically significant for death during hospitalization at a level of HDL-C between 1.0-1.5 mmol/L. Previous studies in AMI populations have demonstrated that a lower HDL-C leads to greater mortality in both STEMI 36 and NSTEMI patients 37 . This observation may be due to the possibility of the presence of dysfunctional HDL-C, which has been described to be present in patients with coronary artery disease, obesity, diabetes mellitus and smokers 38 . It is increasingly recognized that the function and subclass of HDL-C needs to be considered above the plasma concentrations, as plasma concentrations alone cannot account for the epidemiological observations and lack of treatment efficacy when raising HDL-C levels [39][40][41] . Dysfunctional HDL-C has a reduced pro-oxidative effect and increased pro-inflammatory effect. NSTEMI and STEMI patients have different levels of inflammation present, and this difference in inflammatory process can modify HDL-C functionality 41 , thus potentially leading to the observations in our study. Also, STEMI patients in our population  www.nature.com/scientificreports www.nature.com/scientificreports/ were less likely to have a history of AMI/CABG/PCI (a surrogate for CAD), have a lower body mass index and less diabetes mellitus although there were a higher proportion of smokers compared to NSTEMI. This difference in baseline characteristics may also account for differing levels of dysfunctional HDL-C and hence a better outcome in STEMI patients. Further study of HDL-C function and subfractionsin addition to levels in this population would be helpful in the future in understanding this observation. Unfortunately, we did not have information on inflammatory markers in our population including C-reactive protein and total white cell count, nor did we have compliance data to statin use and could not specifically examine inflammation as a factor, but this can be the focus of future studies.
There have been other studies performed to examine the lipid paradox in cardiac patients in non-MI settings. Authors have described the potential pathophysiological mechanisms of a low LDL-C in situations of increased inflammation such as in heart failure. They explain that increased intestinal edema leads to an increase in translocation of bacterial lipoprotein saccharides (LPS) from the intestines into the blood, which induces inflammatory markers such as tumour necrosis factor-alpha. Lipoproteins form micelles around the bacterial LPS to inactivate the bacterial components, hence accounting for lower LDL-C levels 42,43 . While our study did not examine the biological mechanisms of the lipid paradox in post-MI PCI patients, gut bacteria have been linked to myocardial infarction and this could be a postulated mechanism of action 44 . Further potential explanations for the lipid paradox in heart failure patients include statin pre-medication as well as poorer nutritional status 43 , which wasafactor we adjusted for in our study.
The lipid paradox has also been described in non-cardiac conditions. Amezaga Urruela et al. described active rheumatoid arthritis patients having lower lipid levels, and postulated that this may be due to an inflammatory process 45 . A similar inflammatory cytokine release is observed in acute pancreatitis in which the lipid paradox has also been observed 46 . The inflammatory hypothesis is postulated to contribute significantly to the underlying pathophysiology of AMI 47 ; this inflammatory hypothesis has recently been reinforced in the landmark Canakinumab Anti-inflammatory Thrombosis Outcome Study (CANTOS) trial which studied the use of the orphan drug canakinumab to reduce the risk of developing cardiovascular events using anti-inflammatory therapy with interleukin-1β inhibition 48 .
Finally, it might be prudent for clinicians not to be unduly influenced by the low measured LDL-C levels and hence withhold essential statin treatment for patients with acute myocardial infarction. This warning has been mentioned previously for nephrology patients in which the authors argue that despite the presence of the lipid paradox, statins exert an anti-inflammatory pleotropic effect which makes them effective medications for cardiovascular risk reduction 49 . This is supported by the American and European lipid management guidelines which advocate for the use of high intensity statins in AMI patients regardless of LDL-C levels 23,24 .

Strengths and limitations.
To the best of our knowledge, this is the largest study examining the lipid paradox in an unselected population of patients that are post-MI and have undergone PCI. Our study does have some limitations. As this is a cross-sectional analysis of registry data, we could demonstrate associations but not causation. The SMIR did not collect data on liver function tests as hepatic dysfunction is one of the postulated    Table 5. P value for interaction between type of myocardial infarction and lipids. *Adjusted for: oral medication for hyperlipidemia (yes/no/not applicable), age (numeric), sex (male/female), race (chinese/malay/ indian/others), history of diabetes (yes/no), history of hypertension (yes/no), smoking status (never/former/ current), history of AMI/CABG/PTCA (yes/no), BMI (numeric), Killip class on admission (1/2/3/4), CPR in ambulance/ED (yes/no), random blood glucose within 72 h from onset (numeric), creatinine on admission (numeric), haemoglobin on admission (numeric), elevated first troponin within 72 h from MI onset (yes/no), left ventricular ejection fraction <50% during hospitalization (yes/no). Abbreviations: HDL-C, high density lipoprotein cholesterol; HF, heart failure; LDL-C, low density lipoprotein cholesterol; MI, myocardial infarction; TC, total cholesterol; TG, triglycerides.