U-shaped Association Between Apolipoprotein B and All-Cause Mortality: Results from the National Health and Nutrition Examination Surveys

Background Apolipoprotein B (apoB), a signicant component directly reecting the number of atherogenic lipoprotein particles, gradually becomes a more conducive indicator to control blood lipids. However, epidemiological evidence on its relationship with mortality is limited, especially with all-cause mortality. Methods Participants from the National Health and Nutrition Examination Survey during 2007-2014 were grouped according to the apoB quartiles (15-76, 77-92, 93-110, 111-345mg/dL). We performed Cox proportional hazards models and Kaplan-Meier survival curves to evaluate the relationships of apoB with all-cause and cardiovascular mortality. Restricted cubic spline and piecewise linear regression were performed to detect their non-linear relationships.


Introduction
Apolipoprotein B (apoB) is a signi cant component for atherogenic particles, including very-low-density lipoprotein (VLDL), chylomicrons, low-density lipoprotein (LDL), and intermediate-density lipoprotein (IDL), and each particle contains one molecule of Apo B (1). ApoB exists as two isoforms--apoB-100 and apoB-48, originating from hepatocytes and the small intestine, respectively (2). Low-density lipoprotein cholesterol (LDL-C), a cornerstone measurement for guiding lipid-lowering therapy for nearly decades, does not always re ect LDL particle number precisely because of the wide variety of LDL-C concentration among individuals (3). Consequently, although LDL-C-targeted therapies are widely accepted, it is insu cient to measure the blood lipid level and risk factors of atherosclerotic cardiovascular disease (ASCVD) only by LDL-C(4). Meanwhile, more than 90% of apoB in plasma is apo B-100 associated with LDL, leading apoB a good alternate for LDL particle concentration, especially in patients with cholesterolenriched apoB particles (5). Therefore, the assessment of apoB directly re ects the number of atherogenic lipoprotein particles in the circulation. Moreover, focuses on this indicator is more conducive to controlling blood lipids (6,7).
Previous studies have demonstrated that apoB and ASCVD are more closely related (8,9). However, epidemiological evidence on its relationship with all-cause mortality and cardiovascular death is limited.
Therefore, this study aimed to address the knowledge gap of apoB with cardiovascular and all-cause mortality among United States adults in the National Health and Nutrition Examination Surveys (NHANES)

Study Design and Participants
NHANES is a cross-sectional, nationally representative survey that gathers data on the nutrition and health status of the non-institutionalized US population of all ages, conducted by the Centers for Disease Control and Prevention (CDC). This study was based on the NHANES in 2007-2014 (http://wwwn.cdc.gov/nchs/nhanes). After excluding those with CVD or cancer at baseline, a total of 10375 eligible participants with complete apoB and follow-up data were included in this analysis ( Figure   1). The Institutional Review Board approved the survey protocol of the CDC. All participants have provided written informed consent.

Measurement of apoB
ApoB data were available for only those who provided their blood samples in the morning session in a fasting state. ApoB was measured by immunonephelometric assays using a ProSpec Nephelometer (Siemens GMBH. Marburg, Germany D-35041) at the University of Minnesota Medical Center. The immunonephelometric method uses polyclonal antibodies speci c to apoB, in which the intensity of the scattered light is directly proportional to the concentration of apoB in the specimen. Besides, the apoB concentration of the test specimen is determined by comparing its light scatter to that observed using known standards in a stored calibration curve. In this survey, the unit of apoB is mg/dl. During the 1980s and 1990s, the International Federation for Clinical Chemistry (IFCC) reached a collaborative initiative with various diagnostic companies, determining the international standardization and harmonization of apoB methods named WHO/IFCC SP3-07 (10)(11)(12). Examinations of apoB in this study meet the standards established by this meeting. More details are accessible at http://cdc.gov/nchs/nhanes under the lab methods section.

Covariates and Outcomes
Using standardized questionnaires, participants provided information on age (years), sex (male, female), race (classify as white or other), smoking (classify as yes or no), an education level (less than high school, high school, or above) and self-reported medical history. Examinations of height, weight, and body mass index (BMI) were measured by anthropometric measurement. The estimated glomerular ltration rate (eGFR) was calculated by the Modi cation of Diet in Renal Disease (MDRD) formula (13). Hypertension were de ned as systolic blood pressures (SBP) ≥140 mmHg and/or diastolic blood pressure (DBP) ≥90 mmHg, reported the usage of hypotensive drugs, or history of hypertension (14). Diabetes Mellitus was de ned as fasting blood glucose ≥126 mg/dl, hemoglobin A1c ≥6.5%, reported the usage of hypoglycemic drugs, or history of diabetes (15). According to a standardized procedure and protocol, other covariates included total cholesterol (TC), triglyceride (TG), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C). If TG was ≤400 mg/dL, LDL-C was computed by the Friedewald formula (16). Outcomes in this study were all-cause and cardiovascular mortality. And all mortality data were obtained from a publicly available dataset named National Death Index (NDI), which included participants' data and linked with the NHANES dataset. The de nition of cardiovascular mortality was adapted from the International Classi cation of Diseases, 10th Clinical Modi cation (ICD-10) System codes (I00-I09, I11, I13, I20-I51, I60-I69) (17).

Statistical analysis
We grouped participants according to the apoB quartiles (15-76, 77-92, 93-110, 111-345mg/dL) in baseline characteristics. Accordingly, continuous variables are presented as mean ± standard deviation and categorical variables as number with percentage. Depending on the nature of the data, we used Chisquare test, Kruskal-Wallis H-test, or ANOVA to detect subgroup differences. Three sets of cox regression models were performed to investigate the association between apoB and mortality. Model 1 only included apoB as a continuous variable or quartiles. Model 2 was adjusted for age, sex, and race. In addition to the variables in Model 2, smoking status, education level, BMI, SBP, eGFR, TG, HDL-C, energy intake, diabetes, hypertension, antihypertensive drugs, hypoglycemic drugs, and lipid-lowering medication were additionally adjusted in Model 3. Besides, a trend across apoB quartile groups was tested. Kaplan-Meier survival analyses were performed to present all-cause and cardiovascular mortality by apoB quartiles, and the log-rank test compared survival probabilities. The non-linear relationship between apoB and mortality was detected using restricted cubic spline models. Two-piecewise linear regression models were applied to determine the threshold point for the non-linear relationship. We further employ a logarithmic likelihood ratio test to compare the differences between one-line linear regression models and twopiecewise linear regression models. All statistical analyses were performed with R version 3.6.3 (R Foundation for Statistical Computing, Vienna, Austria). Statistical signi cance was accepted at the level of P < 0.05.

Baseline Characteristics
Sociodemographic, laboratory, and clinical characteristics at baseline of 10375 participants are shown in Table 1. Among those participants (mean age 46.3 ± 16.9, 47.88% men, the average apoB was 93.76 (SD, 25.82) mg/dL. With an average follow-up time of 69.2 months, 533 (5.14%) death occurred totally, of which 91 (0.88%) died of cardiovascular diseases during the follow-up period. Higher apoB quartiles were related with males, lower education level, higher BMI and blood lipid level, and comorbidities with hypertension (P < 0.001). Compared to the lower quartiles, individuals in the highest apoB quartile were more likely to have comorbid diabetes (19.23%).
The Relationship of ApoB With All-Cause and Cardiovascular Mortality.
Strati ed by apoB quartiles (15-76, 77-92, 93-110, 111-345 mg/dL), survival probabilities displayed in Kaplan-Meier curves were demonstrated in Figure 2. The accumulative incidence of cardiovascular mortality increased with apoB (log-rank test, P = 0.04) while there were no statistical differences among apoB quartile groups for all-cause mortality (log-rank test, P = 0.72). As shown in Table 2, when using the value of apoB as a continuous variable, apoB was positively correlated (HR, 1.13; 95% CI, 1.03-1.24) with cardiovascular death events and statistically irrelevant (HR, 0.97; 95% CI, 0.93-1.01) with all-cause mortality in fully adjusted model. Similarly, when using apoB quartiles as categorical variables, the trend between elevating apoB and all-cause mortality was not signi cant compared to the reference group (the lowest apoB quartile). However, the third quartile indicated a notable decline in all-cause death after adjusting for confounders (HR, 0.71; 95% CI, 0.56, 0.91). Meanwhile, participants in higher apoB quartiles had increased risks of cardiovascular death ( P for trend < 0.0053) in Model I, especially in the highest apoB group (HR, 2.45; 95% CI, 1.29-4.65). The positive correlation with cardiovascular mortality persisted in Model III (P for trend < 0.0412).

The Analyses of Non-linear Relationship
Restricted cubic spline models were applied to analyze the non-linear relationship between apoB and mortality in Figure 3. Results implied that there was a U-shaped association between apoB and all-cause mortality (Non-linear P = 0.005) , but not between apoB and cardiovascular death (Non-linear P = 0.88). According to two piecewise linear regression models (Table 3), the threshold value was 108 mg/dL for allcause mortality. Below the threshold point, apoB was negatively associated with cardiovascular mortality (HR, 0.90; 95% CI, 0.85-0.95), while above the threshold point, the increment of apoB was positively related with cardiovascular mortality (HR, 1.09; 95% CI, 1.01-1. 19).

Discussion
Our study found that participants in the fourth quartile had more frequent comorbidities with diabetes mellitus at baseline characteristics. When treating apoB as a continuous variable, the HR for cardiovascular mortality per 1 mg/dL increase in apoB was 1.13. Furthermore, a U-shaped association was shown in apoB and all-case motality, with the higher risks of all-cause mortality on both sides of apoB levels, in which the threshold value was 108 mg/dL. The European Society of Cardiology (ESC) and European Atherosclerosis Society (EAS) stated that apoB should be a secondary target of preventive therapy in patients with hypertriglyceridemia, obesity, or diabetes according to cardiovascular mortality risk categories (18). Among the above patients, accumulating small lipid-depleted LDL subfractions containing less cholesterol is produced, leading to high apoB concentration without necessarily high LDL-C(3). Furthermore, targeting only LDL-C could result in missing opportunities for CV risk reduction in type 2 diabetes mellitus (T2DM) patients (19). These researches are consistent with our nding that the fourth quartile had more frequent comorbidities with diabetes, indicating that apoB provides a more consistent goal for statin treatment than the LDL-C (20).
We found that apoB was positively correlated with cardiovascular mortality even after adjusting for the confounders. Zhu et al. (21) also found that ApoB was associated with occurrence rates of adverse cardiovascular events (HR, 1.349; 95% CI, 1.073-1.695). The ndings of our study were not similar to previous studies on the relationship between LDL-C and cardiovascular deaths. In meta-analyses, Navarese et al. (22) claimed that the most signi cant reduction in risk of all-cause and cardiovascular mortality from LDL-C-lowering therapy might occur for patients with higher baseline LDL-C levels not for lower LDL-C levels (less than 100 mg/dL). Other researches also indicated that LDL-C did not perform well at low concentrations, regardless of the relationship with all-cause or cardiovascular mortality (23,24). Additionally, Mendelian randomization studies had reported that the reduction in cardiovascular risk was associated with lowering the number of atherogenic apoB particles rather than the mass of cholesterol as measured by LDL-C (25), which was consistent with our results. Generally speaking, although LDL-C still acts as the primary criterion for lowering blood lipids in clinical practice, apoB, a more sensitive indicator, should be used more frequently for guiding lipid-lowering treatment, especially for cardiovascular events.
A critical nding of our study was that the association between levels of apoB and all-cause mortality was U-shaped, with low and high levels associated with an increased risk of all-cause mortality. This result was different from the trend of cardiovascular mortality. Besides, when apoB was used as a categorical variable, there was a statistically signi cant decrease in all-cause mortality in the Q3 group after adjusting confounders, which was consistent with the non-linear relationship. This nding was contrary to the general assumption that higher mortality always along with high lipoprotein levels. However, numerous studies performed different associations between lipoprotein levels and all-cause mortality. From the Copenhagen General Population Study, Camilla et al. (9) found a J-shaped association between apoB and all-cause mortality among patients receiving statin therapy. Similarly, in the same study population, researchers showed a U-shaped relation between LDL-C and HDL-C and all-cause mortality levels, even after being adjusted for baseline comorbidities (26,27). However, most previous studies on the association between LDL-C and mortality risk showed no associations or inverse associations (28)(29)(30). Although we had adjusted some confounders, whether the association between moderately increased apoB levels and decreased mortality is causal is still an unresolved question to the ndings. Researchers found that, compared to healthy individuals, chronic obstructive pulmonary disease (COPD) patients had higher HDL-C and lower TG, TC, and LDL-C, while those lipid parameters varied reversely in COPD patients with cardiovascular diseases (31). Other disease conditions, including debilitation and wasting in liver diseases, would result in less secretion of lipoproteins (32). Furthermore, these conditions have been hypothesized to cause a decrease in levels of apoB. Besides, discordance analysis demonstrated that apoB was a more accurate marker of all-cause mortality risk in statin-treated patients than LDL-C or non-HDL cholesterol (9). Generally speaking, apoB, a more sensitive indicator, should be discussed and applied more frequently for guiding lipid-lowering treatment in clinical practice.
Another critical nding of this study was that we disclosed the threshold value (108 mg/dL) for all-cause mortality according to two piecewise linear regression models. Nonetheless, the threshold value was 73 mg/dl in the Camilla et al. (9) study, in which the median value for apoB was consistent with our study (92 mg/dl). The reasons for inconsistent thresholds may be related to differences in adjusted variables and study populations. Currently, the standard of apoB > 130 mg/dl multi-society endorsed by the American College of Cardiology (ACC) and American Heart Association (AHA) for patients with an average estimated risk score seems inappropriately high compared with the threshold (108 mg/dl) of our study(33). In contrast, ESC/EAS guideline de nes the secondary target for apoB as < 100 mg/dl in patients with mild-to-moderate hypertriglyceridemia (175-880 mg/dL), including those with obesity or metabolic syndrome, diabetes, or chronic kidney disease seem more reasonable (18).
Despite the critical ndings we mentioned above, several limitations should be taken into consideration. There might be restrictions regarding generalization across other regions as we only included the United States citizens. Another limitation is that our study did not imply causality between apoB-lowering treatment and decreased mortality due to the cross-sectional nature. Further randomized clinical trials (RCTs) should be conducted to con rm this perspective.

Conclusion
There were differences in the association between apoB with cardiovascular and all-cause mortality among the US population. ApoB was signi cantly associated with an increased risk for cardiovascular mortality, while there existed a non-linear relationship of apoB with all-cause mortality with an increased risk of all-cause mortality at both low and high apoB levels. Whether the reduction of apoB may reduce cardiovascular and all-cause mortality risk, in the long run, needs to be further evaluated by intervention studies.

Declarations
Ethics approval and consent to participate     Spline analyses of all-cause (A) and cardiovascular (B) mortality by Apolipoprotein B in the overall cohort, and the probability distribution histogram is represented in the background. (Spline analyses were adjusted for age, gender, race, smoking, education level, body mass index, systolic blood pressure, estimated glomerular ltration rate, triglyceride, high-density lipoprotein cholesterol, energy intake, diabetes and hypertension, antihypertensive drugs, hypoglycemic agents, and lipid-lowering drugs).

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. table1.pptx Table2.pptx Table3.pptx