The association between lipid levels and major adverse cardiovascular events in rheumatoid arthritis compared to non-RA

Abstract

Objectives

Lower levels of low density lipoprotein (LDL-C) may be associated with increased cardiovascular (CV) risk in rheumatoid arthritis (RA). We studied whether the complex relationship between LDL-C and high density lipoprotein cholesterol (HDL-C) levels with CV risk is different in RA compared to non-RA.

Methods

Using data from a US health insurance plan (2003–2012), we conducted a cohort study that included RA and non-RA patients matched on age, sex and index date. Non-linearity between lipid levels and major adverse CV events (MACE) was tested. We used multivariable Cox proportional hazards regression models to examine for an interaction between lipids and RA on the risk of MACE, adjusting for CV risk factors.

Results

We studied 16,085 RA and 48,499 non-RA subjects with mean age 52.6 years and 78.6% women. The relationship between LDL-C and MACE was non-linear and similar between RA and non-RA (p for interaction=0.72). We observed no significant increase in CV risk between the lowest LDL-C quintile (<91.g/dL) and successive quintiles until the highest quintile (>190.0mg/dL) was compared; hazard ratio (HR) 1.40,95%CI 1.17,1.68). The relationship between HDL and MACE was also non-linear and similar in RA and non-RA (p for interaction=0.39). Compared to the lowest HDL-C quintile, each successive quintile was associated with reduced risk of MACE [lowest (<43.0mg/dL) vs highest quintile (>71.0mg/dL), HR 0.45,95%CI 0.48,0.72].

Conclusions

The complex relationship between LDL-C, HDL-C and MACE was non-linear in RA and also not statistically different from an age- and sex-matched non-RA cohort.

RA patients have an overall 1.5–2-fold risk for cardiovascular disease (CVD) compared to the general population^1–3, but also lower low density lipoprotein (LDL-C) cholesterol levels^4–7. Studies examining the association between LDL-C levels with CV risk have observed a U-shaped relationship whereby RA subjects with lower LDL-C levels have a risk of CVD similar to subjects with high LDL-C levels, termed the “lipid paradox”^8,9. More studies are needed to further characterize the relationship between LDL-C and CVD. Moreover, whether this U-shaped relationship in RA is significantly different from age- and sex- matched individuals without RA is unclear.

While there are RA studies which examine the relationship between LDL-C and CV risk, only one has described the relationship between high density lipoprotein cholesterol (HDL-C) levels with CV risk among RA patients⁹. In this study, higher levels of HDL-C were associated with lower CV risk, consistent with findings from the general population^10–12. The fact that HDL-C levels have the expected association with CV risk in RA suggests that HDL-C levels may provide important information for estimating CV risk, independent of LDL-C.

The objectives of this study were to: (1) describe the relationship between LDL-C, HDL-C and CV events in an RA and non-RA cohort, (2) compare whether these relationships are significantly different between RA and non-RA, (3) quantify the associations between lipid levels and CV risk, and (4) determine whether HDL-C levels are associated with CV risk independent from LDL-C. We hypothesize that the relationship between LDL-C and CV risk will not be significantly different from a non-RA cohort, and that HDL-C will be associated with MACE independent of LDL-C levels in RA.

METHODS

Data Source

We studied subjects from the United Healthcare database, a large health insurance plan in the United States primarily covering working adults and their family members, with data from January 1, 2003 through December 31, 2012. The database contains claims data including medical diagnoses, procedures, medication prescriptions and health care visits. Results for outpatient laboratory tests including lipid levels were available on a subset of beneficiaries. Personal identifiers were removed from the dataset before the analysis to protect subject confidentiality. Patient informed consent was therefore not required. The study protocol was approved by the Institutional Review Board of the Brigham and Women’s Hospital.

Study Cohort

Eligible patients were subjects aged 18 years and older who had LDL-C and HDL-C measurements available in the study database. Patients with RA were identified with at least two visits coded with the International Classification of Diseases code, 9^th Revision (ICD9) for RA that are 7 days apart and at least one dispensing for a disease modifying anti-rheumatic drug (DMARD) in any order. The RA date was defined as the date they fulfilled these criteria in any order. This algorithm was previously validated with a positive predictive value of 86%¹³. The index date was the first LDL-C and HDL-C levels measured on the same date after the RA date.

The non-RA cohort consisted of patients without RA who had at least two physician visits and at least one prescription of any kind, in any order (non-RA date). The non-RA cohort was matched to the RA cohort on age, sex and the index date (the first LDL-C and HDL-C measurements after the non-RA date) in a 3:1 ratio.

All subjects were required to have 365 days of continuous medical and pharmacy coverage prior to the index date. We excluded nursing home residents and subjects who received statin therapy 365 days before the index date.

Exposures

The exposures of interest were LDL-C and HDL-C levels at the index date. To study non-linear associations between LDL-C and MACE, we categorized LDL-C levels into quintiles: Quintile 1: ≤91.0mg/dL (reference category), Quintile 2: >91.0 and ≤123.0, Quintile 3: >123.0 and <=142.0, Quintile 4: >142.0 and ≤190.0, Quintile 5: >190.0mg/dL. HDL-C was also categorized into quintiles: Quintile 1: ≤43.0mg/dL (reference category), Quintile 2: >43.0 and ≤51.0, Quintile 3: >51.0 and ≤60.0, Quintile 4: >60.0 and ≤71.0, Quintile 5: >71.0mg/dL.

Study Outcome

The primary outcome was a non-fatal major adverse cardiovascular event (MACE) defined as a myocardial infarction, coronary artery bypass graft (CABG), coronary revascularization or stroke (Supplementary Table 1), validated in previous administrative database studies^14,15. We studied patients in both groups starting one day after the index date to the first of any of the following censoring events: statin initiation, development of MACE, insurance disenrollment, end of study period, nursing home admission, or death.

Covariates

We examined the patients’ baseline variables potentially related to lipid levels, RA or development of MACE using data from the 365 days before the index date. These variables included age, sex, comorbidity index¹⁶, and the CV risk factors and treatments related to management of CV risk factors: hypertension (HTN), diabetes mellitus (DM), hyperlipidemia, smoking, obesity, non-statin lipid lowering agents, anti-platelet agents, anti-hypertensive treatment and history of CVD.

Statistical Analysis

We performed univariable analyses to compare the baseline characteristics between the RA and non-RA cohorts matched on age, sex, and index date. First, to characterize the relationship between LDL-C and MACE, we tested for linearity using the cubic spline test¹⁷ separately in the RA and non-RA cohorts. The cubic spline method provided graphical information on the relationship between LDL-C and MACE and also tested for non-linearity of the relationship. We excluded subjects above the 97.5^th percentile and below 2.5^th percentile of LDL-C levels to reduce the influence of extreme outliers, thus the final numbers of RA case to non-RA was not exactly 3:1. Second, we tested for an interaction between LDL-C and RA on the risk of MACE using a product term in a multivariable Cox proportional hazards model, adjusted for the aforementioned covariates at baseline. We found no statistically significant interaction between LDL-C and RA (p for interaction <0.05). Third, to study the risk of MACE related to each successive LDL-C quintile, we constructed a multivariable Cox proportional hazards model combining the RA and non-RA cohorts and adjusting for CV risk factors and RA status. The first 3 steps were repeated for the analysis of HDL-C and the risk of MACE in RA and non-RA. Finally, we constructed a multivariable Cox proportional hazards model adjusted for LDL-C and HDL-C. This allowed us to test for an independent association between HDL-C and CV risk independent of LDL-C levels. A two-sided p-value< 0.05 was considered a statistically significant result. All analyses used the SAS 9.3 Statistical Software package (SAS Institute Inc., Cary, NC).

RESULTS

We studied 16,085 RA and 48,499 non-RA subjects. The mean age was 52.6 years (Table 1). RA subjects had a higher comorbidity index compared to non-RA, reflected by a higher percentage of RA subjects with CV risk factors compared to non-RA. The mean total cholesterol (TC) was lower in RA (196.3mg/dL) compared to non-RA (200.5mg/dL). LDL-C was also lower in RA (113.8mg/dL compared to non-RA (117.1mg/dL). HDL-C levels were similar between the two groups, 57.3mg/dL in RA and 58.6mg/dL in non-RA.

Table 1.

Baseline characteristics of the study cohorts

Clinical characteristics	RA, n= 16,085	Non-RA, n= 48,499
Mean age (SD)	52.63 (8.68)	52.62 (8.70)
Female (%)	78.55	78.58
Comorbidities, (%)
Cardiovascular dsiease	7.24	4.49
Diabetes	14.61	10.17
Hypertension	43.25	34.61
Hyperlipidemia	46.48	40.84
Obesity	7.69	6.26
Smoking	5.00	4.08
Comorbidity index, mean (SD)	0.37 (1.24)	0.11 (0.10)
Medications
Anti-platelet agent	1.96	1.08
Beta blocker	13.14	9.36
ACE or ARB	21.82	16.50
Non-statin lipid lowering agent	5.14	3.57
NSAID	71.89	38.74
RA medications
Methotrexate	44.16	<1
Other nbDMARD	31.02	<1
Anti-TNF	33.10	<1
Other biologic DMARDs	2.06	<1
Laboratory values
Mean Total cholesterol, mg/dL (SD)	196.34 (34.28)	200.45 (33.39)
Mean LDL-C, mg/dL (SD)	113.78 (28.26)	117.71 (28.31)
Mean HDL-C, mg/dL (SD)	57.41 (17.36)	58.60 (17.13)
Mean triglycerides, mg/dL (SD)	126.06 (66.92)	120.94 (65.60)

In the RA cohort we observed 453 MACE events over 32,007 person-years (PYs) with an incidence rate of 14.2 (95% CI 12.9, 15.5) per 1,000 PYs. In the non-RA cohort, we observed 704 MACE events over 99,098 PYs with an incidence rate of 7.1 (95% CI 6.6, 7.6) per 1,000 PYs.

LDL-C and MACE

We observed a similar U-shaped relationship between LDL-C levels and MACE in the RA and non-RA cohorts, whereby subjects at highest risk for MACE had the lowest and highest LDL-C levels (Figure 1A, B). The relationship in the non-RA cohort was non-linear (p=0.01), while we could not reject linearity in the RA cohort (p=0.13). Due to the non-linearity of the association between LDL-C and MACE in the non-RA cohort, we conducted subsequent analyses with LDL-C categorized by quintiles.

Figure 1.

Relationship between LDL-C and MACE (A, B), and HDL-C and MACE (C, D) and p-value testing for linearity in the RA and non-RA cohorts.

No significant difference in the relationship between LDL-C and MACE was observed between the two cohorts (p for interaction=0.72). Thus, we proceeded to study the association between LDL-C levels and MACE using data from both the RA and non-RA cohorts combined in one Cox proportional hazards regression model. Compared to the lowest LDL-C quintile, subjects with the highest LDL-C quintile had a 40% higher HR (95% CI 1.17, 1.68) for MACE (Figure 2A), adjusted by CV risk factors. No significant difference in risk of MACE was found between the second, third and fourth quintile compared to the lowest quintile.

Figure 2.

Hazard ratios (HR) for MACE with each successive (A) LDL-C quintile and (B) HDL-C quintile, adjusted for age, gender, RA status, comorbidities, and CV risk factors.

(Note: p-values compared to the lowest quintile,*p=0.0002, **p=0.04, ***p<0.0001).

HDL-C and MACE

In both RA and non-RA, the risk of MACE was highest in subjects with low HDL-C compared to higher levels of HDL-C (Figure 1C, D). The relationship between HDL-C levels and MACE in the non-RA cohort was non-linear (p=0.0003), while linearity could not be rejected in the RA cohort (p=0.11). Due to the non-linear relationship in the non-RA cohort, we categorized HDL-C levels into quintiles for the analysis.

Since we found no significant difference between the association of HDL-C levels and MACE in RA and non-RA (p for interaction, p=0.39), the cohorts were combined for the analysis. Each successively higher quintile of HDL-C was associated with a lower risk of MACE compared to the lowest quintile (Figure 2B). Subjects with the highest HDL-C quintile had 55% lower risk for MACE (HR 0.45, 95%CI 0.48,0.72) compared to the lowest quintile, adjusted by CV risk factors.

HDL-C and MACE Controlling for LDL-C

In a model including both LDL-C and HDL-C levels for risk of MACE, HDL-C was associated with the risk of MACE independent of LDL-C levels. Each successive HDL-C quintile was associated with lower risk of MACE compared to the lowest quintile (Table 2). For LDL-C, there was no significant difference in risk of MACE between the lowest LDL-C quintile and each successive quintile, until the highest LDL-C quintile was reached. In the final multivariable Cox model, having RA was associated with 1.7-fold elevated risk of MACE (HR 1.7, 95% CI 1.5, 1.9). Age, male sex and CV risk factors such as HTN, DM and smoking were also associated with elevated risk of MACE (Supplemental Table 2).

Table 2.

Multivariable hazards ratios of LDL-C and HDL-C^* for the risk of MACE.

Variable		Hazard ratio*	95% CI	p-value
LDL-C quintile	1	Ref	-	-
	2	0.98	0.81, 1.19	0.86
	3	1.09	0.90, 1.24	0.84
	4	1.09	0.90, 1.32	0.38
	5	1.38	1.16, 1.66	0.0004
HDL-C quintile	1	Ref	-	-
	2	0.83	0.71, 0.98	0.03
	3	0.69	0.57, 0.82	<0.0001
	4	0.67	0.55, 0.81	<0.0001
	5	0.60	0.49, 0.73	<0.0001

^*adjusted by age, sex, RA status, comorbidity index, and CV risk factors

DISCUSSION

We observed a non-linear relationship between LDL-C and MACE, often described as the lipid paradox, in both age- and sex- matched RA and non-RA patients from the same source population. The quantitative association between LDL-C and MACE was significant only when comparing the lowest LDL-C quintile (LDL-C<91.0mg/dL) with the highest LDL-C quintile (LDL-C>190.0mg/dL), with no significant differences between the lowest quintile and each successive quintile. In contrast, for HDL-C we observed the expected relationship that successively higher levels of HDL-C were associated with lower risk of MACE.

While two RA cohort studies observed a U-shaped relationship between LDL-C levels and CV risk whereby subjects with the lowest and highest LDL-C were at highest risk of a CV event^8,9, no study has formally tested for linearity or directly compared this relationship with a non-RA population. Earlier studies suggest that this U-shaped relationship also exists in the general population. Investigators from the Framingham Heart Study (FHS), a population based prospective cohort study, observed a U-shaped relationship between cholesterol levels and non-CVD related mortality in subjects age 50 and older¹⁸. Additionally, the general accepted relationship between higher cholesterol and CVD mortality was attenuated with each decade, until ages 70–79 where there was no relationship, and age 80 where there was an inverse relationship between cholesterol and CV mortality. The investigators examined extreme illness or outliers as the potential driver for this relationship, but the association remained after adjusting for these factors. In line with the FHS findings, we observed a U-shaped relationship between LDL-C and MACE in both cohorts after adjusting for age, sex, comorbidities and CV risk factors.

In RA, we hypothesize that a subset of the patients with low LDL-C levels represents patients with a higher burden of inflammation. Previous studies have observed that LDL-C levels in RA are lower compared to the general population^5–7, a relationship also observed in this present study (mean LDL-C RA 113.8mg/dL), non-RA 117.7mg/dL). Our group has also studied RA genetic risk alleles as markers of immune dysregulation and found an inverse relationship between the number of RA risk alleles carried (a proxy for a higher degree of immune dysregulation) and LDL-C levels in RA and non-RA¹⁹. Data from tumor necrosis factor inhibitor (TNFi) clinical trials also support the notion that LDL-C may vary with levels of inflammation. Treatment with TNFi’s reducing inflammation are associated with increases in LDL-C by as much as 30%^20,21. Increases in LDL-C have also been reported with treatment using other biologic²² and non-biologic DMARD therapy²³. There are also data to support that the converse is true- that LDL-C levels may decrease with increase in inflammation^6,24.

In our quantitative analyses we observed that subjects with the lowest LDL-C levels were not at significantly higher risk compared to subjects with LDL-C levels in the 2^nd to 4^th quintile in either the RA or non-RA cohorts. Rather, the risk appears similar across all LDL-C quintiles until the lowest quintile (LDL-C≤90mg/dL) was compared with the highest quintile (LDL-C>190mg/dL). These findings are consistent with a previous study of RA subjects where LDL-C levels were categorized based on the Adult Treatment Panel (ATP) III Guidelines^9,25. The investigators observed a significant relationship between LDL-C and incident myocardial infarction only when the lowest category (LDL-C<70mg/dL) was compared with the highest LDL-C category (LDL-C>160mg/dL)⁹. Together these data suggest that the qualitative U-shaped relationship where subjects with the lowest LDL-C levels have higher CV risk than those with moderate LDL-C levels was not statistically significant. CV risk may not change significantly with increasing LDL-C levels until high LDL-C levels (approximately 160mg/dL) were reached.

In contrast to LDL-C, each successively higher quintile of HDL-C was associated with a lower risk of MACE in both RA and non-RA. These findings are consistent with general population studies^11,12 and one study in RA⁹. Moreover, we demonstrate that the association of HDL-C and MACE was independent of LDL-C levels, highlighting the potential benefit of considering both LDL-C and HDL-C levels when estimating CV risk in RA. One potential explanation for the independent significance of HDL-C levels with CV risk among RA patients in particular is that HDL-C levels appear to remain relatively stable with changes in inflammation and may correlate better than LDL-C for CV risk. The data are conflicting regarding whether treatment is associated with higher or lower HDL-C levels and more studies are needed^{20,21,26–28}. In the present study, HDL-C levels were also similar between RA and the age-, sex-matched non-RA patients (mean HDL-C 57.4mg/dL in RA, 58.6mg/dL in non-RA) similar to a previous study⁵.

Despite a similar relationship between LDL-C, HDL-C and MACE between RA and non-RA, RA subjects remained at 1.7-fold risk of CVD compared to age- and sex-matched non-RA subjects from the same population. More specifically, for any given level of LDL-C, RA patients are at higher risk of CVD compared to non-RA after accounting for age, gender, CV risk factors and HDL-C level. This magnitude of increased CV risk is similar to studies published a decade ago^2,29, suggesting that current practice and new RA therapies have not closed the gap between CV risk in RA compared to non-RA.

There were limitations to this study. LDL-C and HDL-C levels were obtained as part of routine care and the fasting status was unknown. However, data from a population based study suggested that non-fasting LDL-C and HDL-C levels do not vary significantly from fasting levels. The variation for LDL-C was 10% while HDL-C varied by less than 2%³⁰. The variations reported for triglyceride levels between fasting and non-fasting status was higher at 20%. Thus we reported mean triglyceride levels in RA vs. non-RA but did not analyze these data because the potential variations in levels would limit interpretation. The RA classification algorithm applied to these data had a PPV of 86% and the influence of potential misclassification on the results is unknown. The non-RA cohort was not the general population, but represented individuals of similar age and sex without RA from the same commercially insured US population. The RA cohort had a higher prevalence of CV risk factors than the non-RA cohort leading to the potential for residual confounding despite adjusting by CV risk factors and comorbidities. Finally, we were unable to directly examine the effect of inflammation due to the low numbers of subjects in both cohorts with concurrent LDL-C, HDL-C and CRP or ESR measurements.

In conclusion, the results of this study found that the U-shaped relationship between LDL-C and CV risk in RA, termed the lipid paradox was not significantly different from age- and sex- matched individuals in a non-RA cohort. However our findings support the notion highlighted by the lipid paradox that LDL-C levels may be a suboptimal measure for CV risk assessment in RA. We observed no association between increasing levels of LDL-C and increased CV risk until the lowest levels were compared with the highest LDL-C levels. In this contemporary cohort we observed a 1.7-fold risk in RA compared to the non-RA cohort, demonstrating a persistently elevated risk of CVD in RA and the need for improved methods for CV risk assessment. A recent study examining the accuracy of the 2013 ACC/AHA Guidelines³¹ for the treatment of blood cholesterol to reduce CV risk in adults suggests that it performs no better than the Framingham Risk Score³² in RA. Both calculators underestimate CV risk in RA^33,34. A TransAtlantic Cardiovascular risk Calculator for RA (ATTAC-RA)³⁵, an international collaborative effort is underway to address this major need in optimizing care for RA patients. More work is needed to improve the accuracy of CV risk estimation in RA and identify effective interventions to reduce CVD in RA patients.