Omega-3 index improves upon the pooled cohort equation in predicting risk for CVD

Abstract

BACKGROUND:

Accurate predictive tools are crucial for identifying patients at increased risk for atherosclerotic cardiovascular disease (ASCVD). The Pooled Cohort Equation (PCE) is commonly used to predict 10-year risk for ASCVD, but its accuracy remains imperfect.

OBJECTIVE:

This study examined the extent to which the omega-3 index (O3I; the proportion of eicosapentaenoic acid+docosahexaenoic acid in erythrocyte membranes) improved the predictive capability of PCE.

METHODS:

The O3I was determined in 2550 participants without ASCVD at baseline from Framingham Offspring Cohort. The extent to which the O3I added to the PCE score was assessed using area under the curve (AUC). We also estimated how much the O3I added predictive power to each standard risk factor (blood pressure, diabetes, smoking, total and high-density lipoprotein cholesterol [HDL-C]) individually when added to the basic age+sex+race model. Mean follow-up was 9.1 years.

RESULTS:

The AUC predicting 10-year ASCVD events using PCE was 0.689. It increased to 0.698 (P < .05) upon addition of the O3I. The AUC additions to the basic model were 0.028 (blood pressure, HDL-C), 0.020 (diabetes), 0.012 (O3I), 0.006 (cholesterol), and 0.004 (smoking); all but smoking were significant (P < .05). Also, the O3I significantly (P < .05) improved the predictive ability of each of these risk factors when analyzed separately.

CONCLUSION:

The O3I improved the PCE prediction, suggesting that it captures risk beyond standard factors. Thus, the O3I may help in ASCVD risk stratification. Further research is needed to extend these findings into more diverse cohorts and to explore the integration of O3I into other existing ASCVD risk assessment tools.

Introduction

Cardiovascular (CV) disease (CVD) is the leading cause of morbidity and mortality in the United States (US) as well as globally.¹ Nearly half of all adults in the US over the age of 20 have some form of CVD with a prevalence of 127 million individuals in 2018.¹ CVD is the leading cause of mortality in the US, claiming more lives than both cancer and lower respiratory disease combined. Additionally, CVD often affects younger patients, with 35% of all life-changing CV events occurring in adults aged 35-65 years. Through improved prevention, risk assessment, and management there has been a decline in the CVD death rate between 1980 and 2010. However, this trend has been reversing over the past decade with the overall death rate due to CVD increasing from 2010 to 2019.¹

Accurate risk assessment of atherosclerotic CVD (ASCVD) continues to be the foundation of the primary prevention strategy, followed by risk mitigation.² The Framingham Risk Score (FRS) was among the first risk assessment tools, combining several risk factors into a single 10-year risk score.^3,4 The FRS has been revised and broadened to be inclusive of all ASCVD outcomes besides just coronary heart disease (CHD) with an updated calculator called the Pooled Cohort Equation (PCE).^5,6 The PCE has become an important tool in assessing risk and is clinically relevant for informing the decision to, for example, initiate statin therapy in the primary prevention of ASCVD.⁷ At a “decision threshold” 10-year risk of 7.5% to 10%, the PCE has been validated through multiple large cohorts and is now considered well-calibrated.^8,9

The PCE, however, remains imperfect. For example, it overpredicts risk when applied to affluent individuals already participating in regular screening and preventive care.⁸ The PCE also tends to underestimate risk in people less engaged with healthcare providers and/or of lower socioeconomic status.⁸ Additionally, it has yet to be thoroughly validated in younger or older patient cohorts (<40 and >75 years of age) and in patients with systolic blood pressure (SBP) >200 mm Hg, which is outside of the level allowed by the PCE.⁸

The red blood cell (RBC) content of the marine-derived omega-3 fatty acids (O3FAs; eicosapentaenoic acid [EPA] + docosahexaenoic acid [DHA], otherwise known as the omega-3 index [O3I]) represents a potential additional input into the PCE. Randomized controlled trials generally show that O3FA treatment decreases major adverse CV events including CVD mortality.^10,11 In the Framingham Offspring cohort, the O3I was inversely related to risk for CVD events and total mortality.^12–14 A more recent meta-analysis of 18 prospective cohort studies comprising 160,404 individuals and 24,342 deaths during a median of 14 years of follow-up showed that higher DHA blood levels were associated with significant risk reductions in CVD death, cancer death, and all-cause mortality.¹⁵ Furthermore, in a similar meta-analysis including 29 cohorts and over 183,000 individuals, higher O3FA blood levels were associated with a decreased stroke risk.¹⁶

In 2009, a suite of 10 RBC FAs (including two O3FAs) was shown to outperform the FRS in distinguishing acute coronary syndrome patients from controls and to add significantly to the ability of the FRS to discriminate between them.¹⁷ The O3I is a particularly attractive target because it is easily modified by dietary change and/or O3FA supplementation.¹⁸ The present study was undertaken to test the hypothesis that the O3I will improve upon the PCE in assessing 10-year risk for CVD.

Methods

Sample

The Framingham Heart Study is an ongoing population-based study from the town of Framingham, Massachusetts, US. The Original cohort was established in 1948 with the aim to identify factors that contribute to the development of CVD.¹⁹ In 1971, the Offspring cohort was established, including children of the Original cohort and their spouses.²⁰ The Offspring cohort enrolled 5124 participants who have been studied over 9 examination cycles, approximately once every 4 years. In 1994, the Omni Cohort I enrolled 507 men and women of African American, Hispanic, Indian, Pacific Islander, and Native American origins who were also residents of Framingham, MA, with exam cycles paralleling that of the Offspring cohort. In the Offspring Cohort, 3021 attended their eighth examination cycle (2005-2008), at which time RBCs were collected. Similarly, 298 individuals in the Omni I cohort attended exam 3 (2007-2008), at which time RBCs were also collected, and both sets were analyzed for FA composition (see below). Participants were excluded if they were missing RBC FA measurements, were over age 80, or were missing any heart risk factor covariates, leaving a combined (Offspring+Omni I) sample size of n = 2894. Of these, 344 had a diagnosis of CVD at the time of RBC collection and were excluded from all subsequent analyses, leaving a sample of n = 2550 for all primary analyses (2296 from the Offspring cohort and 254 from the Omni I cohort). See Supplemental Figure 1. The study protocol was approved by the Institutional Review Board of the Boston University Medical Center. Informed consent was provided by all participants.

FA determination

Blood was drawn after a 10 to 12 hours fast into an EDTA tube; RBCs were isolated by centrifugation and were frozen at −80°C immediately after collection. RBC FA composition was determined as described previously.²¹ Briefly, RBCs were incubated at 100°C with BF₃-methanol and hexane to generate FA methyl esters that were then analyzed by gas chromatography with flame ionization detection. Twentyseven FAs were quantified, and the O3I was computed as the sum of DHA and EPA expressed as a percent of total RBC FAs. The inter-assay coefficient of variation was <4% for both EPA and DHA.^13,16,22

Covariates and CVD outcome

We considered 8 primary baseline demographic and CV risk factors recommended in the American College of Cardiology/American Heart Association Guideline on the Assessment of CV Risk by Goff et al.: sex, age, race, prevalent diabetes, smoking status, total cholesterol, high-density lipoprotein cholesterol (HDL-C), and SBP.⁵ These factors were used to generate a risk score for each participant using the PCE. Total CVD is defined as any stroke, coronary event, or CHD mortality, and was based on outcomes adjudicated by the Framingham investigators.

Statistical analysis

Descriptive metrics for the sample are presented based on standard statistical methods, including means, SD, correlations, etc. Hazard ratios were estimated using the survival package in R,²³ with the area under the curve (AUC) as the primary measure of model predictive ability. Proportional hazards assumptions were verified. Primary analyses predicted incident clinical outcomes (date of event [CVD or death] or date of censoring) by quintile of O3I (as we and others have done before)^13,16,22 and by individual risk factors in models adjusted for age and sex.^{13, 16, 22} Time to censoring occurred at death, CVD, or last follow-up, with follow-up censored at 10 years. In our analysis of individual risk factors, we computed increases in the AUC in a stepwise manner, comparing the predictive value of each individual risk factor when added to age and sex. All analyses used 2-sided tests at the 0.05 significance level.

Results

Sample characteristics

In the analytic sample of 2550 participants free of CVD at baseline, the mean age was 64 years, 42.4% were male and 80% were non-Hispanic white. In this sample the mean 10-year ASCVD risk as predicted by the PCE was 14.3%, which is higher than the observed rate of 8.7% (223 incident CVD events). The mean (SD) time to data censoring was 9.1 (2.2) years (Table 1).

Table 1.

Characteristics of the participants without CVD at baseline (N = 2550).

Characteristic	% (n) or Mean (SD)
Sex - Male	42.4% (n = 1082)
Race/Ethnicity – non-Hispanic whitea	80.0% (n = 2041)
Age - years	64 (SD = 7.8)
Total cholesterol - mg/dL	189.9 (SD = 36.3)
HDL cholesterol - mg/dL	58.4 (SD = 18.2)
Systolic BP - mm Hg	127.4 (SD = 16.7)
Smoking - current	9.4% (n = 239)
Prevalent diabetes - yes	14.4% (n = 367)
Hypertension medications - yes	44.2% (n = 1126)
Statin use – yes	37.7% (n = 962)
Omega 3 Index (%)	5.5 (SD = 1.7)
PCE-predicted 10-year risk of CVDb	14.3% (SD = 14.9%)
Observed incident CVD events	8.7% (n = 223)

^aAmong non-Hispanic whites, there are 292 Hispanics, 103 blacks, 67 Asians, 19 American Indian/native Alaskan, and 28 who refused to answer the question based on self-report.

^bGoff et al. (2014) 2013 ACC/AHA Guideline on the Assessment of CV Risk: A report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines.⁵ Table A. Following Goff et al, the version of the PCE for black individuals was applied to self-identified black participants only, with the white (non-Hispanic) equation applied to all non-black participants in the sample, per the recommendation on page S56, Column 1 which states “…providers may consider using the equations for non-Hispanic whites for these patients. When doing so, the estimated risks may be overestimated, especially for Hispanic and Asian Americans.”).

BP, blood pressure; CVD, cardiovascular disease; HDL, high-density lipoprotein; PCE, Pooled Cohort Equation.

O3I and the PCE risk prediction

The O3I improved upon the PCE with an 8-unit AUC (1 unit = AUC increase x 1000) increase (P = .049) (Table 2). In addition, we found that the O3I produced a 7-unit AUC increase (P = .038) above the AUC generated using the PCE risk factors in a sample-optimized model. That is, a model that included the absolute values (categorical or continuous) for each risk factor instead of following the PCE protocol in which one accumulates points based on risk factor ranges or categories. Because the order in which risk factors are added into the model affects the magnitude of each individual risk factor’s contribution to the AUC, we undertook 2 analyses, 1 in which the O3I was added to the age-sex-race model last and 1 in which it was added first (Fig 1). The latter reflects the AUC changes shown in Table 2. When the O3I was added first, it added to the AUC more (10.64) than when it was added last (5.40).

Table 2.

Predictive ability for CVD of the Pooled Cohort Equation (PCE) with and without the omega-3 index (O3I).

	AUC (x1000)	Models being compared (X vs Z)	Change in AUC ΔAUC = AUCX – AUCZ	Relative change in AUC 100ΔAUC/(1000 – AUCZ)a	P-valueb
Risk factor models c
M0a: PCE	689		–	–	–
M1a: PCE + O3I	698	M1a vs M0a	8	2.7	.049*
Risk score predictors estimated on the sample
M0b: PCE	733		–	–	–
M1b: PCE + O3I	740	M1b vs M0b	7	2.5	.038*

^aThe relative change in AUC illustrates how much of the remaining potentially explainable variation in CVD risk is explained by the more complex Model X as compared to the simpler Model Z.

^bThe P-value for the 2 nested models (X and Z) that are being compared. This can be interpreted as the evidence that the more complex model (X) is a better predictor of CVD risk than the simpler model (Z).

^cModels M0a and M1a include the PCE risk score for each participant as the predictor in the model. Models M0b and M1b include all of the main effects and interaction terms that are included in the PCE but allowing the predictor point estimates to be estimated for this sample.

Abbreviations: AUC, area under the curve and CVD, cardiovascular disease.

Figure 1. Area under the curve (AUC) for prediction of cardiovascular disease events over 10 years in the Framingham Offspring Study using the standard risk factors plus the omega-3 index (O3I). Both panels begin with a model including only age and sex.

The panel on the left then sequentially adds to that model total cholesterol (TC), then high-density lipoprotein (HDL) cholesterol, etc. (reading up the legend scale); the panel on the right adds first the O3I, then TC, then HDL, etc. Abbreviations: BP, blood pressure; Diab, diabetes.

O3I and risk prediction based on individual risk factors

In addition to examining the effect of the O3I on the predictive ability of the PCE per se, we also examined the effect of the O3I on the predictive ability of each component of the PCE individually. We began by assessing the increases in the changes in AUC after the inclusion, 1 at a time, of each of the 5 non-demographic PCE risk factors (ie, all but age, sex, and race) in the Age + Sex + Race model. We then added the O3I to each of these models to determine the extent to which it improved each AUC (Table 3; Fig 2). In all 5 cases, including the O3I increased the AUC to a small but statistically significant extent. When individual PCE risk factors were added to the Age + Sex + Race model, the AUC increased by between 4 units (smoking) and 28 units (HDL-C; BP). The further addition of the O3I increased the AUC units by 8 and 13 in each case.

Table 3.

Effect of the inclusion of the O3I on the predictive ability of the individual PCE risk factors.

	AUC (x1000)	Models being compared (X vs Z)	Change in AUC ΔAUC = AUCX – AUCZ	Relative change in AUC 100ΔAUC/(1000 – AUCZ)a	P-valueb
Risk factor models c
M0: No predictors	500
M1: Age	657	M1 vs. M0	157	31.4%	<.001
M2: Age + Sex	666	M2 vs. M1	9	−2.7%	.019
M3: Age + Sex + O3I	679	M3 vs. M2	13	3.9%	.019
M4: Age + Sex + Race	672	M4 vs. M2	6	1.8%	.14
M5: Age + Sex + Race + O3I	683	M5 vs. M4	12	3.5%	.027
M6: Age + Sex + Race + Total Chol	678	M6 vs. M4	6	2.0%	.514
M7: Age + Sex + Race + Total Chol + O3I	687	M7 vs. M6	9	2.8%	.026
M8: Age + Sex + Race + HDL Chol	700	M8 vs. M4	28	8.5%	<.001
M9: Age + Sex + Race + HDL Chol + O3I	707	M9 vs. M8	7	2.5%	.043
M10: Age + Sex + Race + Diabetes	692	M10 vs. M4	20	6.0%	.002
M11: Age + Sex + Race + Diabetes + O3I	700	M11 vs. M10	9	2.8%	.040
M12: Age + Sex + Race + Blood Pressure	700	M12 vs. M4	28	8.6%	.005
M13: Age + Sex + Race + Blood	709	M13 vs. M12	9	2.9%	.044
Pressure + O3I
M14: Age + Sex + Race + Smoking	676	M14 vs. M4	4	1.3%	.340
M15: Age + Sex + Race + Smoking + O3I	687	M15 vs. M14	11	3.4%	.032

^aThe relative change in AUC illustrates how much of the remaining potentially explainable variation in CVD risk is explained by the more complex Model X as compared to the simpler Model Z.

^bThe P-value for the 2 nested models (X and Z) that are being compared. This can be interpreted as the evidence that the more complex model (X) is a better predictor of CVD risk than the simpler model (Z).

^cDenoting “Age” in the model includes both a linear and quadratic term for age, while denoting “Sex” in the models includes both a main effect as well as an interaction between sex and both linear and quadratic age terms. Denoting Race in the model indicates a main effect as well as interactions with age, sex and age*sex. All remaining variables always include interactions with race and sex along with main effects. In addition, denoting either “Total Chol” or “HDL Chol” also include interaction terms with Age. “Blood Pressure (BP)” includes interactions between systolic BP and both age and hypertension treatment status. “Smoking” includes an interaction with Age. The inclusion of interaction terms is meant to reflect the complexity of the PCE equation which has interactions for these variables. The “O3I” variable in all cases means a single categorical variable with 5 levels – 1 for each quintile of the O3I.

Abbreviations: AUC, area under the curve; HDL Chol, high-density lipoprotein cholesterol; O3I, omega-3 index; PCE, Pooled Cohort Equation; Total Chol, total cholesterol.

Figure 2. The figure depicts the increases in areas under the curve (AUCs) upon adding the omega-3 index (O3I) to the base (Age [A] +Sex [S] + Race [R] +risk factor [RF]) models.

For example, adding high-density lipoprotein cholesterol (HDL Chol) to the A+S model increased the AUC by about 28 units. Adding the O3I to the A+S+ R +HDL model further increased the AUC by an additional 7 units (total of 35 units vs A+S+R). This 7 unit difference in AUC was significantly greater than the change from just adding HDL Chol to the A+S+R model. P-values for the change in AUC (vs A+S model) with the RF alone vs with the RF+O3I: * = .05; ^ = .01 and ^^ = .001.

CVD events by O3I quintile

O3I levels were inversely associated with risk of incident CVD (Table 4), with the strongest statistical evidence of differences in a model adjusting for the existing PCE for comparisons vs Q5. Supplemental Figure 2 shows the related Kaplan-Meier Survival curves.

Table 4.

Observed and predicted incident CVD events by omega-3 index quintile.

O3I quintile	Incident events Unadjusted	Model predicted events (Model 1)a	Model predicted events (Model 2)b
Q1 (<4.14%)	8.6% (45/525)	8.2%†	7.6%*
Q2 (4.14 – 4.89%)	10.0% (52/517)†	9.3%‡	8.8%†
Q3 (4.89 – 5.69%)	10.8% (54/502)†	9.2%‡	8.7%†
Q4 (5.69 – 6.83%)	7.8% (39/503)	6.6%	6.3%
Q5 (>6.83%)	6.6% (33/503) (ref)	5.1% (ref)	5.1% (ref)

^aAge adjusted only.

^bAdjusted for existing PCE. Model predictions are made using the sample’s mean age or mean PCE value.

^*P < .10.

^†P < .05.

^‡P < .01.

Abbreviations: CVD, cardiovascular disease; O3I, omega-3 index; PCE, Pooled Cohort Equation; ref, reference.

Discussion

The 3 primary findings from this study were 1) a model that included only unmodifiable risk factors (ie, age, race and sex) explained approximately two-thirds of the increase (beyond a coin toss) in the ability to predict who will have a CVD event in the ensuing 10 years; the other 6 (including O3I) modifiable variables together contributed only an additional one-third of the increase in AUC; 2) the O3I added a small but statistically significant increase in predictability to both the overall PCE and to each of the individual components of the PCE, and 3) the O3I added more predictive power than smoking or total cholesterol, but less than diabetes, HDL-C, and systolic BP. Thus, the predictive power provided by the O3I was roughly equivalent to that provided by each of the 5 standard risk factors. This suggests that the mechanism(s) by which a higher O3I operates to lower risk is, at least in part, independent of effects on these other risk factors. Furthermore, the O3I also showed a numerically similar and statistically significant increase in the predictive value of the PCE in the sample-optimized model (ie, when it was well-calibrated for the Framingham Offspring cohort (Table 2).

In an earlier analysis from the Framingham Offspring cohort, the O3I was shown to have an inverse relationship with incident CVD after controlling for the 5 risk factors examined here (and several more covariates).¹² This fits with the findings from recent meta-analyses that showed an inverse relationship between dose of omega-3 FA treatment and risk of ASCVD events.¹⁰ Similarly, strong inverse correlations were reported in prospective observational studies between O3I and incident ASCVD during long-term follow-up.^10,11 Outside of age, our findings suggest that the O3I is a comparable prognostic factor to each of the other standard CVD risk factors, and is readily modifiable by simply increasing intake of fish/seafood and/or EPA+DHA (Table 3).

Based on our analysis, the lion’s share of the PCE predictive value was derived from age, race, and sex (AUC = 0.672; Table 2) with 6 other modifiable variables (including O3I) adding an additional to make the overall model have an AUC of 0.740 (Table 2). As noted above, the extent to which the O3I contributed to the improved AUC depended on when it was added during model building, adding less when added last compared to when added first (Fig 1). This is presumably because a small portion of the predictive power of the O3I was already accounted for by the other risk factors. Nevertheless, the O3I effect was significant in either case.

The key decision threshold of 7.5% 10-year ASCVD risk is important because it is the pivotal point for starting statin therapy. O3I could prove to be a useful additional tool for clinicians because it improved the identification of those who were at higher risk for ASCVD (Table 4; Fig 3) who would benefit from statin therapy.

Figure 3.

This figure illustrates the distribution of the predicted 10-year Pooled Cohort Equation (PCE) risk probabilities with and without omega-3 index (O3I), as well as between individuals who did/did not develop cardiovascular disease (CVD) within 10 years.

Omega-3s have wide-ranging systemic effects and have been shown to positively impact cardiac, autonomic, endothelial, and metabolic function in addition to decreasing the chronic levels of BP, triglyceride levels and systemic inflammation.²⁴ O3I therefore likely represents a stand-alone and novel risk factor.

Although the O3I did add statistically significantly to the PCE, the extent to which it added was small but comparable to the other 5 modifiable risk factors. The Framingham Offspring cohort is a single sample representing a specific population with a mean 10-year ASCVD risk calculated by the PCE of 14.3%, which makes this relatively high-risk group, likely because the mean age was around 65 years. However, the observed 10-year risk for ASCVD in this cohort was 8.7%, consistent with the fact that, as noted earlier, the PCE overestimates risk in affluent, health-conscious cohorts. The extent to which the O3I would add predictive power in other cohorts, especially those with younger participants, should be examined. Moreover, a particular focus on increased accuracy of risk assessment in the 7.5% to 10% “decision threshold” range will be important going forward.

The recent development of the PREVENT risk score looks to address some of these foundational issues broadening the scope to include ages ≥30 years and assessing CV, kidney, and metabolic health in the determination of the score.²⁵ Assessing how O3I further adds to this new PREVENT risk score is an exciting new avenue for future research.

Because the average American consumes suboptimal amounts of marine omega-3s, 90% of the US population has an O3I below 8%, above which is the ideal cardio-protective range.¹³ Thus, most individuals would need to increase intake of EPA and DHA to get cardioprotective O3I range of ≥8%.²⁶ In the current study, the upper quintile for O3I was ≥ 6.83%, suggesting that even residents of Framingham, MA, a coastal city, are typically not consuming enough EPA and DHA to reach the optimal O3I.¹²

The assessment of CV risk is a central step in primary prevention, and this study demonstrates that the O3I is a significant and independent marker of ASCVD risk that is on par with the 5 classic modifiable risk factors. When used in conjunction with them in the PCE it provides a significantly more accurate picture of ASCVD risk. More importantly, O3I represents a risk factor that can be easily modified non-pharmacologically.