Prognostic Utility of ApoB/AI, Total Cholesterol/HDL, Non-HDL Cholesterol, or hs-CRP as Predictors of Clinical Risk in Patients Receiving Statin Therapy After Acute Coronary Syndromes
Results From PROVE IT–TIMI 22
Objectives— The purpose of this study was to compare the prognostic utility of apoB/AI, total cholesterol/HDL (TC/HDL) ratio, non-HDL cholesterol (non–HDL-C), or hs-CRP as predictors of clinical risk among patients receiving statin therapy after acute coronary syndromes (ACS).
Methods and Results— Patients with ACS were randomized in the PROVE IT–TIMI 22 trial to either pravastatin 40 mg or atorvastatin 80 mg. Cox regression models adjusting for confounders were used to assess the relationship between on-treatment lipids or hs-CRP and risk of death or acute coronary events. At 4 months a 1 SD increment in apoB/AI (HR 1.10, 95% CI 1.01 to 1.20), TC/HDL (HR 1.12, 95% CI 1.01 to 1.24), and non–HDL-C (HR 1.20, 95% CI 1.07 to 1.35) predicted events to a similar extent as LDL-C (HR 1.20, 95% CI 1.07 to 1.35) with neither apoB/AI, TC/HDL, nor non–HDL-C improving risk prediction models which included LDL-C. In contrast, the addition of hs-CRP significantly improved risk prediction models irrespective of the lipid parameters included, with a 29% to 30% increased risk observed per 1 SD increment in log CRP.
Conclusion— In the present study of ACS patients receiving statin therapy, on-treatment apoB/AI, TC/HDL, and non–HDL-C offered similar prognostic information to LDL-C. However, the addition of hs-CRP to lipid-based measurements significantly improved risk prediction. On treatment CRP measurement may therefore offer additive prognostic information to lipids in ACS patients.
Low density lipoprotein cholesterol (LDL-C) is the traditional marker for identifying risk from hypercholesterolemia and monitoring the efficacy of statin therapy. However, LDL-C incompletely measures atherogenic lipoproteins such as very low density lipoprotein (VLDL) or intermediate density lipoprotein (IDL) and direct measurement of the concentration of apolipoprotein B (apoB), which is a measure of the number of proatherogenic particles and apolipoprotein AI (apoAI), which is a measure of the number of antiatherogenic particles, have been proposed by some as being superior measures of atherogenic dyslipidemia.1 Other approaches use standard lipid parameters such as non–high density lipoprotein cholesterol (non–HDL-C), which reflect the total cholesterol content of atherogenic lipoproteins or the total cholesterol/HDL ratio (TC/HDL).
The debate regarding the choice of the best lipid parameter has further intensified with apparently conflicting evidence between prospective studies.2–6 However, neither the measurement of atherogenic particle number nor atherogenic cholesterol content assesses inflammation which is an important marker of risk.2 In particular, considerable interest has now centered on the use of high-sensitivity C reactive protein (hs-CRP) to screen high-risk individuals for statin therapy.7 To date only one study (conducted in a statin-naïve all-female primary prevention cohort) has assessed the prognostic utility of hs-CRP in addition to apolipoproteins,2 and previous studies in statin-treated populations have only assessed the prognostic utility of hs-CRP to LDL-C.8 Hence a second question has arisen as to whether hs-CRP adds any prognostic utility to the measurement of apoB/AI ratio or other lipid measures among individuals receiving statin therapy. Therefore in the present study, we compared the relationship between on-treatment levels of newer and traditional markers of atherogenic dyslipidemia with and without hs-CRP with the subsequent risk of death or acute coronary events among patients receiving statin therapy after acute coronary syndromes (ACS).
The PROVE IT–TIMI 22 trial was a randomized trial comparing intensive statin therapy with atorvastatin 80 mg versus standard dose statin therapy with pravastatin 40 mg in 4162 patients with ACS and has been previously described.9 Briefly, patients were eligible for enrollment if they had been hospitalized for ACS within the previous 10 days and were clinically stable and were then followed for 24 months with complete follow-up on 99.8% of the cohort.9 The study was approved by the institutional review board of each hospital, and all patients provided written informed consent.
Blood Sampling and Analysis
Details of the blood collection protocol for lipid markers and hs-CRP have been previously reported.8,9 For full details please see the supplemental materials (available online at http://atvb.ahajournals.org).
Clinical End Points
The prespecified end point for this analysis focused on the composite of death or nonfatal acute coronary events, specifically nonfatal myocardial infarction (MI) or unstable angina (UA) requiring rehospitalization as previously described.10 All end points were adjudicated independently by a clinical events committee blinded to the study arm.9
ApoB/AI levels were available in 3683 patients at baseline, 2998 patients at 4 months, and in 1630 patients at the end of study. Complete case analysis was used in this analysis cohort for the purpose of comparing regressions models. The effect of intensive versus standard dose statin therapy on lipids were compared at 4 months and the end of study using a 2-sample t test where the data were normally distributed or a Wilcoxon test otherwise. The correlation between lipid parameters and lipid parameters and CRP (log transformed) were compared using the Pearson’s correlation coefficient.
Cox regression analysis was used to evaluate the association between lipid parameters at baseline and at 4 months (on-treatment) and the subsequent risk of death or nonfatal coronary events. Data are presented as hazard ratios with associated 95% confidence intervals (CI). All models described were constructed identically and were adjusted for age, gender, history of hypertension, history of diabetes, smoking status, body mass index (BMI), type of presenting acute coronary syndrome (ACS), and use of percutaneous intervention (PCI), and were based on models previously published by our group.10
Cox models make no assumption about the baseline hazard distribution. Hence, when assessing the additional contribution of any variable to risk prediction models we chose to use the pseudolog likelihood henceforth referred to as log likelihood of the residuals (LL RES) and the likelihood ratio test (LRT) as recently suggested, rather than receiver operator curves.11 To the basic model described above a larger model was constructed containing the additional parameter of interest, for instance expressed as a 1SD change for any lipid or as a 1SD change in log CRP. A likelihood ratio test (LRT) statistic (calculated as 2X difference in LLRES between models) >3.84 (corresponding to a Chi squared statistic of 3.84) was accepted as providing a nominally significant difference between models at the 5% level with one degree of freedom. To allow for comparisons between models, the dataset was restricted to 2652 subjects in whom information on all variables were available at month 4. Additional analyses are depicted graphically assessing the relationship between lipid level and probability of adverse event using regression models adjusted as previously described and stratified by hs-CRP level. Analyses were performed using SAS v9 (SAS Institute Inc) and S-Plus v6.2 (Insightful Corp); the statistical analyses were performed at the TIMI Data Coordinating Center and the Worldwide Clinical Trials. The authors had full access to and take full responsibility for the integrity of the data. All authors have read and agree to the manuscript as written.
The baseline characteristics of the analysis cohort are shown in supplemental Table I. The average age was 58 years, and the cohort comprised women (22%), subjects with diabetes (19%), subjects with hypertension (51%), and current smokers (36%). Baseline levels of apoB/A I, TC/ HDL and non–HDL-C are shown in Table 1.
Effect of Intensive Statin Therapy on Lipids
At 4 months after ACS, intensive statin therapy reduced apoB to a greater extent than moderate therapy (35% versus 11%), but raised apoAI to a lesser extent (3% versus 9%) (P<0.0001 for each) as shown in Table 1. Overall intensive therapy reduced the ratio apoB/AI by 37% from baseline versus 18% for standard therapy, P<0.0001, with similar differences between statin regimens observed at the end of study (average 2 years). Intensive therapy also reduced both total cholesterol/HDL-C ratio, and non–HDL-C to a greater extent than standard therapy (30% versus 11%, and 37% versus 8% respectively, P for each <0.0001; Table 1). hs-CRP was also lower among subjects allocated intensive therapy at 4 months (1.3 mg/L versus 2.1 mg/L, P<0.0001).
Correlation Between Lipids and hs-CRP
Strong correlations were observed between apoB and LDL-C lipids and between apoB and non–HDL-C, r=0.86 and 0.93 respectively and between non–HDL-C and LDL-C, r=0.91 (all P<0.0001). Similarly the correlation between apoAI and HDL-C and TC/HDL ratio and apoB/AI ratio were 0.81 and 0.88, respectively (all P<0.0001). In contrast, weak but significant correlations were observed between hs-CRP and apoB (r=0.21), non–HDL-C (r=0.21), LDL-C (r=0.18), TC/HDL (r=0.22), and apoB/AI (r=0.23), all P<0.0001. Even weaker correlations between hs-CRP and apoA I and HDL-C were observed r=−0.02 (P>0.05), and −0.05 (P<0.01), respectively.
Relationship Between Baseline Lipid Levels and Clinical Outcomes
After multivariable adjustment, which included the intensity of statin therapy, no significant association was observed between a 1 SD difference in baseline apoB/AI ratio (0.25), or total cholesterol/HDL-C ratio (1.42), or non-HDL cholesterol (0.88 mmol/L) and the subsequent risk of death or acute coronary events (HR 1.07, 95% CI 0.97 to 1.18, P=0.19); HR 1.06, 95% CI 0.97 to 1.17, P=0.21; HR 1.06, 95% CI 0.97 to 1.17, P=0.19 respectively). However, in each of these three models (containing baseline apoB/A1, total cholesterol/HDL-C ratio, or non-HDL cholesterol respectively) use of intensive statin therapy was similarly associated with a reduced risk (HR 0.76, 95% CI 0.63 to 0.92, P=0.005; HR 0.80, 95% CI 0.67 to 0.96, P=0.017; HR 0.80, 95% CI 0.67 to 0.96, P=0.017, respectively, for each lipid model).
Relationship Between 4-Month Lipid Levels and Clinical Outcomes
We initially assessed the relationship between “on treatment” levels of each individual lipid parameter or hs-CRP at 4 months and the subsequent risk of recurrent events using the median of each parameter to allow for comparison (Figure 1). The hazard ratio of death or acute coronary events appeared to be elevated to a similar extent with considerable overlap across lipids. Compared to the basic Cox model the addition of each atherogenic lipid parameter alone broadly tended to improve the models assessed using the likelihood ratio test (LRT statistic >3.84 considered significant for the model). In contrast, markers of nonatherogenic lipid content or particles alone (HDL-C or apoAI) did not significantly improve the multivariable models and tended to be weaker predictors of risk when used alone. The addition of hs-CRP to multivariable models either as a continuous variable or using the predefined cutpoint of 2 mg/L8,12 (Figure 1) improved the basic models to a greater extent than any of the lipid parameters alone, as reflected by a LRT statistic of 12.2 to 13.4 for hs-CRP versus values ranging from 0 to 8.4 for any given lipid parameter.
Incremental Predictive Value of Lipid Parameters to LDL-C
A 1-SD increment in LDL-C (0.82 mmol/L) was significantly associated with recurrent events (HR 1.20, 95% CI 1.07 to 1.35). By way of comparison, the effect estimates for a 1-SD change in apoB/AI ratio (HR 1.10, 95% CI 1.01 to 1.20), TC/HDL ratio (HR 1.12, 95% CI 1.01 to 1.24) or non–HDL-C (HR 1.20, 95% CI 1.07 to 1.35) overlapped considerably. To assess the potential utility of apoB/AI ratio, TC/HDL ratio, or non–HDL-C over and above measurements of LDL-C in predicting recurrent events, we added each additional lipid parameter to models which included LDL-C as well the baseline covariates (Model 1). The addition of either a 1-SD change in apoB/AI ratio (0.22), TC/HDL ratio (1.28), or non–HDL-C (1.0 mmol/L) to models which already included LDL-C produced LLRES values that were virtually identical to models containing LDL-C alone, and hence the LRT statistics used to discriminate improvement of one model over another were not significant (apoB/AI LRT statistic=0.2; TC/HDL LRT statistic=0.4; non–HDL-C LRT statistic 0.6; with an LRT statistic >3.84 indicating nominal statistical significance), providing strong evidence against a significant improvement in risk prediction by using other apoB/AI ratio, TC/HDL ratio, or non–HDL-C over and above LDL-C in this dataset (Table 2). By way of comparison the addition of 1 SD in log hs-CRP to models which included LDL-C significantly improved the model (LRT statistic 18.2). The inclusion of interaction terms lipid parameter*statin regimen or hs-CRP* statin regimen in models were not significant, suggesting no material differences in the relationships observed by statin regimen. Statin regimen was also not independently associated with outcomes when included in models containing lipids and did not improve the LRT statistic.
Incremental Predictive Value of hs-CRP to Lipid Parameters
To test whether hs-CRP offered any incremental risk prediction to each of the lipid parameters, a 1-SD change in log CRP was added to model 1 to create a new model (Model 2), which contained both LDL-C and an additional lipid parameter (1 SD increments of either apoB/AI, TC/HDL, or non–HDL-C). The addition of a 1-SD change in log CRP significantly improved model 2 over model 1, irrespective of the lipid fractions included within the models with LRT statistics ranging from 18 to 18.2 (Table 2). Per 1 SD increment in log CRP risk of recurrent events increased by 29% to 30%. When hs-CRP was added to model 1 as a dichotomous variable using the proposed cut point of 2 mg/L (Model 3),8 the statistical model also improved compared to model 1 irrespective of the lipid parameters included with LRT statistics ranging from 10 to 10.2 (Table 2). A higher hs-CRP (≥2 mg/L) appeared to identify individuals with higher event rates versus those with a lower hs-CRP (<2 mg/L) across the range of lipids (Figure 2). In particular, the difference in risk between subjects with an hs-CRP <2 mg/L versus subjects with hsCRP levels ≥2 mg/L appeared greater than the difference in risk between low and high lipid parameters among subjects with a CRP <2 mg/L.
We further explored the relevance of different hs-CRP cut points. Compared with subjects with a CRP <1 mg/L (median 0.5 mg/L) and lipid levels < median (reference group), the hazard ratio increased progressively both with increasing hs-CRP levels (1 to 2 mg/L, >2 mg/L) and with increasing lipid levels (≥ median) (supplemental Figure I). Importantly, irrespective of whether any of the lipid parameters were high (apoB/AI ≥0.60, TC/HDL ≥3.58, non–HDL-C ≥2.77 mmol/L, respectively) or low (apoB/AI <0.60, TC/HDL <3.58, non–HDL-C <2.77 mmol/L, respectively), the risk of recurrent events was lowest in those with low CRP levels <1 mg/L.
In this large trial of ACS patients, intensive statin therapy with atorvastatin 80 mg reduced the ratios of apoB/AI, TC/HDL ratio, and the level of non–HDL-C at 4 months to a greater extent than moderate dose statin therapy with pravastatin 40 mg. We observed that after adjustment for statins none of the baseline lipids predicted risk of future events. However, on-treatment levels of each lipid were associated with risk of recurrent events to a comparable degree and provided similar information on risk prediction to that provided by LDL-C. In contrast, we observed that hs-CRP significantly improved risk prediction models, providing independent prognostic information to each of the lipid parameters. Moreover, the apoB/AI ratio was found to be highly correlated with the TC/HDL ratio and apoB was highly correlated with both non–HDL-C and LDL-C, whereas hs-CRP was only weakly correlated with each lipid parameter, supporting the concept that hs-CRP provides information independent and complementary to that provided by lipid parameters.2,13
The present findings should be considered in the context of other statin trials. Analyses of the placebo controlled LIPID and TEXCAPS/AFCAPS trials have suggested that on treatment, lipids (rather than baseline lipids) and in particular apoB/AI are better predictors of the risk of recurrent events than LDL-C, although there is considerable overlap around the point estimates for the different lipid parameters.14,15 More recently an analysis of the TNT and IDEAL trials comparing intensive versus standard therapy in stable CHD observed similar findings to this study: specifically LDL-C and non–HDL-C were highly correlated with apoB as were apoB/AI with TC/HDL,6 and the effect estimate for different lipid parameter and risk of events showed significant overlap suggesting that they provided similar information on risk. In contrast to our data, Cox models used in their analysis which included both LDL-C and apoB or LDL-C and non–HDL-C, observed an attenuation in risk with respect to LDL-C, whereas the other lipid parameters remained significant. Use of statistical models which contain highly correlated variables often results in the attenuation of one of the correlated variables and do not test the superiority of one model over another if they focus on the risk estimate. It should therefore be noted that it was unclear whether the statistical models used for the analysis in TNT and IDEAL were statistically improved by the addition of the second lipid parameter over and above LDL-C (eg, by the LRT test).
Studies in statin naïve populations comparing the utlilty of apolipoproteins over other lipid fractions have also produced conflicting findings. For example the INTERHEART case–control study suggests that nonfasting apoB/A1 ratio as compared with LDL-C and TC/HDL provided the greatest odds ratio for myocardial infarction.16 In contrast to the former case controlled retrospective study, two prospective studies2,5 have both suggested that non–HDL-C and the ratio of TC/HDL or even LDL-C were comparable to apolipoprotein fractions in the prediction of incident CHD. Given the high degree of correlation between lipids and lipoproteins it is likely that even larger prospective studies than those currently published will be required to clarify these issues. One approach would be to pool, harmonize, and analyze existing data using an individual participant meta-analysis which would provide greater power than hitherto possible as the number of cases accrued even in individual large prospective studies are small.
In the present study, when hs-CRP was added as a continuous variable it significantly improved risk prediction models irrespective of the lipids included. Significantly, we observed that the improvement in risk prediction provided by the addition of hs-CRP to the basic model plus lipids was much greater than that provided by the addition of lipids to the basic model. Independent of lipids, per 1 SD increment in log CRP there was a 29% to 30% increased risk of death or recurrent coronary events consistent with data from statin naïve populations.2
As a 1-SD increment on a log transformed scale is not immediately intuitive in clinical practice, we explored whether the proposed hs-CRP cut point of 2 mg/L offered meaningful additional risk prediction to lipid parameters. We found that the dichotomous cut point of 2 mg/L for hs-CRP also significantly improved risk prediction models, suggesting that the proposed cut point of hs-CRP <2 mg/L offers significant clinical utility. As individuals in PROVE IT–TIMI 22 had median hs-CRP levels of 1.3 mg/L in the intensive arm and 2.1 mg/L in the moderate therapy arm at 4 months, we were able to explore the relevance of even lower hs-CRP levels than those proposed by the Centre for Disease Control, ie, <1 mg/L, 1 to 2 mg/L and >2 mg/L. In these exploratory analyses we observed that individuals with an hs-CRP <1 mg/L tended to have the lowest risk of recurrent events irrespective of whether lipids were high or low, with the lowest risk being observed among those with both low lipid and hs-CRP levels. Taken together these data provide further support for the notion that both lipids and a marker of inflammation such as hs-CRP should be monitored to provide more complete risk assessment in patients with coronary artery disease, with a view to perhaps addressing adverse lifestyle issues or uncontrolled risk factors that contribute to an elevation of hs-CRP.17 Although the present study does demonstrate that hs-CRP improves risk prediction, further work and larger studies, perhaps using individual participant data, are needed to guide the determination of relevant hs-CRP cut points for potential clinical use.
Inflammation is a complex process consisting of a number of different cell types, a variety of cytokines, chemokines, and chemoattractacts.17 hs-CRP is a downstream marker of inflammation and is regulated by the proinflammatory cytokine IL-6. In the absence of inflammatory conditions, it is postulated that hs-CRP reflects the net balance between pro- and antiinflammatory cytokine (and or cellular) activity in the vessel wall. We have previously reported that hs-CRP tracks with uncontrolled risk factors and that there are synergistic effects of conventional risk factors on inflammation.17 If these synergistic effects translate into increased risk then this may also explain why the addition of hs-CRP improves risk models for incident CHD which otherwise contain traditional risk factors in isolation.13 Thus hs-CRP may be acting as a global barometer of a number of dynamic processes within the body.17
In this analysis apolipoproteins were not available in all subjects at baseline and at 4 months. However, the characteristics of subjects with informative data were not statistically different to those subjects in whom lipid measurements were not available (data not shown). The analyses performed involve Cox regression models which can control for specified confounders but leave the possibility of residual confounding because of unmeasured variables. Finally, the present analysis assesses risk prediction and not causality, and as hs-CRP is a downstream marker of inflammation we make no inference within the present data about any causal relationship but rather we provide information on whether inflammation as assessed by hs-CRP improves risk prediction.
In the present analysis of ACS patients receiving statin therapy apoB/AI, TC/HDL, non–HDL-C, and LDL-C all provided similar information on risk prediction. However, the addition of hs-CRP to lipid-based measurements significantly improved risk prediction. On treatment CRP measurement may therefore offer additive prognostic information to lipids in ACS patients.
Sources of Funding
K.K.R. is funded by a British Heart Intermediate Fellowship. The PROVE IT Trial was funded by Bristol-Myers Squibb, Sankyo Co LTD. Funding sources did not make any contribution to the intellectual content of this manuscript.
Please see the supplemental materials.
Received October 7, 2008; revision accepted December 20, 2008.
Pischon T, Girman CJ, Sacks FM, Rifai N, Stampfer MJ, Rimm EB. Non-high-density lipoprotein cholesterol and apolipoprotein B in the prediction of coronary heart disease in men. Circulation. 2005; 112: 3375–3383.
Meisinger C, Loewel H, Mraz W, Koenig W. Prognostic value of apolipoprotein B and A-I in the prediction of myocardial infarction in middle-aged men and women: results from the MONICA/KORA Augsburg cohort study. Eur Heart J. 2005; 26: 271–278.
Kastelein JJ, van der Steeg WA, Holme I, Gaffney M, Cater NB, Barter P, Deedwania P, Olsson AG, Boekholdt SM, Demicco DA, Szarek M, LaRosa JC, Pedersen TR, Grundy SM. Lipids, apolipoproteins, and their ratios in relation to cardiovascular events with statin treatment. Circulation. 2008; 117: 3002–3009.
Ridker PM. Rosuvastatin in the primary prevention of cardiovascular disease among patients with low levels of low-density lipoprotein cholesterol and elevated high-sensitivity C-reactive protein: rationale and design of the JUPITER trial. Circulation. 2003; 108: 2292–2297.
Ridker PM, Cannon CP, Morrow D, Rifai N, Rose LM, McCabe CH, Pfeffer MA, Braunwald E, the Pravastatin or Atorvastatin Evaluation and Infection Therapy-Thrombolysis in Myocardial Infarction 22 (PROVE IT-TIMI 22) Investigators. C-reactive protein levels and outcomes after statin therapy. N Engl J Med. 2005; 352: 20–28.
Ray KK, Bach RG, Cannon CP, Cairns R, Kirtane AJ, Wiviott SD, McCabe CH, Braunwald E, Gibson CM. Benefits of achieving the NCEP optional LDL-C goal among elderly patients with ACS. Eur Heart J. 2006; 27: 2310–2316.
Cook NR. Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation. 2007; 115: 928–935.
Morrow DA, de Lemos JA, Sabatine MS, Wiviott SD, Blazing MA, Shui A, Rifai N, Califf RM, Braunwald E. Clinical relevance of C-reactive protein during follow-up of patients with acute coronary syndromes in the Aggrastat-to-Zocor Trial. Circulation. 2006; 114: 281–288.
Simes RJ, Marschner IC, Hunt D, Colquhoun D, Sullivan D, Stewart RA, Hague W, Keech A, Thompson P, White H, Shaw J, Tonkin A. Relationship between lipid levels and clinical outcomes in the Long-term Intervention with Pravastatin in Ischemic Disease (LIPID) Trial: to what extent is the reduction in coronary events with pravastatin explained by on-study lipid levels? Circulation. 2002; 105: 1162–1169.
Gotto AM Jr, Whitney E, Stein EA, Shapiro DR, Clearfield M, Weis S, Jou JY, Langendorfer A, Beere PA, Watson DJ, Downs JR, de Cani JS. Relation Between Baseline and On-Treatment Lipid Parameters and First Acute Major Coronary Events in the Air Force/Texas Coronary Atherosclerosis Prevention Study (AFCAPS/TexCAPS). Circulation. 2000; 101: 477–484.
McQueen MJ, Hawken S, Wang X, Ounpuu S, Sniderman A, Probstfield J, Steyn K, Sanderson JE, Hasani M, Volkova E, Kazmi K, Yusuf S. Lipids, lipoproteins, and apolipoproteins as risk markers of myocardial infarction in 52 countries (the INTERHEART study): a case-control study. Lancet. 2008; 372: 224–233.
Ray KK, Cannon CP, Cairns R, Morrow DA, Rifai N, Kirtane AJ, McCabe CH, Skene AM, Gibson CM, Ridker PM, Braunwald E. Relationship between uncontrolled risk factors and C-reactive protein levels in patients receiving standard or intensive statin therapy for acute coronary syndromes in the PROVE IT-TIMI 22 trial. J Am Coll Cardiol. 2005; 46: 1417–1424.