- Email: [email protected]
Low Serum Total Cholesterol Level is a Surrogate Marker, But Not a Risk Factor, for Poor Outcome in Patients Hospitalized With Acute Heart Failure: A Report From the Korean Heart Failure Registry
Journal of Cardiac Failure Vol. 18 No. 3 2012
Low Serum Total Cholesterol Level is a Surrogate Marker, But Not a Risk Factor, for Poor Outcome in Patients Hospitalized With Acute Heart Failure: A Report From the Korean Heart Failure Registry CHANG-HWAN YOON, MD,1 TAE-JIN YOUN, MD,1 SOYEON AHN, PhD,2 DONG-JU CHOI, MD,1 GOO-YOUNG CHO, MD,1 IN-HO CHAE, MD,1 JI CHOI, MS,3 HYUNGJUN CHO, PhD,3 SEONGWOO HAN, MD,4 MYEONG-CHAN CHO, MD,5 EUN-SEOK JEON, MD,6 SHUNG CHULL CHAE, MD,7 JAE-JOONG KIM, MD,8 KYU-HYUNG RYU, MD,9 AND BYUNG-HEE OH, MD,10 ON BEHALF OF THE KOREAN HEART FAILURE REGISTRY Seoul, Seongnam, Cheongju, and Daegu, South Korea
ABSTRACT Background: Hypercholesterolemia is a major risk factor for incident coronary artery disease and the prevalence of heart failure (HF). The causal relationship between low total cholesterol (TC) levels and poor clinical outcome in patients with acute HF has not been investigated. This study evaluated the effect of cholesterol levels on the long-term outcome in patients hospitalized due to acute HF. Methods and Results: We analyzed a cohort of 2,797 HF patients who were eligible for analysis in 3,200 patients of the Korean Heart Failure Registry. Patients were stratified into quartiles of TC (Q1 !133, Q2 133e158, Q3 159e190, and Q4 O190 mg/dL). Propensity score matching was performed with the patients in Q1 and Q4. Patients with lower serum TC had lower blood pressure, lower hemoglobin, lower serum sodium, and higher natriuretic peptide levels than patients with higher TC levels. Low TC was associated with increased risks for death and readmission due to HF; the adjusted hazard ratio (HR) of Q1 compared with Q4 was 1.57 (95% confidence interval [CI] 1.30e1.90). However, propensity score matching analysis revealed that low cholesterol itself did not affect outcome (HR 1.12, 95% CI 0.85e1.48). Conclusions: Low TC is strongly associated with mortality and morbidity in patients with HF. However, low TC seemed to be a secondary result of the patient’s state rather than an independent risk factor for poor outcome. (J Cardiac Fail 2012;18:194e201) Key Words: Heart failure, hypercholesterolemia, risk factors, prognosis.
Heart failure (HF) is a major public health concern as HF incidence, hospitalizations, and costs continue to rise around the world. Recently, cardiovascular diseases, particularly coronary artery disease, have been increasing in Korea in line with a rapid increase in life expectancy, the popularity of westernized diets, and an increased
prevalence of hypercholesterolemia. Along with these changes, the incidence and medical costs of HF have also been increasing in Korea. Hypercholesterolemia is a risk factor for not only coronary artery disease, but also HF.1 However, patients with advanced HF often have a low cholesterol level, which is
From the 1Cardiovascular Center, Seoul National University Bundang Hospital, Seoul, South Korea; 2Medical Research Collaborating Center, Seoul National University Bundang Hospital, Seongnam, South Korea; 3 Department of Statistics, Korea University, Seoul, South Korea; 4Korea University Hospital, Seoul, South Korea; 5Chungbuk University Hospital, Cheongju, South Korea; 6Seongkyunkwan University Seoul Samsung Hospital, Seoul, South Korea; 7Kyungpook National University Hospital, Daegu, South Korea; 8Ulsan University Hospital, Seoul, South Korea; 9 Hallym University Medical Center, Seoul, South Korea and 10Seoul National University Hospital, Seoul, South Korea. Manuscript received March 23, 2011; revised manuscript received December 14, 2011; revised manuscript accepted December 19, 2011.
Reprint requests: Dong-Ju Choi, MD, Director of Cardiovascular Center, Seoul National University Bundang Hospital, Professor of Internal Medicine, College of Medicine, Seoul National University, Gumiro 166, Bundang-gu, Seongnam-si, Gyeonggi-do, Republic of Korea. Tel: 82-31787-7007. E-mail: [email protected] Funding: Korean Society of Circulation (in celebration of its 50th anniversary). See page 201 for disclosure information. 1071-9164/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.cardfail.2011.12.006
194
Cholesterol Effect on Heart Failure
associated with a paradoxically poor prognosis.2 Some investigators speculated that this paradox originates from impaired immunity against endotoxin, because lipoproteins are natural nonspecific buffers of endotoxin.2,3 However, the causative effect of the low cholesterol level on poor outcome in patients with HF has not been assessed, because many confounding factors affecting cholesterol level and outcome make analysis difficult. To understand the epidemiology and clinical characteristics of HF in Korea, and to develop clinical guidelines for the treatment of HF, we established a large prospective multicenter registry of HF patients.4 In contrast to clinical trial data, registry data provide real-time and real-world information, including wide-ranging ages and comorbidities, and nearly equal proportions of men and women. In addition, registry data include a large number of patients, which enables us to perform various statistical analyses to overcome multiple confounding factors. In the present study, we investigated the causal relationship and association between total cholesterol (TC) level and clinical outcome of patients with HF using propensity score matching of the registry data. Methods Study Subjects The Korean Heart Failure (KorHF) Registry is a Korean prospective multicenter registry designed to reflect ‘‘real-world’’ clinical data in Korean patients presenting with acutely decompensated HF. The registry was founded in June 2005 and is supported by the Korean Society of Heart Failure. The web-based online HF registry (www.khfs.or.kr) was established at 28 high-volume university and community hospitals that have facilities for the multidisciplinary treatment of HF. Before the initiation of the KorHF study, several investigator meetings were held, and a practical steering committee from major enrolled hospitals was selected to standardize care given in clinical practice, as well as the study protocol, to minimize the differences in medical care among the different hospitals and across the different time periods. Data were collected at each site by a trained study coordinator using a standardized case report form. Standardized definitions of all patientrelated variables and clinical diagnoses were used. The study protocol was approved by the Ethics Committee at each participating institution. Patients gave written informed consents before study entry. Data were entered into the KorHF Registry database via a web-based electronic data capture system that included an electronic case report form. Data collection and audition were performed by the KorHF Registry Steering Committee at the Korean Society of Heart Failure. Study Population HF was diagnosed on admission according to the Framingham criteria.5 From June 2005 to April 2009, 3,200 patients were diagnosed with HF. For the present study, we retrospectively enrolled patients with HF who had TC measured at admission, a total of 2,797 patients. According to the KorHF study design, all subjects were followed with standard HF management as previously described6 at predetermined intervals.
Yoon et al
195
Follow-Up Information Research coordinators guided by documented definitions used standardized report forms to collect follow-up events until October 2009. Medical records were reviewed whenever possible when patients required repeated hospitalization. In addition to patient telephone interviews, referring physicians and institutions were contacted when necessary for additional information. We obtained information on patient survival, hospital readmission for HF, and heart transplantation. Data Collection Serum TC, high-density lipoprotein cholesterol (HDL-C), and triglyceride (TG) levels were collected at the time of admission. Low-density lipoprotein cholesterol (LDL-C) levels were calculated using the Friedewald formula.7 Low-density lipoprotein cholesterol was not calculated if serum TG was O400 mg/dL; instead, a direct measurement of LDL-C was made. Serum samples were collected after a 12-hour fast. All specimens were analyzed in laboratories that were accredited by the Korean Society for Laboratory Medicine. Hemodynamic variables used in the study were those recorded at initial presentation. Body mass index (BMI) was calculated using the patient’s height and weight after they were stabilized by medical therapy. To be classified into a nonischemic etiology, the patient had no angiographically significant coronary artery disease nor any evidence of previous myocardial infarction (MI), percutaneous coronary intervention, or coronary artery bypass graft surgery. Hypertension, diabetes mellitus (DM), and smoking histories were based on review of medical records. End Points The primary end point was composite of death and readmission due to HF during follow-up. Other end points were individually evaluated: death, readmission due to HF, and heart transplantation. Statistical Analysis Patients were divided into quartiles (Q) of TC. Values are expressed as mean 6 SD for continuous normally distributed variables; for data that are not normally distributed, values are expressed as median and interquartile range (IQR). Differences in baseline characteristics between quartiles were analyzed using the analysis of variance for continuous variables and Pearson chisquare test for categoric variables. Event-free survival analysis was performed using the Kaplan-Meier method, and differences between curves were evaluated using the log-rank test. Baseline variables and event-free survival were analyzed using a Cox proportional hazards model. To check whether the proportional hazard assumption is valid or not, we plotted the Kaplan-Meier curve for each variable. A graphical check of the proportional hazards model showed that the underlying assumption was not violated. We also applied the cox.zph function in R (version 2.13.0). This tested proportionality of all the predictors in the model by creating interactions with time. We thus tested the proportional hazards assumption for all models and confirmed that there existed strong evidence of proportional hazards for covariates except age, either by the individual test or by the global test. The multivariate Cox regression analysis evaluated the following variables, which are known risk factors for HF mortality and/or predictors of mortality on univariate analysis: age, sex, BMI, left ventricular ejection fraction (LVEF), New York Heart Association (NYHA) functional class, HF
196 Journal of Cardiac Failure Vol. 18 No. 3 March 2012 etiology, hypertension history, and DM history. We handled the cases with missing value with listwise deletion. The propensity scores were estimated using multiple logistic regression analysis. All prespecified covariates (age, sex, BMI, LVEF, NYHA functional class, HF etiology, heart rate, presence of atrial fibrillation, systolic (SBP) or diastolic (DBP) blood pressure, hypertension history, DM history, serum sodium, creatinine, blood urea nitrogen (BUN), and hemoglobin) were included in the final models for the TC quartiles: Q1 versus Q4. A propensity score, indicating the predicted probability of belonging to Q1 conditional on the observed covariates, was then calculated from the logistic regression equation for each patient. The predictive ability of each propensity score model was assessed by means of the C statistic. Using the Greedy 8/1 digit match algorithm, we created propensity scoreematched pairs without replacement (a 1:1 match). Specifically, we sought to match each patient in Q1 to one in Q4 who had a propensity score that was identical to 8 digits. If this could not be done, the algorithm then proceeded sequentially to the next highest digit match (a 7-, 6-, 5-, 4-, 3-, 2-, or 1-digit match) on propensity score to make ‘‘next-best’’ matches, in a hierarchic sequence until no more matches could be made. Once a match was made, previous matches were not reconsidered before making the next match. After the propensity score matches were made, we assessed the balance in baseline covariates between the 2 groups with the paired t test or Wilcoxon signed rank test for continuous variables and the McNemar test or marginal homogeneity test for categoric variables. Using these matched groups, we performed the matched Cox proportional hazards regression analysis. In addition, to match whole possible pairs in the entire samples, we performed optimal nonbipartite matching.8 The goal was to match pairs that share as many of the same covariates as possible while the response variable (quartiled TC) was kept apart from each others. First, we fitted the proportional odds regression on quartiled TC and calculated distance for each pairs. Using this distance, we performed the optimal nonbipartite matching. Statistical analyses were completed with SPSS for Windows version 17.0 (SPSS, Chicago, Illinois, USA), SAS version 9.0 (SAS Institute, Cary, North Carolina, USA) or R (version 2.13.0).
Results Baseline Characteristics
Baseline characteristics of the excluded patients without TC (n 5 403) were compared with the patients with TC (n 5 2,797). Most variables were similar except for creatinine and serum sodium, as shown in Table 1. The cohort of 2,797 patients was 49.4% male. The mean age of patients with TC was 67.6 6 14.2 years, with ages ranging from 15 to 98 years. Mean LVEF was 39.6 6 17.3%. Preserved systolic function, which was defined as LVEF O50%, was observed in 24.8% of the subjects. Etiologies of HF were ischemic in 44.3% and nonischemic 55.7%. Nonischemic etiology consisted of hypertension (36.7%), cardiomyopathy (26.5%), valvular heart disease (12.7%), myocarditis (0.7%), and infiltrative disease (0.4%). In this cohort, 31% of patients had DM, 47% had hypertension, and 21.9% had atrial fibrillation. Mean TC was 163.2 6 46.6 mg/dL, with TC ranging from 30 to 474 mg/dL, mean
Table 1. Baseline Characteristics Between the Included and the Excluded Patients Patients without Patients with TC (n 5 403) TC (n 5 2,797) P Value Age (y) Male (%) BMI (kg/m2) SBP (mm Hg) DBP (mm Hg) HR (beats/min) Ischemic etiology (%) Hypertension (%) DM (%) Atrial fibrillation (%) NYHA IV (%) LVEF (%) Preserved LVEF (%) Hemoglobin (g/dL) BUN (mg/dL) Creatinine (mg/dL) Serum sodium (mmol/L)
67.2 6 15.4 54.0 23.2 6 4.1 131.7 6 33.0 78.0 6 18.4 92.1 6 26.3 42.5 42.2 26.6 25.6 20.5 40.0 6 17.4 26.2 12.4 6 2.5 26.0 6 15.6 1.64 6 1.47 137.4 6 5.0
67.6 6 14.2 49.4 23.3 6 4.0 130.3 6 29.8 77.9 6 18.0 91.1 6 25.2 44.3 47.0 31 21.9 20.3 39.6 6 17.3 24.8 12.4 6 2.4 24.8 6 15.8 1.48 6 1.27 138.2 6 5.2
.611 .097 .736 .428 .878 .487 .517 .057 .059 .108 .754 .763 .589 .932 .158 .043 0.012
TC, total cholesterol; BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; HR, heart rate; DM, diabetes mellitus; NYHA, New York Heart Association functional class; LVEF, left ventricular ejection fraction; Preserved LVEF, LVEF O50%; BUN, blood urea nitrogen.
LDL-C was 100 mg/dL, mean HDL-C was 43.3 mg/dL, and median TG was 108.0 mg/dL. Total Cholesterol Levels and Baseline Characteristics
The patient characteristics among the groups according to TC quartiles are presented in Table 2. Groups were similar regarding age, heart rate, LVEF, and percentages of hypertension and NYHA functional class IV patients. However, the lowest quartile (Q1) showed a significantly higher percentage of men, nonischemic etiology, and atrial fibrillation, higher levels of BUN and creatinine, and lower levels of BMI, SBP, DBP, and hemoglobin than the other quartile groups. The percentage of DM and preserved LVEF were highest in Q1 and Q4. Outcome
Event-free survival was compared by Kaplan-Meier curve and log-rank test between the included patients (n 5 2,797) and the excluded (n 5 403) to evaluate whether there was selection bias by missing value affecting the outcome or not. The excluded patients without TC showed insignificantly higher readmission (P 5 .156 by log-rank test). The mortality was similar in both groups (P 5 .329 by log-rank test). Of the included 2,797 consecutively admitted patients with HF, 2,607 (93.4%) of them were discharged alive (in-hospital mortality 6.6%). During the follow-up period (median 510 days, IQR 158e961), 543 patients died and 755 patients were readmitted. The cumulative survival rates for all patients (and for patients with LVEF O50%, and patients with LVEF !50%) after discharge at 1, 2, and 3 years were 86% (87%, 86%), 80% (80%, 81%), and 76% (77%, 75%), respectively. The cumulative readmission
Cholesterol Effect on Heart Failure
Yoon et al
197
Table 2. Baseline Characteristics According to TC Quartiles (Q)
Age (y) Male (%) BMI (kg/m2) SBP (mm Hg) DBP (mm Hg) HR (beats/min) Ischemic etiology (%) Hypertension (%) DM (%) Atrial fibrillation (%) NYHA IV (%) LVEF (%) Preserved LVEF (%) Hemoglobin (g/dL) BUN (mg/dL) Creatinine (mg/dL) Serum sodium (mmol/L) Digoxin, n (%) ACE inhibitor, n (%) ARB, n (%) Beta-blocker, n (%) Aldactone, n (%)
Q1: TC !133 mg/dL (n 5 713)
Q2: TC 133e158 mg/dL (n 5 686)
Q3: TC 159e190 mg/dL (n 5 713)
Q4: TC O190 mg/dL (n 5 685)
P Value
67.2 6 14.3 54.3 22.9 6 4.0 122.9 6 24.6 73.8 6 16.0 88.9 6 24.1 41.2 46.5 34.5 26.3 20.5 36.6 6 14.7 27.5 11.7 6 2.4 28.2 6 19.3 1.60 6 1.30 137.3 6 5.4 123 (19.1) 421 (59.0) 241 (33.8) 351 (49.2) 309 (43.3)
68.1 6 13.8 50.1 23.0 6 4.1 126.1 6 28.4 75.6 6 17.1 89.9 6 25.9 44.6 45.9 27.8 22.3 20.3 36.3 6 13.3 20.9 12.3 6 2.2 23.9 6 15.4 1.54 6 1.41 137.8 6 5.4 100 (15.9) 443 (64.6) 275 (40.1) 373 (54.4) 297 (43.3)
68.3 6 13.7 49.2 23.4 6 4.0 132.2 6 31.2 78.3 6 18.9 91.6 6 24.4 48.1 48.3 29.8 20.9 21.9 35.5 6 12.8 20.5 12.7 6 2.2 23.4 6 13.6 1.42 6 1.12 138.5 6 4.8 101 (15.7) 490 (68.7) 307 (43.1) 410 (57.5) 293 (41.1)
66.6 6 14.0 43.2 23.8 6 3.8 135.7 6 31.5 80.3 6 18.8 91.1 6 23.3 48.8 47.8 34.9 13.1 18.6 35.0 6 12.8 24.2 12.8 6 2.3 23.5 6 13.7 1.51 6 1.39 139.1 6 4.9 64 (10.3) 478 (69.8) 281 (41.0) 430 (62.8) 282 (41.2)
.066 !.001 !.001 !.001 !.001 .123 .016 .764 .003 !.001 .754 .573 .010 !.001 !.001 .048 !.001 !.001 !.001 .003 !.001 .712
ACE, angiotensin-converting enzyme; ARB, angiotensin receptor blocker; other abbreviations as in Table 1. P values of chi-square or analysis of variance among the 4 groups.
rates after discharge at 1, 2, and 3 years were 78% (80%, 77%), 69% (71%, 69%), and 61% (63%, 61%), respectively. Event-free survival rates at 1, 2, and 3 years were 69% (71%, 68%), 59% (59%, 58%), and 49% (51%, 49%), respectively. Heart transplantation was performed in 27 patients. Low TC Predicted Adverse Outcome
Low TC was associated with an increased risk for combined events of death or readmission due to HF (P ! .001
by log-rank test) (Fig. 1A). Death was also higher in the low TC group (P ! .001), as was readmission due to HF (P ! .001; Fig. 1B and 1C) According to the Cox proportional hazards regression analysis, the hazard ratios (and 95% confidence interval [CI]) of the combined events, death, and readmission due to HF for Q1 compared with Q4 were 1.46 (1.23e1.72), 1.67 (1.30e2.13), and 1.47 (1.20e1.80), respectively. After adjusting for multiple known HF prognostic factors, as well as variables found to be significant predictors of outcome on univariate analysis (Table 3), TC
Fig. 1. Kaplan-Meier survival analysis. (A) Event-free survival curve of the patients per quartile for the primary outcome: death and readmission due to heart failure. Low total cholesterol was associated with an increased risk for combined events of death or readmission due to HF (P ! .001 by log-rank test). (B) Survival curve for death (P ! .001). (C) Survival curve for readmission due to heart failure (P ! .001).
198 Journal of Cardiac Failure Vol. 18 No. 3 March 2012 Table 3. Univariate Cox Proportional Hazards Analysis
Age (O65 y) Sex (male) Ischemic etiology Hypertension Diabetes mellitus NYHA III/IV Low LVEF (!45%) BMI
HR
95% CI
Significance
1.58 0.98 1.28 1.26 1.26 1.40 0.97 0.97
1.39e1.81 0.88e1.11 1.14e1.44 1.12e1.42 1.12e1.43 1.20e1.63 0.91e1.03 0.95e0.99
!.001 .795 !.001 !.001 !.001 !.001 .285 !.001
Abbreviations as in Table 1.
(1.14e1.87), respectively (Table 4). The hazard ratio of the combined events was not changed by various risk adjustment models, suggesting a strong association between TC and outcome. The relationship of TC level to HF outcome in pertinent subsets of patients with HF was also analyzed. When the cohort was divided into subgroups (ischemic or nonischemic etiology and LEF O45% or #45%), low TC was still associated with an increased risk of death or readmission due to HF (Table 5). Propensity Score Matching to Evaluate the Risk of Low TC Itself on the Outcome
remained an independent prognostic factor for the combined events. We performed a Cox regression adjusted by TC quartile, sex, HF etiology (ischemic vs nonischemic), NYHA functional class (I vs IIeIV), DM, hypertension, LVEF (O45%), and some continuous variables (age, BMI, TC). We tested a residual plot, and there were no notable tendencies among continuous predictors. We assessed whether missing patterns were associated with TC quartiles. We found no significant association between the missing pattern and the quartile (P 5 .95 by chi-square test). The hazard ratios (95% CI) of the combined events, death, and readmission due to HF for Q1 compared with Q4, after adjustment were 1.57 (1.30e1.90), 1.81 (1.33e2.47), and 1.46
Baseline characteristics of patients with a low TC level were significantly different from those with higher TC levels, as presented in Table 2, which possibly led to worse outcome in the lowest TC group. To assess the causal relationship between low TC level and clinical outcome in patients with HF, we performed propensity score matching within the cohort. The predictive ability of the propensity score model was assessed by means of the C statistic, which was 0.787, indicating good discrimination between the 2 quartiles. The balanced baseline covariates between the 2 groups, after the propensity score matches were made, are presented in Table 6. Box-plot graphs of propensity score showed
Table 4. Clinical Outcome and Cox Proportional Hazards Analysis Q1 (TC !133 mg/dL) n No. of patients with missing value Median follow-up duration, d (IQR) Cumulative heart transplantation at 6 mo/1 y/2 y Cumulative death or readmission at 6 mo/1 y/2 y (estimated proportion, %) Cumulative death at 6 mo/1 y/2 y (estimated proportion,%) Cumulative readmission at 6 mo/1 y/2 y (estimated proportion,%) For composite end point Unadjusted HR (95% CI) Risk-adjusted A HR Risk-adjusted B HR Risk-adjusted C HR Risk-adjusted D HR For death Unadjusted HR (95% CI) Risk-adjusted A HR Risk-adjusted B HR Risk-adjusted C HR Risk-adjusted D HR For readmission Unadjusted HR (95% CI) Risk-adjusted A HR Risk-adjusted B HR Risk-adjusted C HR Risk-adjusted D HR
Q2 Q3 (TC 133e158 mg/dL) (TC 159e190 mg/dL)
Q4 (TC O190 mg/dL)
713 167 452 (126e929) 8/10/12
686 158 530 (168e991) 6/8/10
713 158 545 (170e995) 2/2/3
685 157 497 (170e949) 1/1/1
189/245/288 (29/39/49)
157/195/251 (25/32/44)
149/190/235 (23/30/40)
126/175/213 (20/29/38)
110/140/160 (18/24/30)
87/100/124 (14/17/24)
86/99/118 (14/16/22)
67/85/97 (11/15/19)
122/161/192 (19/28/37)
107/139/176 (17/24/34)
93/128/161 (15/22/30)
73/110/141 (12/20/28)
P Value
1.46 1.43 1.46 1.57 1.33
(1.23e1.72) (1.21e1.69) (1.23e1.72) (1.30e1.90) (1.08e1.66)
1.18 1.14 1.16 1.20 1.10
(0.99e1.40) (0.96e1.36) (0.97e1.38) (0.99e1.47) (0.88e1.37)
1.06 1.04 1.04 1.10 1.09
(0.89e1.26) (0.87e1.24) (0.92e1.18) (0.90e1.34) (0.88e1.35)
1.00 1.00 1.00 1.00 1.00
!.001 !.001 !.001 !.001 .048
1.67 1.60 1.63 1.81 1.34
(1.30e2.13) (1.25e2.05) (1.27e2.09) (1.33e2.47) (0.96e1.87)
1.21 1.13 1.15 1.18 0.99
(0.93e1.58) (0.87e1.47) (0.88e1.49) (0.85e1.65) (0.70e1.39)
1.10 1.06 1.06 1.10 1.03
(0.85e1.44) (0.81e1.38) (0.81e1.38) (0.79e1.54) (0.73e1.44)
1.00 1.00 1.00 1.00 1.00
!.001 !.001 !.001 !.001 .125
1.47 1.45 1.48 1.46 1.39
(1.20e1.80) (1.18e1.78) (1.20e1.81) (1.14e1.87) (1.07e1.79)
1.23 1.21 1.23 1.29 1.25
(1.00e1.52) (0.98e1.50) (0.99e1.52) (1.00e1.65) (0.97e1.61)
1.07 1.06 1.06 1.10 1.11
(0.86e1.32) (0.85e1.31) (0.85e1.31) (0.85e1.42) (0.86e1.44)
1.00 1.00 1.00 1.00 1.00
.001 !.001 !.001 !.001 .077
IQR, interquartile range; HR, hazard ratio; other abbreviations as in Table 1. Risk-adjusted A: age and sex; risk-adjusted B: age, sex, and HF etiology; risk-adjusted C: age, sex, HF etiology, NYHA class, DM, hypertension, BMI, and LVEF; risk-adjusted D (for comparison to propensity score matching): age, sex, BMI, SBP, DBP, HR, ischemic etiology, hypertension, DM, atrial fibrillation, NYHA IV, LVEF, hemoglobin, BUN, creatinine, and serum sodium.
Cholesterol Effect on Heart Failure
Yoon et al
199
Table 5. Subgroup Analysis According to HF Etiology or LV Systolic Function
Unadjusted HR (95% CI) Risk-adjusted A HR Risk-adjusted B HR
Ischemic Etiology
Nonischemic Etiology
LVEF #45%
LVEF O45%
1.43 (1.08e1.90) 1.38 (1.04e1.84) 1.35 (1.01e1.81)
1.67 (1.25e2.25) 1.66 (1.24e2.23) 1.65 (1.23e2.23)
1.42 (1.11e1.82) 1.42 (1.11e1.82) 1.42 (1.11e1.82)
1.73 (1.20e2.48) 1.69 (1.18e2.43) 1.75 (1.21e2.52)
Abbreviations as in Tables 1 and 4. Hazard ratio of Q1 compared with Q4. Risk-adjusted A: age and sex; risk-adjusted B: age, sex, NYHA class, DM, hypertension, BMI, and either HF etiology or LVEF.
a significant portion of overlapped IQR after matching (Fig. 2). Standardized difference was also decreased after matching (data not shown). We found that the hazard ratio of low TC was no longer significant, as presented in Table 7. Additionally we performed optimal nonbipartite matching to create 1,079 pairs. We restricted cases of which the calculated distances were !2,000. The resulting pairs represented 398 cases. A matched Cox regression using those pairs again did not reveal any statistically significant association between the TC quartiles and the events. Discussion Low TC is associated with worse outcome in patients with advanced HF, according to data from a large multicenter HF registry in Korea. The association was still significant after adjustment for other previously reported risk factors that predict worse outcome in HF. However, patients with low TC showed different baseline characteristics from the patients with higher TC. Propensity score matching to assess the direct effect of TC on HF prognosis in comparable groups revealed that low TC did not increase the hazard ratio. In Korea, as cardiovascular disease has increased, the prevalence of HF has risen owing to increased coronary artery disease. A high cholesterol level is a risk factor for Table 6. Baseline Characteristics of Q1 and Q4 After Propensity Score Matching
coronary artery disease, but the effectiveness of controlling the cholesterol level in patients with HF to improve prognosis is debated. To improve cholesterol and HF management, we need to understand whether high cholesterol leads to poor outcome in patients with HF. Earlier studies report the opposite, a reverse epidemiology in patients with HF known as the ‘‘cholesterol paradox.’’9 Low cholesterol level was associated with worse longterm survival in several studies.2,10,11 In-hospital survival was also reduced in patients with HF and a low cholesterol level.12 The reverse epidemiology was not affected by separating the HF etiology (ischemic HF vs nonischemic HF)2 or HF pathophysiology (systolic HF vs diastolic HF).13 The cholesterol paradox in patients with established HF was also observed in a specific group of patients with DM, obesity, or hypertension.9,14,15 The association of low cholesterol level and poor outcome seems to be strong and consistent, substantiated by many earlier studies. The Korean population is a homogeneous ethnic group. Among this specific population, a low cholesterol level was associated with poor prognosis of patients with HF, regardless of the etiology or pathophysiology. However, a cause-effect relationship between low cholesterol level and poor outcome in patients with HF is difficult to establish. First, there are many possible confounding factors. For example, a low cholesterol level is often associated with a low albumin level, which reflects cachexia or low anabolic status of patients with chronic consuming diseases.
Q1 Q4 (TC !133 (TC O190 mg/dL; n 5 252) mg/dL; n 5 252) P Value Age (y) Male (%) BMI (kg/m2) SBP (mm Hg) DBP (mm Hg) HR (beats/min) Ischemic etiology (%) Hypertension (%) DM (%) Atrial fibrillation (%) NYHA IV (%) LVEF (%) Hemoglobin (g/dL) BUN (mg/dL) Creatinine (mg/dL) Serum sodium (mmol/L)
67.2 6 14.9 52.0 23.4 6 4.1 129.4 6 25.4 79.8 6 16.1 89.7 6 26.5 46.7 48.3 36.1 21.8 19.9 39.9 6 16.9 12.3 6 2.4 25.6 6 16.1 1.43 6 0.96 138.4 6 4.8
Abbreviations as in Table 1.
66.8 6 13.6 50.7 23.5 6 3.5 128.7 6 26.9 77.5 6 15.5 89.5 6 22.9 48.3 45.9 35.8 19.6 14.9 39.6 6 15.5 12.4 6 2.3 24.9 6 15.4 1.56 6 1.30 138.6 6 5.2
.284 .790 .967 .930 .908 .586 .375 .929 1.000 .800 .352 .986 .913 .868 .516 .611
Fig. 2. Box plots of the propensity scores of total cholesterol quartiles (Q) 1 and 4.
200 Journal of Cardiac Failure Vol. 18 No. 3 March 2012 Table 7. Clinical Outcome and Cox Proportional Hazards Analysis After Propensity Score Matching Q1 (TC !133 mg/dL; n 5 252) Cumulative heart transplantation at 6 mo/1 y/2 y Cumulative death or readmission at 6 mo/1 y/2 y (estimated proportion, %) Cumulative death at 6 mo/1 y/2 y (estimated proportion, %) Cumulative readmission at 6 mo/1 y/2 y (estimated proportion, %) Death or readmission Death Readmission
Q4 (TC O190 mg/dL; n 5 252)
3/4/4 53/71/91 (22/31/42)
1/1/1 48/67/86 (20/30/41)
19/29/45 (8/13/23) 39/52/63 (17/24/31)
19/28/36 (8/13/18) 33/46/61 (15/21/31)
1.12 (0.85e1.48) 1.26 (0.83e1.91) 1.13 (0.81e1.57)
1.00 1.00 1.00
P Value
0.422 0.274 0.488
Abbreviations as in Table 1.
The cachexic or wasting status is known to be an independent marker of poor prognosis.16,17 Second, a prospective randomized controlled experimental study design is required to demonstrate the cause and effect, but this design is not possible to achieve. To overcome these limitations, we introduced propensity score matching from the registry data to include a large number of patients with acute HF. The TC level was significantly associated with age, BMI, hypertension, DM, and the levels of hemoglobin, platelet, serum BUN, serum creatinine, and sodium, which are known to be surrogate markers indicating poor prognosis in patients with HF. Therefore, it is probable that low cholesterol is a consequence of poor cardiac function or comorbidity, rather than a causative factor leading to poor outcome in patients with HF. Earlier study designs could not clarify this point. To eliminate the effect of the confounding factors on HF outcome even after multivariate analysis, we designed the propensity score matching with factors affecting the TC level. For this score, we matched patients with a low cholesterol level (from Q1) to others with a high cholesterol level (from Q4). After matching, both groups showed similar baseline characteristics in contrast to the original population. Using these 2 groups, we found that low cholesterol level was no longer significant in predicting poor outcome in patients with HF. Therefore, the low cholesterol level could be a consequence of poor cardiac function or comorbidity rather than a leading cause of poor outcome. In line with our results, 2 randomized prospective studies (CORONA [Controlled Rosuvastatin Multinational Trial in Heart Failure] and GISSI-HF [Gruppo Italiano per lo Studio della Sopravvivenza nell’Infarto Miocardicoe Insufficienza Cardiaca]) that lowered TC with statin therapy18,19 showed no effect of the cholesterol-lowering treatment on the outcome, which also suggests that there may be no direct effect of low cholesterol level on the prognosis of patients with HF. A future study may properly answer one key question in reverse epidemiology research, namely, whether improved nutrition (and consistent increase in surrogate markers of nutrition) by nutritional supplementation may exert survival benefits in patients with advanced heart failure.
Study Limitations
The major limitation of the present study is the absence of data regarding the use of statin therapy, which was not included in the database. The use of a statin was thought to have a beneficial effect in patients with HF owing to the pleiotrophic effects. Accordingly, 2 earlier studies to determine outcome in HF patients based on statin use at baseline demonstrated modest benefit in mortality over a mean 2-year follow-up.20,21 However, 2 randomized studies reported that the use of a statin did not affect the prognosis of patients with acute HF.18,19 Therefore, we think that the lack of the data regarding statin use did not significantly affect the conclusion of this study. The loss of significance might be due to the loss of statistical power after matching. The hazard ratio, however, approached 1 after propensity score matching, and each 10 mg/dL increase in TC showed a hazard ratio of 0.97e1.03 (95% CI), which strongly suggests that there is no association between cholesterol level and outcome in patients with HF. By matching patients on the basis of their cholesterol levels (low and high quartiles) with similar disease severity and comorbidity burden, a highly selected group of patients was created, namely, patients at high risk without signs of malnutrition. Therefore, although propensity matched, the direct comparability is still compromised. In addition, results of more sophisticated matching (optimal nonbipartite matching) utilizing all 4 TC quartiles revealed that the risk of the composite outcome was generally lower at higher TC levels, with hazard ratios of 1.44, 1.48, 0.88, and 1.00 for TC Q1eQ4, respectively, although the hazard ratios comparing Q1 with Q2 and Q3 with Q4 were not statistically significant. It did not result in HRs near unity, which provides a possible explanation for the discrepancy between propensity score matching and multivariate Cox regression in the present study. In conclusion, the hazard of low TC on poor outcome of patients with HF was not significant in a propensity score matching analysis in the present study. Therefore, a low level of TC appears to be a surrogate marker of poor cardiac function or poor functional status of patients with HF.
Cholesterol Effect on Heart Failure
Acknowledgments The authors express their sincere gratitude to the secretarial staff and physicians at the 24 participating university and community medical centers for their support in this study: Catholic University Seoul St Mary’s Hospital, Chungnam National University Hospital, Chungbuk National University Hospital, Chonnam National University Hospital, Ewha Woman’s University Hospital, Eulji University Daejeon Hospital, Gacheon University Gil Hospital, Hallym University Sacred Heart Hosptial, Hanyang University Guri Hospital, Jeju National University Hospital, Konkuk University Medical Center, Keimyung University Hosptial, Korea University Guro Hospital, Kyungpook National University Hospital, Sungkyunkwan University Samsung Medical Center, Seoul National University Bundang Hospital, Seoul National University Hosptial, Ulsan University Asan Medical Center, Wonkwang University Hosptial, Yonsei University Wonju Christian Hospital, Yeungnam University Hospital, Yonsei University Severance Hospital, Dongguk University Ilsan Hosptial, Soonchunhyang University Cheonan Hospital, and Inje University Busan Paik Hospital.
7.
8. 9.
10.
11.
12.
13.
Disclosures 14.
None. 15.
References 1. Schocken DD, Benjamin EJ, Fonarow GC, Krumholz HM, Levy D, Mensah GA, et al. Prevention of heart failure: a scientific statement from the American Heart Association Councils on Epidemiology and Prevention, Clinical Cardiology, Cardiovascular Nursing, and High Blood Pressure Research; Quality of Care and Outcomes Research Interdisciplinary Working Group; and Functional Genomics and Translational Biology Interdisciplinary Working Group. Circulation 2008;117:2544e65. 2. Rauchhaus M, Clark AL, Doehner W, Davos C, Bolger A, Sharma R, et al. The relationship between cholesterol and survival in patients with chronic heart failure. J Am Coll Cardiol 2003;42:1933e40. 3. Rauchhaus M, Coats AJ, Anker SD. The endotoxin-lipoprotein hypothesis. Lancet 2000;356:930e3. 4. Choi DJ, Han S, Jeon ES, Cho MC, Kim JJ, Yoo BS, et al. Characteristics, outcomes and predictors of long-term mortality for patients hospitalized for acute heart failure: a report from the Korean Heart Failure Registry. Korean Circ J 2011;41:363e71. 5. Ho KK, Anderson KM, Kannel WB, Grossman W, Levy D. Survival after the onset of congestive heart failure in Framingham Heart Study subjects. Circulation 1993;88:107e15. 6. Hunt SA, Abraham WT, Chin MH, Feldman AM, Francis GS, Ganiats TG, et al. ACC/AHA 2005 guideline update for the diagnosis and management of chronic heart failure in the adult: a report of the
16.
17.
18.
19.
20.
21.
Yoon et al
201
American College of Cardiology/American Heart Association Task Force on Practice Guidelines (Writing Committee to Update the 2001 Guidelines for the Evaluation and Management of Heart Failure): developed in collaboration with the American College of Chest Physicians and the International Society for Heart and Lung Transplantation: endorsed by the Heart Rhythm Society. Circulation 2005;112:e154e235. Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin Chem 1972;18:499e502. Lu B, Greevy R, Xu X, Beck C. Optimal nonbipartite matching and its statistical applications. Am Stat 2011;65:21e30. Kalantar-Zadeh K, Block G, Horwich T, Fonarow GC. Reverse epidemiology of conventional cardiovascular risk factors in patients with chronic heart failure. J Am Coll Cardiol 2004;43:1439e44. Afsarmanesh N, Horwich TB, Fonarow GC. Total cholesterol levels and mortality risk in nonischemic systolic heart failure. Am Heart J 2006;152:1077e83. Sakatani T, Shirayama T, Suzaki Y, Yamamoto T, Mani H, Kawasaki T, et al. The association between cholesterol and mortality in heart failure. Comparison between patients with and without coronary artery disease. Int Heart J 2005;46:619e29. Horwich TB, Hernandez AF, Dai D, Yancy CW, Fonarow GC. Cholesterol levels and in-hospital mortality in patients with acute decompensated heart failure. Am Heart J 2008;156:1170e6. Guder G, Frantz S, Bauersachs J, Allolio B, Wanner C, Koller MT, et al. Reverse epidemiology in systolic and nonsystolic heart failure: cumulative prognostic benefit of classical cardiovascular risk factors. Circ Heart Fail 2009;2:563e71. Horwich TB, Fonarow GC, Hamilton MA, MacLellan WR, Woo MA, Tillisch JH. The relationship between obesity and mortality in patients with heart failure. J Am Coll Cardiol 2001;38:789e95. Dries DL, Sweitzer NK, Drazner MH, Stevenson LW, Gersh BJ. Prognostic impact of diabetes mellitus in patients with heart failure according to the etiology of left ventricular systolic dysfunction. J Am Coll Cardiol 2001;38:421e8. Anker SD, Negassa A, Coats AJ, Afzal R, Poole-Wilson PA, Cohn JN, et al. Prognostic importance of weight loss in chronic heart failure and the effect of treatment with angiotensin-converting-enzyme inhibitors: an observational study. Lancet 2003;361:1077e83. Jankowska EA, Biel B, Majda J, Szklarska A, Lopuszanska M, Medras M, et al. Anabolic deficiency in men with chronic heart failure: prevalence and detrimental impact on survival. Circulation 2006;114:1829e37. Kjekshus J, Apetrei E, Barrios V, Bohm M, Cleland JG, Cornel JH, et al. Rosuvastatin in older patients with systolic heart failure. N Engl J Med 2007;357:2248e61. Tavazzi L, Maggioni AP, Marchioli R, Barlera S, Franzosi MG, Latini R, et al. Effect of rosuvastatin in patients with chronic heart failure (the GISSI-HF trial): a randomised, double-blind, placebo-controlled trial. Lancet 2008;372:1231e9. Krum H, Latini R, Maggioni AP, Anand I, Masson S, Carretta E, et al. Statins and symptomatic chronic systolic heart failure: a post-hoc analysis of 5010 patients enrolled in Val-HeFT. Int J Cardiol 2007; 119:48e53. Huan Loh P, Windram JD, Tin L, Reddy P, Velavan P, Rigby AS, et al. The effects of initiation or continuation of statin therapy on cholesterol level and all-cause mortality after the diagnosis of left ventricular systolic dysfunction. Am Heart J 2007;153:537e44.
Low Serum Total Cholesterol Level is a Surrogate Marker, But Not a Risk Factor, for Poor Outcome in Patients Hospitalized With Acute Heart Failure: A Report From the Korean Heart Failure Registry CHANG-HWAN YOON, MD,1 TAE-JIN YOUN, MD,1 SOYEON AHN, PhD,2 DONG-JU CHOI, MD,1 GOO-YOUNG CHO, MD,1 IN-HO CHAE, MD,1 JI CHOI, MS,3 HYUNGJUN CHO, PhD,3 SEONGWOO HAN, MD,4 MYEONG-CHAN CHO, MD,5 EUN-SEOK JEON, MD,6 SHUNG CHULL CHAE, MD,7 JAE-JOONG KIM, MD,8 KYU-HYUNG RYU, MD,9 AND BYUNG-HEE OH, MD,10 ON BEHALF OF THE KOREAN HEART FAILURE REGISTRY Seoul, Seongnam, Cheongju, and Daegu, South Korea
ABSTRACT Background: Hypercholesterolemia is a major risk factor for incident coronary artery disease and the prevalence of heart failure (HF). The causal relationship between low total cholesterol (TC) levels and poor clinical outcome in patients with acute HF has not been investigated. This study evaluated the effect of cholesterol levels on the long-term outcome in patients hospitalized due to acute HF. Methods and Results: We analyzed a cohort of 2,797 HF patients who were eligible for analysis in 3,200 patients of the Korean Heart Failure Registry. Patients were stratified into quartiles of TC (Q1 !133, Q2 133e158, Q3 159e190, and Q4 O190 mg/dL). Propensity score matching was performed with the patients in Q1 and Q4. Patients with lower serum TC had lower blood pressure, lower hemoglobin, lower serum sodium, and higher natriuretic peptide levels than patients with higher TC levels. Low TC was associated with increased risks for death and readmission due to HF; the adjusted hazard ratio (HR) of Q1 compared with Q4 was 1.57 (95% confidence interval [CI] 1.30e1.90). However, propensity score matching analysis revealed that low cholesterol itself did not affect outcome (HR 1.12, 95% CI 0.85e1.48). Conclusions: Low TC is strongly associated with mortality and morbidity in patients with HF. However, low TC seemed to be a secondary result of the patient’s state rather than an independent risk factor for poor outcome. (J Cardiac Fail 2012;18:194e201) Key Words: Heart failure, hypercholesterolemia, risk factors, prognosis.
Heart failure (HF) is a major public health concern as HF incidence, hospitalizations, and costs continue to rise around the world. Recently, cardiovascular diseases, particularly coronary artery disease, have been increasing in Korea in line with a rapid increase in life expectancy, the popularity of westernized diets, and an increased
prevalence of hypercholesterolemia. Along with these changes, the incidence and medical costs of HF have also been increasing in Korea. Hypercholesterolemia is a risk factor for not only coronary artery disease, but also HF.1 However, patients with advanced HF often have a low cholesterol level, which is
From the 1Cardiovascular Center, Seoul National University Bundang Hospital, Seoul, South Korea; 2Medical Research Collaborating Center, Seoul National University Bundang Hospital, Seongnam, South Korea; 3 Department of Statistics, Korea University, Seoul, South Korea; 4Korea University Hospital, Seoul, South Korea; 5Chungbuk University Hospital, Cheongju, South Korea; 6Seongkyunkwan University Seoul Samsung Hospital, Seoul, South Korea; 7Kyungpook National University Hospital, Daegu, South Korea; 8Ulsan University Hospital, Seoul, South Korea; 9 Hallym University Medical Center, Seoul, South Korea and 10Seoul National University Hospital, Seoul, South Korea. Manuscript received March 23, 2011; revised manuscript received December 14, 2011; revised manuscript accepted December 19, 2011.
Reprint requests: Dong-Ju Choi, MD, Director of Cardiovascular Center, Seoul National University Bundang Hospital, Professor of Internal Medicine, College of Medicine, Seoul National University, Gumiro 166, Bundang-gu, Seongnam-si, Gyeonggi-do, Republic of Korea. Tel: 82-31787-7007. E-mail: [email protected] Funding: Korean Society of Circulation (in celebration of its 50th anniversary). See page 201 for disclosure information. 1071-9164/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.cardfail.2011.12.006
194
Cholesterol Effect on Heart Failure
associated with a paradoxically poor prognosis.2 Some investigators speculated that this paradox originates from impaired immunity against endotoxin, because lipoproteins are natural nonspecific buffers of endotoxin.2,3 However, the causative effect of the low cholesterol level on poor outcome in patients with HF has not been assessed, because many confounding factors affecting cholesterol level and outcome make analysis difficult. To understand the epidemiology and clinical characteristics of HF in Korea, and to develop clinical guidelines for the treatment of HF, we established a large prospective multicenter registry of HF patients.4 In contrast to clinical trial data, registry data provide real-time and real-world information, including wide-ranging ages and comorbidities, and nearly equal proportions of men and women. In addition, registry data include a large number of patients, which enables us to perform various statistical analyses to overcome multiple confounding factors. In the present study, we investigated the causal relationship and association between total cholesterol (TC) level and clinical outcome of patients with HF using propensity score matching of the registry data. Methods Study Subjects The Korean Heart Failure (KorHF) Registry is a Korean prospective multicenter registry designed to reflect ‘‘real-world’’ clinical data in Korean patients presenting with acutely decompensated HF. The registry was founded in June 2005 and is supported by the Korean Society of Heart Failure. The web-based online HF registry (www.khfs.or.kr) was established at 28 high-volume university and community hospitals that have facilities for the multidisciplinary treatment of HF. Before the initiation of the KorHF study, several investigator meetings were held, and a practical steering committee from major enrolled hospitals was selected to standardize care given in clinical practice, as well as the study protocol, to minimize the differences in medical care among the different hospitals and across the different time periods. Data were collected at each site by a trained study coordinator using a standardized case report form. Standardized definitions of all patientrelated variables and clinical diagnoses were used. The study protocol was approved by the Ethics Committee at each participating institution. Patients gave written informed consents before study entry. Data were entered into the KorHF Registry database via a web-based electronic data capture system that included an electronic case report form. Data collection and audition were performed by the KorHF Registry Steering Committee at the Korean Society of Heart Failure. Study Population HF was diagnosed on admission according to the Framingham criteria.5 From June 2005 to April 2009, 3,200 patients were diagnosed with HF. For the present study, we retrospectively enrolled patients with HF who had TC measured at admission, a total of 2,797 patients. According to the KorHF study design, all subjects were followed with standard HF management as previously described6 at predetermined intervals.
Yoon et al
195
Follow-Up Information Research coordinators guided by documented definitions used standardized report forms to collect follow-up events until October 2009. Medical records were reviewed whenever possible when patients required repeated hospitalization. In addition to patient telephone interviews, referring physicians and institutions were contacted when necessary for additional information. We obtained information on patient survival, hospital readmission for HF, and heart transplantation. Data Collection Serum TC, high-density lipoprotein cholesterol (HDL-C), and triglyceride (TG) levels were collected at the time of admission. Low-density lipoprotein cholesterol (LDL-C) levels were calculated using the Friedewald formula.7 Low-density lipoprotein cholesterol was not calculated if serum TG was O400 mg/dL; instead, a direct measurement of LDL-C was made. Serum samples were collected after a 12-hour fast. All specimens were analyzed in laboratories that were accredited by the Korean Society for Laboratory Medicine. Hemodynamic variables used in the study were those recorded at initial presentation. Body mass index (BMI) was calculated using the patient’s height and weight after they were stabilized by medical therapy. To be classified into a nonischemic etiology, the patient had no angiographically significant coronary artery disease nor any evidence of previous myocardial infarction (MI), percutaneous coronary intervention, or coronary artery bypass graft surgery. Hypertension, diabetes mellitus (DM), and smoking histories were based on review of medical records. End Points The primary end point was composite of death and readmission due to HF during follow-up. Other end points were individually evaluated: death, readmission due to HF, and heart transplantation. Statistical Analysis Patients were divided into quartiles (Q) of TC. Values are expressed as mean 6 SD for continuous normally distributed variables; for data that are not normally distributed, values are expressed as median and interquartile range (IQR). Differences in baseline characteristics between quartiles were analyzed using the analysis of variance for continuous variables and Pearson chisquare test for categoric variables. Event-free survival analysis was performed using the Kaplan-Meier method, and differences between curves were evaluated using the log-rank test. Baseline variables and event-free survival were analyzed using a Cox proportional hazards model. To check whether the proportional hazard assumption is valid or not, we plotted the Kaplan-Meier curve for each variable. A graphical check of the proportional hazards model showed that the underlying assumption was not violated. We also applied the cox.zph function in R (version 2.13.0). This tested proportionality of all the predictors in the model by creating interactions with time. We thus tested the proportional hazards assumption for all models and confirmed that there existed strong evidence of proportional hazards for covariates except age, either by the individual test or by the global test. The multivariate Cox regression analysis evaluated the following variables, which are known risk factors for HF mortality and/or predictors of mortality on univariate analysis: age, sex, BMI, left ventricular ejection fraction (LVEF), New York Heart Association (NYHA) functional class, HF
196 Journal of Cardiac Failure Vol. 18 No. 3 March 2012 etiology, hypertension history, and DM history. We handled the cases with missing value with listwise deletion. The propensity scores were estimated using multiple logistic regression analysis. All prespecified covariates (age, sex, BMI, LVEF, NYHA functional class, HF etiology, heart rate, presence of atrial fibrillation, systolic (SBP) or diastolic (DBP) blood pressure, hypertension history, DM history, serum sodium, creatinine, blood urea nitrogen (BUN), and hemoglobin) were included in the final models for the TC quartiles: Q1 versus Q4. A propensity score, indicating the predicted probability of belonging to Q1 conditional on the observed covariates, was then calculated from the logistic regression equation for each patient. The predictive ability of each propensity score model was assessed by means of the C statistic. Using the Greedy 8/1 digit match algorithm, we created propensity scoreematched pairs without replacement (a 1:1 match). Specifically, we sought to match each patient in Q1 to one in Q4 who had a propensity score that was identical to 8 digits. If this could not be done, the algorithm then proceeded sequentially to the next highest digit match (a 7-, 6-, 5-, 4-, 3-, 2-, or 1-digit match) on propensity score to make ‘‘next-best’’ matches, in a hierarchic sequence until no more matches could be made. Once a match was made, previous matches were not reconsidered before making the next match. After the propensity score matches were made, we assessed the balance in baseline covariates between the 2 groups with the paired t test or Wilcoxon signed rank test for continuous variables and the McNemar test or marginal homogeneity test for categoric variables. Using these matched groups, we performed the matched Cox proportional hazards regression analysis. In addition, to match whole possible pairs in the entire samples, we performed optimal nonbipartite matching.8 The goal was to match pairs that share as many of the same covariates as possible while the response variable (quartiled TC) was kept apart from each others. First, we fitted the proportional odds regression on quartiled TC and calculated distance for each pairs. Using this distance, we performed the optimal nonbipartite matching. Statistical analyses were completed with SPSS for Windows version 17.0 (SPSS, Chicago, Illinois, USA), SAS version 9.0 (SAS Institute, Cary, North Carolina, USA) or R (version 2.13.0).
Results Baseline Characteristics
Baseline characteristics of the excluded patients without TC (n 5 403) were compared with the patients with TC (n 5 2,797). Most variables were similar except for creatinine and serum sodium, as shown in Table 1. The cohort of 2,797 patients was 49.4% male. The mean age of patients with TC was 67.6 6 14.2 years, with ages ranging from 15 to 98 years. Mean LVEF was 39.6 6 17.3%. Preserved systolic function, which was defined as LVEF O50%, was observed in 24.8% of the subjects. Etiologies of HF were ischemic in 44.3% and nonischemic 55.7%. Nonischemic etiology consisted of hypertension (36.7%), cardiomyopathy (26.5%), valvular heart disease (12.7%), myocarditis (0.7%), and infiltrative disease (0.4%). In this cohort, 31% of patients had DM, 47% had hypertension, and 21.9% had atrial fibrillation. Mean TC was 163.2 6 46.6 mg/dL, with TC ranging from 30 to 474 mg/dL, mean
Table 1. Baseline Characteristics Between the Included and the Excluded Patients Patients without Patients with TC (n 5 403) TC (n 5 2,797) P Value Age (y) Male (%) BMI (kg/m2) SBP (mm Hg) DBP (mm Hg) HR (beats/min) Ischemic etiology (%) Hypertension (%) DM (%) Atrial fibrillation (%) NYHA IV (%) LVEF (%) Preserved LVEF (%) Hemoglobin (g/dL) BUN (mg/dL) Creatinine (mg/dL) Serum sodium (mmol/L)
67.2 6 15.4 54.0 23.2 6 4.1 131.7 6 33.0 78.0 6 18.4 92.1 6 26.3 42.5 42.2 26.6 25.6 20.5 40.0 6 17.4 26.2 12.4 6 2.5 26.0 6 15.6 1.64 6 1.47 137.4 6 5.0
67.6 6 14.2 49.4 23.3 6 4.0 130.3 6 29.8 77.9 6 18.0 91.1 6 25.2 44.3 47.0 31 21.9 20.3 39.6 6 17.3 24.8 12.4 6 2.4 24.8 6 15.8 1.48 6 1.27 138.2 6 5.2
.611 .097 .736 .428 .878 .487 .517 .057 .059 .108 .754 .763 .589 .932 .158 .043 0.012
TC, total cholesterol; BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; HR, heart rate; DM, diabetes mellitus; NYHA, New York Heart Association functional class; LVEF, left ventricular ejection fraction; Preserved LVEF, LVEF O50%; BUN, blood urea nitrogen.
LDL-C was 100 mg/dL, mean HDL-C was 43.3 mg/dL, and median TG was 108.0 mg/dL. Total Cholesterol Levels and Baseline Characteristics
The patient characteristics among the groups according to TC quartiles are presented in Table 2. Groups were similar regarding age, heart rate, LVEF, and percentages of hypertension and NYHA functional class IV patients. However, the lowest quartile (Q1) showed a significantly higher percentage of men, nonischemic etiology, and atrial fibrillation, higher levels of BUN and creatinine, and lower levels of BMI, SBP, DBP, and hemoglobin than the other quartile groups. The percentage of DM and preserved LVEF were highest in Q1 and Q4. Outcome
Event-free survival was compared by Kaplan-Meier curve and log-rank test between the included patients (n 5 2,797) and the excluded (n 5 403) to evaluate whether there was selection bias by missing value affecting the outcome or not. The excluded patients without TC showed insignificantly higher readmission (P 5 .156 by log-rank test). The mortality was similar in both groups (P 5 .329 by log-rank test). Of the included 2,797 consecutively admitted patients with HF, 2,607 (93.4%) of them were discharged alive (in-hospital mortality 6.6%). During the follow-up period (median 510 days, IQR 158e961), 543 patients died and 755 patients were readmitted. The cumulative survival rates for all patients (and for patients with LVEF O50%, and patients with LVEF !50%) after discharge at 1, 2, and 3 years were 86% (87%, 86%), 80% (80%, 81%), and 76% (77%, 75%), respectively. The cumulative readmission
Cholesterol Effect on Heart Failure
Yoon et al
197
Table 2. Baseline Characteristics According to TC Quartiles (Q)
Age (y) Male (%) BMI (kg/m2) SBP (mm Hg) DBP (mm Hg) HR (beats/min) Ischemic etiology (%) Hypertension (%) DM (%) Atrial fibrillation (%) NYHA IV (%) LVEF (%) Preserved LVEF (%) Hemoglobin (g/dL) BUN (mg/dL) Creatinine (mg/dL) Serum sodium (mmol/L) Digoxin, n (%) ACE inhibitor, n (%) ARB, n (%) Beta-blocker, n (%) Aldactone, n (%)
Q1: TC !133 mg/dL (n 5 713)
Q2: TC 133e158 mg/dL (n 5 686)
Q3: TC 159e190 mg/dL (n 5 713)
Q4: TC O190 mg/dL (n 5 685)
P Value
67.2 6 14.3 54.3 22.9 6 4.0 122.9 6 24.6 73.8 6 16.0 88.9 6 24.1 41.2 46.5 34.5 26.3 20.5 36.6 6 14.7 27.5 11.7 6 2.4 28.2 6 19.3 1.60 6 1.30 137.3 6 5.4 123 (19.1) 421 (59.0) 241 (33.8) 351 (49.2) 309 (43.3)
68.1 6 13.8 50.1 23.0 6 4.1 126.1 6 28.4 75.6 6 17.1 89.9 6 25.9 44.6 45.9 27.8 22.3 20.3 36.3 6 13.3 20.9 12.3 6 2.2 23.9 6 15.4 1.54 6 1.41 137.8 6 5.4 100 (15.9) 443 (64.6) 275 (40.1) 373 (54.4) 297 (43.3)
68.3 6 13.7 49.2 23.4 6 4.0 132.2 6 31.2 78.3 6 18.9 91.6 6 24.4 48.1 48.3 29.8 20.9 21.9 35.5 6 12.8 20.5 12.7 6 2.2 23.4 6 13.6 1.42 6 1.12 138.5 6 4.8 101 (15.7) 490 (68.7) 307 (43.1) 410 (57.5) 293 (41.1)
66.6 6 14.0 43.2 23.8 6 3.8 135.7 6 31.5 80.3 6 18.8 91.1 6 23.3 48.8 47.8 34.9 13.1 18.6 35.0 6 12.8 24.2 12.8 6 2.3 23.5 6 13.7 1.51 6 1.39 139.1 6 4.9 64 (10.3) 478 (69.8) 281 (41.0) 430 (62.8) 282 (41.2)
.066 !.001 !.001 !.001 !.001 .123 .016 .764 .003 !.001 .754 .573 .010 !.001 !.001 .048 !.001 !.001 !.001 .003 !.001 .712
ACE, angiotensin-converting enzyme; ARB, angiotensin receptor blocker; other abbreviations as in Table 1. P values of chi-square or analysis of variance among the 4 groups.
rates after discharge at 1, 2, and 3 years were 78% (80%, 77%), 69% (71%, 69%), and 61% (63%, 61%), respectively. Event-free survival rates at 1, 2, and 3 years were 69% (71%, 68%), 59% (59%, 58%), and 49% (51%, 49%), respectively. Heart transplantation was performed in 27 patients. Low TC Predicted Adverse Outcome
Low TC was associated with an increased risk for combined events of death or readmission due to HF (P ! .001
by log-rank test) (Fig. 1A). Death was also higher in the low TC group (P ! .001), as was readmission due to HF (P ! .001; Fig. 1B and 1C) According to the Cox proportional hazards regression analysis, the hazard ratios (and 95% confidence interval [CI]) of the combined events, death, and readmission due to HF for Q1 compared with Q4 were 1.46 (1.23e1.72), 1.67 (1.30e2.13), and 1.47 (1.20e1.80), respectively. After adjusting for multiple known HF prognostic factors, as well as variables found to be significant predictors of outcome on univariate analysis (Table 3), TC
Fig. 1. Kaplan-Meier survival analysis. (A) Event-free survival curve of the patients per quartile for the primary outcome: death and readmission due to heart failure. Low total cholesterol was associated with an increased risk for combined events of death or readmission due to HF (P ! .001 by log-rank test). (B) Survival curve for death (P ! .001). (C) Survival curve for readmission due to heart failure (P ! .001).
198 Journal of Cardiac Failure Vol. 18 No. 3 March 2012 Table 3. Univariate Cox Proportional Hazards Analysis
Age (O65 y) Sex (male) Ischemic etiology Hypertension Diabetes mellitus NYHA III/IV Low LVEF (!45%) BMI
HR
95% CI
Significance
1.58 0.98 1.28 1.26 1.26 1.40 0.97 0.97
1.39e1.81 0.88e1.11 1.14e1.44 1.12e1.42 1.12e1.43 1.20e1.63 0.91e1.03 0.95e0.99
!.001 .795 !.001 !.001 !.001 !.001 .285 !.001
Abbreviations as in Table 1.
(1.14e1.87), respectively (Table 4). The hazard ratio of the combined events was not changed by various risk adjustment models, suggesting a strong association between TC and outcome. The relationship of TC level to HF outcome in pertinent subsets of patients with HF was also analyzed. When the cohort was divided into subgroups (ischemic or nonischemic etiology and LEF O45% or #45%), low TC was still associated with an increased risk of death or readmission due to HF (Table 5). Propensity Score Matching to Evaluate the Risk of Low TC Itself on the Outcome
remained an independent prognostic factor for the combined events. We performed a Cox regression adjusted by TC quartile, sex, HF etiology (ischemic vs nonischemic), NYHA functional class (I vs IIeIV), DM, hypertension, LVEF (O45%), and some continuous variables (age, BMI, TC). We tested a residual plot, and there were no notable tendencies among continuous predictors. We assessed whether missing patterns were associated with TC quartiles. We found no significant association between the missing pattern and the quartile (P 5 .95 by chi-square test). The hazard ratios (95% CI) of the combined events, death, and readmission due to HF for Q1 compared with Q4, after adjustment were 1.57 (1.30e1.90), 1.81 (1.33e2.47), and 1.46
Baseline characteristics of patients with a low TC level were significantly different from those with higher TC levels, as presented in Table 2, which possibly led to worse outcome in the lowest TC group. To assess the causal relationship between low TC level and clinical outcome in patients with HF, we performed propensity score matching within the cohort. The predictive ability of the propensity score model was assessed by means of the C statistic, which was 0.787, indicating good discrimination between the 2 quartiles. The balanced baseline covariates between the 2 groups, after the propensity score matches were made, are presented in Table 6. Box-plot graphs of propensity score showed
Table 4. Clinical Outcome and Cox Proportional Hazards Analysis Q1 (TC !133 mg/dL) n No. of patients with missing value Median follow-up duration, d (IQR) Cumulative heart transplantation at 6 mo/1 y/2 y Cumulative death or readmission at 6 mo/1 y/2 y (estimated proportion, %) Cumulative death at 6 mo/1 y/2 y (estimated proportion,%) Cumulative readmission at 6 mo/1 y/2 y (estimated proportion,%) For composite end point Unadjusted HR (95% CI) Risk-adjusted A HR Risk-adjusted B HR Risk-adjusted C HR Risk-adjusted D HR For death Unadjusted HR (95% CI) Risk-adjusted A HR Risk-adjusted B HR Risk-adjusted C HR Risk-adjusted D HR For readmission Unadjusted HR (95% CI) Risk-adjusted A HR Risk-adjusted B HR Risk-adjusted C HR Risk-adjusted D HR
Q2 Q3 (TC 133e158 mg/dL) (TC 159e190 mg/dL)
Q4 (TC O190 mg/dL)
713 167 452 (126e929) 8/10/12
686 158 530 (168e991) 6/8/10
713 158 545 (170e995) 2/2/3
685 157 497 (170e949) 1/1/1
189/245/288 (29/39/49)
157/195/251 (25/32/44)
149/190/235 (23/30/40)
126/175/213 (20/29/38)
110/140/160 (18/24/30)
87/100/124 (14/17/24)
86/99/118 (14/16/22)
67/85/97 (11/15/19)
122/161/192 (19/28/37)
107/139/176 (17/24/34)
93/128/161 (15/22/30)
73/110/141 (12/20/28)
P Value
1.46 1.43 1.46 1.57 1.33
(1.23e1.72) (1.21e1.69) (1.23e1.72) (1.30e1.90) (1.08e1.66)
1.18 1.14 1.16 1.20 1.10
(0.99e1.40) (0.96e1.36) (0.97e1.38) (0.99e1.47) (0.88e1.37)
1.06 1.04 1.04 1.10 1.09
(0.89e1.26) (0.87e1.24) (0.92e1.18) (0.90e1.34) (0.88e1.35)
1.00 1.00 1.00 1.00 1.00
!.001 !.001 !.001 !.001 .048
1.67 1.60 1.63 1.81 1.34
(1.30e2.13) (1.25e2.05) (1.27e2.09) (1.33e2.47) (0.96e1.87)
1.21 1.13 1.15 1.18 0.99
(0.93e1.58) (0.87e1.47) (0.88e1.49) (0.85e1.65) (0.70e1.39)
1.10 1.06 1.06 1.10 1.03
(0.85e1.44) (0.81e1.38) (0.81e1.38) (0.79e1.54) (0.73e1.44)
1.00 1.00 1.00 1.00 1.00
!.001 !.001 !.001 !.001 .125
1.47 1.45 1.48 1.46 1.39
(1.20e1.80) (1.18e1.78) (1.20e1.81) (1.14e1.87) (1.07e1.79)
1.23 1.21 1.23 1.29 1.25
(1.00e1.52) (0.98e1.50) (0.99e1.52) (1.00e1.65) (0.97e1.61)
1.07 1.06 1.06 1.10 1.11
(0.86e1.32) (0.85e1.31) (0.85e1.31) (0.85e1.42) (0.86e1.44)
1.00 1.00 1.00 1.00 1.00
.001 !.001 !.001 !.001 .077
IQR, interquartile range; HR, hazard ratio; other abbreviations as in Table 1. Risk-adjusted A: age and sex; risk-adjusted B: age, sex, and HF etiology; risk-adjusted C: age, sex, HF etiology, NYHA class, DM, hypertension, BMI, and LVEF; risk-adjusted D (for comparison to propensity score matching): age, sex, BMI, SBP, DBP, HR, ischemic etiology, hypertension, DM, atrial fibrillation, NYHA IV, LVEF, hemoglobin, BUN, creatinine, and serum sodium.
Cholesterol Effect on Heart Failure
Yoon et al
199
Table 5. Subgroup Analysis According to HF Etiology or LV Systolic Function
Unadjusted HR (95% CI) Risk-adjusted A HR Risk-adjusted B HR
Ischemic Etiology
Nonischemic Etiology
LVEF #45%
LVEF O45%
1.43 (1.08e1.90) 1.38 (1.04e1.84) 1.35 (1.01e1.81)
1.67 (1.25e2.25) 1.66 (1.24e2.23) 1.65 (1.23e2.23)
1.42 (1.11e1.82) 1.42 (1.11e1.82) 1.42 (1.11e1.82)
1.73 (1.20e2.48) 1.69 (1.18e2.43) 1.75 (1.21e2.52)
Abbreviations as in Tables 1 and 4. Hazard ratio of Q1 compared with Q4. Risk-adjusted A: age and sex; risk-adjusted B: age, sex, NYHA class, DM, hypertension, BMI, and either HF etiology or LVEF.
a significant portion of overlapped IQR after matching (Fig. 2). Standardized difference was also decreased after matching (data not shown). We found that the hazard ratio of low TC was no longer significant, as presented in Table 7. Additionally we performed optimal nonbipartite matching to create 1,079 pairs. We restricted cases of which the calculated distances were !2,000. The resulting pairs represented 398 cases. A matched Cox regression using those pairs again did not reveal any statistically significant association between the TC quartiles and the events. Discussion Low TC is associated with worse outcome in patients with advanced HF, according to data from a large multicenter HF registry in Korea. The association was still significant after adjustment for other previously reported risk factors that predict worse outcome in HF. However, patients with low TC showed different baseline characteristics from the patients with higher TC. Propensity score matching to assess the direct effect of TC on HF prognosis in comparable groups revealed that low TC did not increase the hazard ratio. In Korea, as cardiovascular disease has increased, the prevalence of HF has risen owing to increased coronary artery disease. A high cholesterol level is a risk factor for Table 6. Baseline Characteristics of Q1 and Q4 After Propensity Score Matching
coronary artery disease, but the effectiveness of controlling the cholesterol level in patients with HF to improve prognosis is debated. To improve cholesterol and HF management, we need to understand whether high cholesterol leads to poor outcome in patients with HF. Earlier studies report the opposite, a reverse epidemiology in patients with HF known as the ‘‘cholesterol paradox.’’9 Low cholesterol level was associated with worse longterm survival in several studies.2,10,11 In-hospital survival was also reduced in patients with HF and a low cholesterol level.12 The reverse epidemiology was not affected by separating the HF etiology (ischemic HF vs nonischemic HF)2 or HF pathophysiology (systolic HF vs diastolic HF).13 The cholesterol paradox in patients with established HF was also observed in a specific group of patients with DM, obesity, or hypertension.9,14,15 The association of low cholesterol level and poor outcome seems to be strong and consistent, substantiated by many earlier studies. The Korean population is a homogeneous ethnic group. Among this specific population, a low cholesterol level was associated with poor prognosis of patients with HF, regardless of the etiology or pathophysiology. However, a cause-effect relationship between low cholesterol level and poor outcome in patients with HF is difficult to establish. First, there are many possible confounding factors. For example, a low cholesterol level is often associated with a low albumin level, which reflects cachexia or low anabolic status of patients with chronic consuming diseases.
Q1 Q4 (TC !133 (TC O190 mg/dL; n 5 252) mg/dL; n 5 252) P Value Age (y) Male (%) BMI (kg/m2) SBP (mm Hg) DBP (mm Hg) HR (beats/min) Ischemic etiology (%) Hypertension (%) DM (%) Atrial fibrillation (%) NYHA IV (%) LVEF (%) Hemoglobin (g/dL) BUN (mg/dL) Creatinine (mg/dL) Serum sodium (mmol/L)
67.2 6 14.9 52.0 23.4 6 4.1 129.4 6 25.4 79.8 6 16.1 89.7 6 26.5 46.7 48.3 36.1 21.8 19.9 39.9 6 16.9 12.3 6 2.4 25.6 6 16.1 1.43 6 0.96 138.4 6 4.8
Abbreviations as in Table 1.
66.8 6 13.6 50.7 23.5 6 3.5 128.7 6 26.9 77.5 6 15.5 89.5 6 22.9 48.3 45.9 35.8 19.6 14.9 39.6 6 15.5 12.4 6 2.3 24.9 6 15.4 1.56 6 1.30 138.6 6 5.2
.284 .790 .967 .930 .908 .586 .375 .929 1.000 .800 .352 .986 .913 .868 .516 .611
Fig. 2. Box plots of the propensity scores of total cholesterol quartiles (Q) 1 and 4.
200 Journal of Cardiac Failure Vol. 18 No. 3 March 2012 Table 7. Clinical Outcome and Cox Proportional Hazards Analysis After Propensity Score Matching Q1 (TC !133 mg/dL; n 5 252) Cumulative heart transplantation at 6 mo/1 y/2 y Cumulative death or readmission at 6 mo/1 y/2 y (estimated proportion, %) Cumulative death at 6 mo/1 y/2 y (estimated proportion, %) Cumulative readmission at 6 mo/1 y/2 y (estimated proportion, %) Death or readmission Death Readmission
Q4 (TC O190 mg/dL; n 5 252)
3/4/4 53/71/91 (22/31/42)
1/1/1 48/67/86 (20/30/41)
19/29/45 (8/13/23) 39/52/63 (17/24/31)
19/28/36 (8/13/18) 33/46/61 (15/21/31)
1.12 (0.85e1.48) 1.26 (0.83e1.91) 1.13 (0.81e1.57)
1.00 1.00 1.00
P Value
0.422 0.274 0.488
Abbreviations as in Table 1.
The cachexic or wasting status is known to be an independent marker of poor prognosis.16,17 Second, a prospective randomized controlled experimental study design is required to demonstrate the cause and effect, but this design is not possible to achieve. To overcome these limitations, we introduced propensity score matching from the registry data to include a large number of patients with acute HF. The TC level was significantly associated with age, BMI, hypertension, DM, and the levels of hemoglobin, platelet, serum BUN, serum creatinine, and sodium, which are known to be surrogate markers indicating poor prognosis in patients with HF. Therefore, it is probable that low cholesterol is a consequence of poor cardiac function or comorbidity, rather than a causative factor leading to poor outcome in patients with HF. Earlier study designs could not clarify this point. To eliminate the effect of the confounding factors on HF outcome even after multivariate analysis, we designed the propensity score matching with factors affecting the TC level. For this score, we matched patients with a low cholesterol level (from Q1) to others with a high cholesterol level (from Q4). After matching, both groups showed similar baseline characteristics in contrast to the original population. Using these 2 groups, we found that low cholesterol level was no longer significant in predicting poor outcome in patients with HF. Therefore, the low cholesterol level could be a consequence of poor cardiac function or comorbidity rather than a leading cause of poor outcome. In line with our results, 2 randomized prospective studies (CORONA [Controlled Rosuvastatin Multinational Trial in Heart Failure] and GISSI-HF [Gruppo Italiano per lo Studio della Sopravvivenza nell’Infarto Miocardicoe Insufficienza Cardiaca]) that lowered TC with statin therapy18,19 showed no effect of the cholesterol-lowering treatment on the outcome, which also suggests that there may be no direct effect of low cholesterol level on the prognosis of patients with HF. A future study may properly answer one key question in reverse epidemiology research, namely, whether improved nutrition (and consistent increase in surrogate markers of nutrition) by nutritional supplementation may exert survival benefits in patients with advanced heart failure.
Study Limitations
The major limitation of the present study is the absence of data regarding the use of statin therapy, which was not included in the database. The use of a statin was thought to have a beneficial effect in patients with HF owing to the pleiotrophic effects. Accordingly, 2 earlier studies to determine outcome in HF patients based on statin use at baseline demonstrated modest benefit in mortality over a mean 2-year follow-up.20,21 However, 2 randomized studies reported that the use of a statin did not affect the prognosis of patients with acute HF.18,19 Therefore, we think that the lack of the data regarding statin use did not significantly affect the conclusion of this study. The loss of significance might be due to the loss of statistical power after matching. The hazard ratio, however, approached 1 after propensity score matching, and each 10 mg/dL increase in TC showed a hazard ratio of 0.97e1.03 (95% CI), which strongly suggests that there is no association between cholesterol level and outcome in patients with HF. By matching patients on the basis of their cholesterol levels (low and high quartiles) with similar disease severity and comorbidity burden, a highly selected group of patients was created, namely, patients at high risk without signs of malnutrition. Therefore, although propensity matched, the direct comparability is still compromised. In addition, results of more sophisticated matching (optimal nonbipartite matching) utilizing all 4 TC quartiles revealed that the risk of the composite outcome was generally lower at higher TC levels, with hazard ratios of 1.44, 1.48, 0.88, and 1.00 for TC Q1eQ4, respectively, although the hazard ratios comparing Q1 with Q2 and Q3 with Q4 were not statistically significant. It did not result in HRs near unity, which provides a possible explanation for the discrepancy between propensity score matching and multivariate Cox regression in the present study. In conclusion, the hazard of low TC on poor outcome of patients with HF was not significant in a propensity score matching analysis in the present study. Therefore, a low level of TC appears to be a surrogate marker of poor cardiac function or poor functional status of patients with HF.
Cholesterol Effect on Heart Failure
Acknowledgments The authors express their sincere gratitude to the secretarial staff and physicians at the 24 participating university and community medical centers for their support in this study: Catholic University Seoul St Mary’s Hospital, Chungnam National University Hospital, Chungbuk National University Hospital, Chonnam National University Hospital, Ewha Woman’s University Hospital, Eulji University Daejeon Hospital, Gacheon University Gil Hospital, Hallym University Sacred Heart Hosptial, Hanyang University Guri Hospital, Jeju National University Hospital, Konkuk University Medical Center, Keimyung University Hosptial, Korea University Guro Hospital, Kyungpook National University Hospital, Sungkyunkwan University Samsung Medical Center, Seoul National University Bundang Hospital, Seoul National University Hosptial, Ulsan University Asan Medical Center, Wonkwang University Hosptial, Yonsei University Wonju Christian Hospital, Yeungnam University Hospital, Yonsei University Severance Hospital, Dongguk University Ilsan Hosptial, Soonchunhyang University Cheonan Hospital, and Inje University Busan Paik Hospital.
7.
8. 9.
10.
11.
12.
13.
Disclosures 14.
None. 15.
References 1. Schocken DD, Benjamin EJ, Fonarow GC, Krumholz HM, Levy D, Mensah GA, et al. Prevention of heart failure: a scientific statement from the American Heart Association Councils on Epidemiology and Prevention, Clinical Cardiology, Cardiovascular Nursing, and High Blood Pressure Research; Quality of Care and Outcomes Research Interdisciplinary Working Group; and Functional Genomics and Translational Biology Interdisciplinary Working Group. Circulation 2008;117:2544e65. 2. Rauchhaus M, Clark AL, Doehner W, Davos C, Bolger A, Sharma R, et al. The relationship between cholesterol and survival in patients with chronic heart failure. J Am Coll Cardiol 2003;42:1933e40. 3. Rauchhaus M, Coats AJ, Anker SD. The endotoxin-lipoprotein hypothesis. Lancet 2000;356:930e3. 4. Choi DJ, Han S, Jeon ES, Cho MC, Kim JJ, Yoo BS, et al. Characteristics, outcomes and predictors of long-term mortality for patients hospitalized for acute heart failure: a report from the Korean Heart Failure Registry. Korean Circ J 2011;41:363e71. 5. Ho KK, Anderson KM, Kannel WB, Grossman W, Levy D. Survival after the onset of congestive heart failure in Framingham Heart Study subjects. Circulation 1993;88:107e15. 6. Hunt SA, Abraham WT, Chin MH, Feldman AM, Francis GS, Ganiats TG, et al. ACC/AHA 2005 guideline update for the diagnosis and management of chronic heart failure in the adult: a report of the
16.
17.
18.
19.
20.
21.
Yoon et al
201
American College of Cardiology/American Heart Association Task Force on Practice Guidelines (Writing Committee to Update the 2001 Guidelines for the Evaluation and Management of Heart Failure): developed in collaboration with the American College of Chest Physicians and the International Society for Heart and Lung Transplantation: endorsed by the Heart Rhythm Society. Circulation 2005;112:e154e235. Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin Chem 1972;18:499e502. Lu B, Greevy R, Xu X, Beck C. Optimal nonbipartite matching and its statistical applications. Am Stat 2011;65:21e30. Kalantar-Zadeh K, Block G, Horwich T, Fonarow GC. Reverse epidemiology of conventional cardiovascular risk factors in patients with chronic heart failure. J Am Coll Cardiol 2004;43:1439e44. Afsarmanesh N, Horwich TB, Fonarow GC. Total cholesterol levels and mortality risk in nonischemic systolic heart failure. Am Heart J 2006;152:1077e83. Sakatani T, Shirayama T, Suzaki Y, Yamamoto T, Mani H, Kawasaki T, et al. The association between cholesterol and mortality in heart failure. Comparison between patients with and without coronary artery disease. Int Heart J 2005;46:619e29. Horwich TB, Hernandez AF, Dai D, Yancy CW, Fonarow GC. Cholesterol levels and in-hospital mortality in patients with acute decompensated heart failure. Am Heart J 2008;156:1170e6. Guder G, Frantz S, Bauersachs J, Allolio B, Wanner C, Koller MT, et al. Reverse epidemiology in systolic and nonsystolic heart failure: cumulative prognostic benefit of classical cardiovascular risk factors. Circ Heart Fail 2009;2:563e71. Horwich TB, Fonarow GC, Hamilton MA, MacLellan WR, Woo MA, Tillisch JH. The relationship between obesity and mortality in patients with heart failure. J Am Coll Cardiol 2001;38:789e95. Dries DL, Sweitzer NK, Drazner MH, Stevenson LW, Gersh BJ. Prognostic impact of diabetes mellitus in patients with heart failure according to the etiology of left ventricular systolic dysfunction. J Am Coll Cardiol 2001;38:421e8. Anker SD, Negassa A, Coats AJ, Afzal R, Poole-Wilson PA, Cohn JN, et al. Prognostic importance of weight loss in chronic heart failure and the effect of treatment with angiotensin-converting-enzyme inhibitors: an observational study. Lancet 2003;361:1077e83. Jankowska EA, Biel B, Majda J, Szklarska A, Lopuszanska M, Medras M, et al. Anabolic deficiency in men with chronic heart failure: prevalence and detrimental impact on survival. Circulation 2006;114:1829e37. Kjekshus J, Apetrei E, Barrios V, Bohm M, Cleland JG, Cornel JH, et al. Rosuvastatin in older patients with systolic heart failure. N Engl J Med 2007;357:2248e61. Tavazzi L, Maggioni AP, Marchioli R, Barlera S, Franzosi MG, Latini R, et al. Effect of rosuvastatin in patients with chronic heart failure (the GISSI-HF trial): a randomised, double-blind, placebo-controlled trial. Lancet 2008;372:1231e9. Krum H, Latini R, Maggioni AP, Anand I, Masson S, Carretta E, et al. Statins and symptomatic chronic systolic heart failure: a post-hoc analysis of 5010 patients enrolled in Val-HeFT. Int J Cardiol 2007; 119:48e53. Huan Loh P, Windram JD, Tin L, Reddy P, Velavan P, Rigby AS, et al. The effects of initiation or continuation of statin therapy on cholesterol level and all-cause mortality after the diagnosis of left ventricular systolic dysfunction. Am Heart J 2007;153:537e44.