Cardiovascular Drugs and Therapy 1992;6:101-102 Kluwer Academic Publishers, Boston. Printed in U.S.A.
Cholesterol Lowering and the Reduction of CHD Incidence and Total Mortality: Results from a Meta-Analysis of Randomized Trials Ingar Holme Life Insurance Companies" Institute for Medical Statistics, Ullevaal Hospital, Oslo, Norway
Summary. This editorial reports on a meta-analysis done on cholesterol-lowering trials. Coronary heart disease incidence is reported for 16 trials and total mortality for 19 trials. The cholesterol benefit ratio, i.e., the percent risk reduction divided by the percent cholesterol reduction, was analyzed in terms of total vs. potential modifying factors. For coronary heart disease the benefit ratio was 2.8 (95% CL: 1.5, 4.2), and for total mortality it was 1.2 (95% CL: - 0 . 1 , 2.3). Cholesterol reduction had to be at least 8-9% to outweigh the negative impact of treatment on total mortality. Cardiovasc Drugs Ther 1992;6:101-102
Key Words. coronary heart disease, analysis
cholesterol,
meta-
T h e history of cholesterol-lowering trials designed to show a reduction in the risk of coronary heart disease goes back to the mid 1960s. It was not until the Lipids Research Clinics Program Coronary Primary Preventive Trial released its results in 1984 that the hypothesis that cholesterol reduction decreased the incidence of coronary heart disease (CHD) became accepted by most cardiologists [1]. Before that time results had been varying and somewhat disappointing [2]. The Lipid Research Clinics Coronary Primary Prevention Trial (LRC-CPPT) claimed that for every 1% achieved cholesterol reduction, a 2% CHD risk reduction took place. This was a rather broad statement, containing few reservations with respect to covariation with age, sex, level of baseline cholesterol, type of cholesterol regimen, or duration of therapy. Potentially, the effectiveness of cholesterol reduction could be dependent on such variables. In a meta-analysis published recently, Holme addresses some of these issues. There are 16 randomized trials with CHD as the end point aimed at cholesterol lowering published in the literature. Each of those has been used as an experimental unit in a cross-trial statistical analysis [3]. The effectiveness of treatment is assessed by computing the odds ratio for each trial, i.e., the odds in the actively treated groups divided by the similar odds in the control group (odds: cases divided by non-
cases). Thereafter the natural logarithm of the odds ratio is used for analytical purposes instead of the odds ratio itself (because of better statistical properties), and it is related to the percent net reduction in cholesterol. For a given percent cholesterol reduction, the CHD risk reduction can be assessed through a weighted linear regression line. The slope of this line will be the cholesterol benefit ratio, which gives the CHD risk reduction in percent for each percent cholesterol reduction. In the meta-analysis this percentage was estimated to be 2.8, i.e., on the higher side of the LRC estimate. However, due to a large interstudy variation, the 95% confidence interval covered the range 1.5-4.1. (The data are slightly changed due to an error in the original paper.) MacMahon et al. have published a meta-analysis of several non-experimental epidemiological studies of the cholesterol-CHD relationship [4]. They claimed that the maximum benefit ratio of cholesterol reduction would be approximately 3.0. This means that for every percent cholesterol reduction a maximum 3% CHD reduction would take place. The estimation was done accounting for a within-patient measurement error of cholesterol. Considering the limited median time until the CHD event in these trials (approximately 3 years), combined with the fact that benefit is hardly seen until a delay time of approximately 11/2 years, this meta-analysis finding is extraordinarily consistent with the model of 100% effective risk reduction by cholesterol lowering. It should be noted, though, that the cholesterol reduction should be maintained for at least 5 years and that the results apply for middle-age subjects (mostly men) with elevated cholesterol at baseline. The power of the meta-analysis to find interaction effects between cholesterol lowering efficiency and other covariates was low. However, the analyses did show that the drug trials tended to be more efficient
Address correspondence and reprint requests to Ingar Holme, Institute for Medical Statistics, P.O. Box 6, Ullevaal Hospital, Kirkevn 166, 0407 Oslo 4, Norway. 10/
102
Holme
than the dietary trials with respect to CHD risk reduction. Caution is needed, however, in the interpretation of this finding because the cholesterol benefit ratio seems to be dependent on the level of baseline cholesterol. In his analysis, Holme showed that the cholesterol benefit ratio was significantly higher in those patients with the more elevated baseline cholesterol values. Because baseline cholesterol was somewhat higher in the drug than in the dietary trials, an adjustment for baseline cholesterol was made to account for this, but the result was essentially the same. The cholesterol benefit ratio of 2.8 for CHD was mostly unaffected by factors such as sex, age, primary or secondary preventive trials, and multiple or single-factor trials. Thus, the global statement made by the LRC research group concerning the cholesterol benefit ratio is not contradicted by the findings in this meta-analysis. Holme also did a parallel meta-analysis with total mortality as the end point based on data from 19 trials. A borderline significant relationship was found between the log odds ratio and the percent net reduetion in cholesterol. However, the regression line crossed the ordinate axis significantly above zero, meaning that no cholesterol reduction was associated with a significant excess mortality. The regression line sloped downwards and remained above zero for cholesterol values below 8-9%. This means that for net reductions in cholesterol below that magnitude, an associated excess in total mortality was observed on average. Above that level the regression line had negative log odds ratio levels, which resulted in an associated decreased total mortality. It is a.fact that total mortality has not been effectively reduced in the lipid-lowering trials. One reason may be that the net cholesterol reduction between the
trials has not been large enough, in fact, "only" 5-6% of the average in the reported trials. This is far less than observed using modern drugs such as hydroxymethyl-glutaryl coenzyme A (HmG CoA) reductase inhibitors [5]. Another reason may be that interventions have also taken place in populations with rather normal cholesterol values, where CHD makes up a smaller percentage of the total mortality, resulting in reduced power to detect significant relationships. Finally, cholesterol lowering may also produce countereffects on noncardiovascular end points. It is to be expected from the large trials underway using HMG CoA reductase inhibitors that total mortality should be significantly reduced if the regression line reported by Holme is reflecting a true biological mechanism.
References 1. Lipids Research Clinics Program. The Lipid Research Clinics Coronary Primary Preventive Trial results: I. Reduction in incidence of coronary heart disease. JAMA 1984;251:351364. 2. Committee of Principal Investigators. WHO cooperative trial on primary prevention of ischaemic heart disease using clofibrate to lower serum cholesterol: Mortality follow-up, Lancet 1980;2:379-385. 3. Holme I. An analysis of randomized trials evaluating the effect of cholesterol reduction on total mortality and coronary heart disease. Circulation 1990;82:1916-1924. 4. MacMahon S, Peto R, Cutler J, et al. Blood pressure, strokes and coronary heart disease. Part I, prolonged differences in blood pressure: Prospective observational studies corrected for the regression dilution bias. La~cet 1990;335:765-774. 5. The Lovastatin Study Group III. A multicenter comparison of lovastatin and cholestyramin therapy for severe hypercholesterolemia. JAMA 1989;260:359-366.
Cholesterol Lowering and the Reduction of CHD Incidence and Total Mortality: Results from a Meta-Analysis of Randomized Trials Ingar Holme Life Insurance Companies" Institute for Medical Statistics, Ullevaal Hospital, Oslo, Norway
Summary. This editorial reports on a meta-analysis done on cholesterol-lowering trials. Coronary heart disease incidence is reported for 16 trials and total mortality for 19 trials. The cholesterol benefit ratio, i.e., the percent risk reduction divided by the percent cholesterol reduction, was analyzed in terms of total vs. potential modifying factors. For coronary heart disease the benefit ratio was 2.8 (95% CL: 1.5, 4.2), and for total mortality it was 1.2 (95% CL: - 0 . 1 , 2.3). Cholesterol reduction had to be at least 8-9% to outweigh the negative impact of treatment on total mortality. Cardiovasc Drugs Ther 1992;6:101-102
Key Words. coronary heart disease, analysis
cholesterol,
meta-
T h e history of cholesterol-lowering trials designed to show a reduction in the risk of coronary heart disease goes back to the mid 1960s. It was not until the Lipids Research Clinics Program Coronary Primary Preventive Trial released its results in 1984 that the hypothesis that cholesterol reduction decreased the incidence of coronary heart disease (CHD) became accepted by most cardiologists [1]. Before that time results had been varying and somewhat disappointing [2]. The Lipid Research Clinics Coronary Primary Prevention Trial (LRC-CPPT) claimed that for every 1% achieved cholesterol reduction, a 2% CHD risk reduction took place. This was a rather broad statement, containing few reservations with respect to covariation with age, sex, level of baseline cholesterol, type of cholesterol regimen, or duration of therapy. Potentially, the effectiveness of cholesterol reduction could be dependent on such variables. In a meta-analysis published recently, Holme addresses some of these issues. There are 16 randomized trials with CHD as the end point aimed at cholesterol lowering published in the literature. Each of those has been used as an experimental unit in a cross-trial statistical analysis [3]. The effectiveness of treatment is assessed by computing the odds ratio for each trial, i.e., the odds in the actively treated groups divided by the similar odds in the control group (odds: cases divided by non-
cases). Thereafter the natural logarithm of the odds ratio is used for analytical purposes instead of the odds ratio itself (because of better statistical properties), and it is related to the percent net reduction in cholesterol. For a given percent cholesterol reduction, the CHD risk reduction can be assessed through a weighted linear regression line. The slope of this line will be the cholesterol benefit ratio, which gives the CHD risk reduction in percent for each percent cholesterol reduction. In the meta-analysis this percentage was estimated to be 2.8, i.e., on the higher side of the LRC estimate. However, due to a large interstudy variation, the 95% confidence interval covered the range 1.5-4.1. (The data are slightly changed due to an error in the original paper.) MacMahon et al. have published a meta-analysis of several non-experimental epidemiological studies of the cholesterol-CHD relationship [4]. They claimed that the maximum benefit ratio of cholesterol reduction would be approximately 3.0. This means that for every percent cholesterol reduction a maximum 3% CHD reduction would take place. The estimation was done accounting for a within-patient measurement error of cholesterol. Considering the limited median time until the CHD event in these trials (approximately 3 years), combined with the fact that benefit is hardly seen until a delay time of approximately 11/2 years, this meta-analysis finding is extraordinarily consistent with the model of 100% effective risk reduction by cholesterol lowering. It should be noted, though, that the cholesterol reduction should be maintained for at least 5 years and that the results apply for middle-age subjects (mostly men) with elevated cholesterol at baseline. The power of the meta-analysis to find interaction effects between cholesterol lowering efficiency and other covariates was low. However, the analyses did show that the drug trials tended to be more efficient
Address correspondence and reprint requests to Ingar Holme, Institute for Medical Statistics, P.O. Box 6, Ullevaal Hospital, Kirkevn 166, 0407 Oslo 4, Norway. 10/
102
Holme
than the dietary trials with respect to CHD risk reduction. Caution is needed, however, in the interpretation of this finding because the cholesterol benefit ratio seems to be dependent on the level of baseline cholesterol. In his analysis, Holme showed that the cholesterol benefit ratio was significantly higher in those patients with the more elevated baseline cholesterol values. Because baseline cholesterol was somewhat higher in the drug than in the dietary trials, an adjustment for baseline cholesterol was made to account for this, but the result was essentially the same. The cholesterol benefit ratio of 2.8 for CHD was mostly unaffected by factors such as sex, age, primary or secondary preventive trials, and multiple or single-factor trials. Thus, the global statement made by the LRC research group concerning the cholesterol benefit ratio is not contradicted by the findings in this meta-analysis. Holme also did a parallel meta-analysis with total mortality as the end point based on data from 19 trials. A borderline significant relationship was found between the log odds ratio and the percent net reduetion in cholesterol. However, the regression line crossed the ordinate axis significantly above zero, meaning that no cholesterol reduction was associated with a significant excess mortality. The regression line sloped downwards and remained above zero for cholesterol values below 8-9%. This means that for net reductions in cholesterol below that magnitude, an associated excess in total mortality was observed on average. Above that level the regression line had negative log odds ratio levels, which resulted in an associated decreased total mortality. It is a.fact that total mortality has not been effectively reduced in the lipid-lowering trials. One reason may be that the net cholesterol reduction between the
trials has not been large enough, in fact, "only" 5-6% of the average in the reported trials. This is far less than observed using modern drugs such as hydroxymethyl-glutaryl coenzyme A (HmG CoA) reductase inhibitors [5]. Another reason may be that interventions have also taken place in populations with rather normal cholesterol values, where CHD makes up a smaller percentage of the total mortality, resulting in reduced power to detect significant relationships. Finally, cholesterol lowering may also produce countereffects on noncardiovascular end points. It is to be expected from the large trials underway using HMG CoA reductase inhibitors that total mortality should be significantly reduced if the regression line reported by Holme is reflecting a true biological mechanism.
References 1. Lipids Research Clinics Program. The Lipid Research Clinics Coronary Primary Preventive Trial results: I. Reduction in incidence of coronary heart disease. JAMA 1984;251:351364. 2. Committee of Principal Investigators. WHO cooperative trial on primary prevention of ischaemic heart disease using clofibrate to lower serum cholesterol: Mortality follow-up, Lancet 1980;2:379-385. 3. Holme I. An analysis of randomized trials evaluating the effect of cholesterol reduction on total mortality and coronary heart disease. Circulation 1990;82:1916-1924. 4. MacMahon S, Peto R, Cutler J, et al. Blood pressure, strokes and coronary heart disease. Part I, prolonged differences in blood pressure: Prospective observational studies corrected for the regression dilution bias. La~cet 1990;335:765-774. 5. The Lovastatin Study Group III. A multicenter comparison of lovastatin and cholestyramin therapy for severe hypercholesterolemia. JAMA 1989;260:359-366.
