Abstract Encouraging intervention trials drive our expectations toward more aggressive cholesterol-lowering therapies, lower target levels, and less severe hypercholesterolemia. Available studies may predict which patients, degrees of total cholesterol (TC) reduction, and baseline and target levels of TC provide the most clinical benefit. Data were pooled from seven primary and nine secondary controlled trials with major coronary heart disease (CHD) events as primary endpoints. The analysis showed that we can expect large reductions in CHD from TC reduction in primary and secondary prevention. However, the reduction is much larger in subjects with high TC and/or previous CHD events. The percent reduction in CHD increased exponentially with increasing percent TC reductions, which predicted >70% of the change in CHD. Consequently, we cannot expect cost-effective clinical benefits from mean reductions in TC >15 (LDL cholesterol >20)%. The TC level at the study endpoint correlated with CHD incidence irrespective of the study group and explained almost 45% of CHD incidence. The relationship was progressive and leveled off at a TC level below about 150 mg/dL (3.9 mmol/L) (LDL cholesterol ≈110 mg/dL [≈2.8 mmol/L]). Little extra clinical benefit can be expected from further reductions. We can expect an average 2% reduction in CHD events per percent reduction in TC. We can also expect a 2-fold greater clinical benefit among subjects with high initial TC levels than among those with low levels. Finally, we can expect that the cholesterol-attributable risk is reset to that predicted by the TC level achieved within 4 to 6 years.
- Received June 18, 1997.
- Accepted August 6, 1997.
The conventional risk factors of hyperlipidemia, smoking, and hypertension are the main causes of premature atherosclerotic CHD in prospective population studies.1 2 3 4 This is true for men and women although women generally develop their symptoms about 10 years later than men.
Within a number of populations, the risk of CHD increases exponentially with increasing levels of TC.5 Moreover, the incidence of CHD varies in a similar way also among populations in relation to their mean TC levels.
TC predicts about half of the CHD incidence and up to half of the incidence occurs among subjects who lack conventional risk factors3 (see also Fig 3⇓). Consequently, this group is not easily accessible for dedicated preventive measures. Should we then expect that cholesterol-lowering alone could reduce CHD by more than about 50%?
The reduction by 24 to 35% in the incidence of CHD shown in the recent primary6 and secondary7 8 intervention trials with hydroxymethylglutaryl-CoA reductase inhibitor (statin) therapy in hypercholesterolemia is impressive. Furthermore, non-CHD mortality did not increase in these trials. This suggests that about two-thirds of the predictable CHD incidence can be reversed with maintained safety by a 20 to 35% reduction in LDL cholesterol (corresponding to about a 15 to 25% reduction in TC).
Extrapolation from these results drives our expectations toward more aggressive cholesterol-lowering therapies and toward lower target levels among subjects with less and less severe hypercholesterolemia. Registrations of new potent drugs and increased dosages of old drugs fuel these trends. However, increasing cost-consciousness and cost-containment put definite limits on such trends.
Previous meta-analyses have weighted the importance for the conclusions of quantitative and qualitative properties of the individual trials.9 10 11 12 Large trials have been given a larger weight than small trials. However, a small, well-performed trial may provide more reliable results than a larger trial. Various quality rankings are subjective as well as biased. However, discrepancies between individual trials act as conservative confounders, which blunt rather than strengthen statistical conclusions. Other obvious confounders, which generate different CHD incidences, are age and sex. More or less arbitrary adjustments of CHD incidences for age and/or sex have been used to improve the possibilities of conclusive results. However, adjustments may introduce new biases. Consequently, significant relationships, which emerge from unweighted unadjusted meta-analyses in spite of confounders, may provide robust conclusions.
Therefore, the aim of this overview was (1) to equally value all studies fulfilling defined minimum criteria, (2) to regard every trial as an independent and closed entity, (3) to use published data straightforwardly without statistical weightings or adjustments, and (4) to analyze where the most clinical benefit could be expected from cholesterol-lowering regimens.
Published data from 16 randomized cholesterol-lowering trials were derived from the literature.6 7 8 13 14 15 16 17 18 19 20 21 22 23 24 25 Trials were included only if cholesterol-lowering and major CHD events (CHD death and fatal myocardial infarction) were primary objectives (Table 1⇓). Trials were excluded if monitoring of progression of atherosclerosis was the major objective or if they only reported mortality data. One intervention study was excluded due to its mixed primary and secondary cohort.26 Trials were included irrespective of whether6 7 8 14 15 16 18 19 20 21 22 23 24 25 or not13 17 they showed a relationship between cholesterol-lowering treatment and CHD outcomes.
Seven studies were primary6 13 14 15 16 17 18 and nine were secondary7 8 19 20 21 22 23 24 25 prevention trials (Table 2⇓). Six studies used dietary intervention to reduce TC.14 15 17 19 20 21 Three of these also aimed to control tobacco smoking and hypertension.14 15 17 Clofibrate with25 or without13 22 23 24 nicotinic acid was used in five studies. In one of these,24 nicotinic acid and clofibrate were used in different arms, but the results were combined in this analysis. Cholestyramine was used in one study,16 gemfibrozil in one,18 pravastatin in two,6 8 and simvastatin in one.7 For the purpose of this analysis, however, the type of cholesterol-lowering regimen was disregarded.
Whenever possible, only definite CHD deaths and nonfatal myocardial infarction have been included as major CHD events in this analysis (Table 1⇑). Consequently, the CHD incidences used here occasionally19 21 23 differ from the reported total incidence. Minor variations in CHD event definitions were otherwise ignored.
All studies reported mean TC at baseline and either percent reduction to study endpoint or percent or absolute difference between control and treatment groups at the end of study. Reported percent values were used to calculate absolute levels at endpoint (achieved TC) when necessary. The achieved TC level in the intervention group was defined as treated TC (Table 2⇑). In the control group, the corresponding estimate was denoted control TC. Control TC rather than baseline TC was chosen as reference because the control group was reasonably stable only in some studies.6 7 8 13 15 24 25 The studies were regarded as intention-to-treat studies. The WOSCOPS trial6 only published on-treatment reductions in TC of 20%, but a 5% lower intention-to-treat effect has been communicated orally (Table 2⇑). Therefore, it was set to 15% in this analysis.
The mean follow-up times were 6.2 (range 4.9 to 10) years in primary prevention trials and 4.4 (3.3 to 6.2) years in secondary prevention trials (Table 2⇑). The projected total numbers of PYE were almost 355 000 in the control groups and 348 000 in the intervention groups with about 12% in secondary prevention trials. Altogether 4456 major CHD events (54% in primary prevention) were registered in the control groups and 3642 (59% in primary prevention) in the treatment groups. In the control groups, the average incidences of major CHD events in secondary prevention trials were almost six times that in primary prevention trials.
For comparison, three prospective epidemiological studies, which reported definite CHD mortality and nonfatal myocardial infarctions in strata of baseline TC in essentially healthy populations, were included in this analysis.2 3 4 These studies had mean durations of 7.1 to 8.6 years and covered almost 176 500 PYE. One study2 included a small number (6%) of myocardial infarction patients, but this was disregarded. Another4 contained diabetic subjects, who could be excluded.
Handling of Pooled Data
The percent difference between control and treatment groups (reduction) in major CHD events (ΔCHD%) in the nine largest (n>2000) trials was plotted against percent difference (reduction) in TC (ΔTC%). The regressions for these trials and for all trials were calculated separately on log-transformed data. Curves derived from the regression equations were introduced in the plot.
The observed incidences of major CHD events in the control and treated groups were plotted against the corresponding TC values at the end of each trial. The two data points of each study were joined. The regressions including control as well as intervention group data were calculated on log-transformed values for primary and secondary prevention trials separately and the equations obtained were used to calculate the fitted curves.
A standardized benefit (ΔCHD%:ΔTC%) was constructed from the data in Table 2⇑ to show which cumulative or additional percentage of CHD reduction could be expected per percent reduction in TC in strata of relative TC reductions or of reference (ie, control) TC values.
The observed CHD incidences were compared with incidences estimated from the nonintervened prospective studies by level of TC achieved in primary prevention trials. More precisely, the achieved TC values were introduced into the regression equation obtained from pooled data of the three prospective studies to give estimated incidences. The estimated and observed incidences in the control groups were first compared, and then the corresponding comparison was made for the intervention groups. For the secondary prevention trials, a different approach was used, since appropriate data from corresponding prospective trials are not available. Here, the results from control groups of the trials were used as nonintervened data to calculate the regression equation, and then the estimated incidences in the treated groups were calculated as above. A ratio between observed and estimated CHD incidences of 1 would suggest that the most clinical benefit had been recovered during the study period and that TC levels achieved by therapeutic interventions equaled spontaneous TC levels in predictive value.
Descriptive statistics were calculated according to standard procedures. In the main analysis, differences in CHD incidences between study groups were tested with two-way ANOVA with studies as covariates. Paired data were tested using Student’s paired t tests. Relationships were investigated with linear regression analysis with logarithmic transformations as indicated. Values in parentheses denote 95% confidence intervals throughout this article. A value of P≤.05 (two-tailed tests) was regarded as statistically significant.
Results and Discussion
In the primary prevention trials, ANOVA of the incidences of major CHD events showed significant differences between control and intervention groups (P=.021) as well as between individual studies (P=.011). The mean incidence in the control groups was 10.0 (6.7 to 13.2) and in the intervention groups 7.6 (5.3 to 9.9) events per 1000 PYE.
Also, in the secondary prevention trials, ANOVA showed similar differences between control and intervention groups (P=.012) and between individual studies (P=.005). The mean incidence in the control groups was 59.2 (44.1 to 74.3) and in intervention groups 46.9 (36.9 to 57.0) events per 1000 PYE.
Percent Changes in Total Cholesterol and Major CHD Events
Evidently, the difference between studies was partly due to confounders inherent in protocol- and cohort-specific discrepancies. However, the studies achieved widely varying degrees of TC reductions and had very different cholesterol-attributable risks. Indeed, ΔCHD% increased with increasing averages in ΔTC% (Fig 1⇓). Type of trial (primary or secondary prevention), type of therapy, and other differences between trials seemed not to matter much in this regard. This suggested a rather robust relationship between the degrees of TC and CHD reductions. Furthermore, trials reporting no significant difference in major CHD events also showed no difference in TC between the study groups.13 17 Only trials with ΔTC% of >7 reported significant reductions in major CHD events.
ΔCHD% leveled off exponentially with increasing ΔTC%. The regression was highly significant for the nine large trials (R2=0.73, P=.003) (Fig 1⇑) and suggested that >70% of the variation in ΔCHD% was explained by ΔTC%. A similar relationship appeared when all 16 trials were included in the regression analysis (R2=0.39, P=.010), and the two regression curves were almost superimposable (Fig 1⇑). The curves predicted a decrease in CHD by 38% from an unrealistic 90% reduction in TC. However, most of this benefit was recovered already at a ΔTC% of 15. Indeed, the cumulative standardized benefit (ΔCHD%:ΔTC%) was 4.9 for TC reductions up to 3%, 3.7 up to 5%, and only 1.2 up to 25% (Table 3⇓). The additional standardized benefit from 1% further reduction in TC was 2 at a ΔTC% of 3 and only 0.27 at a ΔTC% of 25 (Table 3⇓). These results indicate that less and less additional clinical benefit should be expected from average TC reductions >15%. Indeed, if studies with TC reductions ≥10%6 7 8 18 are considered, there is no definite trend between studies in actually observed CHD reductions (Fig 1⇑). Recently, posthoc stratifications of CARE data showed that ΔTC% >10 provided no further clinical benefit.27
Before the WOSCOPS6 and CARE8 studies were reported, Holme11 arrived at a similar conclusion regarding major CHD events and Gould et al12 did also for mortality from CHD from meta-analyses using linear regression of log-transformed weighted-odds ratios.
The dose-effect relationship for statins is also exponential. Each doubling of dose provides about 6% additional decrease in LDL cholesterol.28 29 This indicates that very large reductions in TC may be achieved by progressively increasing doses and costs with less and less further clinical benefit. The clinical and health-economic dilemma is obvious.
Achieved Levels of Total Cholesterol and CHD Incidence
The above conclusions were based on mean reductions in different study cohorts and say little about the need in the individual patient. Severely hypercholesterolemic subjects may benefit from larger relative reductions in cholesterol and vice versa. This possibility was supported by the fact that the incidence of CHD events increased with increasing levels of achieved TC (Fig 2⇓). When control and intervention groups of all secondary prevention trials were included, the regression analysis (R2=0.44, P=.003; cf, legend to Fig 2⇓) indicated that the risk increased progressively and that the achieved TC explained about 44% of the observed CHD incidence. A 10% reduction from a TC of 300 mg/dL (7.8 mmol/L) predicted a reduction in major CHD events from 86 to 68 per 1000 PYE (ie, about 21%), whereas the same percent reduction from 200 mg/dL (5.2 mmol/L) suggested a reduction from 39 to 34 per 1000 PYE (ie, about 14%).
There was a similar trend in the seven primary prevention trials (R2=0.24, P=.078; cf legend to Fig 2⇑). As expected, the risk was lower, the curve was flatter, and TC explained only 24% of the CHD incidence. The curve suggested that a decrease in TC of 10% from 300 mg/dL (7.8 mmol/L) predicted a risk reduction from 10.5 to 9.0 CHD events per 1000 PYE (ie, 18%), whereas the same reduction from 200 mg/dL (5.2 mmol/L) suggested a reduction from 6.2 to 5.6 events per 1000 PYE (ie, 10%). Clearly, TC reductions determine CHD reduction in relative as well as absolute terms.
Although there was a considerable variation between studies, the lines connecting control and intervention groups in the individual trials were roughly parallel to the fitted curves in primary as well as secondary prevention trials (Fig 2⇑). They were steep in the high end and flat in the low end of the TC distribution. Obviously, TC reduction is more cost-effective among severely hypercholesterolemic than among normocholesterolemic subjects with or without previous infarction, and there is a lower limit below which further reduction is not cost-effective.
Posthoc stratifications of data in the CARE trial8 in myocardial infarction patients with rather normal TC levels support these conclusions directly. The results showed that patients having TC above the study average (209 mg/dL [5.4 mmol/L]) experienced a greater reduction in major CHD events from pravastatin treatment than those below average. These analyses also showed that subjects with initial LDL cholesterol levels of 150 to 175 mg/dL (3.9 to 4.5 mmol/L) showed a 35% reduction and those with 125 to 150 mg/dL (3.2 to 3.9 mmol/L) showed a 26% reduction, whereas those with <125 mg/dL (3.2 mmol/L) experienced no clinical benefit (3% increase) from treatment. Based on CARE data, an LDL cholesterol level of 125 mg/dL would correspond to a TC of about 150 mg/dL (3.9 mmol/L).
In the secondary prevention 4S trial with simvastatin among CHD patients with moderately elevated TC (mean 6.75, range 5.5 to 8 mmol/L or mean 260, range 210 to 310 mg/dL), posthoc stratification according to baseline LDL cholesterol have been published.30 The results showed a similar relative risk reduction of about 35% across the quartiles of baseline LDL cholesterol. No patients in the 4S trial represented the very low levels of LDL cholesterol <125 mg/dL shown in the CARE trial to experience no clinical benefit from cholesterol reduction. The LDL cholesterol levels in the 4S cohort were more comparable with the levels of the upper stratum in the CARE trial, with which it shared a closely similar CHD reduction. This was positioned in the flat part of the curve in Fig 1⇑. Therefore, much variation may not be expected in this region. Consequently, the 4S trial does not invalidate the conclusion from the CARE study and the present analysis that CHD patients with low TC levels experience low clinical and cost benefits from cholesterol-lowering therapy.
Clinical Benefit in Strata of Total Cholesterol
Holme9 showed a significant (P=.001) relationship between TC and the percent CHD reduction per percent TC reduction in a weighted analysis on log-transformed data. He concluded that the clinical benefit from every percent TC reduction is about twice higher in the upper than in the lower range of distribution of baseline TC levels.
Here, the standardized benefit (ΔCHD%:ΔTC%) showed an overall mean of 1.8. This suggested close to 2% return in CHD reduction per percent TC reduction. The difference between strata of reference (control) TC was not significant in the unweighted analysis performed here (P=.152) but tended to increase from 1.3 in the lowest to 2.6 in the highest quartile of mean TC. The simplistic approach may have blunted the current statistics and, certainly, they do not invalidate the conclusion of Holme. Consequently, the same degree of cholesterol-lowering provides less clinical benefit among subjects with low than with high TC levels.
Estimation of Recoverable Benefits From Cholesterol Reduction
The conclusions in the present analysis were based on the cumulative number of clinical endpoints and included the first year(s) when little/no difference occurred between the two groups. It is likely that the differences were most pronounced between the groups beyond the 5th year as suggested by Law et al,5 but neither study was sized to provide reliable data for parts of the projected study period. A separate analysis was undertaken to test to what extent an achieved TC value reflected the recoverable clinical benefit during the duration of the trials.
Pooled data from the three prospective epidemiological studies2 3 4 showed the expected progressive increase in CHD incidence with increasing TC levels. The relationship was highly significant (R2=0.67, P=.0002; Fig 3⇓ and legend). This suggested that baseline TC predicted about 65% of future CHD events. When achieved TC from the primary prevention trials were introduced into the regression equation, estimated incidences were obtained (Table 4⇓). These estimates represented predictions from the unintervened populations. Clearly, observed (mean 10.0) and estimated (mean 9.8) incidences per 1000 PYE were closely similar in the control groups (P=.917), and this supported the validity of the approach. However, this was true also for the treated groups (means 7.6 versus7.5 events per 1000 PYE, P=.950). The mean ratio between observed and estimated incidences was 1.07 (0.77 to 1.37) in the control groups and 1.06 (0.70 to 1.43) in the intervention groups. This was not significantly different from 1. Consequently, these results suggested that most clinical benefit had been recovered within the duration (mean 6 years) of the primary preventive trials.
For secondary prevention, results from the control groups were used to obtain the regression equation for untreated cohorts ([log incidence]=0.7514+0.0042× [achieved TC (mg/dL)], R2=0.50, P=.032). Assuming that baseline and achieved TC were roughly the same, they also predict about 50% of future CHD after a major CHD event. This equation was subsequently used to calculate estimated incidences in the secondary intervention groups (Table 5⇓). There was no significant difference between observed (mean 46.9) and estimated (mean 41.7) incidences per 1000 PYE (P=.187). The mean ratio was 1.13 and the 95% confidence interval (0.91 to 1.35) included 1, suggesting that most clinical benefit had been recovered also within the duration (mean 4 years) of the secondary prevention trials.
The clinical benefit from cholesterol-lowering therapy is well established. The current results strongly support this conclusion and show that type of therapy is less important than degree of cholesterol reduction. They also show that most of the TC-attributable CHD incidence is reduced in proportion to relative as well as absolute cholesterol reductions and that this clinical benefit is recovered within about 4 to 6 years. The clinical benefit is 6 times better in secondary than in primary prevention and double in cohorts with severe compared with mild hypercholesterolemia. This means that 6 times as many need to be treated to avoid one clinical event in primary as in secondary prevention and twice as many in mild compared with severe hypercholesterolemia.
Mean relative reductions in TC >15 (LDL cholesterol >20)% are not meaningful since they require large doses and costs of drugs and provide little extra clinical benefit. In contrast, a TC reduction of 3%, which is possible to achieve by changes in eating habits, provides a 15% reduction in CHD. Applied to the large population without obvious hypercholesterolemia or CHD, dietary changes may mean a lot in reducing the incidence of CHD in the community. Fewer and fewer major CHD events are avoided by reducing TC much below an absolute level of about 150 mg/dL (3.9 mmol/L) (LDL cholesterol≈110 mg/dL≈2.8 mmol/L), and the large numbers of subjects in these strata make drug costs per prevented event very high. In contrast, the clinical benefit is high among the rather few subjects, who have very high levels of TC and need very large absolute and relative reductions in TC. Ideally, efficient cholesterol-lowering drugs should provide absolute and relative TC reductions, which increase progressively with increasing baseline TC levels, but the mean reduction does not have to be extreme.
This leaves us with a difficult choice; should we try to maintain full health by preventing a first attack among great numbers at a high cost of drugs or try to prevent a second attack among few, who already suffer sequelae from a previous attack, at a lower cost?
Selected Abbreviations and Acronyms
|CHD||=||coronary heart disease|
This analysis was supported by grants from the Swedish Medical Research Council (4531) and the Swedish Heart-Lung Foundation (51044); this support is gratefully acknowledged.
Kannel WB. Clinical misconceptions dispelled by epidemiological research. Circulation. 1995;92:3350–3360.
Pocock SJ, Shaper AG, Phillips AN. Concentrations of high density lipoprotein cholesterol, triglycerides, and total cholesterol in ischaemic heart disease. Br Med J. 1989;298:998–1002.
Rosengren A, Welin L, Tsipogianni A, Wilhelmsen L. Impact of cardiovascular risk factors on coronary heart disease and mortality among middle aged diabetic men: a general population study. Br Med J. 1989;299:1127–1131.
Law MR, Wald NJ, Thompson SG. By how much and how quickly does reduction in serum cholesterol concentration lower risk of ischaemic heart disease? Br Med J. 1994;308:367–372.
Sacks FM, Pfeffer MA, Moye LA, Rouleau JL, Rutherford JD, Cole TG, Brown L, Warnica JW, Arnold JMO, Wun C-C, Davis BR, Braunwald E. The effect of pravastatin on coronary events after myocardial infarction in patients with average cholesterol levels. N Engl J Med. 1996;335:1001–1009.
Holme I. An analysis of randomized trials evaluating the effect of cholesterol reduction on total mortality and coronary heart disease incidence. Circulation. 1990;82:1916–1924.
Holme I. Relation of coronary heart disease incidence and total mortality to plasma cholesterol reduction in randomised trials: use of meta-analysis. Br Heart J.. 1993;69:S42–S47.
Gould AL, Rossouw JE, Santanello NC, Heyse JF, Furberg CD. Cholesterol reduction yields clinical benefit: a new look at old data. Circulation. 1995;91:2274–2282.
Committee of Principal Investigators. A cooperative trial in the prevention of ischaemic heart disease using clofibrate. Br Heart J. 1978;40:1069–1118.
Hjermann I, Holme I, Byre KV, Leren P. Effect of diet and smoking intervention on the incidence of coronary heart disease. Lancet. 1981;ii:1303–1310.
World Health Organization European Collaborative Group. Mutifactorial trial in the prevention of coronary heart disease: 3. Incidence and mortality results. Eur Heart J. 1983;4:141–147.
Lipid Research Clinics Program. The Lipid Research Clinics Coronary Primary Prevention Trial results. I. Reduction in incidence of coronary heart disease. JAMA. 1984;25:351–364.
Wilhelmsen L, Berglund G, Elmfeldt D, Tibblin G, Wedel H, Pennert K, Vedin A, Wilhelmsson C, Werkö L. The multifactorial primary prevention trial in Göteborg, Sweden. Eur Heart J. 1986;7:279–288.
Frick MH, Elo O, Haapa K, Heinonen OP, Heinsalmi P, Helo P, Huttunen JK, Kaitaniemi P, Koskinen P, Manninen V, Mäenpää H, Mälkönen M, Mänttäri M, Norola S, Pasternack A, Pikkarinen J, Romo M, Sjöblom T, Nikkilä EA. Helsinki Heart Study: primary-prevention trial with gemfibrozil in middle-aged men with dyslipidemia. N Engl J Med. 1987;317:1237–1245.
Research Committee to the Medical Research Council. Low fat diet in myocardial infarction: a controlled trial. Lancet. 1965;ii:501–504.
Leren P. The effect of plasma cholesterol lowering diets in male survivors of myocardial infarction. Acta Med Scand. 1966;(suppl)466:1–92.
Research Committee to the Medical Research Council. Controlled trial of soya-bean oil in myocardial infarction. Lancet. 1968;ii:693–700.
Research Committee of the Scottish Society of Physicians. Ischaemic heart disease: a secondary prevention trial using clofibrate. Br Med J. 1971;4:775–784.
Group of Physicians of the Newcastle upon Tyne Region. Trial of clofibrate in the treatment of ischaemic heart disease. Br Med J. 1971;4:767–775.
Dayton S, Pearce ML, Hashimoto S, Dixon WJ, Tomiyasu U. A controlled clinical trial of a diet high in unsaturated fat in preventing complications of atherosclerosis. Circulation. 1969;40:1–63.
Sacks FM, Wun C-C, Cole TG, et al. The influence of baseline lipid levels on reduction in coronary events by pravastatin: the Cholesterol and Recurrent Events Trial. Presented at the 69th Scientific Sessions of the American Heart Association; November 11–13, 1996, New Orleans, La.