Cost-Effectiveness of Populationwide Educational Approaches to Reduce Serum Cholesterol Levels

Guidelines from the National Cholesterol Education Program highlight the importance of dietary therapy as a first approach to cholesterol control and suggest that drug therapy be used only among those at high risk for CHD (coronary heart disease).¹ Although as many as 35% to 40% of men and ≈20% of women have serum cholesterol levels of >6.21 mmol/L (240 mg/dL), a substantial proportion of CHD cases occur in persons with relatively low serum cholesterol levels.²³⁴ According to one estimate, ≈70% of new CHD cases in men and ≈50% of new cases in women occur in persons with cholesterol levels of <6.47 mmol/L (<250 mg/dL).⁵ Thus, it has been suggested that populationwide cholesterol intervention programs would be required to achieve substantial reductions in national CHD event rates and death rates.⁵⁶

Although numerous studies have analyzed the cost-effectiveness of targeted programs with medications to lower serum cholesterol levels,^789101112 data on the cost and cost-effectiveness of populationwide strategies are limited.¹³¹⁴¹⁵ We used published data on the costs and effectiveness of populationwide risk factor reduction programs¹⁴¹⁶¹⁷ and our Coronary Heart Disease Policy Model,¹⁸¹⁹ a comprehensive computer simulation of CHD, to analyze the potential costs and effectiveness of population-based strategies for cholesterol reduction in the United States from 1995 through 2020.

Methods

The Coronary Heart Disease Policy Model, which has been described in detail elsewhere,⁵⁷¹⁸¹⁹ is a state-transition, computer-based model consisting of three integrated submodels: the Demographic-Epidemiologic Submodel, the Bridge Submodel, and the Disease History Submodel. All model projections are based on the assumption that data from observational cohort studies²⁰ and experimental clinical trials²¹²² can be applied broadly to estimate the risk of CHD events and that ultimately the incidence of such events will influence mortality.²¹²²

The Demographic-Epidemiologic Submodel applies to the US population of persons 35 to 84 years old who are free of CHD. Each year, a new cohort of 35-year-olds, as estimated by the US Bureau of the Census,²³²⁴ enters the submodel, and persons turning age 85 exit the submodel. To exit before age 85, persons must either die or develop CHD, at which time they enter the Bridge Submodel.

The Demographic-Epidemiologic Submodel is used to assess each individual's risk of developing CHD and death from all causes on the basis of risk strata defined by age (age 35 to 84 in 10-year intervals), sex, smoking status (no or yes), diastolic blood pressure (≤94, 95 to 104, or ≥105 mm Hg), serum cholesterol (≤240 mg/dL [6.21 mmol/L], 240 to 299 mg/dL [6.21 to 7.75 mmol/L], or ≥300 mg/dL [7.76 mmol/L]), HDL (<35 mg/dL [0.91 mmol/L], 35 to 49 mg/dL [0.91 to 1.27 mmol/L], or ≥50 mg/dL [1.3 mmol/L]), and LDL (<160 mg/dL [4.16 mmol/L], 160 to 189 mg/dL [4.16 to 4.91 mmol/L], or ≥190 mg/dL [4.94 mmol/L]). The initial multivariate distributions of risk factors, conditional on age and sex, and transitions for these risk factors with aging were modeled with data from the Second National Health and Nutrition Examination Survey (NHANES II) (unpublished data from the National Center for Health Statistics, NHANES II, 1976-1980). These distributions were further updated to reflect 1986 population averages based on observed trends in risk factors²⁵²⁶ and more recent data from NHANES III.²⁷

To predict the annual probability of a CHD event or non-CHD death on the basis of risk factors, multiple logistic risk functions that controlled for all risk factors in the model simultaneously and allowed for interaction terms between risk factors and age were estimated using 36-year follow-up data from the Framingham Heart Study (personal communication, Dr Ralph D'Agostino, Boston University, 1994). Because recent studies suggest the lack of detrimental noncoronary effects from cholesterol reductions,²¹²² our baseline analysis assumed that serum cholesterol did not affect noncoronary mortality, but we investigated the impact of varying this assumption in sensitivity analyses. We assumed that the effect of risk factor reduction on coronary mortality was mediated by its impact on coronary events such as new-onset angina, acute myocardial infarction, and cardiac arrest and on subsequent case-fatality rates.²¹²² Coronary mortality and overall mortality were derived from US vital statistics.²⁸

The Bridge Submodel characterizes subjects during the first 30 days after they develop CHD. For each age range and sex, the Bridge Submodel categorizes each incident case of CHD into angina, myocardial infarction, or cardiac arrest and applies mortality probabilities and 30-day resource costs to persons in each group.¹⁸¹⁹

The Disease History Submodel tracks the subsequent development of recurrent CHD events, case-fatality rates, and resource costs, including the annual rate of coronary revascularization procedures in persons with prevalent CHD who survive the Bridge Submodel or who enter the model at age 35 with preexisting CHD. Events include subsequent myocardial infarction, coronary revascularization, and cardiac arrest.

In summary, in each model year, persons can develop new events. Persons leave the model when they die from CHD or non-CHD at the appropriate age- and sex-specific rate or when they reach age 85. The Coronary Heart Disease Policy Model can be run indefinitely into the future by adding a new cohort of 35-year-old persons in each subsequent year. The long-term, steady-state effect of interventions can be modeled, and cost-effectiveness ratios can be calculated.

Cost-Effectiveness Assumptions

Our simulations used the baseline assumptions of the updated Coronary Heart Disease Policy Model, including costs updated to 1993 using the general medical care component of the Consumer Price Index.²⁹ We followed the population for a 25-year period from 1995 through 2020, with a new 35-year-old cohort entering each year. In the baseline analysis, cholesterol-lowering interventions and other interventions in 1995 were assumed to begin having an effect on CHD events in 1998 and to have an impact only on persons who are initially free of CHD. Thus, although costs were applied across the entire population, we assumed that persons already diagnosed with CHD were receiving state-of-the-art recommendations regarding cholesterol reduction and would not be affected by the populationwide program. Only the direct medical costs related to CHD were included in our analyses.

Costs and person-years were discounted at 5% per year. The net effect of discounting is to weigh future costs and life-years less heavily than those encountered in the near term.¹² The cost-effectiveness ratio comprises a numerator and denominator, which represent the net change in total discounted costs and total discounted life-years, respectively, for a populationwide intervention program relative to no intervention. Net changes in cost reflect decreases that are a result of fewer cases of CHD as well as increases in costs associated with the program. To allow comparison with the cost-effectiveness of accepted medical interventions, the ratio estimates the cost per year of life saved.¹²

Assumptions Regarding Intervention

In the baseline analyses, we estimated the costs, life-years saved, and costs per year of life saved under the range of cholesterol reductions (1% to 4%) and costs ($4.95 to $9) estimated in North Karelia,¹⁶ the Stanford Three-Community Study,¹⁷ and the Stanford Five-City Project.¹⁴ These interventions consisted of education through media campaigns, including television, radio, newspaper, and other printed material, and direct education through community activities and face-to-face instruction.¹⁴¹⁶¹⁷

Data from North Karelia¹⁶ showed that a communitywide intervention program lowered serum cholesterol levels by 3% in men and 1% in women after 10 years. As part of the program, there were also reductions in diastolic blood pressure (1% in men and 2% in women) and in smoking rate (28% in men and 14% in women). In the Stanford Three-Community Study,¹⁷ cholesterol reductions of 3% were achieved. In the Stanford Five-City Project,¹⁴ there were reductions of 2% in serum cholesterol, 4% in systolic and diastolic blood pressure, and 13% in smoking rate in intervention communities.

In the baseline analyses, logistic risk equations included only total serum cholesterol and HDL, not LDL. The coefficient for HDL was assumed to be zero in baseline analyses. In sensitivity analyses, risk equations including HDL and LDL were used based on the assumption that they would be reduced proportionately by the intervention.

The costs of communitywide intervention programs are more difficult to estimate. The primary costs of the Stanford Five-City Project related to television and radio advertisements, with an estimated cost of $4.95 per person reached per year.¹⁴ In the North Karelia study, it was estimated that ≈$10 was spent (in 1985 US dollars) for each person reached by the program,³⁰ not including a substantial amount of volunteer activity. An additional estimate of $5.00 per person per year was given for the cost of continuing media coverage in North Karelia.³⁰ When inflated to 1993 US dollars, the program would cost an estimated $16.55 per person reached in year 1 and an additional $8.28 per person per year thereafter. Thus, the average cost per person per year over a 25-year period is ≈$8.66.

Sensitivity Analyses

Separate analyses assessed the impact of alternate assumptions regarding the effectiveness of a populationwide program in changing LDL and HDL cholesterol levels, in reducing cholesterol levels in persons with prevalent CHD, and in reducing multiple risk factors on estimates of cost-effectiveness. The impacts of lowering the discount rate to 3% and introducing quality adjustments for persons with a history of CHD were also investigated in separate analyses. The effects of the direct and induced costs of the program on cost-effectiveness were also examined.

The impact of changes in the strength of the relation between serum cholesterol levels and the incidence of CHD and CHD deaths was assessed in simulations that raised or lowered the logistic regression coefficients in the model by 1 SEM. The impact of including cholesterol as a risk factor for non-CHD death was also investigated. In these analyses, we modeled the association between non-CHD death based on the natural logarithm of serum cholesterol using pooled data from several studies.³¹³² Thus, changes in serum cholesterol would have a greater impact on non-CHD death among individuals with low serum cholesterol than among individuals with high serum cholesterol. For example, among men age 45 to 54 with serum cholesterol levels of 280 or 220 mg/dL, a 40 mg/dL decrease would increase the odds of non-CHD death by 4.4% or 5.8%, respectively. As in the baseline analyses, all sensitivity analyses assumed that cholesterol-lowering interventions in 1995 did not begin to affect CHD events or non-CHD deaths until 1998.

Results

Under baseline epidemiological assumptions and by varying costs per person per year, the cost per year of life saved would range from a reduction in costs, ie, a net savings in costs as well as lives, to a cost of $88 100 per year of life saved (Figure). A nationwide program with the costs ($4.95 per year) and cholesterol-lowering benefits (2%) of the Stanford Five-City Project would save a total of ≈624 000 life-years, and its net discounted costs, including both the costs of the program and the costs averted because of a decrease in CHD incidence, would be ≈$2.1 billion. When both costs and benefits were discounted to adjust for the differential timing of expenditures and benefits, the cost per year of life saved for such a program was estimated at $3200. A nationwide program, with costs ($16.55 in year 1 and $8.28 per year thereafter) and cholesterol-lowering benefits (3%) similar to those found in the North Karelia study, would cost ≈$6100 per year of life saved. Even if the program cost were $16.55 per year, nationwide programs that result in cholesterol reductions of ≥2% would be relatively cost effective compared with most accepted medical interventions, with estimated costs per year of life saved of ≤$38 500.

Impact on Persons With CHD

To make the most favorable cost-effectiveness estimates for the populationwide programs, we computed the costs per year of life saved assuming that all persons would achieve a reduction in serum cholesterol level regardless of their history of CHD. Not surprisingly, all programs would then have much more favorable cost-effectiveness ratios (Table 1).

Multiple Risk Factors

When the impact of multiple risk factors was modeled according to the results of the populationwide programs, programs were projected to save resources or to be associated with very favorable cost-effectiveness ratios (Table 1). For example, the Stanford Five-City Project, which cost $4.95 per person per year, reported a 4% decrease in diastolic blood pressure and a 13% decrease in smoking in addition to the 2% decrease in serum cholesterol. The savings from averting CHD events would more than offset the costs of this program, so there would be an estimated discounted savings of $16.6 billion in addition to a savings of 16.8 million life-years. Thus, when reductions in multiple risk factors were modeled, populationwide programs appeared to be likely to produce good value for resources expended.

Non-CHD Mortality

When reduced serum cholesterol was considered as a risk factor for non-CHD death, the costs per year of life saved for interventions that reduced serum cholesterol increased (Table 1). Nevertheless, the benefits of cholesterol reduction for avoidance of coronary deaths outweighed any assumed increase in non-CHD deaths. Furthermore, for cholesterol reductions of ≥2%, at the costs of the Stanford Five City Project the ratio was ≤$15 200, even if the program had no effect on other risk factors. If programs also have a beneficial effect on risk factors other than cholesterol, the cost-effectiveness ratios were very favorable (Table 1).

Impact on HDL and LDL

If it is assumed that populationwide programs would yield proportional reductions in both HDL and LDL cholesterol, the estimated costs per year of life saved remained similar. For example, cost-effectiveness ratios varied from $5200 to $5400 and from $6300 to $6500 per year of life saved for reductions of 1% to 4% in both HDL and LDL cholesterol when the annual per-person costs of the program were estimated at $4.95 and $8.28, respectively.

Quality-Adjusted Life Years

Quality of life adjustments were made for persons with a history of CHD according to whether angina or congestive heart failure was present using time trade-off utility data from the Acute Myocardial Infarction (AMI) Patient Outcome Research Team (PORT) and the Beaver Dam Health Outcomes Study.³³ Using costs per quality-adjusted year of life saved, the cost-effectiveness ratios improved by ≈10%. For example, when a 2% reduction in serum cholesterol was modeled, the cost-effectiveness ratio dropped from $3200 to $2900 and from $38 500 to $34 800 for interventions costing $4.95 and $16.55 per year, respectively.

Discount Rate

When a 3% discount rate was used rather than the baseline rate of 5%, the cost per year of life saved for all interventions decreased modestly (Table 1).

Induced Costs

The baseline analysis considered only the direct costs of the intervention program. Because intervention programs may increase physician visits and testing, thereby increasing induced costs, we identified thresholds for the annual cost of the program at which the program would be cost saving or have a cost per year of life saved of <$50 000 (Table 2). Because the upper thresholds were well above the estimated costs of the Stanford Five-Community ($4.95 per person per year) and North Karelia studies (≈$8.66 per person per year), populationwide programs with serum cholesterol reductions of ≥2% were estimated to remain reasonably cost effective even if associated with total program and induced costs of ≤$20 per person per year.

Model Coefficients

In simulations, the coefficients associated with serum cholesterol were either raised or lowered, thereby simulating the impact that serum cholesterol would have on the results if it were a more important or less important risk factor than originally modeled (Table 3). If cholesterol reductions were ≥2% or the annual per-capita cost of the program was <$16.55, programs would be reasonably cost effective, with ratios remaining at <$50 000 per year of life saved regardless of the coefficients used.

Discussion

Our analyses suggest that educational interventions to lower serum cholesterol are likely to be reasonably cost effective and possibly cost saving over a broad range of assumptions, especially if total serum cholesterol is reduced by ≥2%. The relative attractiveness of populationwide cholesterol interventions has been noted^16133435 and was emphasized in a previous analysis using the Coronary Heart Disease Policy Model.⁵

Americans are increasingly recognizing the potential for dietary changes to lower blood cholesterol levels. Serum cholesterol levels declined by an average of ≈3.5 mg/dL (0.09 mmol/L) from 1973 through 1974 to 1980 through 1982 in Minnesota.³⁶ Between 1980 through 1982 and 1985 through 1987, a further decrease in age-adjusted mean serum total cholesterol levels of ≈5 mg/dL (0.13 mmol/L) in men and ≈5.5 mg/dL (0.15 mmol/L) in women was reported in Minnesota³⁷ ; from 1985 through 1987 to 1990 through 1992, additional declines of ≈5.1 mg/dL (0.14 mmol/L) and ≈3.8 mg/dL (0.1 mmol/L) were reported in men and women, respectively.³⁸ These changes in Minnesota between 1973 through 1974 and 1990 through 1992 would be estimated to achieve more than half the reduction in CHD that would be achieved through targeted programs that would be 100% successful in lowering all elevated cholesterol levels to a target of 240 to 250 mg/dL (6.21 to 6.47 mmol/L).⁵ Reductions of ≈7 mg/dL (0.18 mmol/L) were found in the Framingham Heart Study when two cohorts of individuals who were separated by an interval of 20 years were compared, and it was estimated that reductions in risk factors played a substantial role in the reduction in coronary mortality.³⁹ Using assumptions and analytic models very different from ours, Kristiansen et al¹³ also suggested that populationwide programs would be favorable from a cost-effectiveness standpoint, but their analysis was limited to men age 40 to 59.

As the trend toward lower cholesterol levels continues, it is possible that a populationwide intervention program would be less effective than those reported to date, which would lead to somewhat less favorable cost-effectiveness ratios. However, the incremental impact of communitywide intervention programs has been modest but real, with incremental reductions generally in the 2% to 3% range in adults, which would translate into a ≈5 mg/dL (0.13 mmol/L) reduction.¹⁴¹⁶ Interventions in school-age children have also been effective. Such programs resulted in a 2% reduction in cholesterol levels in Finnish seventh- and eighth-grade students⁴⁰ ; a 4% reduction in Norwegian fifth- to seventh-grade students⁴¹ ; an 8.5 mg/dL (0.22 mmol/L) reduction in Westchester County, NY, fourth- to eighth-grade students; and a 5 mg/dL (0.13 mmol/L) reduction in Bronx, NY, fourth- to eighth-grade students.⁴² Although these are relatively modest reductions in cholesterol, the low per-person cost of populationwide programs makes them attractive from a cost-effectiveness standpoint. Even with the use of very high estimates of the direct and induced costs of communitywide programs, the cost per year of life saved compares favorably with other medical interventions.¹²

Medications to reduce cholesterol levels aggressively may be very cost effective when used as secondary prevention in persons with preexisting CHD,⁷ but the aggressive use of medications for primary prevention has generally not been projected to have a favorable cost-effectiveness ratio, except in persons who are at high risk as based on age, sex, and other risk factors.^789101112 An exception would be the use of niacin, which is substantially less expensive than other available medications.

It should be emphasized that small populationwide changes in cholesterol levels may have relatively small effects on the risk of coronary events in an individual patient⁴³ but in aggregate have probably contributed to as much as one third of the US decline in CHD mortality since the mid-1960s.⁴⁴ Reductions in fat intake could defer ≈2% of the deaths in US adults each year and increase average life expectancy by 3 to 4 months,⁴⁵ which is approximately one tenth of the potential life expectancy gain if CHD were totally eliminated.⁴⁶

Although it is difficult to estimate the costs of a nationwide program, the North Karelia and Stanford Three-Community Study¹⁴³⁰ preceded the era of aggressive medical treatment for elevated cholesterol levels and, after correction for inflation, appear to be reasonable cost estimates. As shown in our sensitivity analyses, even a substantial per-person induced cost of additional medical evaluation and/or treatment would not alter our basic results, even if these induced costs had no incremental impact on cholesterol levels.

Our baseline analysis did not include any adverse side effects of cholesterol reduction because none have been shown in the two most recent large-scale randomized trials.²¹²² Even if there are such adverse effects, our estimated cost-effectiveness ratios remained very reasonable. We also did not include any costs associated with following a diet (eg, a possible higher cost of low-cholesterol foods) or any potential increase in non-CHD medical costs incurred by individuals who would be projected to live longer because of lower cholesterol levels. However, we also did not include any potential benefits of cholesterol reduction for other diseases, such as peripheral vascular disease. Our analyses cannot be extrapolated to other populations if factors not considered in our analyses would substantially affect the risks associated with serum cholesterol.

As shown in our sensitivity analyses, if cholesterol reduction programs also have an effect in individuals with preexisting heart disease, the programs become substantially more cost effective. Furthermore, educational programs need not be limited to cholesterol information, and our sensitivity analyses demonstrate the remarkable cost-effectiveness of programs that may have an impact on several risk factors simultaneously.

Any populationwide benefits of cholesterol reduction would be substantially blunted if dietary changes reduced HDL cholesterol as well as LDL cholesterol. However, the cost-effectiveness of populationwide programs should remain favorable even with symmetrical proportional reductions in both HDL and LDL cholesterol levels.

There is no reason why the United States must choose between a purely targeted or a purely populationwide approach to reduce serum cholesterol levels. Populationwide approaches are likely to have favorable cost-effectiveness ratios and should be part of any comprehensive US program to reduce CHD.¹³⁵

• Download figure
• Download PowerPoint

Footnotes