Using Serum Cholesterol To Identify High Risk and Stimulate Behavior Change: Will It Work?

Abstract

If the measurement of serum cholesterol is to be used to identify a sub-population at high risk of subsequent disease and to stimulate behavior change in this group to lower serum cholesterol, the test must be able to both discriminate and motivate the group. The ability of the test to motivate has been documented in a randomized trial. This paper therefore tests, in a cohort of United States men ages 40-59 at entry, the ability of serum cholesterol to discriminate between individuals who would and would not die from coronary heart disease. While risk of death increased with increasing serum cholesterol, the values for men who developed coronary heart disease overlapped the values of the men who did not develop the disease. Overall test accuracy declined from 92% correctly classified when 5% of the population was defined as “at high risk” to 55% when 50% of the population was defined as “at high risk”. On a receiver operating curve, defining 5% of the men as “high risk” yielded a false positive rate of 5% and a true positive rate of 8%. Increasing the proportion of men defined as “at high risk” to 50% increased the true positive rate to 75% but also increased the false positive rate to over 50%. Monte Carlo simulation demonstrated that lowering mean serum cholesterol 0.78 mmol/l in the entire population would lower deaths from coronary heart disease by 28%. The same effect could be achieved by lowering the serum cholesterol of all people in the top 20% of the distribution to 4.66 mmol/l. Lowering the average total serum cholesterol by 0.78 mmol/l and the serum cholesterol among the top 5% to 4.66 mmol/l could be expected to lower death rates from coronary heart disease by 33%. The estimated per capita cost of achieving a 28% reduction in deaths from coronary heart disease using the population approach is $20 per person per year. Using pharmacological therapy for the top 20% of the population to achieve the same approach would generate a per capita cost of $400. A combined strategy to treat the top 5% with drug while lowering the average total serum cholesterol by 0.78 mmol/l would cost $120 per person per year. While a pure population-based approach is most cost-effective, medical ethics probably requires that the high risk are treated with drug when they are identified. If may also be true that treating the high risk with drug may increase the salience of dietary change for the remainder of the population.

Key Words:

• coronary heart disease
• individualized logistic regression
• prevention
• intervention
• Monte Carlo simulation
• epidemiological model
• serum cholesterol.