Appropriate Cohorts for Cost-Effectiveness Analysis: To Mix or Not to Mix?

The majority of the best practice guidelines for conducting economic analyses addresses the importance of time horizon and the base case population. A related issue, but one that has not been addressed in detail, is the choice of cohort v. population projections for costs and benefits. Several studies have shown that one can arrive at different conclusions depending on whether one evaluates single cohorts, multiple cohorts, or a dynamic population, including Hoyle and Anderson in this issue of the journal, and generally make claims for one approach being optimal over another. In this editorial, we briefly review these alternative analytic approaches, discussing the implications for decision making, and propose that the issues raised be more formally addressed by future best practice guidelines for economic analyses and modeling. The majority of cost-effectiveness analyses is based on modeling a ‘‘single-aged’’ cohort over a specified time horizon (e.g., 50-year-old women over a lifetime). Sensitivity analyses are often presented separately for other ages, with the implication that an intervention may be cost effective for a certain age range but not for all ages. Recent criticism of the single-aged cohort approach notes that it is not realistic because it does not reflect the impact of an intervention on the entire eligible population that would be affected by the policy decision. Specifically, Dewilde and Anderson advocate for a multiple cohort approach, where the costs and benefits are projected for multiple existing cohorts of various ages, and the overall cost and benefit of an intervention is a weighted average of the cohort-specific results. The conclusion that the multiple cohort approach yields optimal decisions is predicated on the assumption that an intervention would only be considered cost effective if it was cost effective for all potentially eligible individuals. The most common way of approaching this issue is to focus on the decision that is being addressed through the description of the intervention strategies, and use this to identify the cohort that needs to be modeled (cohort of a single age v. cohort of all eligible ages). As an example, consider the strategy of biennial screening for cervical cancer between the ages of 15 and 70. The decision maker may want to know the costs and benefits associated with screening 15-year-old girls for 55 years (a single cohort analysis) to represent the expected outcomes associated with that particular screening program. An alternative framing of the strategy would be to define what would happen if screening were to start tomorrow (in a previously unscreened population), which would include adding ‘‘catch-up’’ screening of women aged 16 to 70 years. Finding a cost-effective result using the former (single cohort) approach would not imply anything about the cost effectiveness of screening in women older than age 15 (i.e., catchup screening). At the same time, finding that screening is not cost effective using the latter (multiple cohort) analysis would certainly not imply that screening in general is not cost effective. An extension of the multiple birth cohort model is a population-based model, in which not only the current eligible population is modeled but all future incident cohorts enter the model for T years (T is defined by the analyst). In this issue of the Journal, Hoyle and Anderson describe the mathematics of conducting a cost-effectiveness analysis using a population-based approach. The prevalent cohorts are those individuals who are eligible for the intervention the year that it is introduced (i.e., the multiple birth cohorts described by Dewilde and Anderson), From the Division of Health Policy and Management, University of Minnesota, School of Public Health, Minneapolis, Minnesota (KMK), and the Section of Public Health and Health Policy, University of Glasgow, Glasgow, UK (EF, AB).