A STARTLING LACK OF TRANSPARENCY: A REVIEW OF RECENTLY PUBLISHED MARKOV SCREENING MODELS

Purpose:   The discount rate can have a substantial impact on cost-effectiveness analyses (CEA), particularly when savings in costs and life-years occur years into the future, such as with disease screening. Most recommendations stress transparency when reporting results by including a range of discount rates, often with an upper bound of 7 to 10 percent. This review investigated the transparency with which the discount rate and results are reported in the literature.

Methods:   A total of 262 published between 2011 and 2014 were identified through a Medline search. We excluded 142 articles not meeting the inclusion criteria (CEA, Markov model, screening study) and carefully extracted information about the discount rate, sensitivity analyses, and other details from the remaining articles (N=120). We only considered information available in the main article, excluding previously published details and supplementary information.

Results:   A lifetime horizon was reported in 83% of the studies. The median value for the base-case discount rate was 3% while the base-case range for costs was 1.5 to 5 percent and for effects was 0 to 5 percent. Only 14% of studies reported the discount rate in the abstract, 28% in a variables table, and 23% in the main results table, roughly half reported in at least one of these three. While nearly all of the studies included deterministic sensitivity analyses, less than half (n=54) included discount rates. Of these 54 studies, only 9 included an upper bound of at least 7 percent (against screening) while nearly all included a discount rate of 0 percent (favor screening). Further, for most of these 54 studies, the discount rate significantly impacted the ICERs and study conclusions.

Conclusions:    The costs and life-years saved for the studies in this review nearly always occur years after screening. Consequently, the discount rate plays a significant role in the magnitude of CEA ratios and study conclusions. However, transparency in these studies is poor, likely for one of two reasons: 1) authors did not realize the importance of the discount rate or 2) the discount rate clearly changed the conclusions of the study and authors decided not to report the findings. We recommend that authors should present their results with a range of discount rates and journal editors must require transparency in CEAs of screening models.