An Illustration of the Impact of Confounding Bias On Economic Evaluations in the Context of Real-World Evaluations
Author(s)
Simon LaRue, MSc1, Mike Paulden, PhD2, Denis Talbot, PhD3, Jason R. Guertin, PhD3;
1Centre de recherche du CHU de Québec-Université Laval, Axe santé des populations et pratiques optimales en santé, Quebec City, QC, Canada, 2University of Alberta, School of Public Health, Edmonton, AB, Canada, 3Université Laval, Département de médecine sociale et préventive, Quebec City, QC, Canada
1Centre de recherche du CHU de Québec-Université Laval, Axe santé des populations et pratiques optimales en santé, Quebec City, QC, Canada, 2University of Alberta, School of Public Health, Edmonton, AB, Canada, 3Université Laval, Département de médecine sociale et préventive, Quebec City, QC, Canada
Presentation Documents
OBJECTIVES: The increasing use of real-world data in economic evaluation raises concerns about the potential for confounding bias. Methods to control for this bias have been developed over recent decades, but there is still much to be understood about how confounding influences economic evaluations. This research aims to illustrate the impact of unadjusted confounding variables in economic evaluations based on non-randomized data.
METHODS: We simulated the costs and effectiveness of two treatments across nine possible confounding effect scenarios (i.e., none [1 scenario], either solely on the cost or the effectiveness component [4 scenarios], and both on the cost and effectiveness components [4 scenarios]). We considered two settings: 1) where the two treatments were equivalent, and 2) where one treatment is more costly but more effective (resulting ICER = $50,000 per unit of effectiveness). Cost-effectiveness planes, cost-effectiveness acceptability curves and expected value of perfect information graphs were plotted over multiple willingness-to-pay thresholds to assess the simulated confounders’ impact on the results.
RESULTS: Our simulations found that confounding bias can have a significant impact on incremental costs and incremental effectiveness estimates. Confounders were shown to cause a shift in the expected value of perfect information and cost-effectiveness acceptability curves. In both of our settings, we found that confounders combination that either reduced the incremental effectiveness and increased the incremental cost or vice versa have the most potential to switch the conclusion of the economic evaluation. Inadequate bias uncertainty can potentially lead to flawed judgments about the cost-effectiveness of a treatment, resulting in incorrect decisions that could have negative consequences.
CONCLUSIONS: This study advances our understanding of the influence of confounders in non-randomized studies. Such an understanding is important to ensure an accurate assessment of the economic value of treatments and to prevent potential losses of population health.
METHODS: We simulated the costs and effectiveness of two treatments across nine possible confounding effect scenarios (i.e., none [1 scenario], either solely on the cost or the effectiveness component [4 scenarios], and both on the cost and effectiveness components [4 scenarios]). We considered two settings: 1) where the two treatments were equivalent, and 2) where one treatment is more costly but more effective (resulting ICER = $50,000 per unit of effectiveness). Cost-effectiveness planes, cost-effectiveness acceptability curves and expected value of perfect information graphs were plotted over multiple willingness-to-pay thresholds to assess the simulated confounders’ impact on the results.
RESULTS: Our simulations found that confounding bias can have a significant impact on incremental costs and incremental effectiveness estimates. Confounders were shown to cause a shift in the expected value of perfect information and cost-effectiveness acceptability curves. In both of our settings, we found that confounders combination that either reduced the incremental effectiveness and increased the incremental cost or vice versa have the most potential to switch the conclusion of the economic evaluation. Inadequate bias uncertainty can potentially lead to flawed judgments about the cost-effectiveness of a treatment, resulting in incorrect decisions that could have negative consequences.
CONCLUSIONS: This study advances our understanding of the influence of confounders in non-randomized studies. Such an understanding is important to ensure an accurate assessment of the economic value of treatments and to prevent potential losses of population health.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
MSR154
Topic
Methodological & Statistical Research
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
No Additional Disease & Conditions/Specialized Treatment Areas