This presentation is part of: Poster Session - CEA: Methods and Applications; Health Services Research
THE IMPACT OF FAILURE TO CALIBRATE ON RESULTS IN ECONOMIC EVALUATION
Douglas Coyle, Ottawa Health Research Institute, Clinical Epidemiology, Ottawa, ON, Canada
Purpose: To demonstrate who failure to calibrate economic models can lead to biased estimates of the cost effectiveness of health interventions Methods: The Canadian Economic Model of Osteoporosis has been used in several previous evaluations of osteoporosis treatments. It is fully calibrated in that the model replicates population data for the risk of fracture and mortality. Many previous models of osteoporosis are not calibrated: e.g. the population risk of fractures are weighted by the relative risks of fracture with osteoporosis or previous fracture history. This will lead to an overestimation of the risk of fracture for such groups. Analysis assessed the cost effectiveness of alendronate compared to no therapy for a 75 year old women with previous fracture history with and without calibration for fracture risk and mortality. Analysis identified the optimal age at which treatment with alendronate becomes cost effective. Results: For a 75 year women with previous fracture, the annual probability of fracture without therapy is 1.31% with calibration and 1.58% without calibration. The incremental cost per QALY gained (ICUR) of alendronate is $35 600 when calibrating the model with respect to fracture risks and mortality and $19 400 without. Assuming a QALY was worth $50000, it would be cost effective to treat with alendronate women with previous fracture who were aged 75 and over when the model was calibrated. Without calibration, alendronate would be cost effective for women aged 73 and over. Conclusions: Although recommended, decision models used for economic analysis are often not calibrated to replicate population data. Failure to calibrate models can lead to substantially different estimates of an ICUR and can lead to differences in policy recommendations. Studies based on decision models need to report what means of calibration were undertaken.
