Journal Article

Sample-Level and Individual-Level Risk-Tolerance Estimates from a DCE and a TT Exercise

Gonzalez, Juan Marcos, Sutphin, Jessie, Reed, Shelby, Wallace, Matthew, Secemsky, Eric A, Corriere, Matthew A, Johnson, F Reed, Yang, Jui-Chen

2024

Overview

Background Quantitative-estimates of risk tolerance can be useful for both regulatory and individualized clinical decision-making. However, little is known about how sample-level and individual-level measures of risk tolerance compare across preference-elicitation methods. Using a survey instrument designed for this purpose, we evaluated the convergent validity of sample-level estimates of maximum- acceptable risk (MAR) from a discrete-choice experiment (DCE) and a threshold technique (TT) exercise. Methods Patients with lower-limb intermittent claudication (IC) completed an online survey designed to elicit benefit risk tradeoffs between devices with varying risks of a repeat revascularization procedure and risks of death, both at 2 and 5 years. A 292 factorial design randomized patients to complete the DCE or TT first and to risk levels shown with or without icon arrays. The fixed level of benefit offered in the TT exercise was a reduction in the risk of a repeat procedure from 30% to 10% by 2 years, and from 40% to 30% by 5 years. Risk of death by 5 years with both decide options began at 8%, and increased by 2 percentage points (up to 20%) for the more effective device until the patient chose the less effective device. The TT directly provided lower bound risk-tolerance values. Interval regression also was applied to TT data to estimate sample-level MARs. MARs from the DCE were calculated with a random-parameters logit (RPL) model using the sample level distributions from benefits and risks of devices from which individual-level estimates were derived and truncated at 8% and 20%. Results Seven US medical centres recruited 272 patients with IC to complete the survey. Mean age was 70 years (SD = 10.1), 68% were male, and 92% were White. Sample-level MARs from the TT were 12.9% (95% CI 12.413.04) using direct responses and 14.1% (95% CI 13.514.6) with interval regression compared to 12.7% (95%CI 11.913.6) from the DCE. Individual-level MARs different between the ITT and DCE by less than 2 percentage-points for 53% of patients. The mean absolute difference between patients DCE and TT MARs was 2.5 (SD = 2.4). Results were consistent across versions with or without icon arrays. Conclusions Group-level mean MARs from the DCE and TT (direct response) were very similar. At the individual level, however, differences in MARs between the DCE and TT were observed. Uncertainty associated with individual-level estimates from the RPL mat contribute to individual-level differences.

• Publisher: Springer Nature B.V
• Subject: Estimates