DO NUMBERS MAKE A DIFFERENCE?

Purpose: Patients frequently overweigh the risks associated with rare adverse events (AEs). As a result, significant efforts have been made to determine how best to present numerical information when describing the probabilities of rare AEs. What is unclear, however, is whether precise numeric estimates influence decision making or behavioral intentions.

Methods: We administered a survey to college students to examine the influence of probabilities and risk perceptions (worry, riskiness and gist evaluations) on subjects’ stated likelihood of starting a medication. We first described the impact of rheumatoid arthritis and subsequently asked subjects to imagine themselves as a patient with this disease in a clinical encounter during which a physician described a new treatment option to them. Route of administration, benefit, and cost were held constant. We varied current health state (able versus unable to maintain current level of activity and responsibilities), type of AE (pneumonia versus cancer) and probability of the AE (1 in 100, 1 in 1,000, 1 in 10,000 and 1 in 100,000). Each subject responded to a single, randomly-assigned scenario. Linear regression models were constructed to predict likelihood of starting the medication (measured on a 5-point scale) using status quo, type of AE, the probability of the AE and risk perceptions (worry, riskiness, and gist evaluations) after adjusting for age, gender, ethnicity, and numeracy. Levels of gist evaluations and ethnicity were treated as dummy variables.

Results: 415 subjects completed the survey. Their mean (SD) age was 19.8 (1.8). 72.5% were woman, and 55.7% were Caucasian, 5.5% Black and 25.7% Asian. The mean (SD) numeracy score was 0.74 (0.20). Health state, type of AE, probability, worry, riskiness, and gist evaluations were associated with likelihood of taking the medication when evaluated separately. In the full model (containing all predictors and covariates), current health state and all three risk perceptions remained significantly associated with likelihood of taking the medication, while numeric probability was not (See Table).

Conclusion: Risk perceptions predict subjects’ willingness to take medication, while probabilistic information does not. The results suggest that decision support must extend beyond presentation of probabilistic information in order to ensure informed choice.