A slide about how to use utility-based signal detection to assess people's understanding of whether they need antibiotics.
METHODS
Stimuli
Hypothetical antibiotic scenarios of respiratory tract infections with varying symptoms and illness duration modelled after the NHS and NICE guidelines
16 scenarios (within-subjects)
• 20% require antibiotics (e.g. pneumonia)
• 80% don't (e.g. bronchitis)
Design & Procedure
3 conditions (between-subjects)
- Control (N = 264)
- Communication (N = 254)
- Communication + CRP (N = 251)
Uncertainty manipulated (NN, UN, UY, YY)
DV = antibiotic expectations (yes or no)
Antibiotic Expectations results: people often expected antibiotics in situations where they weren't needed.
Notably, many people seemed to realize when they didn't need antibiotics.
Pairwise contrasts for Bias and Sensitivity
Using a multilevel Bayesian probit regression, we estimated bias and sensitivity
By analysing the posterior distribution, we estimated the 95% Bayesian credible interval of the difference in the criterion and the sensitivity between the three communication conditions
A difference is significant when its 95% credible interval does not include zero (which was the care for CRP vs. control and Comm vs. control)
And, finally, Andriana Theodoropoulou shared a #signalDetection experiment.
Simple, evidence-based communication reduced inappropriate #antibiotic expectations, especially when combined with diagnostic test results.
Follow Andriana on @researchgate.bsky.social: www.researchgate.net/profile/Andr...