Drug product dispensing and estimates of use in a general population survey as a signal detection problem

Purpose Understanding potential bias due to rarity of the outcome is important when monitoring newly approved drugs and drugs with low availability to the general public. Although there is an increasing use of online surveys to investigate health outcomes, the limits of inference due to drug availability have not been studied. The goal of this study was to quantify the relationship between dispensing of prescription drugs and estimates of use in an online general population survey. Methods An online repeated, cross‐sectional survey from 2018 to 2020 was used to estimate the number of adults in the United States who used prescription drugs in the general population and compared to estimated number of prescriptions dispensed over an equivalent time period. Joinpoint regression was used to quantify thresholds. A sample of respondents was retested to estimate reliability statistics. Results A model with a single threshold was the best fit, with the estimated threshold of 565 000 (95% CI: 9500‐11 600 000) prescriptions dispensed per year. Above the threshold, there was a significant association between dispensing and estimates (p