Using expert opinion to quantify unmeasured confounding bias parameters

Objective: To develop and apply a method to quantify bias parameters in the case example of the association between alcohol use and HIV-serodiscordant condomless anal sex with potential confounding by sensation seeking among men who have sex with men (MSM), using expert opinion as an external data source. Methods: Through an online survey, we sought the input of 41 epidemiologist and behavioural scientists to quantify six parameters in the population of MSM: the proportion of high sensation seeking among heavy-drinking MSM, the proportion of sensation seeking among low-level drinking MSM, and the risk ratio (RR) of the association between sensation seeking and condomless anal sex, for HIV-positive and HIV-negative MSM. Results: Eleven experts responded. For HIV-positive heavy drinkers, the proportion of high sensation seeking was 53.6% (beta distribution [α=5.50, β=4.78]), and 41.1% (beta distribution [α=3.10, β=4.46]) in HIV-negative heavy drinkers. In HIV-positive low-level alcohol drinkers, high sensation seeking was 26.9% (beta distribution [α=1.81, β=4.92]), similar to high sensation seeking among HIV-negative low-level alcohol drinkers (25.3%) (beta distribution [α=2.00, β=5.89]). The lnRR for the association between sensation seeking and condomless anal sex was ln(2.4) (normal distribution [μ=0.889, σ=0.438]) in HIV-positive and ln(1.5) (normal distribution [μ=0.625, σ=0.391]) in HIV-negative MSM. Conclusion: Expert opinion can be a simple and efficient method for deriving bias parameters to quantify and adjust for hypothesized confounding. In this test case, expert opinion confirmed sensation seeking as a confounder for the effect of alcohol on condomless anal sex and provided the parameters necessary for probabilistic bias analysis.