Ashleigh R Tuite1, David N Fisman. 1. Division of Epidemiology, Dalla Lana School of Public Health, University of Toronto, 155 College Street, Toronto, Ontario, Canada.
Abstract
BACKGROUND: Discordant couple studies are frequently used to evaluate preventive interventions for sexually transmitted infections (STI). This study design may be vulnerable to spectrum bias when transmission risk is heterogeneous. METHODS: We used Markov models to assess the effect of heterogeneous transmission risk on the ability to detect effective interventions using a discordant couple study design. We also evaluated the implications that such bias may have for statistical power. Models incorporated potential health states in a population of initially infection-discordant couples, according to infection status with a hypothetical STI and participation in a hypothetical clinical research study. We evaluated the effect of length of discordant relationship at time of study enrollment, the shape of distribution describing transmission risk among couples, and the effect of sex-specific differential transmission probabilities, on model outcomes. RESULTS: The results demonstrate that discordant couple studies are prone to spectrum bias, the degree of which is affected by the shape of the underlying transmission probability density function. CONCLUSIONS: Such bias could lead to unexpected study findings, including gender-specific vaccine effects, and loss of statistical power, making this an important and underrecognized consideration in the design and interpretation of discordant couple studies.
BACKGROUND: Discordant couple studies are frequently used to evaluate preventive interventions for sexually transmitted infections (STI). This study design may be vulnerable to spectrum bias when transmission risk is heterogeneous. METHODS: We used Markov models to assess the effect of heterogeneous transmission risk on the ability to detect effective interventions using a discordant couple study design. We also evaluated the implications that such bias may have for statistical power. Models incorporated potential health states in a population of initially infection-discordant couples, according to infection status with a hypothetical STI and participation in a hypothetical clinical research study. We evaluated the effect of length of discordant relationship at time of study enrollment, the shape of distribution describing transmission risk among couples, and the effect of sex-specific differential transmission probabilities, on model outcomes. RESULTS: The results demonstrate that discordant couple studies are prone to spectrum bias, the degree of which is affected by the shape of the underlying transmission probability density function. CONCLUSIONS: Such bias could lead to unexpected study findings, including gender-specific vaccine effects, and loss of statistical power, making this an important and underrecognized consideration in the design and interpretation of discordant couple studies.