Adjusting for hidden biases in sexual behaviour data: a mechanistic approach

Area of Science:

• Epidemiology
• Mathematical Biology
• Biostatistics

Background:

• Mathematical models for sexually transmitted infections (STIs) rely on accurate estimates of duration in risk states and sexual partnership change rates.
• Aggregate estimates from cross-sectional studies often contain biases (distributional, sampling, censoring, measurement), affecting model inputs.

Approach:

• This study develops bias adjustment methods using data from a 2011 survey of female sex workers in Eswatini.
• Bayesian hierarchical models were constructed to address biases in duration of sex work and sexual partnership rates.

Key Points:

• Failure to account for bias mechanisms can lead to over- or underestimation of duration in sex work by up to twofold.
• Conventional interpretation of sexual partner numbers is biased; unbiased estimators for partnership change rates can be defined by incorporating partnership duration.
• The importance of unbiased estimators increases when survey recall periods and partnership durations are similar.

Conclusions:

• The developed bias adjustment methods are crucial for refining inputs for STI mathematical models.
• The approach and insights are applicable to diverse datasets and research aiming to quantify sexual behavior data accurately.