Abstract
OBJECTIVES: In addition to researcher-designed sampling biases, population-representative surveys for biomarker measurement of STIs often have substantial missingness due to non-contact, non-consent and other study-implementation issues. STI prevalence estimates may be biased if this missingness is related to STI risk. We investigated how accounting for sampling, interview non-response and non-provision of biological samples affects prevalence estimates in the third National Survey of Sexual Attitudes and Lifestyles (Natsal-3). METHODS: Natsal-3 was a multistage, clustered and stratified probability sample of 16-74 year-olds conducted between 2010 and 2012. Individuals were sampled from all private residential addresses in Britain; respondents aged 16-44 were further sampled to provide a urine specimen based on characteristics including self-reported sexual behaviours. We generated prevalence estimates and confidence intervals for six STIs in five stages: first without accounting for sampling or non-response, then applying inverse-probability weights cumulatively accounting for interview sampling, interview non-response, urine sampling and urine non-response. RESULTS: Interview non-completion occurred for 42.3% of interview-sampled individuals; urine non-completion occurred for 43.5% of urine-sampled individuals. Interview-sampled individuals, interview respondents, those selected for urine samples and those providing urine samples were each in turn slightly more at-risk for most STIs, leading to lower prevalence estimates after incorporating each set of weights. Researcher-controlled sampling had more impact than respondent-controlled response. CONCLUSIONS: Accounting for both sampling structures and willingness to interview or provide urine specimens can affect national STI prevalence estimates. Using both types of weights, as was done in Natsal-3, is important in reporting on population-based biomarker surveys.