| Summary: | Abstract Background Monitoring abortion rates is highly relevant for demographic and public health considerations, yet its reliable estimation is fraught with uncertainty due to lack of complete national health facility service statistics and bias in self-reported survey data. In this study, we aim to test the confidante methodology for estimating abortion incidence rates in Nigeria, Cote d’Ivoire, and Rajasthan, India, and develop methods to adjust for violations of assumptions. Methods In population-based surveys in each setting, female respondents of reproductive age reported separately on their two closest confidantes’ experience with abortion, in addition to reporting about their own experiences. We used descriptive analyses and design-based F tests to test for violations of method assumptions. Using post hoc analytical techniques, we corrected for biases in the confidante sample to improve the validity and precision of the abortion incidence estimates produced from these data. Results Results indicate incomplete transmission of confidante abortion knowledge, a biased confidante sample, but reduced social desirability bias when reporting on confidantes' abortion incidences once adjust for assumption violations. The extent to which the assumptions were met differed across the three contexts. The respondent 1-year pregnancy removal rate was 18.7 (95% confidence interval (CI) 14.9–22.5) abortions per 1000 women of reproductive age in Nigeria, 18.8 (95% CI 11.8–25.8) in Cote d’Ivoire, and 7.0 (95% CI 4.6–9.5) in India. The 1-year adjusted abortion incidence rates for the first confidantes were 35.1 (95% CI 31.1–39.1) in Nigeria, 31.5 (95% CI 24.8–38.1) in Cote d’Ivoire, and 15.2 (95% CI 6.1–24.4) in Rajasthan, India. Confidante two’s rates were closer to confidante one incidences than respondent incidences. The adjusted confidante one and two incidence estimates were significantly higher than respondent incidences in all three countries. Conclusions Findings suggest that the confidante approach may present an opportunity to address some abortion-related data deficiencies but require modeling approaches to correct for biases due to violations of social network-based method assumptions. The performance of these methodologies varied based on geographical and social context, indicating that performance may be better in settings where abortion is legally and socially restricted.
|
|---|
