Assessing induced abortion underreporting in restrictive settings using prospective morbidity surveys from Kenya, Liberia, and Sierra Leone.

Accurate estimation of induced abortion rates in legally restrictive contexts remains a challenge because fear of prosecution and stigma hinder truthful reporting among women. We examined induced abortion reporting accuracy by assessing the agreement between women's self-reports of induced abortion and provider diagnosis of women's abortions. We used data from three prospective morbidity surveys conducted in health facilities in Kenya (surveyed in 2022), Liberia (surveyed in 2021), and Sierra Leone (surveyed in 2021). The data were collected from post-abortion care (PAC) patients and healthcare providers in health facilities across Liberia (137 facilities), Sierra Leone (294 facilities), and Kenya (74 facilities). The data collectors were trained enumerators who were health providers and mostly females. Data analysis involved Kappa statistics to assess the agreement between PAC patients' self-reports and provider diagnosis regarding whether the abortion for which women sought PAC were induced or spontaneous, accounting for interviewer and respondent socio-demographic characteristics. Overall, 1888 pairs of PAC patient and provider prospective morbidity survey interviews were completed: 965 (Kenya), 401 (Liberia), and 522 (Sierra Leone). Fewer women (26.2% in Liberia, 28.5% in Kenya, and 28.7% in Sierra Leone) self-reported induced abortions compared to the provider diagnoses (Kenya, 42.3%; 41.3% in Liberia; 43.5% in Sierra Leone). Across the three countries, there was a 13.8-15.2% difference in the proportions of induced abortions based on women's self-reports and provider diagnoses. In cases where providers reported induced abortions while patients indicated miscarriages, 23% of such cases had clinical evidence of induction (e.g., presence of foreign bodies in the genital tract or signs of trauma in the cervix), whereas 77% did not have any recorded symptoms. Interviews conducted by males showed substantial to almost perfect agreement (κ = 0.73). The level of agreement between women's self-report and provider diagnosis was substantially greater among women with a secondary or tertiary education compared to those with primary education (κ = 0.68 vs. κ = 0.67). The agreement levels did not vary significantly by age and marital status. Direct methods for estimating the incidence of induced abortion are unlikely to generate accurate data because women underreport, and providers may misdiagnose induced abortions. Researchers should acknowledge the limitations of direct abortion estimation methods and carefully recruit and train interviewers to minimize biases and enhance reporting accuracy. We emphasize combining indirect methods with direct methods to improve the reliability and precision of induced abortion estimates.