Measurement of stillbirths and neonatal deaths in standardised population-based surveys

BACKGROUND: Worldwide, 4.4 million stillbirths and neonatal deaths (SB&NND) are estimated to occur annually. Household surveys, notably the Demographic and Health Surveys (DHS), are an important source of SB&NND data. This PhD aimed to review the DHS’s evolution for SB&NND data capture and compare the full birth history with additional questions on pregnancy losses (FBH+) and full pregnancy history (FPH) approaches in terms of data quality and potential measurement errors. METHODS: A literature and programmatic review of the DHS programme and HDSS pregnancy surveillance system was conducted. A randomised comparison of FBH+ versus FPH modules (the EveryNewborn-INDEPTH (EN-INDEPTH) survey) was undertaken in five health and demographic surveillance sites (HDSS). Reported stillbirth rates (SBR) and neonatal mortality rates (NMR); time for completion of survey modules; evidence for heterogeneity between sites; patterns of corrections between question types and structures, and modules were assessed. Survey data were compared to and HDSS data in four sites. RESULTS: Both FPH and FBH+ modules have been used within the DHS programme, but there is limited evidence concerning their accuracy for SB&NND. The FBH+ was the core model questionnaire between DHS-phases I-VII. A total of 69,176 women consented. 34,805 (50·3%) were randomised to FBH+ and 34,371 (49·7%) to FPH. There was little difference between the average time to administer questions in FBH+ (9.1 minutes) and FPH (10.5 minutes). The SBR was 15.2/1000 and 17.4/1000 total births for FBH+ and FPH, respectively. SBR was 21% (95% CI (-10% - 62%)) higher in FPH than in FBH+ with strong evidence of heterogeneity across the sites (I-squared=80·9% (p<0.001)). The NMR was similar in FPH (25.1/1000 births) and FBH+ (25.4/1000 births) with no evidence of heterogeneity between the sites (I-squared=0.0% (p=0.48)). Corrections were similar by survey module and occurred in 84% of survey interviews. Single corrections were the most common, multiple-select, and free-text questions increased response time by two minutes on average and had the most corrections. Pregnancy surveillance systems differed between sites (frequency of surveillance visits, main respondent reporting pregnancy outcomes, pregnancy testing, and data capture and surveillance modes). In three of the four sites, both arms of the EN-INDEPTH survey reported more pregnancies than the HDSS Matlab being the exception. Overall, the survey data produced higher estimated SBRs (FBH+: RR=1.13, 95%CI (0.79, 1.63), p=0.519; FPH: RR=1.20, 95%CI (0.76, 1.90), p=0.444) and NMRs (FBH+: RR=1.19, 95%CI (0.98, 1.43), p=0.071; FPH: RR=1.15, 95%CI (1.01, 1.30), p=0.030) than the HDSS data. In Matlab, the HDSS recorded more stillbirths and miscarriages than the survey in the FPH arm only. In Kintampo, the HDSS recorded more stillbirths than the survey, even though the HDSS recorded fewer pregnancies. CONCLUSIONS: Evidence from this PhD of improved capture of stillbirths using the FPH influenced the DHS’s switch to the FPH module in its eighth phase; however, limited evidence concerning SB&NND data accuracy and quality in standardised surveys is available. Therefore, additional efforts towards improved survey implementation of the FPH approach, including training, interviewer prompts, translations, and developing and testing standard data quality criteria for SB&NND in surveys and routine data, are warranted.