TY - GEN
T1 - Measuring early neonatal mortality in low-income countries
AU - Jensen, Andreas Møller
PY - 2023/3/17
Y1 - 2023/3/17
N2 - During the past 20 years, child mortality has decreased markedly in low-income
countries, especially for the oldest children. Thus, despite declines in neonatal
mortality, the proportion of under-5 mortality occurring in the neonatal period is
estimated to have increased from 40% to 47% worldwide. Accurate data are
needed to monitor progress towards the Sustainable Development Goals and to
identify areas in need of interventions to reduce child mortality. However, complete vital statistics are not available for low-income countries. Therefore, we
currently have to rely on other data sources to estimate child mortality. The
sources include cross-sectional demographic surveys collecting retrospective
mortality information and Health and Demographic Surveillance Systems
(HDSS) following the same population longitudinally through repeated household visits. Both data collection methods and estimation methodologies differ
between data sources which may affect the resulting mortality estimates.Through a series of studies, this thesis explores the association between data collection methods and obtained estimates for early neonatal mortality, neonatal
mortality, and stillbirth rates, in terms of magnitude, precision and timing as well
as the effect of refractory deaths (deaths which cannot be prevented by the intervention under study) on intervention effect estimates.The Bandim Health Project (BHP) runs a rural and an urban HDSS in GuineaBissau, West Africa. Within the BHP’s rural HDSS the magnitude of differences
in mortality and stillbirth estimates was evaluated by comparing estimates derived by two different methods of analysing HDSS data: a) the commonly used
method of assuming that full information on births can be obtained from women
under surveillance, and therefore assume that all their children are under risk
from the date of birth (retrospective analysis) and b) the prospective method of
counting children born after registered pregnancies from birth or date of registration for children registered after birth.The precision and timing in mortality estimates were examined by comparing
reported births and deaths among women who were both part of either the BHP’s
urban or rural HDSS and interviewed in a retrospective demographic survey.
Reported births by the women in the survey were matched to recorded births to
the same women in the HDSS data. Difference in date of birth and age at death was assessed and potential risk factors associated with displacement was investigated.In a modelling study, it was explored how the proportion of refractory deaths
and the distribution over time affected effect estimates and estimated power under different scenarios where also intervention coverage over time and the magnitude of the treatment effect varied. Refractory deaths were treated as competing risk events in the analyses and cause-specific cumulative incidences were
compared.In the assessment of methods to estimate mortality and stillbirth rates within the
HDSS data, it was found that the early neonatal mortality rate was underestimated by 4% (1%-7%) and the stillbirth rate by 9% (7%-10%) by the method
assuming full information compared with the prospective method. A larger underestimation was observed when analyses were limited to data collected at biannual household visits.Comparing the reported date of birth with HDSS data showed that 9% of the
reported survey births differed from the recorded month or year of birth in HDSS
data. The date of birth of children who were dead at time of the interview was
more likely to be displaced compared with children who were alive at time of
the interview. More than 60% of the reported age at dead differed compared with
the recorded age in the HDSS data.Exploring the impact of refractory deaths revealed that the power of the effect
estimates was more affected if the proportion of refractory deaths was high and
that power seemed to increase as the proportion of refractory deaths decreased
over time. While a 1:1 allocation of treatment group resulted in the highest power
of effect estimates, a linear increase in coverage to 100% resulted in the lowest
levels of power. A delayed treatment effect of one day diminished power substantiallyIn conclusion, stillbirth and mortality estimates are affected by the choice of
method applied when HDSS data is analysed. The early neonatal mortality and
stillbirth rates are substantially underestimated when retrospective data is assumed to provide full information on births. Furthermore, when information on
births and child deaths is collected in surveys information on both date of birth
and age at death are less precise than HDSS data. Especially date of birth for
dead children is more likely to be misreported. Age at death is also likely to be
misreported, which could affect estimates of early neonatal mortality. When
evaluating interventions on neonatal mortality, researchers should consider modelling the expected intervention coverage distribution and the potential treatment effect under different scenarios of refractory deaths to qualify the sample size
and power calculations.
AB - During the past 20 years, child mortality has decreased markedly in low-income
countries, especially for the oldest children. Thus, despite declines in neonatal
mortality, the proportion of under-5 mortality occurring in the neonatal period is
estimated to have increased from 40% to 47% worldwide. Accurate data are
needed to monitor progress towards the Sustainable Development Goals and to
identify areas in need of interventions to reduce child mortality. However, complete vital statistics are not available for low-income countries. Therefore, we
currently have to rely on other data sources to estimate child mortality. The
sources include cross-sectional demographic surveys collecting retrospective
mortality information and Health and Demographic Surveillance Systems
(HDSS) following the same population longitudinally through repeated household visits. Both data collection methods and estimation methodologies differ
between data sources which may affect the resulting mortality estimates.Through a series of studies, this thesis explores the association between data collection methods and obtained estimates for early neonatal mortality, neonatal
mortality, and stillbirth rates, in terms of magnitude, precision and timing as well
as the effect of refractory deaths (deaths which cannot be prevented by the intervention under study) on intervention effect estimates.The Bandim Health Project (BHP) runs a rural and an urban HDSS in GuineaBissau, West Africa. Within the BHP’s rural HDSS the magnitude of differences
in mortality and stillbirth estimates was evaluated by comparing estimates derived by two different methods of analysing HDSS data: a) the commonly used
method of assuming that full information on births can be obtained from women
under surveillance, and therefore assume that all their children are under risk
from the date of birth (retrospective analysis) and b) the prospective method of
counting children born after registered pregnancies from birth or date of registration for children registered after birth.The precision and timing in mortality estimates were examined by comparing
reported births and deaths among women who were both part of either the BHP’s
urban or rural HDSS and interviewed in a retrospective demographic survey.
Reported births by the women in the survey were matched to recorded births to
the same women in the HDSS data. Difference in date of birth and age at death was assessed and potential risk factors associated with displacement was investigated.In a modelling study, it was explored how the proportion of refractory deaths
and the distribution over time affected effect estimates and estimated power under different scenarios where also intervention coverage over time and the magnitude of the treatment effect varied. Refractory deaths were treated as competing risk events in the analyses and cause-specific cumulative incidences were
compared.In the assessment of methods to estimate mortality and stillbirth rates within the
HDSS data, it was found that the early neonatal mortality rate was underestimated by 4% (1%-7%) and the stillbirth rate by 9% (7%-10%) by the method
assuming full information compared with the prospective method. A larger underestimation was observed when analyses were limited to data collected at biannual household visits.Comparing the reported date of birth with HDSS data showed that 9% of the
reported survey births differed from the recorded month or year of birth in HDSS
data. The date of birth of children who were dead at time of the interview was
more likely to be displaced compared with children who were alive at time of
the interview. More than 60% of the reported age at dead differed compared with
the recorded age in the HDSS data.Exploring the impact of refractory deaths revealed that the power of the effect
estimates was more affected if the proportion of refractory deaths was high and
that power seemed to increase as the proportion of refractory deaths decreased
over time. While a 1:1 allocation of treatment group resulted in the highest power
of effect estimates, a linear increase in coverage to 100% resulted in the lowest
levels of power. A delayed treatment effect of one day diminished power substantiallyIn conclusion, stillbirth and mortality estimates are affected by the choice of
method applied when HDSS data is analysed. The early neonatal mortality and
stillbirth rates are substantially underestimated when retrospective data is assumed to provide full information on births. Furthermore, when information on
births and child deaths is collected in surveys information on both date of birth
and age at death are less precise than HDSS data. Especially date of birth for
dead children is more likely to be misreported. Age at death is also likely to be
misreported, which could affect estimates of early neonatal mortality. When
evaluating interventions on neonatal mortality, researchers should consider modelling the expected intervention coverage distribution and the potential treatment effect under different scenarios of refractory deaths to qualify the sample size
and power calculations.
U2 - 10.21996/2whh-pj66
DO - 10.21996/2whh-pj66
M3 - Ph.D. thesis
PB - Syddansk Universitet. Det Sundhedsvidenskabelige Fakultet
ER -