Deriving causes of child mortality by re–analyzing national verbal autopsy data applying a standardized computer algorithm in Uganda, Rwanda and Ghana

Abstract

BackgroundTo accelerate progress toward the Millennium Development Goal 4, reliable information on causes of child mortality is critical. With more national verbal autopsy (VA) studies becoming available, how to improve consistency of national VA derived child causes of death should be considered for the purpose of global comparison. We aimed to adapt a standardized computer algorithm to re–analyze national child VA studies conducted in Uganda, Rwanda and Ghana recently, and compare our results with those derived from physician review to explore issues surrounding the application of the standardized algorithm in place of physician review.Methods and FindingsWe adapted the standardized computer algorithm considering the disease profile in Uganda, Rwanda and Ghana. We then derived cause–specific mortality fractions applying the adapted algorithm and compared the results with those ascertained by physician review by examining the individual– and population–level agreement. Our results showed that the leading causes of child mortality in Uganda, Rwanda and Ghana were pneumonia (16.5–21.1%) and malaria (16.8–25.6%) among children below five years and intrapartum–related complications (6.4–10.7%) and preterm birth complications (4.5–6.3%) among neonates. The individual level agreement was poor to substantial across causes (kappa statistics: –0.03 to 0.83), with moderate to substantial agreement observed for injury, congenital malformation, preterm birth complications, malaria and measles. At the population level, despite fairly different cause–specific mortality fractions, the ranking of the leading causes was largely similar.ConclusionsThe standardized computer algorithm produced internally consistent distribution of causes of child mortality. The results were also qualitatively comparable to those based on physician review from the perspective of public health policy. The standardized computer algorithm has the advantage of requiring minimal resources from the health care system and represents a promising way to re–analyze national or sub-national VA studies in place of physician review for the purpose of global comparison.

Causes of death and predictors of childhood mortality in Rwanda: a matched case-control study using verbal social autopsy

“…Although the proportion of neonatal deaths in the first week of life in this cohort (75.9%) was similar to previous reports based on vital registration data, [20, 21] our results showed a lower proportion of neonatal asphyxia and prematurity-related deaths than both regional estimates and previous results in Rwanda [22, 23]. These findings may reflect the contribution of programs in these districts targeting survival of premature and full-term infants through improved neonatal resuscitation techniques and immediate newborn care, including introduction of continuous positive airway pressure at the hospital level [24].…”

“…The relative proportion of deaths attributable to pneumonia was consistent with both regional and historical rates [21, 23]. In contrast, a lower proportion of deaths were attributable to diarrhea than previously reported [23], likely due in part to the expansion of community IMCI and efforts to improve quality of facility-based IMCI [27].…”

“…Despite national malaria control efforts, there are higher rates of malaria incidence in the Eastern region of Rwanda [25]; in addition, the country experienced an upsurge in cases of malaria during the time of data collection, attributed to climate changes, pyrethroid resistance, sub-standard insecticide-treated bednets, and inconsistent application of indoor residual spraying [18, 26]. The relative proportion of deaths attributable to pneumonia was consistent with both regional and historical rates [21, 23]. In contrast, a lower proportion of deaths were attributable to diarrhea than previously reported [23], likely due in part to the expansion of community IMCI and efforts to improve quality of facility-based IMCI [27].…”

Validating hierarchical verbal autopsy expert algorithms in a large data set with known causes of death

“…Because the algorithms determine all contributing causes, in the event that more than one cause was identified the primary cause was chosen according to a pre–specified hierarchy. We determined the primary causes of neonatal death utilizing the same algorithms across five hierarchies for neonatal deaths that are currently in use: Arifeen et al [ 22 ], Baqui et al [ 23 ], Kalter et al [ 21 ], Lawn et al [ 24 ], and Liu et al [ 25 ]; and the primary causes of child death (1–59 months of age) utilizing the hierarchies for this age group described by Arifeen et al [ 22 ], Kalter et al [ 21 ], and Liu et al [ 25 ]. Other things being equal, estimating more causes at once will yield lower accuracy than estimating fewer causes [ 26 ].…”

Integrating community-based verbal autopsy into civil registration and vital statistics (CRVS): system-level considerations

“…There is extensive experience in the use of VA in research settings for determining the COD [ 21 – 23 ]. This experience has led to advances in the VA questionnaire design [ 19 , 24 , 25 ], in data capture on mobile devices [ 19 , 26 ], and in the use of computer algorithms for determining and coding the probable COD [ 13 , 20 , 27 – 30 ]. These advances make paper-based and mobile tablet-based VA at community level an increasingly available and effective substitute for physician-certified COD in settings where a significant proportion of deaths occur outside hospitals.…”