Under 10 mortality patterns, risk factors, and mechanisms in low resource settings of Eastern Uganda: An analysis of event history demographic and verbal social autopsy data

Background Globally, the under-10 years of age mortality has not been comprehensively studied. We applied the life-course perspective in the analysis and interpretation of the event history demographic and verbal autopsy data to examine when and why children die before their 10th birthday. Methods We analysed a decade (2005-2015) of event histories data on 22385 and 1815 verbal autopsies data collected by Iganga-Mayuge HDSS in eastern Uganda. We used the lifetable for mortality estimates and patterns, and Royston-Parmar survival analysis approach for mortality risk factors’ assessment. Results The under-10 and 5-9 years of age mortality probabilities were 129 (95% Confidence Interval [CI] = 123-370) per 1000 live births and 11 (95% CI = 7-26) per 1000 children aged 5-9 years, respectively. The top four causes of new-born mortality and stillbirth were antepartum maternal complications (31%), intrapartum-related causes including birth injury, asphyxia and obstructed labour (25%), Low Birth Weight (LBW) and prematurity (20%), and other unidentified perinatal mortality causes (18%). Malaria, protein deficiency including anaemia, diarrhoea or gastrointestinal, and acute respiratory infections were the major causes of mortality among those aged 0-9 years-contributing 88%, 88% and 46% of all causes of mortality for the post-neonatal, child and 5-9 years of age respectively. 33% of all causes of mortality among those aged 5-9 years was a share of Injuries (22%) and gastrointestinal (11%). Regarding the deterministic pattern, nearly 30% of the new-borns and sick children died without access to formal care. Access to the treatment for the top five morbidities was after 4 days of symptoms’ recognition. The childhood mortality risk factors were LBW, multiple births, having no partner, adolescence age, rural residence, low education level and belonging to a poor household, but their association was stronger among infants. Conclusions We have identified the vulnerable groups at risk of mortality as LBW children, multiple births, rural dwellers, those whose mother are of low socio-economic position, adolescents and unmarried. The differences in causes of mortalities between children aged 0-5 and 5-9 years were noted. These findings suggest for a strong life-course approach in the design and implementation of child health interventions that target pregnant women and children of all ages. The study used 2005–2015 anonymised Iganga-Mayuge Health Demographic and Surveillance Site (HDSS) event histories and Verbal Social Autopsy (VSA) data. The HDSS is in central-eastern Uganda and it collects information on health, household socioeconomic status, and migration. The HDSS covers at least 185 villages in seven sub-counties within the districts of Iganga and Mayuge with at least 100,000 registered population, of which 59% live in rural areas. Details profiling the site have been reported elsewhere [24,25]. These data have not been used for this kind of study before. Table 1 summarises the estimated number of household and population in the two districts. Details about the Iganga-Mayuge HDSS can be found on the site website (http://www.muchap.mak.ac.ug). *Sub-counties covered by the Iganga-Mayuge HDSS. Source: 2014 Uganda National Population and Housing Census report [26]. Each household within the HDSS is geo-referenced, and every household member has a unique identifier. The household registration, pregnancy registration, pregnancy outcome and verbal autopsy forms are the main tools used for data collection. The household registration form is used to capture information on household assets, individual status (education level and marital status) and residence status (migration) twice a year. The HDSS uses community health workers known as the village health team to identify and register pregnancies, and pregnancy/birth outcome using HDSS standard registers. The variables collected in these registers include HDSS household ID, expected date of delivery, maternal age, marital status, birth outcome and date, place of birth and sex of the child. The HDSS employees Field Assistants who are responsible for verifying the pregnancy and birth outcome registers. The HDSS uses WHO/INDEPTH Network Verbal Social Autopsy tools to collect causes of death data for every death that occurs [27]. The tools capture VA data for different age groups of the deceased. These include Neonates (0–28 days), Child (29 days to 14 years), and Adult (15 years and above). The HDSS uses WHO ICD 10 coding guidelines [28] for assigning the causes of death and there is a responsible team of trained physicians who assign the causes of death using the ICD-10 classification of diseases. The verbal autopsy is normally conducted after mourning days–usually 3–4 weeks after the event. The data elements under stillbirth and new-born mortality verbal autopsy tool include places of birth, maternal morbidity’s experience, childbirth status (live or stillbirth), time the baby died after birth, birth weight, gestation period and causes of death. The data elements for the 29 days-15 years mortality’s verbal autopsy tool include the place of treatment before death, the time taken to receive treatment in days, and the causes of death. In each village, there is a community health worker who notifies and reports all community pregnancies, births, and deaths events within two weeks. Thereafter, the reported events are verified by the DHSS Field Assistants to check for residential status and henceforth use a structured respective standard tool for actual data collection. The HDSS has standard operating procedures for administering each tool, training, and data management. Refresher training is always conducted before the actual days of data collection and while in the field each team of 4 Field Assistants is assigned to a supervisor whose role is to ensure that the data collection standards are adhered to. The forms that only bare the supervisor signature are then submitted to the Field Manager, who later submits them to the data management office. The data is managed by a team of data scientist who include two statisticians, two computer scientists, one data entry supervisor and four data entrants who make sure that the data is well entered and cleaned. We use the 22,385-event history demographic data records to estimate the age-specific mortality patterns and risk factors. The residents’ live births, migration out of the DHSS for less than 6 months and in-migration for at least 6 months were considered for analysis. The HDSS residents are those that have stayed within the site for at least 6 months and those that out-migrated are considered residents for a period of fewer than 6 months before returning to the HDSS. The study main outcome is an event (death) that is experienced after “a live” birth by time t (10th birthday). The study covariates are LBW(<2.5 Kgs)–that is captured from the child immunization card for those that delivered at the health facility, maternal age(grouped as <20 years, 20–29 years and 30 years and above), marital status(1 –having or staying with a partner and 0 –having no partner), household wealth index(least poor for indices 4–5, 3 middle poor and 1–2 poorer), maternal education level, child sex, place of residence, and birth category. The data was managed and analysed in STATA v.15. The life table approach was used to estimate age-specific mortality patterns (S1 File). To assess the under-10 mortality risk factors, Royston-Parmar flexible parametric model for survival analysis was performed using stpm2 user-written STATA command [29] with individual-level clustering while controlling for all covariates after multiple imputations. The approach was used because of its flexibility as compared to other survival parametric models such as the exponential and Weibull models [29]. Further, death among children has monotone hazard rate that reduces with time, which is suitable for the Royston-Parmar method. Since the focus was on the hazard rates, the scale category considered was a hazard at 3 degrees of freedom. The 3 degree of freedom is the same as a spline with 2 interior knots [29], which provide better estimates [29]. Further, the hazard probability graphs were generated to assess how the mortality varies across the child’s age for each of the risk factors. The censoring was done at 120 months (10 years) and the analysis was restricted to children who were born alive. The extent and pattern of missing data were scrutinised to guide the modelling strategy. Birth weight, wealth index, and Marital status records were missing among 64%, 22% and 20% of the registered birth records respectively (S1 Table). For all other variables, data were missing for <1% of the study participants (S1 Table). The place of delivery, the time of death, place of residence, education level and maternal age were associated with the birth weight missingness–indicating a possibility of birth weight missingness at random (S2 Table). Similarly, maternal age, marital status and education level were associated with wealth index missingness–indicating the possibility of wealth index missing at random (S2 Table). Marital status missingness was associated with the place of resident and maternal age (S2 Table). Our consideration of missing at random is based on the assumption that the missingness of variables’ fields depends on some of the observable variables [30,31]. Given the determinants of missingness, we ran multiple imputations (m = 100) not only controlling for identified factors that contribute to the likelihood of missingness, but also all other variables (as auxiliary) in the dataset. The multiple imputation method has been indicated as an important approach for minimising the missingness bias [32] and this does not depend on the magnitude of missingness [33,34]. There were modest differences between the imputed and complete discriptive statitics for variables with missing field (S1 Table). We use the VASA data on 844 neonatal deaths including stillbirths and 986 children (1 month-10 years) deaths to understand the mortality mechanism by assessing the causes of death and access to the required services at birth and after birth for sick children. The VASA data were provided as a separate file and one of the limitations was that the data could not be linked to the event histories dataset. Based on the discussion with the HDSS data manager, this limitation arises from the fact that during the collection of verbal autopsy data, the collectors were not assigning the VASA form with the standard HDSS household and individual identifiers, which makes it hard to merge the two. Because of the limitation of linking the VASA data to the event histories data, the cause-specific mortality probabilities and rates could not be estimated. Additionally, the dataset did not include all the variables that could be used to assign the causes of death using the available software algorithm such as InSilicoVA [35] or SmartVA [36] or InterVA [37] and thus, the results only depend on the causes of death that were assigned by the Physicians.