Menu
Menu
Menu
Menu
Introduction Community and individual sociodemographic characteristics play an important role in child survival. However, a question remains how urbanisation and demographic changes in sub-Saharan Africa affect community-level determinants for child survival. Methods Longitudinal data from the Iganga/Mayuge Health and Demographic Surveillance Site was used to obtain postneonatal under-5 mortality rates between March 2005 and February 2015 in periurban and rural areas separately. Multilevel survival analysis models were used to identify factors associated with mortality. Results There were 43 043 postneonatal under-5 children contributing to 116 385 person years of observation, among whom 1737 died. Average annual crude mortality incidence rate (IR) differed significantly between periurban and rural areas (9.0 (8.1 to 10.0) per 1000 person-years vs 18.1 (17.1 to 19.0), respectively). In periurban areas, there was evidence for decreasing mortality from IR=11.3 (7.7 to 16.6) in 2006 to IR=4.5 (3.0 to 6.9) in 2015. The mortality fluctuated with no evidence for reduction in rural areas (IR=19.0 (15.8 to 22.8) in 2006; IR=15.5 (13.0 to 18.6) in 2015). BCG vaccination was associated with reduced mortality in periurban and rural areas (adjusted rate ratio (aRR)=0.45; 95% CI 0.30 to 0.67 and aRR=0.56; 95% CI 0.41 to 0.76, respectively). Maternal education level within the community was associated with reduced mortality in both periurban and rural sites (aRR=0.83; 95% CI 0.70 to 0.99; aRR=0.90; 95% CI 0.81 to 0.99). The proportion of households in the poorest quintile within the community was associated with mortality in rural areas only (aRR=1.08; 95% CI 1.00 to 1.18). In rural areas, a large disparity existed between the least poor and the poorest (aRR=0.50; 95% CI 0.27 to 0.92). Conclusion We found evidence for a mortality decline in peri-urban but not rural areas. Investments in the known key health (eg, vaccination) and socio-economic interventions (education, and economic development) continue to be crucial for mortality declines. Focused strategies to eliminate the disparity between wealth quintiles are also warranted. There may be equitable access to health services in peri-urban areas but improved metrics of socioeconomic position suitable for peri-urban residents may be needed.
Set up in 2004, the IMHDSS is an open population cohort, located in Iganga and Mayuge districts, Eastern Uganda, a 2 hours drive east of the capital Kampala. Since the baseline census in 2005, all residents in 65 villages within a clearly demarcated area have been prospectively followed up at biannual censuses, during which an adult member of each household is interviewed to collect information about births, deaths and migration. In addition, the IMDSS relies on trusted members of the communities ‘village scouts’ who report pregnancies and births to the IMDSS office throughout the year, in order to improve the capture of the key events as they occur. Roughly one-third of the residents in the IMHDSS live in urbanised parts of the rural districts and the rest in rural parts. Further details of the surveillance site have been previously described.21 The current study includes all postneonatal under-5 children (ie, >28 days old and <60 months) who were residents in the IMDSS at any time during the 10 years between the 1 March 2005 and 28 February 2015. They were retrospectively entered into the current study either at 28 days old if they had been born to resident women, or at the time they became resident of the IMDSS if they had been born to non-resident women and moved into the surveillance site, and were followed up until their fifth birthday or censored at death or migration. A resident is an individual who has lived in the same location within the IMHDSS for more than 4 months. While immortal time bias is a threat to cohort studies, measures have been put in place to reduce such bias. They include the short interval between biannual update rounds, the recording of pregnancies during update rounds, and the use of key informants village scouts. The entry to the current study at 28 days was chosen to facilitate interpretation because the main causes of deaths during neonatal period and their determinants differ from those during 1–59 months. The first component of a principal component analysis (PCA) was used to calculate the wealth quintiles for periurban and rural areas.22 Data from the socio-economic surveys were used. Variables with more than two categories were recoded into binary variables which were then included in the PCA. Because the socioeconomic surveys are conducted every 3–4 years, the socioeconomic status of the household at the nearest to the time of the entry into the cohort was assigned to each individual. BCG vaccination data had been collected at every or every other update round from 2010. The presence or absence of BCG scar was observed by an interviewer, if the child was present, and the date of vaccination was copied from the child’s vaccination card if available. As BCG vaccinations are normally given at birth or soon after, we assumed that children had received the vaccination by the time they entered into the current study (at 28 days or when they became residents thereafter). Other vaccinations such as measles, polio and DTP were also collected but only BCG was used because it did not rely on the availability of vaccination card alone to collect data in this rural setting. Furthermore, vaccinations normally given later in the infancy period were not considered in the current study in which the outcome of interest may occur before exposure, such as to measles vaccination at 9 or 12 months, in order to ensure that the exposure and outcome relationship was causal. Mother’s education was recoded to a binary variable to indicate the mother completed 7 years of primary education or not. Population density, building structure and facilities (modernised vs unmodernised), and main occupation of residents (ie, trading vs farming) were used by the HDSS team to determine periurban or rural areas. Depending on the location of their residence, periurban or rural was assigned to individuals. If one migrated from rural to periurban area or vice versa during the study period, and lived in the new location for more than 4 months, their residence status also changed. Maternal education in community is the percentage of the mothers educated at least to primary school (year 7) per village. The proportion of households in the poorest quintiles was calculated per village. Overcrowded households in community are the percentage of households with at least four people per sleeping room; households with improved sanitation in community indicate the percentage of households with own flush toilet or vip pit latrine; households with improved water source in community is the percentage of households with tap or piped water, well water on residency and protected spring. Mosquito net ownership in community is the percentage of households who own a mosquito net. All the community-level variables were time-varying covariates to take into consideration changes that may have taken place over the study period of 10 years. All the community-level continuous variables were standardised to have mean zero and unit variance. After describing the study subjects, the incidence mortality rate for each year and for each age band was calculated using the stsplit command and the strate commands in STATA V.13. As BCG vaccination status was available in 5405 children and maternal education in 25 062 children only, missing values were imputed using multiple imputation by chained equations (‘mi impute chained’ in STATA V.13) because data were missing in more than one variable.23 Variables that are predictive of missingness as well as the variables correlated with the variables used in the data analysis (birth year, residence village, rural–urban residence, wealth quintiles, child survival status) were included in the imputation model. Twenty imputed datasets were created which were then combined using Rubin’s rule. Then, piecewise exponential mixed effects multilevel survival analysis models, which are equivalent to a Poisson regression model, were used to estimate mortality incidence.24 The analysis methods incorporate the change in exposure status by splitting the exposure time into shorter time scales so that the appropriate exposure value may be assigned to each of the time scales. In addition, the multilevel modelling techniques take into account the hierarchical structure of our data where individual children were nested within villages. The analysis was conducted separately for periurban and rural samples to assess whether explanatory variables differ between the two before deciding to combine them. Several models were fitted. Model 0 (empty model) contained no explanatory variable, which provides an estimation of the degree of correlation in mortality that existed at the village level. In the next models (models 1, 2, 3), each of the individual factors was included while adjusting for the age and year. Finally, community factors were included (models 4 and 5). Fixed effects were expressed as rate ratios (RRs). The random effects, that is measures of variations in mortality across communities, were expressed as proportional change in variance (PCV) and general contextual effects were quantified by the median rate ratio (MRR).25 The MRR compares mortality incidence rates between identical children from two randomly selected different clusters. As the IMHDDS is an open cohort surveillance site, recruitment of new participants is key. For this, IMHDSS work particularly closely with community volunteers who identify and report pregnancies and births. For this particular study, patients and the public were not involved in the study design, data analysis or writing of the manuscript.
N/A
