Study area
The study area is Egypt, which is located in North Africa. The GPS coordinates of Egypt are 26° 49′ 13.991″ N 30° 48′ 8.993″ E. The whole area of Egypt is about 1,000,000 km2. However, much of Egypt is desert, only 6.8% of its area is inhabited [14]. Administratively, Egypt is divided into 27 governorates, four of them are urban governorates (Cairo, Alexandria, Port Said, and Suez). Nine of these governorates are located in the Nile Delta (Lower Egypt) (Demietta, Dakahlia, Sharkia, Kalyoubia, Kafr EL Sheikh, Gharbia, Menoufia, Behera, Ismailia), nine are located in the Nile Valley (upper Egypt) (Giza, Beni-Suef, Fayoum, Menia, Asyout, Suhag, Qena, Aswan, Luxor), and the remaining five (Red Sea, ELWadi ELGidid, Matrouh, North Sinai, South Sinai) (frontier governorates) are located on the eastern and western boundaries of Egypt [15].
Data management and population data source
The data sources were census [16], Annual Bulletin of Births and Deaths statistics [17], Income, Expenditure and Consumption survey [18], and Egypt in Figures [14] from Central Agency for Public Mobilization and Statistics (CAPMAS, 2018) and Egypt Demographic and Health Survey Datasets (EDHS 2008,2014) [19, 20]. The website from the Ministry of Planning, Monitoring and Administrative Reform (MPMAR) was used to get the number of heath offices for each governorate [21]. The calculation of the completeness was done by Excel. Spatial analysis was analyzed using the Esri ArcGIS 10.1 software.
Estimating the completeness of death registration
Random-effects models were used to calculate the logit of death registration completeness at the national and sub-national level of Egypt which overcome the limitations of the other traditional methods. The models were developed from 2451 country-years in 110 countries (1970 to 2015) using the Global Burden of Disease 2015 database [8].
The models are as follow:
$$ \mathrm{Logit}(X)=-0.0177\ast \mathrm{RegCDR}\mathrm{sq}+0.6375\ast \mathrm{RegCDR}+-13.89144\ast \left(\%65+\right)+-1.1136\ast \ln \left(5\mathrm{q}0\right)+2.2063\ast \mathrm{Compl}\ 5\mathrm{q}0+-0.0174\ast \mathrm{Year}+29.3677+\upgamma $$
$$ \mathrm{Logit}(X)=-0.0174\ast \mathrm{RegCDR}\mathrm{sq}+0.5957\ast \mathrm{RegCDR}+-12.9528\ast \left(\%65+\right)+-1.1266\ast \ln \left(5\mathrm{q}0\right)+2.0030\ast \mathrm{Compl}\ 5\mathrm{q}0+-0.0188\ast \mathrm{Year}+32.3442+\gamma $$
$$ \mathrm{Logit}(X)=-0.0198\ast \mathrm{RegCDR}\mathrm{sq}+0.6959\ast \mathrm{RegCDR}+-17.4154\ast \left(\%65+\right)+-1.1720\ast \ln \left(5\mathrm{q}0\right)+1.9387\ast \mathrm{Compl}\ 5\mathrm{q}0+-0.0144\ast \mathrm{Year}+23.5542+\upgamma $$
where
RegCDR is the registered crude death rate
RegCDRsq is the square of RegCDR
%65+ is the fraction of the population aged 65 years and over
ln(5q0) is the natural log of the estimated under-five mortality rate
Compl 5q0 is the completeness of the registered 5q0. This is calculated as the 5q0 from registration data divided by the estimated actual level of 5q0.
Year is a calendar year (2017)
γ is the random effect of Egypt (both sexes = 0.0907, male = 0.1982, female = 0.1291)
X is the completeness of registration at all ages
$$ \mathrm{Logit}(X)=\ln \left(\frac{X}{1-X}\right) $$
Estimation of under-five mortality can be done with different methods which require information obtained exclusively from censuses or surveys. One of them is known as the Brass method [22] which needs three detailed information: the number of children ever born, the number of children dead, and the total female population in the reproductive age (15 to 49, usually) [23]. Because of the lack of this information at census 2018, DHS rates R package [24] was used to estimate the under-five mortality rate (U5MR) for each governorate based on the DHS datasets using a reference period of 10 years preceding the survey of Egypt Demography and Health Survey 2008 and 2014 (EDHS 2008, 2014). The estimated U5MR was calculated by averaging the EDHS 2008 and 2014. After that, the estimated U5MR was scaled into the Inter-agency Group for Child Mortality Estimation (IGME) under-five mortality rates in Egypt for 2017. South Sinai had no values for the estimated U5MR, so the value of the adjacent governorate that is North Sinai was used.
Spatial analysis
Hot spot analysis
Moran’s I was used to define the spatial autocorrelation and it ranges from − 1 to + 1. If Moran’s I is near to + 1, strong spatial autocorrelation occurs, which means that values cluster together. On the other hand, weak spatial autocorrelation occurs when Moran’s I is near to − 1, which means dissimilar values occur next to each other. Moran’s I can be converted into a Z score to test the statistical significance of spatial autocorrelation [25].
Getis-Ord Gi* spatial statistics tool was used to detect clusters of both hot spots (high values) and cold spots (low values). A hot spot happens when a high value is surrounded by other features with high values as well, while the cold spot happens when a low value is surrounded by other low values [26].
Modeling spatial relationships
Ordinary least square (OLS) regression and geographically weighted regression (GWR) [27] using the fixed Gaussian model were performed to explore the spatial relation between the completeness of death registration and explanatory variables (poverty, illiteracy, health offices density (number of health offices per 100,000 population)). The corrected Akaike information criterion (AICc) [28] was used in comparison between the two models. If the difference in the AICc values between OLS model and GWR model is more than 3, the GWR model can be considered more appropriate than the OLS model, even though it is more complex [29]. The GWR model has been found superior to the OLS model, so the results of GWR will be only represented (Appendix 1).
