Survival analysis of under-five mortality using Cox and frailty models in Ethiopia
- Dawit G. Ayele^{1}Email author,
- Temesgen T. Zewotir^{2} and
- Hemry Mwambi^{2}
DOI: 10.1186/s41043-017-0103-3
© The Author(s). 2017
Received: 1 February 2016
Accepted: 23 May 2017
Published: 2 June 2017
Abstract
Background
The risk of a child dying before reaching 5 years of age is highest in sub-Saharan African countries. But in Ethiopia, under-five mortality rates have shown a substantial decline.
Methods
For this study, the Cox regression model for fixed and time-dependent explanatory variables was studied for under-five mortality in Ethiopia. We adapted survival analysis using the Cox regression model with 2011 Ethiopian Demographic and Health Survey data.
Results
From the results, it was found that under-five children who live in Addis Ababa had a lower hazard (risk) of death (p value = 0.048). This could be as a result of higher health facilities and living standards in Addis Ababa, compared to other regions. Under-five children who lived in rural areas had a higher hazard (risk) of death compared to those living in urban areas. In addition, under-five children who lived in rural areas had 18% (p value = 0.01) more hazard (risk) of death than those living in urban areas. Furthermore, with older mothers, the chance of a child dying before reaching the age of 5 is lower.
Conclusion
The chances of a child dying before reaching the age of 5 are less if the mother does not become pregnant again before the child reaches the age of 5. Therefore, giving birth when older and not becoming pregnant again before the child reaches the age of 5 is one means of reducing under-five mortality.
Keywords
Hazards Interaction effect EDHS Frailty Under-five mortalityBackground
In sub-Saharan African countries, under-five mortality is higher when compared to other countries and eight times higher than WHO European Region [1, 2]. Even though different measures are taken to reduce under-five mortality, under-five mortality is still higher (including Ethiopia) and the rate is above 100 deaths per 1000 live births [3, 4].
Based on the 2000, 2005, and 2011 Ethiopian Demographic and Health Surveys, the risk of under-five mortality shows decline over time. This decline is due to the progressive and consistent implementation of the health interventions since 1960 [5–8]. But the risk is till high in Ethiopia. In addition to under-five mortality, trends of crude death rates also show decreases [9–12].
The pattern of under-five mortality is known to have increasing trend. However, the trends of under-five mortality rate at global level showed decreasing pattern by 53%,. This is the estimated rate of 91 deaths per 1000 live births in 1990 to 43 deaths per 1000 live births in 2015. Based on the records from WHO, average annual rate of reduction for under-five mortality is from 1.8% a year over the period 1990-2000 to 3.9% for 2000-2015. This reduction is not sufficient to meet the MDG 4. [28]. Having this in mind, using survival analysis for lifetime distributions is one of the main difficulties in the investigation of life data. Survival analysis can be used to determine life potentiality presumptions in different times. Comparing the lifetime distributions of various investigated groups [13, 14] is one of the main problems in the analysis of life data. In most cases, the impact of explanatory variables on the lifetime as the dependent variable is modelled using regression models. These models are an important part in survival analysis [15, 16]. Therefore, this kind of model is known as the Cox regression model or the proportional hazards regression model [17]. It is advantageous to use the Cox regression model because it has both nonparametric and parametric parts at the same time: the parametric part being due to the parameter β in the model. But the distribution of failure time is assumed known. It is nonparametric in the sense that λ _{0} is an unspecified function in the form of the baseline undefined hazard function. Therefore, this part of the model is the nonparametric part. Because of this characteristic, the model is more flexible as measurements are always done with error involved. When the response time is not known, the proportional hazards model (PHM) can be carried out. The Cox proportional hazards model, compared to traditional analysis of variance (ANOVA) approaches and event-time models, has several advantages. These advantages are that when individuals are recorded at the tailrace, they can be explicitly included in the modelling, and the covariates may vary through time, therefore allowing passage hazards to change daily or seasonally.
However, this model also has its disadvantages. The main disadvantage is that hazards are calculated using the ranks of covariate values because the models are semi-parametric. For this reason, quantitative differences among treatments cannot be modelled as in traditional regression [18–20]. Therefore, hazard ratios compare the probability of the event occurring within a given time interval for an individual belonging to one group versus another. On the other hand, for continuous predictor variables, it compares for an increase in 1 unit of the variable.
The objective of this study is to investigate the socio-economic, demographic, and geographic predictors of under-five mortality in Ethiopia.
Methods
The Demographic and Health Survey (DHS) is conducted in 2000, 2005, and 2011 in Ethiopia. This survey is a periodic, cross-sectional survey administered at the household level. For this study, data from the 2011 Ethiopian Demographic and Health Survey were used. The survey consisted of 624 selected enumeration areas (EAs). Complete household listings were carried out in each of the 624 EAs. For the survey, a sample of 17,817 households was selected. To estimate at the national level, all data of the survey were weighted and interviews were conducted with 9096 women aged 15–49 and 6033 men aged 15–59. Therefore, the 2011 EDHS sample was designed to provide estimates for the health and demographic variables of interest for Ethiopia as a whole: urban and rural areas of Ethiopia and 11 geographical areas [9–11].
In the model, there are two unknown components, the regression parameter β and the baseline hazard function h _{0}(t). The model component h _{0}(t) is called the baseline hazard function.
The Cox regression model has a key assumption. The assumption is related to proportional hazards. The proportional hazards assumption states that the hazard ratio is constant over time or the hazard for an individual is proportional to the hazard for any other individual.
For proportional hazard function of β’s can be interpreted as time invariant shifters of the hazard function. Because of this property, the result can be interpreted as factors that affect risk, relative to the baseline risk where S _{0}(t) is the essential life function t [20, 21].
where H _{0}(t) is the baseline cumulative hazard function.
The baseline hazard function includes a function of time. In the model, there may be time-dependent explanatory variables, which is any variable whose value can change in time. The value for a time-independent variable remains fixed [20, 22, 23].
where V _{ i } is the grouped-level (household) frailty. The frailties in the model assumed unobservable unit mean and variance θ which is unknown. In each probabilistic sampling unit, which is kebele, households have different values of random effects. Here, V _{ i } ' s reflect variability, and this shows heterogeneity of risks between households. From the model, the value of o frailty reflects that children from the same households are independent [24–26]. Therefore, the variance of the random effects lies between o and α. Therefore, large variance values indicate high heterogeneity between households. It also shows greater correlation between children with in the same households. Therefore, the frailty model follows the Gamma distribution and from the model correct Z-ratios expected [17, 20].
Results
For the 2011 EDHS, the Cox regression analysis was fitted to the data to find factors affecting under-five mortality in Ethiopia. PROC PHREG in SAS 9.3 was used with additional options for model diagnosis like access option. The response variable for this analysis is child survival before the age of 5 and age at death. The covariate/predictors included in the model were social, economic, demographic, and geographic variables. These covariates are region, place of residence, religion, education, current age, mother’s age at first marriage, mother’s age at child’s birth, family size, relationship, husband’s education, working status, marital status, sex of child, pregnancy, time to collect water, source of drinking water, type of cooking facilities, toilet facilities, wealth index, smoking habits, main material of floor, roof, and walls. As well as the fixed effects, cluster/probability sampling unit (PSU) was considered as a random effect in the model.
Type III tests for socio-economic, demographic, and geographic variables
Effect | DF | Chi-square | Pr > ChiSq |
---|---|---|---|
Region | 10 | 32.0529 | 0.0004 |
Type of place of residence | 1 | 4.5863 | 0.0322 |
Religion | 5 | 6.222 | 0.2852 |
Education in single years | 5 | 1.1057 | 0.9536 |
Respondent’s current age | 1 | 206.2717 | <0.0001 |
Age of respondent at 1st birth | 1 | 81.9094 | <0.0001 |
Family size | 1 | 4.151 | 0.0416 |
Husband/partner’s education level | 4 | 0.9479 | 0.9176 |
Current marital status | 1 | 0.2692 | 0.6039 |
Age at 1st sex | 1 | 5.8519 | 0.0156 |
Currently pregnant | 1 | 0.9127 | 0.3394 |
Respondent currently working | 1 | 2.9779 | 0.0844 |
Total children ever born | 1 | 35.69 | <0.0001 |
Sex of child | 1 | 7.5047 | 0.0062 |
Time to get water | 3 | 3.6774 | 0.2985 |
Source of drinking water | 2 | 2.3263 | 0.3125 |
Type of cooking fuel | 3 | 3.7077 | 0.2948 |
Type of toilet facility | 2 | 2.3533 | 0.3083 |
Main floor material | 2 | 3.6333 | 0.1626 |
Main roof material | 3 | 0.3193 | 0.9564 |
Main wall material | 2 | 0.3712 | 0.8306 |
Smoking | 1 | 0.9341 | 0.3338 |
Age of respondent at 1st birth and currently pregnant | 1 | 4.1467 | 0.0417 |
Education in single years and currently pregnant | 1 | 4.6186 | 0.0316 |
Family size and respondent currently working | 1 | 4.2749 | 0.0987 |
Total children ever born and current marital status | 1 | 2.2638 | 0.0389 |
Respondent’s current age and family size | 1 | 5.0619 | 0.0245 |
Age of respondent at 1st birth and family size | 1 | 8.4761 | 0.0036 |
Age of respondent at 1st birth and total children ever born | 1 | 7.0545 | 0.0079 |
Family size and total children ever born | 1 | 14.6694 | 0.0001 |
Supremum test for proportional hazards assumption
Variable | Maximum absolute value | Replications | Seed | p value |
---|---|---|---|---|
Respondent’s current age | 39.9485 | 1000 | 7548 | 0.219 |
Education in single years | 19.7343 | 1000 | 7548 | 0.241 |
Number of household members | 34.8539 | 1000 | 7548 | 0.250 |
Total children ever born | 41.6613 | 1000 | 7548 | 0.089 |
Age of respondent at 1st birth | 52.6433 | 1000 | 7548 | 0.033 |
Age at 1st sex | 43.7919 | 1000 | 7548 | 0.201 |
Estimates from the CPH on the effect of socio-economic, demographic and geographic variables on under-five mortality
Parameters | Estimates | Hazard ratio | CI | p value | |
---|---|---|---|---|---|
Lower | Upper | ||||
Region (Ref. Tigray) | |||||
Addis Ababa | −0.383 | 0.681 | 0.421 | 0.796 | 0.048 |
Affar | −0.296 | 0.744 | 0.564 | 0.841 | 0.001 |
Amhara | −0.04 | 0.961 | 0.285 | 1.087 | 0.539 |
Benishangul-Gumuz | −0.258 | 0.772 | 0.586 | 0.881 | 0.001 |
Dire Dawa | −0.192 | 0.825 | 0.645 | 0.981 | 0.044 |
Gambela | −0.067 | 0.936 | 0.458 | 1.179 | 0.465 |
Harari | 0.007 | 1.007 | 0.047 | 1.085 | 0.947 |
Oromiya | −0.178 | 0.837 | 0.687 | 0.965 | 0.014 |
SNNP | −0.239 | 0.787 | 0.649 | 0.857 | 0.002 |
Somali | −0.377 | 0.686 | 0.611 | 0.749 | <0.0001 |
Type of place of residence (Ref. urban) | |||||
Rural | 0.165 | 1.18 | 1.085 | 1.254 | 0.01 |
Currently pregnant (Ref. no) | |||||
Yes | −0.145 | 0.865 | 0.805 | 0.930 | <0.0001 |
Respondent’s current age | −0.051 | 0.95 | 0.897 | 0.967 | <0.0001 |
Age of respondent at 1st birth | 0.05 | 1.051 | 1.012 | 1.154 | <0.0001 |
Family size | −0.019 | 0.981 | 0.785 | 0.997 | 0.022 |
Age at 1st sex | 0.012 | 1.012 | 1.004 | 1.124 | 0.004 |
Total children ever born | 0.065 | 1.067 | 1.011 | 1.189 | <0.0001 |
Sex of child | |||||
Female | −0.089 | 0.915 | 0.759 | 0.995 | 0.002 |
Age of respondent at 1st birth and currently pregnant (Ref. yes) | |||||
No/do not know | −0.028 | 0.972 | 0.743 | 0.987 | 0.045 |
Education in single years and currently pregnant (Ref. yes) | |||||
No/do not know | −0.254 | 0.776 | 0.574 | 0.857 | 0.0013 |
Family size and respondent currently working (Ref. yes) | |||||
No | 0.1105 | 1.117 | 1.087 | 1.204 | 0.0031 |
Respondent’s current age and family size | −0.005 | 0.995 | 0.847 | 0.998 | 0.0118 |
Age of respondent at 1st birth and family size | 0.007 | 1.007 | 1.002 | 1.106 | 0.0105 |
Age of respondent at 1st birth and total children ever born | −0.003 | 0.997 | 0.785 | 0.999 | 0.0439 |
Family size and total children ever born | 0.025 | 1.025 | 1.005 | 1.158 | <0.0001 |
Frailty (cluster or PSU) | 1.721 | 1.487 | 1.983 | 0.004 |
The proportional hazards regression model for these data with age at first sex as the predictor is found to be 1.012 with C.I. (1.004, 1.124). Therefore, with each yearly increase in age of a mother at first sex, the risk of her child dying before reaching the age of 5 is increased by 1.2%.
In addition to the main effects, there were interaction effects which have an influence on the under-five mortality of Ethiopia. Among these effects, the first one is the interaction between the age of the respondent at first birth and whether currently pregnant. The result is presented in Table 3. From the table, it can be seen that as the age of respondents at first birth increases, the risk of the child dying before the age of 5 decreases, for women who are not pregnant (HR = 0.972, C.I. (0.743, 0.987)). But the risk of a child dying before the age of 5 is lower for women who were not pregnant.
The other interaction effect is between education in single years and whether currently pregnant. The result is presented in Table 3. From the table, it can be seen that as the respondents’ educational level increases, the risk of the child dying before the age of 5 decreases, for both not pregnant women (HR = 0.776, C.I. (0.574, 0.857)). As with the interaction between the age of the respondent at first birth and whether currently pregnant, the risk of a child dying before the age of 5 was lower for women who were not pregnant.
The interaction between family size and respondent’s current work status was found to be significant. The result is presented in Table 3. The result indicates that the risk for a child to not reach the age of 5 increases as family size increases for unemployed women. Moreover, for respondents who are not working, the result is higher than for respondents who are working (HR = 1.117, C.I. (1.087, 1.204)).
The joint effect of a respondent’s current age and family size was one of the interaction effects (Table 1). As the result indicates, the chance of a child dying before reaching age 5 increases as the age of the respondent increases for an increase in family size. Moreover, the chance of a child dying before reaching the age of 5 is highest for maximum family size, followed by median family size and lastly minimum family size.
The effect of the interaction between age of respondent at first birth and family size was found to be significant (Table 1). Unlike the interaction effect between current age and family size, the interaction between age of respondent at first birth and family size shows increase. Therefore, for an increase in respondents age at first birth and family size, the risk of a child to reach age 5 increases by 7% (HR = 1.007, C.I. (1.002, 1.106)). The other two-way significant interaction effects are between age of respondent at first birth and total children ever born and between family size and total children ever born. This result is presented in Table 3.
Discussion and conclusions
The risk of a child dying before reaching 5 years of age is highest in sub-Saharan African countries. The under-five mortality rate is greater than 100 deaths per 1000 live births [3, 4], which is about eight times higher than that in the WHO European Region where the mortality rate is 12 deaths per 1000 live births [27, 28]. In addition, large inequities exist between high-income and low-income countries regarding the deaths of under 5-year-olds. In 2012, the under-five age mortality rate in low-income countries was 82 deaths per 1000 live births, more than 13 times the average rate in high-income countries (6 deaths per 1000 live births). Reducing these inequities across countries and saving more children’s lives by ending preventable child deaths are important priorities [29, 30].
Under-five mortality rates have shown a substantial decline in Ethiopia. This has been achieved since the 1960s implementation of health interventions [5]. Mortality trends can be studied by associating mortality rates for the three 5-year periods for a single survey or from several surveys over time. The recently conducted 2011 Ethiopian Demographic and Health Survey results show a decline in all levels of childhood mortality. Infant mortality has declined by 42% over the 15-year period preceding the survey, from 101 deaths per 1000 live births to 59 deaths per 1000 live births. Similarly, under-five mortality also shows a decline by 47% over the same period, from 166 deaths per 1000 live births to 88 deaths per 1000 live births. Based on the three DHS surveys (2000, 2005, and 2011), mortality trends can also be examined. According to these surveys, infant and under-five mortality rates show a continuous declining trend. Under-five mortalities decreased from 166 deaths per 1000 live births in the 2000 survey to 88 in 2011, while infant mortality decreased from 97 deaths per 1000 live births in the 2000 survey to 59 in the 2011 survey. Furthermore, crude death rates also show substantial declines over the past 10 years [9–12]. As the record shows, infant and under-five mortalities in Ethiopia have continued to decline over the past 25 years. But, in the last 10 years, there has been a more pronounced reduction [11, 31]. Recent study conducted by Ayele, Zewotir, and Mwambi in 2015 supported the reduction of under-five mortality [32]. However, despite these decreases, mortality rates are still high in Ethiopia. Therefore, to identify the socio-economic and demographic factors influencing under-five mortality, survival analysis has been used.
For this study, the nationally representative data from the 2011 Ethiopian Demographic and Health Survey was used. The objective was to investigate the socio-economic, demographic, and geographic predictors of under-five mortality in Ethiopia. The Cox regression analysis technique has been applied to identify the important socio-economic, demographic, and geographic predictors of under-five mortality. A Cox regression model is used when the relative risk values for different levels of variables measured at different levels of socio-economic, demographic, and geographic variables. Even though infant and under-five mortality rates have been decreasing in many parts of sub-Saharan Africa in recent decades, they remain among the highest in the world. In the literature, the decline in mortality is well documented, but it has been difficult to determine the various socio-economic, demographic, and geographic factors associated with this decline.
Therefore, for this study, the Cox regression model was used as it is a procedure which is useful for modelling the time to a specified event, based upon the values of given covariates. For this study, 23 covariates were used to predict the status of under-five mortality. Moreover, the death of a child (status variable) is the dependent variable in Cox regression and is a binary variable. Time variable (age at death) measures the duration to the event defined by the status variable. The covariates (independent variables) for this study contain both the categorical and continuous variables.
The results of this model which was applied to under-five mortality in Ethiopia showed that under some conditions, socio-economic, demographic, and social factors all have a great effect on a child’s survival up to age 5. Additional to this, the Cox regression model determined that the data obtained during the 2011 Ethiopian Demographic and Health Survey is of great importance. From the results of this present study, it can be concluded that children under-five, who live in Addis Ababa, have a lower hazard (risk) of death. Moreover, for older mothers, the chances of the child dying before reaching age 5 are low. In addition, with each year’s increase in the age of a mother at first sex, the risk of a child dying before reaching the age of 5 decreases. For mothers who gave birth at an older age, and for educated mothers, and for women who are not pregnant, the chances of a child reaching age 5 is higher. Moreover, for large households and for mothers who are not working, the chances of a child not reaching age 5 is higher. Therefore, by giving birth at an older age, and not being pregnant again before the child reaches age 5, is one means of reducing under-five mortality.
Declarations
Acknowledgements
We thank, with deep appreciation, ORC macro and Measure DHS for giving us the access for the data file.
Authors’ contributions
DGA acquired the data, performed the analysis, and drafted the manuscript. DGA, TZ, and HM designed the research problem. All authors discussed the results and implications and commented on the manuscript at all stages. All authors contributed extensively to the work presented in this paper. All authors read and approved the final manuscript.
Competing interests
The authors declare that they have no competing interests.
Ethics approval and consent to participate
Ethical clearance for the survey was provided by the EHNRI Review Board, the National Research Ethics Review Committee (NRERC) at the Ministry of Science and Technology, the Institutional Review Board of ICF International, and the CDC.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Authors’ Affiliations
References
- Global Health Observatory (GHO). Reports. 2017. http://www.who.int/gho/publications/en/. Accessed March 2017.
- WHO. Under-five mortality. 2014. http://www.who.int/gho/child_health/mortality/mortality_ under_five_text/en/index.html[cited 2013.Google Scholar
- Adjuik M, et al. "Cause-Specific Mortality Rates in Sub-Saharan Africa and Bangladesh." Bull World Health Organ. 2006;84(3):181-88.Google Scholar
- Misselhorn M, Harttgen K. A multilevel approach to explain child mortality and undernutrition in South Asia and sub-Saharan Africa. In: Proceedings of the German Development Economics Conference, Berlin/Verein für Socialpolitik, Research Committee Development Economics. 2006. No. 20.Google Scholar
- Susuman AS. Child mortality rate in Ethiopia. Iran J Public Health. 2012;41(3):9–19.PubMedPubMed CentralGoogle Scholar
- Ayele DG, Zewotir TT. Childhood mortality spatial distribution in Ethiopia. J Appl Stat. 2016;43(15):2813–28.View ArticleGoogle Scholar
- Ayele DG, Zewotir TT. Comparison of under-five mortality for 2000, 2005 and 2011 surveys in Ethiopia. BMC Public Health. 2016;16(1):930.View ArticlePubMedPubMed CentralGoogle Scholar
- Ayele DG, Zewotir T, Mwambi H. Indirect child mortality estimation technique to identify trends of under-five mortality in Ethiopia. Afr Health Sci. 2016;16(1):18–26.View ArticlePubMedPubMed CentralGoogle Scholar
- CSA, Central Statistics Agency of Ethiopia and ORC Macro. Ethiopia demographic and health survey 2000. Addis Ababa and Calverton, MD: Central Statistics Agency and ORC Macro; 2000.Google Scholar
- CSA, Central Statistics Agency of Ethiopia and ORC Macro. Ethiopia demographic and health survey 2005. Addis Ababa and Calverton: Central Statistics Agency and ORC Macro; 2006.Google Scholar
- CSA, Central Statistics Agency of Ethiopia and ORC Macro. Ethiopia demographic and health survey 2011. Addis Ababa and Calverton, MD: Central Statistics Agency and ORC Macro; 2012.Google Scholar
- United Nations. World Population Prospects: The 2006 Revision. New York; 2006.Google Scholar
- Collet D. Modelling survival data in medical research. London: Chapman and hall; 1994.View ArticleGoogle Scholar
- Biswas A, et al. Statistical advances in the biomedical sciences: clinical trials, epidemiology, survival analysis, and bioinformatics. New Jersey: John Wiley & Sons, Inc.; 2008.Google Scholar
- Roberts RD, Lapworth C, Barker RJ. Effect of starvation on the growth and survival of post-larval abalone (Haiotis iris). Aquaculture. 2001;200:323–38.View ArticleGoogle Scholar
- Cox DR, Oakes D. Analysis of survival data. New York: Chapman and Hall; 1984.Google Scholar
- Fisher LD, Lin DY. Time-dependent covariates in the Cox proportional hazards regression model. Annu Rev Public Hlth. 1999;20:145–57.View ArticleGoogle Scholar
- Allison PD. Survival analysis using SAS practical guide. Cary, NC, USA: SAS Institute Inc.; 1995.Google Scholar
- Castro-Santos T, Haro A. Quantifying migratory delay: a new application of survival analysis methods. Can J Fish Aquat Sci. 2003;60:986–96.View ArticleGoogle Scholar
- Selvin S. Survival analysis for epidemiologic and medical research: a practical guide. New York: Cambridge university press; 2008.View ArticleGoogle Scholar
- Mason C. Cox proportional hazard models. 2005. http://www.demog.berkeley.edu/213/Week14/welcome.pdf.Google Scholar
- Lee ET. Statistical methods for survival data analysis. Second Editionth ed. New York: John Wiley and Sons. Inc; 1992.Google Scholar
- Lee ET, Wang JW. Statistical methods for survival data analysis. Third Editionth ed. New Jersey: John Wiley and Sons; 2003.View ArticleGoogle Scholar
- Guo G. Use of sibling data to estimate family mortality effects in Guatemala. Demography. 1993;30(1):15–32.View ArticlePubMedGoogle Scholar
- Klein JP. Semiparametric estimation of random effects using Cox model based on the EM algorith. Biometrics. 1992;48(3):795–806.View ArticlePubMedGoogle Scholar
- Parner E. Asymptotic theory for the correlated gamma-frailty models. Ann Stat. 1998;26(1):183–214.View ArticleGoogle Scholar
- WHO. Infant mortality: situation and trend 2017. http://www.who.int/gho/child_health/mortality/neonatal_infant_text/en/index. Accessed Mar 2017.
- Global Health Observatory (GHO) data, 2017. http://www.who.int/gho/child_health/mortality/mortality_under_five_text/en/. Accessed Mar 2017.
- Ewbank DC, Gribble JN. Effects of health programs on child mortality in sub-Saharan Africa. Washington, D.C.: National Academic Press; 1993.View ArticleGoogle Scholar
- Rutherford ME, et al. Access to health care and mortality of children under 5 years of age in the Gambia: a case-control study. Bull World Health Organ. 2009;87:216–24. doi:10.2471/BLT.08.052175.View ArticlePubMedPubMed CentralGoogle Scholar
- Ministry of Health. National Strategy for Child Survival in Ethiopia. Addis Ababa, Ethiopia; 2004.Google Scholar
- Ayele DG, Zewotir TT, Mwambi HG. Structured additive regression models with spatial correlation to estimate under-five mortality risk factors in Ethiopia. BMC Public Health. 2015;15(1):268.View ArticlePubMedPubMed CentralGoogle Scholar