Among the major effective child survival interventions, breast feeding and immunization were found to be major determinants of child survival. Pneumonia contributed to the largest deaths in the study population followed by acute diarrhea and malaria.
We used an unsatisfied basic need assessment in this study. This composite scale was used for the following reasons: (I) Income measurements are rarely reliable measures of poverty in low income countries. (ii) Furthermore this measure also excludes other elements of deprivation such as housing, clothing, education and health careThis index is based on different dimensions: housing quality, dependency ratio, availability of water supply and latrine  The wealth score which is a modification of the unsatisfied basic need assessment modified for local use was found to be a significant predictor of child survival.
Contrary to studies elsewhere [22–24] maternal education was not significantly associated to child mortality. In some studies the effect of education [on mortality] was significant among families with poor socioeconomic status. Taking this into consideration, the analysis by stratification with wealth index did not show any significant difference. This might be attributed to the similarities that exist in the level of education between the comparison groups. Among environmental factors, a floor made of earth was significantly related to an excess in mortality when compared with cement floor. Its effect was lost as adjusted by other factors.
Breast feeding was found to be a strong predictor of child survival in this study. Breast feeding protects of infants from enteric infections by elimination of exposure to food- and water-borne pathogens; also, mature breast milk contains several compounds which increase immunity .
Higher level of protection by breast feeding was found among less educated women. Infants who were not breast fed had highest risk of dying during the first 2 months of life. Mean age of weaning among cases was at earlier age than that of controls. Earlier weaning was also related to excess in child mortality in other studies [1, 28]
Among preventive health services sampled, children's vaccination status was a strong predictor of child survival. The risk of mortality was about 6 fold higher among those who were not vaccinated, compared with those who received vaccination. The protective effect of vaccine is discussed in literatures [29, 30]. The national coverage of immunization in Ethiopia differs by antigen. The BCG coverage has reached 70%, while the rate of fully immunized children was 36% . The utilization of immunization service indirectly indicates caretakers' use of preventive health services.
Having more than five children was associated with increased risk of losing babies; the significance was kept even after adjusted for potential confounding variables. Other reproductive factors were not found to be significantly related to child survival.
Higher parity, short birth interval and low birth weight are known to be interrelated. The growth and development of young children are affected by their mothers' past nutritional histories and their well being during pregnancy. Birth weight is affected by maternal nutrition and health. High parity and short birth interval are known to deplete maternal nutrition and influencing the child survival through birth weight [32, 33]
Measurement of causes of death is needed for several purposes: (i) To establish the relative public health importance of the different causes of death; (ii) To evaluate trends over time especially as a method of evaluating the probable impact of the intervention programs; and (iii) To investigate the circumstance surrounding the deaths of children from specific causes and to devise effective actions to decrease mortality.
Pneumonia contributes to the largest mortality in the study population, followed by acute diarrhea and malaria contributing death similar to national report . The sensitivity and specificity of verbal autopsy diagnosis for malaria, acute respiratory infection and meningitides is low because these diseases share similar symptoms. Use of these results warrants caution. Measles, neonatal tetanus, malnutrition and accidents were detected by verbal autopsy with sensitivities and specificities of greater than 75%. Malnutrition contributed for 37.8% of all deaths. The contribution could be underestimated due to lack of ability of the tool to identify mild and moderate malnutrition. Identification of mild to moderate malnutrition using verbal autopsy is important since these groups are implicated in many more child deaths than previously recognized. Hospital based data in developing countries suggest that 55-75% of perinatal deaths are associated with prematurity and low birth weight. Because of the majority of deaths occur at home, birth weight measures are seldom available. Estimates of gestational age are particularly problematic in societies where often women do not know the dates of their last menstrual period. Birth trauma or asphyxia due to complications of delivery/poor obstetrical care are also thought to be major problems. But there exist little information regarding births that take place outside hospitals ..
HIV/AIDS was found to have caused 2(2.8%) of deaths using WHO pediatric HIV/AIDS definition. In other studies the ability to use verbal autopsy to distinguish deaths associated with maternal HIV infection from death by other causes was poor The estimated contribution of HIV to under five mortality in Ethiopia was 8.1% which was based on data from sentinel surveillance and considering the rate of mother to child transmission, and survival time of infected children . The difference between these two estimates could be due to limited ability of verbal autopsy to identify HIV-related death.
In conducting verbal autopsy different considerations should be taken into account. This study used an open approach in classification of death, broader category, and checklist without filter. The respondents were care takers. The interviewers used were lay persons trained intensively to conduct the study, which is known to increase the repeatability of the diagnosis when compared to medically trained people . Recall period advisable was for 12 months since the event is relatively more common than adult deaths. On the other hand, others argue that mothers are intimately involved in the care of sick children and so they may report the symptoms preceding the death of a child more accurately than a relative caring for an adult [15, 39].
With regard to validation of the tool, the option which was feasible was hospital-based reference which has many limitations. Besides this, we used verbal autopsy to estimate major causes of death.
Limitations of the study
Most case control studies, except for nested case control design, fail to delineate the temporal relationship of exposure and outcome variables. Recall bias might have been introduced as some events and exposures are difficult to be remembered over one year period.
To minimize selection bias all remote and accessible kebeles/units were included in the baseline survey. The definition for cases and controls were strictly adhered to during the data collection. Some cases could be missed because of maternal recall of events that occurred in early neonatal period.
Low level of sensitivity and specificity were reported for acute respiratory infections, malaria and meningitis.
Difficult to make comparison of health care utilization between the groups as controls might not have experienced an illness that is comparable to cases.