Settings and population
The study was conducted in NEMMH, which is located in Hossana town, southern Ethiopia. Hossana town is located 232 kilometres far from Addis Ababa, the capital city of Ethiopia. The hospital is serving about 2 million people. The total number of under-five children found in catchment area are 222,336.
The paediatrics department is one of inpatient department found in NEMMH. The department has nutrition rehabilitation unit. The nutrition rehabilitation unit admits children diagnosed SAM from Hossana town and surrounding areas. The admission criteria for infant less than 6 months or less than 3 kg being in breast-fed includes: the infant is too weak or feeble to suckle effectively (independently of his/her weight-for-length) or W/L (Weight-for-Length) less than 70% or Presence of bilateral oedema.
The admission criteria for 6 months to 59 months includes: W/H or W/L < 70% or MUAC < 110 mm with a Length > 65 cm or Presence of bilateral pitting oedema [10]. All children are managed using WHO standard guideline for Managing SAM.
The Study population was all under five children admitted to the inpatient nutrition unit between January 2012 and December 2015. Children with incomplete data were excluded from the study.
Study design and sampling
Four years retrospective cohort study was conducted on under-five children admitted with the diagnosis of SAM from January 2012 to December 2015. The study was conducted from May to June, 2016. Total sampling method was used; where all under-five children admitted with diagnosis of SAM was considered for analysis. About 500 children were included in the study.
Data collection and measurement
Data on variables of interest were extracted from patient charts, using a predesigned data collection form. The data collection forms were completed legibly by trained data collectors. The data collection form contains socio-demographic factors, diagnosis of SAM, co-morbid diseases, length of stay, admission type and outcomes of discharge. Patients’ chart numbers were collected from the paediatrics ward registration book. By using the chart numbers, charts were drawn by card room workers. The data collectors were trained on the requirements of the protocol and data to be collected. Completeness and legibility of each data collection form was audited at the end of each day by the principal investigators and supervisors to ensure accuracy.
In this study, the dependent variable was time to death of children with SAM while the independent variables included sex, age, length of stay, morbidity, residence, co-morbidity, and admission type.
Mortality was defined as death due to SAM and other comorbid disease while morbidity was referred to SAM. The data on SAM types (Marasmus, Kwashiorkor, and marasmic-Kwashiorkor) was reviewed from patient chart. Length of stay in the hospital was cross-checked by calculating the difference between date of admission and date at which the patient dead/discharged and corrections were made where inconsistency were found. Comorbid disease was defined as co-existence of any other disease/s with SAM. Admission type was assessed by two options: new and repeat admissions.
Data processing and analysis
Data were entered using Epi-Data version 3.1 and exported to SPSS version 16 for analysis. Trend analysis was done by using STATA version 11. Descriptive statistics was used to summaries study variables.
Model diagnostics was done by using the maximum likelihood estimation. Cox regression model assumption of proportional hazards was checked by Kaplan-Meier hazard plots and testing an interaction of covariate with time. Multi-collinearity among independent variables was checked and showed no significance.
Kruskal Wallis was used to compare if means of “length of stay on the ward” and “age” for the different types of SAM were different. For categorical variables, chi-square was used to show association. A Kaplan-Meier curve was used to estimate survival probability of different types of SAM and admission types.
Bivariate analysis was done to identify associations between dependent and independent variables. Variables significant at P <0.05 level in the bivariate analysis were included in the final cox proportional regression analysis, to identify independent predictors of mortality. Finally, co-morbidity, morbidity groups and admission type were included in cox proportional hazards regression analysis. Cuzick, a non-parametric test was used to examine trend patterns for both morbidity and mortality. A p-value of less than 0.05 was considered as statically significant.