Study area
The study was carried out at Alupe Sub-County Hospital. The hospital is a level 4 hospital located in Angorom ward, Teso South Sub-County in Busia County serving a catchment population of 34,321 persons [Figure 1] [7].
Study Design
We conducted a facility-based unmatched case control study carried out between May 2017 and June 2017. The study population consisted of all children below five years attending the child welfare clinic and the outpatient clinic within the hospital during the study period.
Study hypothesis
We hypothesized that improper infant and young child feeding practices as well as inconsistent child welfare clinic follow-up do not contribute a significant proportion of malnutrition in children <5 years attending Alupe Hospital in western Kenya, using the UNICEF conceptual framework for malnutrition [Figure 2].
Case Definitions
Under-nourished child was defined as a child aged 6-59 months attending the hospital as an inpatient or outpatient whose anthropometric measurements were not appropriate for their age with z-scores (Weight-for-Height WHZ, Weight-for-Age WAZ, Height-for-Age HAZ) of less than or equal to -2 SD. Weight for age Z-Scores from the WHO Charts were used to define presence of under-nutrition [8].
Over-nourished child was defined as a child aged 6-59 months attending the hospital as an inpatient or outpatient whose weight for height index (WHZ) and weight for age (WAZ) Z score is more than or equal to +2 SD.
A participant was classified as stunted if height-for-age z-score was less than -2SD and severely stunted if height-for-age z-score was less than -3SD. Wasting was defined as weight-for-height z-score of less than -2SD while severe wasting was weight-for-height z-score of less than -3SD. Any participant with weight-for-age z-score less than -2SD was said to be underweight while -3SD was be said to be severely underweight. On the contrary, those with weight-for-age or weight for height z-scores more than +2SD were classified as overweight, while those with more than +3SD were obese.
MUAC calculations
For the MUAC cut-points to determine whether a child was under- or over-nourished, we used the cut-points of any child with MUAC <126 mm was classified as under-nourished and those with MUAC>179 mm were classified as over-nourished [8].
Definition of controls
Any child aged 6-59 months attending the hospital as an inpatient or outpatient whose anthropometric measurements are appropriate for their age with z-scores between -2SD and +2SD [9].
Sample size determination
The sample size was calculated using statistical software Epi Info® version 7.2.0. The study assumed a 95% confidence interval, 80% power, 10% wasting among controls [10], an Odds ratio of 2.5 and the ratio of cases to controls of 1:3. Using these assumptions, the minimum sample size was 375 (94 cases and 282 controls).
Selection of cases and controls
The cases were sampled consecutively given the low number seen each day for eligible children enrolled for nutritional support in welfare clinic. The controls were selected through systematic random sampling. The average number of children below five years visiting the outpatient section of the child welfare clinic on a daily basis was used as a sampling frame. This was determined by obtaining the number of children visiting the out-patient clinic between April and June of three preceding years before the study. The study was conducted during weekdays within the duration of the study period hence the number of controls to be enrolled in the study on any single day was pre-determined. Using the average number of patients seen each day at the clinic and number of controls to be enrolled in the study each day, a sampling interval was determined and the first control was picked randomly between one and the sampling interval. The sampling interval was then added to enroll the remaining controls. Any eligible participant whose legal parent/guardian did not give consent was replaced by the next available participant whose legal parent/guardian consented to the study.
Data Collection
Triage was carried out by the hospital staffs as is the norm and all critically and severely ill patients were urgently attended to by the hospital clinicians as per procedures and guidelines of the hospital. The weight was measured using electronic digital weighing scale (Seca®). For height/length, children <2 years were measured lying down (recumbent length) while those who were >2 years were measured standing up. For mid-upper arm circumference (MUAC) and head circumference, a non-stretch tape was used.
A pre-tested interviewer-administered questionnaire was used for each study participant to obtain demographic, clinical, nutritional, social and economic information. Each patient was de-identified by a unique code to ensure their privacy and maintenance of confidentiality.
Data management
Data entry, cleaning, validation and analysis was done using Microsoft Excel (Microsoft, Seattle, WA, USA), and Epi info version 7 (CDC, Atlanta, GA, USA). Anthropometric data was analyzed using WHO Anthro® software version 3.2.2 (WHO Anthro®) to assess nutritional indicators like weight-for-length/weight-for-height (wasting), weight-for-age (underweight or overweight), length-for-age/height-for-age (stunting), MUAC-for-age and HC-for-age. The software then provided the z-scores based on gender, age and the anthropometric measurements. We calculated measures of central tendency and dispersion for the continuous variables and proportions for categorical variables. For univariable analysis, we calculated odds ratios (OR), 95% confidence intervals (CI), chi-square statistics and p-values. Variables with p-value ≤ 0.05 were considered to be statistically significant. We carried out unconditional logistic regression with variables that had p-values of <0.2 at univariable analysis. A backward elimination stepwise method was used to identify independent factors associated with malnutrition. During model building, any variable that caused an insignificant increase in deviance on removal from the model were left out of the model while the variable that caused a significant increase in deviance on removal were retained in the model. All variables removed from the model when a backward stepwise method was performed and those known to be potential cofounders or factors associated with malnutrition from previous studies were tested for confounding, any of the mentioned variables that had a more than ten percentage change (>10%) between the crude and adjusted odds ratio was considered as a confounder. The final model after testing for all biologically and statistically plausible interactions had only variables with p-value of ≤0.05.