Improving case-detection of wasting among under-five-year-old children in Ethiopia: A secondary analysis of community-based surveys in humanitarian settings

Abstract

WHO recommends weight-for-height Z-score (WHZ) <-3 or Mid-Upper Arm Circumference (MUAC) < 125 mm as criteria for diagnosing wasting in children aged 6-59 months. In humanitarian settings, MUAC provides a simpler alternative than WHZ measurements, requiring only a tape measure. However, using MUAC alone may miss many at-risk children, causing discrepancies in wasting estimates. We analyzed 31 Standardized Monitoring and Assessment of Relief and Transitions (SMART) surveys from 2022-2025 across eight Ethiopian regions. The sample included 23,419 children aged 6-59 months with complete data. MUAC's diagnostic performance for GAM was evaluated using WHZ as reference, applying standard parameters including sensitivity, specificity, PPV, NPV, Youden index, and ROC curve analyses. We assessed MUAC cut-offs' accuracy nationally by age group (6-23 and 24-59 months) and administrative region. The optimal diagnostic MUAC cutoff for WHZ <-2 was 139 mm, with a Youden Index of 43%. This cutoff varied by age: 128 mm for children under 24 months (Youden Index: 54.5%) and 140 mm for children 24 months and older (Youden Index: 42.5%). MUAC diagnostic performance varied across regions, with optimal thresholds from 126.5 mm in Tigray to 143.5 mm in Gambella. Global acute malnutrition (GAM) prevalence differs based on the indicator used (MUAC versus WHZ-based), with MUAC sometimes overestimating or underestimating GAM compared to WHZ. Increasing the MUAC cutoff from 125 mm to 139 mm improves case detection but substantially increases false positives, showing a sensitivity-specificity trade-off. A single universal MUAC cutoff may under- and overestimate global acute malnutrition. Increasing the cutoff to 139 mm improves detection but raises false positives, highlighting the trade-off between sensitivity and specificity and emphasizing the need for context-specific thresholds to optimize screening. However, implementing context-specific thresholds remains uncertain.