We aimed at comparing acute malnutrition estimates as measured by WHZ, absolute MUAC and MUACAZ. A total of 255,623 children aged 6 to 59 months from 19 surveys were included in the analysis. We found an overall GAM prevalence of 16.1% measured by WHZ− 2, 7.8% by MUAC115 and 15.8% by MUACAZ− 2. Similar results were obtained by livelihood system except for the prevalence of GAM in agro-pastoralists and riverine children that was higher using MUACAZ− 2 as compared to WHZ− 2, and in pastoralists where the prevalence calculated by MUAC125 was a third of the prevalence by either WHZ− 2 or MUACAZ− 2. However, the children identified by either indicator were not the same, being the overlap around 28% between WHZ− 2 and MUACAZ− 2 diagnosis, and 18% between WHZ− 2 and MUAC125.
In 1995 the WHO proposed a set of thresholds to classify the severity of malnutrition in emergency situations according to the prevalence of GAM, measured by WHZ− 2, that are still widely used (< 5% Acceptable, 5–10% Poor, 10–15% Serious, > 15% Critical) . Up to date there are no thresholds to be used at global level when absolute MUAC is used for nutrition surveillance, and using the same thresholds designed for WHZ is problematic. For the majority of the surveys analysed, the classification of the nutrition situation according to WHO thresholds would have been “Critical” when measured by WHZ, and “Poor” when measured by absolute MUAC.
In response to this limitation, FSNAU developed absolute MUAC emergency thresholds adapted to the Somalia context and based on the calculation of the quintiles of the distribution of absolute MUAC. According to those thresholds, the situation is considered “Critical” if the prevalence of GAM as measured by MUAC125 is above 10.7% . However, in eighteen out of the nineteen surveys analysed the situation would nevertheless have been described as “Critical” by WHO thresholds when using WHZ− 2 and only “Serious” by FSNAU threshold using MUAC125, thus triggering different responses.
According to our results if the GAM is measured by MUACAZ− 2 the population estimates are similar to those obtained by WHZ− 2, and so they are the resulting classifications according to WHO thresholds in sixteen out of the nineteen surveys analysed.
One of the main arguments for not using MUACAZ it is the age dependency of the indicator, as it is reported that in certain contexts with a high burden of acute malnutrition, age is difficult to ascertain accurately . This may not be the case in Somalia as the field teams have extensive experience in collecting this information due to the high frequency and long history of surveys being conducted in the country. In the sample analyzed the proportion of children 6–23 months old compared to the 24–59 months age group followed the 1:2 expected proportion ratio. Moreover, in other emergency nutrition surveillance systems data is collected on both weight-for-age and absolute MUAC, suggesting that the age collection is feasible to collect with accuracy . There is also evidence from Bangladesh showing that MUAC-for-age provides comparable information to weight-for-height in the context of nutritional surveillance of populations, and especially for assessing seasonal changes .
We recommend to revise the age dependency limitation in view of the importance of bringing age in the estimation of acute malnutrition when using the MUAC measurement. As other authors have already suggested, it could also be possible to construct tally sheets for MUAC divided by age groups in order to minimise the constraints regarding age ascertainment and its accuracy .
In relation to the individual screening of acutely malnourished children the results obtained by either indicator did not converge. The discrepancy observed between absolute MUAC and by WHZ in children screening had already been studied [17, 18] and taken into account in WHO-UNICEF guidelines [19, 20], but seldom is published on the comparison of children diagnosed as acutely malnourished according to MUACAZ or to WHZ.
The results obtained by Grellety and Golden in their recent review of 47 countries  are consistent with our findings regarding the WHZ /MUAC discrepancy in Somalia overall, and the differences we found according to livelihood system were similar to those found by these authors in Ethiopia. Previous studies conducted in Ethiopia found similar results and argued that the prevalence of acute malnutrition by WHZ was overestimated due to differences in body shape such as relatively longer legs that contribute to the total height . However, recent studies refute that this factor alone can explain the discrepancies observed, concluding that they may be due to the combination of different anthropometric characteristics that are expressed differently across populations [8, 11].
Moreover, our results disagree with the hypothesis of pastoralist populations showing systematic higher estimates for acute malnutrition as compared to other populations in the same context  as they show that the GAM prevalence (according to any of the three measurements) among the pastoralist children was lower than in any of the other livelihood populations studied.
Finally, our results show that age, sex and stunting status of the children also play a role in their likelihood of being diagnosed as acutely malnourished depending on the indicator to be used. Female, younger, and stunted children were more likely to be identified as acutely malnourished if the indicator to be used was absolute MUAC, whereas boys older than two years and stunted were more likely to be diagnosed as GAM by WHZ. Our results confirm those of other studies conducted in countries as Philippines, Chad, South Sudan and Bangladesh , Kenya  or South Sudan , and some authors have offered the explanation that MUAC is mainly a measure of muscle mass, and that muscle mass is reduced in stunted children [24, 25]. Interestingly, we found that older boys were more likely to be identified as experiencing GAM by MUACAZ (similar to WHZ results), but also that MUACAZ was the indicator to be more strongly associated with stunting. Consistently, a study in Kenya reported that upper arm muscle area Z-score, an age and gender adjusted measure of absolute lean body mass, explained most of the variability in the progression of stunting among school-aged children over a two year-period , suggesting that age adjusted measurements may be more appropriate in the interpretation of weight-for-height and MUAC measurements discrepancies in relation to stunting.
Strengths and limitations
The sample characteristics in terms of size and number of surveys allowed for a high precision in the analysis, and for the stratification by livelihood system. The so called IDPs livelihood was over-represented in the overall population as recent data indicated that the proportion of IDPs in Somalia population was around 14% , thus the importance of the stratified analysis.
The data were collected in field conditions which may have an impact on the accuracy of measurements, although the FSNAU survey enumerators long time experience and routinely training may have minimized this limitation.
Another limitation of our study is that the analysis was performed on a single country and results cannot be generalizable for other countries.