As the release of WHO growth standards changed the definition of severe malnutrition, which now includes many children formerly identified as moderately malnourished with the NCHS reference, studying this new category of children is crucial. Within the 32,933 children identified in our sample as severely malnourished according to the WHO standards (see ), 28,093 (85.3%) were not classified as such by the former NCHS reference (i.e. they were “WHO severes” with WH<−3 WHO standards Z-score, but “NCHS moderates” with WH≥70% of the NCHS median). Feeding programs aimed at severe malnutrition are thus likely to include larger numbers of children with better overall nutritional status in the near future. Nutritional actors will need to think about appropriate strategies to manage these “new severes”.
The first model reported here, without clinical observation and based on basic data recorded at admission in most nutritional programs, gives a result close to the model proposed by Prudhon et al
(who used the NCHS reference). The risk factors for death identified in the multivariate analysis including clinical data hint either towards concomitant infection (i.e., fever) or known features of malnutrition (e.g. low weight-for-height). In light of the magnitude of this database, the absence of association between death and sex, diarrhea, vomiting and Paracheck® positivity can be regarded as a robust finding, which is also consistent with previously published work 
based on the NCHS growth references.
A potentially innovative aspect of our results is the good performance of the “appetite test” for the assessment of anorexia. Plumpy'nut® has shown to be popular with malnourished children in previous trials 
. This simple test, easy to implement in therapeutic feeding programs, still requires evaluation in an appropriately designed study comparing against conventional criteria for anorexia.. Quantity of RUTF intake, time allowed for eating and method of observation by caregivers should be standardized (for example quantity of RUTF expected to be eaten per 10 minutes for a given age).
We did not find a positive Paracheck® to be associated with a higher risk of death. This is likely because unlike other potential concomitant infections, malaria benefits from a reliable rapid diagnostic test. This allowed for rapid and effective treatment (artesunate+amodiaquine). We also did not find children originating from Nigeria to be at higher risk of death. We may have suspected Nigerian children to be at higher risk as they traveled long distances to seek treatment and may have been in worse condition.
This analysis highlights the importance of identifying simple and robust indicators for determining whether a child should be placed in outpatient or inpatient care, a key strategy for rational management of community-based programs. In such context, both formula derived from these models may help to build a “risk score” for early triage and rapid referral to inpatient care. One option could be to establish differential management for different types of “new severes” according to anthropometrical and clinical criteria. This could imply up-stream screening by less medically qualified staff, based on precise criteria, associated with a greater risk of death. As a monitoring tool, these formula could also be helpful to examine mortality trends (e.g. after a modification of a treatment protocol).
We note the limitations of using program monitoring data to investigate epidemiological questions in nutrition such as risk factors for death. Although a large number of additional indicators upon admission were collected, variations between clinicians in how these data were recorded prevented their analysis. Breastfeeding practice is a notable example. This is of course important to consider, but the information recorded was not reliable enough to be included in our analyses. Some clinical signs may be due not to acute malnutrition but to a concomitant pathology, (for example apathy can be linked to severe wasting or to another cause like infection or dehydration, as well as pallor can be due to nutritional micronutrient deficiency or to malaria anemia without any malnutrition component). They are moreover “subjective” signs, and their finding and reporting may be clinician-dependant. Detailed prospective studies of co-morbidities and clinical signs at admission are required to improve further the treatment of malnutrition. We limited our analysis to examining indicators where we felt the data collection and the clinical criteria were sufficiently strong over this large database. Future analysis should include quality documentation of metabolic and clinical status, and response to therapy, particularly in the first few days following admission.
Fatality was low in the MSF program, far below the 5% case fatality rate often expressed as a threshold for acceptability in hospitals 
. However, case fatality is likely to be underestimated as some children who were lost to follow up from the outpatient program might have died at home. This is an area for additional improvement with more rigorous tracing of defaulters. In accordance with previous findings 
, most deaths occurred within 7 days of admission. This highlights the importance of ensuring adequate access to effective treatment, community awareness activities encouraging parents to bring their children early for treatment and strategies to detect children at earlier stages of malnutrition. The median age of children admitted to the program was approximately 18 months, with a median weight of approximately 6 kg. Detecting and treating these children earlier is a clear priority. These early deaths also reinforce the importance of studies examining the etiology of death among malnourished children, in particular concomitant infection(s).
As the transition from the NCHS growth reference to the WHO standards will increase the number of children classified as severely malnourished, up-stream screening based on anthropometrical and clinical signs may be crucial in adapting the management strategy. Because the triage must rely on basic-level health workers, criteria developed should be easy, precise and clear, based on robust evidence. Our findings provide risk factors for mortality– weight, height, edema, pallor, anorexia, apathy and fever – that can be proposed for building criteria for the orientation of children towards either hospitalization or ambulatory care