I'll try to answer some of this ... The SAM figure is high. The first things to check are: (1) Is this almost all oedema? Sometimes this is real and sometimes it is the result of poor training / quality control. Does the survey report tell about training and report a kappa statistic for the oedema sign? As a rule-of-thumb ... unless this survey is from central or west Africa (or a related population) then kwashiorkor should be pretty rare. (2) Is a similar prevalence found by applying a MUAC-based case-definition (e.g. MUAC < 115 mm). If the MUAC case-definition returns a much lower prevalence then the survey is probably from a pastoralist population living close to sea-level in a warn climate. W/H is strongly biased by body-shape and is known to produce highly inflated prevalence estimates in some populations (e.g. most Somali clans, parts of Ethiopia, parts of Sudan, parts of Niger, parts of Nigeria ... perhaps in parts of India and Myanmar too). If this is the case then age-specific prevalence by W/H will increase with increasing age (as opposed to peaking between one and two year of age as is the general case). So ... your result could be due to poor recognition of oedema or the use of an inappropriate anthropometric indicator. If this is the case then the best you can do will be to use a simple MUAC-only (i.e. without oedema) case-definition such as MUAC < 115 mm. This will underestimate prevalence only slightly because most kwashiorkor cases will have low MUAC. With regard to "highest recorded" ... you have to be aware that use the OMS2006 (WGS) reference will return considerably higher prevalence of SAM than if the NCHS reference were used in the same population. This makes answering this question by review of past survey reports a little difficult. I did some work with Arabella Duffield on this issue a few years ago and we found the following formula for converting SAM prevalence by OMS2006 to SAM prevalence by NCHS: p(NCHS) = -0.21 + 0.60 * p(OMS) For 20% prevalence: p(NCHS) = -0.21 + 0.60 * 20 = 11.8% Still very high! I have a database of 560 survey datasets. I looked at the prevalence of SAM using OMS2006 and the WHZ < - 3 or oedema case-definition. In that database, the distribution of SAM prevalences by OMS2006 was: Minimum : 0.0% Q1 : 1.6% Median : 2.7% Mean : 3.7% Q3 : 4.8% Max : 18.7% <- Note : long tail to the right (> 75% of surveys below 5%) So ... your 20%+ figure is very high. The problem with a threshold for prevalence by W/H is that W/H base case-definitions have no universal meaning. It is strongly biased by body-shape which depends on genetics, diet (particularly milk), temperature, altitude, &c. Using W/H, mountain-dwelling agrarians tend to APPEAR obese and plains-dwelling pastoralists in tropical or sub-tropical regions tend to APPEAR wasted regardless of their true nutritional status. The problem is at the individual level (e.g. case-definitions for CTC / CMAM admission) and the population level (e.g. case-definition for prevalence surveys). MUAC is better but not perfect in this regard. Until we start using MUAC (or anything but W/H) we will be stuck on this issue.

From Assaye Tolla: Dear Anonymous 303, Yes, it is possible. Whay not? Regarding the palusability of the result, we need to see a lot of other things and the result by it self can not tell you about that. You might need to see the raw data, trends in the area and other parameters that will help you to question the result. I have done a survey in South Sudan with the prevalence of SAM rate at 24%, I don't think whether there is a threshold or not but it depends on the area that you found such high prevalence rate. Good luck!

Hi all, And thanks a lot for your answers. I should have given more details in my post but the results are not yet officially validated, sorry about that. But I can still add some information to answer to some of your question: - it's in the sahel region, so very few oedemas and mostly agro/pastoralist populations - no MUAC data neither quality control information are available for the moment - GAM is around 30% which means that MAM is 'just' 10% (compared to 20% SAM). So the question is: is it plausible that more MAS than MAM cases are found? (maybe I should review math notes about integers in a gaussian normal distribution). Thanks again,

Dear Anonymous 303 , You may use a spreadsheet with standard math functions to solve your problem without reviewing calculus. Two hypotheses: 1 - The WFH distribution of your population is normal You can calculate the theoretical proportion of children < - 3 and < - 2 using the normal distribution function of your spreadsheet for any mean WFH with a SD of 1. My own spreadsheet speaks French, and the function to calculate this is loi.normale using the option "cumulative=vrai (true)". Should not look very different in English. To see what WFH mean gives the same proportion of MAM and SAM, you calculate the SAM /MAM ratio and use the solver function to determine for which mean WFH this ratio is 1. You set mean WFH as variable, SAM /MAM ratio as target cell and ask the solver to get 1 for the target cell. You will see this is achieved when mean WFH is - 2.68. To see what mean WFH gives twice as much SAM as MAM, you use 2 as target for the cell SAM MAM ratio. You get a mean WFH of -3.24. To be frank, all these figures are quite unrealistic. A mean WFH of -2.68 means a SAM prevalence of 38 %. A mean WFH of -3.24 means a SAM prevalence of 59 %. 2 - The WFH distribution of your population is not normal This is the most plausible interpretation. It is quite possible to have a higher proportion of SAM than suggested by the normal distribution. This means a specific group is especially hit. And then no way to predict the SAM MAM ratio. But still, this means this specific group would have a very high SAM prevalence to get your results. I think you better check again how reliable your data are. I am ready to believe the high SAM rate you got, but a higher SAM rate than MAM rate looks strange. Any idea about the shape of the WFH curve distribution ?

A minor comment about a side note in Dr. Myatt's comments. Just wanted to share that we have a different experience in Malawi with regards to the MUAC of children with kwashiorkor. I just pulled up the data from one of our current studies -- we have 1965 children with kwashiorkor (aged 6-59 months). The mean MUAC at enrollment was 12.6, standard deviation is 1.4. Of these 1965, 370 (18.8%) had a MUAC less than 11.5, and 250 (12.7%) had a MUAC less than 11.0. So from our experience at least, it is typical actually for children in central-southern Africa with kwashiorkor to have MUACs in a range that would not necessarily be considered as particularly low. I don't know if this has something to do with the body-shape data he mentions, but that is what we are seeing on the ground.

It is a real shame that you do not have MUAC. You need this for needs assessments. I think you have bigger problems ... The MAM : SAM ratio looks very wrong to me. In a standard normal distribution we would expect c. 2.28% to be below -2 z and 0.13% to be below -3 z. The MAM : SAM ratio is about 16:1. We tend to get fatter tails than that. Looking at the database of 560 surveys I find the the mean ratio is 9.5:1. That's also what I'd expect from the surveys that I have done. This looks wrong to me. There are some scenarios that might explain this ... perhaps you are surveying two very different populations with one survey. What does the distribution of WHZ look like?

Dear Indi, Thanks for the response. Data from Malawi (Sandiford P, Paulin FH. Use of mid-upper-arm circumference for nutritional screening of refugees. Lancet 1995; 345(8957):1120) and Kenya (Berkley J, Mwangi I, Griffiths K, Ahmed I, Mithwani S, English M, Newton C, Maitland K. Assessment of severe malnutrition among hospitalized children in rural Kenya: comparison of weight for height and mid upper arm circumference. JAMA 2005; 294:591-7) suggest that MUAC is sensitive for Kwashiorkor. There is, as far as I know, very little published data on this so you should seriously consider publishing your findings. Mark

Thanks for sharing these references Mark. We will definitely publish our results and compare them to these prior studies. The Sandiford/Paulin letter does not mention what cutoffs they used for MUAC or WHZ, so it is hard for me to compare with our data, but in principle, yes i agree that WHZ is not a very sensitive or specific screen for kwashiorkor. The Berkley et al paper is fantastic, but unconvincing with regards to this topic for a couple reasons: First, the data i cited is for children with SAM treated as outpatients, not inpatients (should have mentioned that before, sorry), so they are almost certainly a different population that can't be directly compared. Second (if i am reading it correctly), i think the data in Table 5 also suggests that even a majority of kids in their inpatient population with bipedal edema do not have MUAC < 11.5. They show that 388+37=425 children with edema had MUAC > 11.5 (1st and 4th columns), whereas 146+207=353 children with edema had MUAC < 11.5 (2nd and 3rd columns). So 55% of their children had MUAC > 11.5. (I am focusing on edema since this is what we are using to diagnose kwashiorkor, not the skin/hair changes that they also studied.) But overall, they also show also that MUAC is more sensitive than WHZ for identifying children with kwashiorkor. MUAC is better than WHZ, but still has a sensitivity less than 50%. This is an interesting discussion for me and i apologize to the original poster about going this far off-topic. I guess i just want to make sure that in our desire to promote MUAC as an efficient screening tool that we do not forget to take the time to teach local health workers and other colleagues how to check for edema and to take the time to do so accurately with each child so that no cases of kwashiorkor are missed. MUAC is better than WHZ for this type of quick screen, but not good enough to replace this small bit of the physical exam.

Indi, These are the only published data relating to this topic that I am aware of. There is a definite need to get more evidence published. I'm glad that you intend to publish. I can let you have my survey database and you could work up something from that too. Let me know. I'd guess that oedema in many datasets is not well collected though. I agree with your reading of both publications. The issue here is, I think, that we must not mistake a rough-and-ready remedy to a survey problem (initial question in this thread) for a general solution to case-finding of kwashiorkor. I agree ... MUAC goes some of the way but not all of the way and we should not pretend that it does ... we need to train health workers to detect kwashiorkor using the oedema sign. It's not so easy to teach (I can find references if you'd like). Mark

Dear Indi & Mark The issue of kwsahiorkor (hydration) and MUAC is interesting. In our JAMA paper, MUAC was better than WHZ for detecting kids with kwashiorkor, which may be due to one or both of two things: i) MUAC changes due to hydration status are less than WHZ ii) Kwashiorkor may occur without severe wasting. We have just completed a study of the effect of dehydation on MUAC and submitted for publication. We took children with clinical signs of dehydration admitted to hospital (mostly with diarrhoea). We performed careful anthopometry at admission, and again after 48h of usual rehydration fluids and other treatments. Percentage short term change in hydration was assumed to be equal to the percentage weight change over 48h. Following rehydration of 5 to 10% dehydration, WHZ changes were dramatic (as could easily be estimated from standard charts). MUAC also changed with hydration status (muscle is at least 75% water), but to a far smaller degree. Interestingly, in some children, diarrhoea worsened during 48h and they became more dehydrated (lost weight)... their MUACs became smaller. We did not include children with kwashiorkor in this study, but I can have a look at the database (or start prospectively collecting) MUACs before and after kwashiorkor has resolved. Given our dehydration data, I would suspect that tissue over-hydration would, on average, result in an increased MUAC. I agree with Mark, health workers need to do both MUAC and look for oedema.

Dear Mark and Andre, thanks for your usual and thorough contribution to this forum, I just hope you understand how much important is for us to have such support. You definitely confirm my doubts about plausibility of those results. However, since it's a preliminary report I don't have further details to evaluate data quality (SD, WFH distribution, etc). Tamsin, in the case of the Sudan survey with 24% SAM, what was the GAM? I'd expect a higher MAM/SAM ratio in severe famine due to high mortality among MAS cases.

To clarify my initial suggestion (MUAC only case-definition) using a simple example ... If we split SAM into wasting and kwashiorkor and assume that 10% of all SAM is kwashiorkor (this is probably an overestimate for areas where kwashiorkor is uncommon). Working with, as an example a SAM prevalence of 10% ... 90% of this 10% is simple wasting and 10% of this 10% is kwashiorkor. If we detect 100% of wasting and 50% of kwashiorkor using MUAC then we would estimate prevalence to be: 0.1 * 0.9 * 1 + 0.1 * 0.1 * 0.5 = 0.095 (9.5%) So the underestimation is slight. This is a rough and ready fix to a specific survey problem. I do NOT mean to suggest that we remove the oedema sign from nutritional anthropometry surveys. I do NOT mean to suggest that we do not use the oedema sign in therapeutic feeding programs. Some comments ... The issue of hydration is interesting. In CTC programs I encourage giving sugared water, fruit juice, or diluted ORS to children waiting to be seen for the clinical screen. Children are often dehydrated during the, sometimes lengthy, journey from home to clinic. With this simple practice we reduce inappropriate referral to inpatient care. I've always thought that MUAC was better at detecting kwashiorkor than W/H because: (1) In many populations nutritional oedema occurs with wasting. (2) Because the weight of retained fluid tends to mask what would otherwise be low W/H values. Modelling the upper-arm as a cylinder of muscle (simplistic but good enough for illustration) we have: V = pi * r^2 * h C = 2 * pi * r gives: V = pi * (C / (2 * pi))^2 * h Someone should check my algebra! ... volume increases much faster than the circumference ... I suppose that MUAC is less sensitive to changes in hydration because it is a circumferential measure and increases much more slowly than volume. And ... thank you for your kind comments.

Dear Anonymous 303 Apologies, but there seems to have been an error in the report above of the Sudan study. 24% was actually the GAM rate and the SAM was rate was 6.1%. Sorry for that, Tamsin

can anyone suggest me the reference of 90% sensitivity of mid upper arm circumference with respect to weight for height to determine SAM?

The sensitivity of using MUAC to detect wasting by W/H will vary with age and with population. This is because W/H is strongly affected by body shape which changes with age and with location. This means that the degree of overlap between children with MUAC < 115 mm and children with WHZ < -3 will vary from place to place. In (e.g.) cold / mountainous areas the overlap will be small since children in these areas will tend to have short limbs and large trunks and chests (very difficult for them to get a very low W/H using international reference populations). Best for us to forget about W/H altogether and use MUAC which is more associated with subsequent morbidity and mortality than W/H and has a pretty constant interpretation from place to place. BTW ... if we change the MUAc threshold we change the sensitivity of your test. A high MUAC threshold will have 100% sensitivity. If you intend to use MUAC as part of a two stage screen then beware that rejection will damage program coverage. This means avoiding the use of high MUAC thresholds. Again ... better to use MUAC alone.