Indices such as HAZ, WAZ, and WHZ are not functional indicators in the sense that they reflect function or mortality risk. They are all about the statistical properties of a reference population. We can work the probability of finding a child with a WAZ < -6 in the reference population (is is a little below 9.9E-10 ... so improbable as to be just about impossible). We reject cases with WAZ < -6 because they are very improbable. You should see a child with a WAZ < -6 once in something like 10 million surveys. Even if we assume that the reference population consists of heavy for age kids which are (on average) two z-scores heavier than the population under study then we would expect to see a child with W/A as extreme as in your question about 3 times in every 10,000 children. Pretty rare. We can (sort of) answer the question ... we can ask "Is WAZ < -6 compatible with life?" and get the answer "Very probably not!". But we do see kids with WAZ < -6 more often than that. What is happening? The problem with W/A, H/A, MUAC/A ... all the "... for age indicators" is that they are very sensitive to errors in age and age is often very difficult to measure accurately. Research suggests that any indicator that includes an age component requires that age be ascertained accurately and are more sensitive to errors in age than to random in anthropometry. See: Bairagi R, Effects of bias and random error in anthropometry and in age on estimation of malnutrition, Am J Epidemiol, 1986;123(1):185-91. What this means is that a lot of the children we see with WAZ < -6 probably have a WAZ > -6 but we have got their age wrong or we have recorded or entered wrong data. This is a particular problem in cross-sectional surveys where age is often reported to nearest years so (e.g) a 30 month old child might be called 24 months or 36 months. If you plot WAZ from a survey you often see a complex pattern of ripples when you should see a bell-curve. This is due to age lumping. We could do some checks with an independent measure. If (e.g.) we collect MUAC then a WAZ < -6 with a MUAC > 110 mm is indicative of error. W/A performs much better in longitudinal applications (e.g. growth monitoring) in which age is known precisely or the measure of interest is tracking of the median growth curve. I hope this is of some help.

It is not really uncommon to find children of <-6 WFA! Here are the values (WHO standards): Age (mo) -6Z WfA 0 1.18 1 1.84 2 2.49 3 3.01 4 3.42 5 3.74 6 3.99 7 4.20 8 4.38 9 4.54 10 4.68 11 4.81 12 4.94 Mark is only correct if he would say that "it is exceedingly unlikely (his calculation) that a child of -6Z is normal and healthy from an elite family". As you can see a 6 month old child of about 4kg is -6Z WFA. If a normal birth weight child has gained very little weight in the first 6 months of life s/he will readily become - 6Z. If you visit a malnutrition ward you will certainly see such children. The standards are based upon a very healthy ELITE poplulation - and the Z-score depends upon the SD within this population - ie the more homogeneous the standard healthy elite population is in terms of its growth the greater will be the deficite in terms of Z-score of a malnourished child. Do not forget that WFA is a composite - if you have a child that is -4Z height for age and -2Z weight-for-height (not very rare) we would only classify this child as moderately wasted - but s/he will be almost -6Z WFA. And a deficit in WFA,HFA or WFH is associated with an increased risk of death. However, your questions of "what is compatible with life" and "should we reject this child's results from the survey analysis" are quite different. If we look at newborn children then -6Z is 1,176g, -7Z wfa is 946g and -8Z is 746g - in modern intensive care neonatal ICUs some of these infants can survive - but uncommonly. The same numbers for a 6 month old are: -6Z -3.99g, -7Z=3.54, -8Z=3.13kg, -9Z=2.77kg. I am not sure at what level we will find that all the children are dead - but in Western medicine such children do very occasionally occur (secondary to major illness). Your second quesion of should we accept these measurements in a survey is quite different. Mark is correct in his statements about the accuracy of age reporting - in fact in most of the world the age is ADVANCED by the respondent (not retarded as we normally do in Europe/North America) so that the reporting is of a child older than s/he actually is. Even in the Scottish census early last century about 10% of ages of infants were wrongly reported! The most interesting example is children whose ages are given on the Chinese calender - traditionally they could a new born as 1year and increment the age by 1 at every chinese new year - so that a child born just before the chinese new year would be reported as 2yrs old just after the chinese new year! Examination of age distribution in nearly all surveys shows heaping at 12month intervals. Should we reject such children - the answer is twofold - first we want to include all children who are "representative" of the population and reject those that are abberant (if we can define this in some way) - for example a "visitor" to a rural area of Africa from the USA may be there but is not representative of the resident population - and perhaps children with congentital defects etc, should be omitted if we want to use the data to determine "nutritional status". But second, we also want to reject children whose measurements are much more likely to be due to recording/measurement error than to be real values. On this basis perhaps all children who are more than a certain deviation from the mean of the sample should be excluded from the analysis - this will probably reject a tiny fraction of children whose measurements are real - and a much larger proportion of children whose measurements are erroneous.

Mike is quite right. The WGS is a reference based on a sample with very restrictive entry requirements. The sample could be classified as coming from an "elite" but the WHO would probably not want to use that term. In most of the settings that we work a large proportion of children would not have been eligible for inclusion on the reference. The issues of criteria vs. norm referenced standards and global vs. local standards are interesting. I'd be interested to hear Mike's take on these issues. I think that Mike's suggestion (I think it is implemented in SMART) of excluding cases based on the observed mean and SD is interesting. I think, however, that the assumption that populations are relatively homogenous in terms or body shape and that they shift en-masse are problematic. You may (e.g.) end up excluding a vulnerable minority.

Thank you, this is very helpful! I will pass it along to my colleauge.