This is an interesting report. Most of the experience using MUAC for CMAM admissions has been with the more extreme 110 mm threshold and we need more reports form programs using the 115 mm threshold. The first point to make is that we have to be very careful when we use terms such as gender bias. Just because an indicator selects more girls than boys does not mean that it is "gender biased". If girls are more at risk of an adverse outcome than boys then the indicator should select more girls than boys. MUAC is used in CMAM programs because it is the best practical predictor of near term mortality. If, in your setting, girls have a bad deal compared to boys when it comes to nutrition, infection, access to health services, &c. and this puts them at risk of morbidity and mortality then you would expect MUAC to select more girls than boys. This is not a gender bias (it is the opposite of a gender bias). In such a setting, an indicator that did not select more girls than boys would be gender biased. You have to ask yourself (and consult census and epidemiological reports) whether, in your setting, girls probably get a worse deal than boys. They tend do in many settings than I have worked in. The second point to make is that comparing WHZ and MUAC is not useful as they measure different things and the predictive power (i.e. for near term mortality) of WHZ is usually the weakest of all practical indicators. It is possible that, in your setting, WHZ is exhibiting a gender bias in favour of boys. Looking at your data ... The observed sex ration (boys : girls) is 349 / 666 = 0.5240. I have compares this with the sex ratios observed in 560 nutritional anthropometry survey datasets from 39 countries and find: Complete database (458951, 15014 cases with MUAC < 115 mm) : Number of boys with MUAC < 115 mm := 6852 Number of girls with MUAC < 115 mm := 8162 Sex ratio (boys : girls) := 0.8395 I calculated the sex ratio in 87 datasets with at least 50 cases (so as not to confuse myself with very low prevalence datasets in which (e.g) 1 boy and 2 girls gives me a sex ratio of 0.5) and got: Minimum := 0.3182 Q1 : = 0.6937 Median := 0.8571 Mean := 0.8725 Q3 := 1.0170 Max := 1.5530 This confirms your finding of more girls than boys being selected. Your sex ratio is quite extreme (i.e in the bottom 3.5% of the distributions of sex ratios observed in the 87 datasets above). It may be that girls in Nepal get a really bad deal compared to boys but ... we have to be very careful about assumptions here ... your data are workload data not population data. It could be that, somehow, your program has a bias against boys. I would definitely check out case-finding and recruitment activities as well as trying to get some idea of cultural attitudes to the program and to malnutrition (e.g. it might be very shameful to have a thin boy but not shameful to have a thin girl and this might effect attendance at the program). NOTE : I am not ruling out the possibility that MUAC may have a gender bias. If you look at the WHO MUAC/A reference table: [url]http://www.who.int/childgrowth/standards/second_set/acfa_girls_3_5_zscores.pdf[/url] [url]http://www.who.int/childgrowth/standards/second_set/acfa_boys_3_5_zscores.pdf[/url] You will see that, at younger ages and under conditions ideal for growth, girls tend to have slightly lower MUACs than boys. This means that a fixed MUAC threshold will tend to select slightly more girls than boys (we have to be wary of how we use reference data since we seldom work in settings where conditions are ideal for growth). I don't think this difference alone can account for your data though. I hope this helps.

This is an interesting report. Most of the experience using MUAC for CMAM admissions has been with the more extreme 110 mm threshold and we need more reports form programs using the 115 mm threshold. The first point to make is that we have to be very careful when we use terms such as gender bias. Just because an indicator selects more girls than boys does not mean that it is "gender biased". If girls are more at risk of an adverse outcome than boys then the indicator should select more girls than boys. MUAC is used in CMAM programs because it is the best practical predictor of near term mortality. If, in your setting, girls have a bad deal compared to boys when it comes to nutrition, infection, access to health services, &c. and this puts them at risk of morbidity and mortality then you would expect MUAC to select more girls than boys. This is not a gender bias (it is the opposite of a gender bias). In such a setting, an indicator that did not select more girls than boys would be gender biased. You have to ask yourself (and consult census and epidemiological reports) whether, in your setting, girls probably get a worse deal than boys. They tend do in many settings than I have worked in. The second point to make is that comparing WHZ and MUAC is not useful as they measure different things and the predictive power (i.e. for near term mortality) of WHZ is usually the weakest of all practical indicators. It is possible that, in your setting, WHZ is exhibiting a gender bias in favour of boys. Looking at your data ... The observed sex ration (boys : girls) is 349 / 666 = 0.5240. I have compares this with the sex ratios observed in 560 nutritional anthropometry survey datasets from 39 countries and find: Complete database (458951, 15014 cases with MUAC < 115 mm) : Number of boys with MUAC < 115 mm := 6852 Number of girls with MUAC < 115 mm := 8162 Sex ratio (boys : girls) := 0.8395 I calculated the sex ratio in 87 datasets with at least 50 cases (so as not to confuse myself with very low prevalence datasets in which (e.g) 1 boy and 2 girls gives me a sex ratio of 0.5) and got: Minimum := 0.3182 Q1 : = 0.6937 Median := 0.8571 Mean := 0.8725 Q3 := 1.0170 Max := 1.5530 This confirms your finding of more girls than boys being selected. Your sex ratio is quite extreme (i.e in the bottom 3.5% of the distributions of sex ratios observed in the 87 datasets above). It may be that girls in Nepal get a really bad deal compared to boys but ... we have to be very careful about assumptions here ... your data are workload data not population data. It could be that, somehow, your program has a bias against boys. I would definitely check out case-finding and recruitment activities as well as trying to get some idea of cultural attitudes to the program and to malnutrition (e.g. it might be very shameful to have a thin boy but not shameful to have a thin girl and this might effect attendance at the program). NOTE : I am not ruling out the possibility that MUAC may have a gender bias. If you look at the WHO MUAC/A reference table: [url]http://www.who.int/childgrowth/standards/second_set/acfa_girls_3_5_zscores.pdf[/url] [url]http://www.who.int/childgrowth/standards/second_set/acfa_boys_3_5_zscores.pdf[/url] You will see that, at younger ages and under conditions ideal for growth, girls tend to have slightly lower MUACs than boys. This means that a fixed MUAC threshold will tend to select slightly more girls than boys (we have to be wary of how we use reference data since we seldom work in settings where conditions are ideal for growth). I don't think this difference alone can account for your data though. I hope this helps.

I think Mark spotted the two reasons why MUAC may return a higher number of girls with SAM First, boys and girls have a different growth, and this is reflected also in MUAC. At any time, girls on average will have slightly lower MUAC than boys, even if well nourished. If you define SAM based on a fixed cut off for both sexes, you will always get, all other factors being equal, a slight excess of girls. Second in some societies, boys and girls do no thave the same risk of SAM for issues related to gender bias. Gender bias has been reported previously in south Asia and is well documented for mortality, an objective criteria. It is likely to occur as well for the risk of SAM. A few years ago, I took part in a study where we could assess the proportions of deaths related to SAM in rural Bangladesh. The study suggested that girls had more than twice the risk of dying from SAM than boys. See: Fauveau et al. The contribution of severe malnutrition to child mortality in rural Bangladesh. Food Nutr Bull 1990; 2 (3): 216-9. Available on the Food and Nutrition website. I suspect gender bias is likely to be involved when you have a major sex imbalance in your SAM patient. In this situation, you should not consider this as a diagnostic bias but rather as an argument to give more support to girls. Diffference in growth may however explain small differences usually seen as reported by Mark (sex ration 0.8).

I think Mark spotted the two reasons why you can have a higher number of girls than boys using MUAC. First, boys and girls grow differently, and this is reflected in MUAC growth. Even in well nourished cohort such as the WHO MGRS cohort, at all ages, on average, girls have smaller MUAC. When using MUAC with a fixed cut off to detect SAM, you are likely to get slightly more girls than boys. Second, there maybe situations in which there are indeed more SAM in girls for issues related to gender bias. Gender bias has been extensively documented in south Asia and this is even reflected in a different mortality in many areas. This is likely to occur as well for SAM. A few years ago, I took part in a study in rural Bangladesh which attempted to assess the contribution of SAM to child mortality. The study suggested that girls had more than twice the risk of dying from SAM than boys. See: Fauveau et al. The contribution of severe malnutrition to child mortality in rural Bangladesh: Implications for targeting nutritional interventions. Food Nutr Bull 1990; 12 (3) 215-9. http://www.unu.edu/unupress/food/8F123e/8F123E05.htm#The%20contribution%20of%20severe%20malnutrition%20to%20c An excess of SAM related deaths in girls in this context suggest that there is something else than a diagnostic bias and that girls do need more support. Your finding that a nutritional survey returned the same number of SAM boys and girls should be taken with caution. Traditional 30 x 30 surveys have a design poorly adapted to understand SAM epidemiology, as SAM numbers found typically in a survey are usually too small to make meaningful comparisons between different groups. Your finding that boys and girls have the same WFH on admission is intriguing, but it would be interesting to know how they are recruited. Do you have an active case finding in the community ? I doubt you can do that with WFH. Or do you take patients which turn up ? You may have as well a problem of different access to care for boys and girls.

I have also the same issue. of the total SAM cases admitted (MUAC <115mm), the number of boys and girls in District1 was 214 and 333 respectively. In District2, it was 40 and 77 respectively.

Dear Regine, Did you meausure MUAC in anthropometric surveys in Nepal? What was the prevalence of <115mm MUAC among boys and girls in your surveys? I think that the sex ratio of severe cases under MUAC in a ideally unbiased sample such the survey's one and from the same population (nepalese), may rule out the bias during screening and admission and help in interpreting the sex ratio you reported.

The reference that André posted and a couple of the further references in that article make for interesting reading regarding sex-differential in SAM and mortality in rural Bangladesh. In such a setting as male:female sex ratio in SAM cases of 0.5 seems reasonable. Anonymous 303 is correct. You can check for diagnostic / selection bias by comparing your workload sex ratio to the population sex ratio for SAM. BUT ... I agree with André regarding the difficulty of using two stage cluster sampled surveys (e.g. 30-by-30 or SMART surveys) to estimate the population sex ratio for SAM because the sample size is usually too small to find enough cases of SAM to make an accurate and reliable estimate of the sex ratio. For example, if prevalence is 1% and the sample size is 900 you will find about 9 SAM cases in the survey sample. Also, a cluster sampled design is a poor choice for investigating a clustered phenomena (in such settings the "design effect" is high and confidence intervals very wide). With a condition such as SAM which is often associated with infectious disease there may be considerable clustering of cases. A better method might be to use a survey design like the CSAS coverage method which uses an "optimally biased" sampling method to find SAM cases. It may be worth looking at any survey data you do have but be aware of its limitations. It might be possible to estimate the population sex ratio using higher MUAC thresholds (see below). Kiross ... where is this data from? The sex-ratios are District 1 = 0.6426, District 2 = 0.5195, Combined : 0.6195. Not as extreme as the Nepalese data (see above). More data (from what I have to hand - most of the routine program data that I have does not include sex) ... Data from OTP in SNNPR, Ethiopia admitting on MUAC < 110 mm - the sex ratio is 0.8424 (246 boys to 292 girls) which is close the mean found in my analysis (above). Data from a case-finding (door-to-door screening) exercise (small numbers) in Wollo, Ethiopia using MUAC < 115 mm had a sex ratio is 0.6667 (12 boys and 18 girls). Data from another case finding exercise is the same area using MUAC < 115 mm had a sex ratio of 0.6857 (48 boys and 70 girls). Data from an OTP program in central DRC using MUAC < 110 mm had a sex ratio of 0.9080 (987 boys and 1087 girls). So we have ... Regine (Nepal) : 0.5240 Kiross (?) : 0.6195 Mark (Ethiopia 1) : 0.8424 Mark (Ethiopia 2) : 0.6667 Mark (Ethiopia 3) : 0.6857 Mark (DRC) : 0.9080 Weighted average : 0.7610 taken from program or case-finding data. There does appear to be a tendency for these programs to admit more girls than boys. The "Ethiopia 2" and "Ethiopia 3" datasets suggests that in Wollo (in April - June 2003) the population sex ratio for SAM is somewhere about 0.68 (data from door-to-door screening in 43 villages). Does anyone else have program or screening data that they can share with us? It does not matter if the MUAC threshold is 115 mm or 110 mm. It seems to me that we might be able to use this as an indicator for equitable coverage ... we could probably use survey data with a higher MUAC threshold (e.g. < 125 mm) to estimate a sex-ratio and then compare out program intake to the population estimate. If (e.g.) we had a population sex ratio of 0.68 and a program sex ration of 0.85 we might suspect that we were not admitting enough girls. Just an idea. Perhaps we will need better survey methods for this to be useful.

I looked at the boy:girl ratios for the same district in Nepal during the 30x30 nutrition survey and the coverage survey done using SLEAC. Nutrition Survey 30x30 (n=961) MUAC<115 boys:girls=0.9 but case load very low for interpretation! Coverage Survey SLEAC (n=3316) random selection of clusters and all children<5 screened house-to-house Sex ratio (boys:girls) for MUAC<115 =0.3 case load very low for interpretation! From the 41 SAM cases found during the SLEAC only 7 cases were currently admitted in the OTP (5 girls and 2 boys). Low case load. I also checked on average age of boys and girls admitted into OTP: Girls are in average 2 months younger than admitted boys which could explain some difference in MUAC. Average age for OTP admissions girls=15.66 months boys=17.74 months Question? Why are girls SAM at an earlier stage than boys? Assumption: feeding practices are different. I checked breastfeeding according to age and gender (no other related data available on this): Children (%) breastfed in relation to gender and age 6-11months: boys:80.6%, girls: 77.8% 12-23 months: boys 80.4%, girls: 78.4% 24-35 months: boys 70.3%, girls: 65.2% 36-47 months: boys: 27.3%, girls: 45.5% 48-59 months: boys: 20.0%, girls: 40.0% Breastfeeding here means breastfed in general and is not saying anything about exclusive, frequency etc. Differences within the first 3 years are minimal. In age 3-5 years girls are actually more often breastfed than boys. Data comes from the OTP admissions and is not population data. I also checked on a possible link to illnesses and with this a possibly higher mortality risk associated with the lower MUAC for girls. out of all 614 OTP admissions analysed 40.4% had also an illness recorded. Out of this 40.4% girls are the majority with 60.9% and boys making up for only 39.1%. Single or multiple symptoms recorded? out of the sick girls 28.5% show multiple illnesses recorded similar for boys (33%) SAM girls admitted in OTP are suffering more often from health problems than boys. However they are not more often suffering from multiple medical problems. In regard to the screening process and a possible gender bias: Main screening is done by community volunteers using MUAC during their normal daily duties as a health volunteer (MOHP system). Assumption1: they identify equal number boys and girls à care takers with girls tend to follow referral advice more often and therefore more girls are admitted in OTP by MUAC à with boys having a much higher status in the Nepal society I would assume that boys are better taken care of (and with this less often sick/malnourished) but in case of any problems immediate consultation at health facility. Assumption 2: more girls are identified with MUAC because girls are more often SAM (MUAC) à gender discrimination of girls à yes most likely Screening at health post level is done by MUAC and also WHZ for (in theory) all <5 children seen during consultations for whatever reason Assumption 1: girls are more often sick and therefore brought more often for health check and at that time nutritionally screened and admitted à possible as girls seem to be sick more often (see data) à check HMIS data to confirm boy:girl ratio for health consultations! What becomes clear to me is that in Bardiya district malnutrition associated with higher mortality risk seems to be linked to gender discrimination. Discrimination of girls/ women is a common practice in Nepal, more hidden than in some other countries I worked in but it is definitely practiced. However I am really surprised that at the age of 15 months this is already that visible. Will try to find other data indicating gender related IYCF practices. We did another nutrition survey 30x30 in Jajarkot district in Nepal. I will check on the sex ratio to see whether there are similarities.

A short comment following additional info provided by Regine. Gender discrimination can start very early in some societies, as shown by differential abortion rates in places where there is severe gender bias + access to in utero sex determination with ultrasound. Often reported all over south Asia. Breastfeeding does not mean necessarily better nutritional status. Actually the opposite if often seen after 12 months of age: breastfeeding is associated with poorer nutritional status. This has been reported in many places. See: Grummer-Strawn LM. Does prolonged breast-feeding impair child growth? A critical review. Pediatrics. 1993 Apr;91(4):766-71. This seems related to reverse causality: when children are malnourished, mothers tend to breastfeed them longer. We saw that in Bangladesh years ago in a longitudinal study, showing that nutritional status in children that will soon stop breastfeeding was better than those who continued breastfeeding. And also that survival of breastfed children was better than for non breastfed, despite a worse nutritional status. Briend A, Bari A. Breastfeeding improves survival, but not nutritional status, of 12-35 months old children in rural Bangladesh. Eur J Clin Nutr. 1989 Sep;43(9):603-8. This problem of reverse causality has been confirmed by others. See: Marquis GS, Habicht JP, Lanata CF, Black RE, Rasmussen KM. Association of breastfeeding and stunting in Peruvian toddlers: an example of reverse causality. Int J Epidemiol. 1997 Apr;26(2):349-56. http://ije.oxfordjournals.org/cgi/reprint/26/2/349 So more breastfeeding in girls in your settings does not rule out discrimination.

Thanks, Regine, for the comprehensive follow-up. I understand the problem with the small numbers from the 30-by-30 surveys. Can you do the same analysis but with MUAC < 125 mm? This will show the sex ratio for GAM. You have 41 cases from SLEAC using an exhaustive sampling strategy. You find a M:F sex ratio of 0.3. This suggests that you found 10 boys and 31 girls with SAM. Is that right? We can do something with this. I calculated a 95% CI using this data and got sex Ratio = 0.3226 (95% CI = 0.1411, 0.6752). This is consistent with your admissions data. Do I have the basic numbers correct here? The age difference could explain some of the difference. In terms of MUAC/A, the boys are probably more wasted than the girls. Approximate MUAC/A for 115 mm for 16 month old and 18 month old boys are: 16 month girls : Median = 144 SD = 10.3 115 mm = (115 - 144) / 10.3 = -2.8 z-scores 18 month boys : Median = 148 SD = 10.0 115 mm = (115 - 148) / 10.0 = -3.3 z-scores But we need to bear in mind that MUAC/A is a worse predictor of near-term mortality than a simple MUAC threshold. The thing about the MUAC/A reference is that it is al about rich and healthy kids. Do you see many of these? The breastfeeding data are interesting but difficult to interpret WRT this question. We'd really want population data. The concurrent illness data are interesting. It does seem that girls tend to be more likely to have a coincidental illness than boys. This should set your mind at rest a little since it suggests that many of the girls you admit will probably get worse without your intervention. If you think in terms of MUAC/A then you can characterise these girls as "borderline severe wasting with complications". This would be a legitimate CTC admission. The health seeking behaviour data is also interesting. It seems to me that, in your setting, malnutrition is often an outcome of infection. If boys are treated better than girls then you would expect to see more SAM girls than SAM boys. It might be useful to ask around about this. You may also be able to find material on gender and heath in Nepal in (e.g.) NGO and UN reports. The Asian Development Bank report shows a mixed picture. I just did a quick literature search and the retrieved articles all support your hypothesis that boys have preferential access to health services including earlier treatment seeking. It seems that you are picking up some of the mess caused by gender inequality. It is a good job that someone is!

30-by-30 cluster GAM with MUAC<125mm: n=961, boys:45, girls:60 coverage survey: 41SAM cases, out of this: boys: 6: MUAC<115mm, WHZ<-3:11 girls: 19 MUAC<115mm, WHZ<-3: 3, oedema:2 (MUAC is used as a priority admission criteria and in case MUAC and WHZ meets admission criteria the child is registered under MUAC admission), oedema is always registered seperate. Rich and healthy kids: caste/ethnic group belonging is a big issue in Nepal and therefore we included a question to this into the OTP admission card but on voluntarily basis. From 614 OTP admissions 314 did not give this information but from the remaining 300 we know that we have admissions: no information: 314 (51.1%) low caste: 22 (3.6%) middle caste: 237 (38.6%) upper caste: 41 (6.7%) I put this into correlation with the actual caste distribution in the district. According to population data and using the 300 children with caste information as 100% the following comes up: low caste: pop in the district: 3.94% but 7.33% of OTP admissions middle caste: pop:45.32% but reflecting 79.0% of OTP admissions upper caste: pop. 50.74% but only 13.67% of our admissions. This shows that the middle caste in the pilot district is having the majority of SAM children admitted but it does not really say whether we have highest malnutrition among this group. Maybe we just reach them best. During the coverage survey only 1 OTP reached >50% coverage, for 3 no SAM case was found in the cluster and the remaining 7 were below target of 50%. So from this I do not have the data I would need to make a statement. However in another district we analysed malnutrition in correlation to caste and found that low caste children had a three times higher risk of malnutrition than the middle caste with the upper caste coming second. But this is a hill district and not from the terai as above data and therefore a comparison very tricky.

Cluster survey data : 0.75 (95% CI = 0.58; 0.93) Coverage survey data (using 6 and 19) : Sex ratio = 0.32 (95% CI = 0.12; 0.59) Your observed sex ratio for cases was 0.52 (consistent with the coverage data). I was being facetious about rich and healthy kids. Sorry. the point I was trying (and failing) to make is that a reference built around ideal growth might not be the best thing to use when working with kids who are extremely unhealthy (e.g. over half of the kids in your program would probably die within a few months without you intervention). Your epidemiology is correct ... it is very difficult to conclude much from admissions data alone. This is an interesting issue ... is there anyone out there with more data?