Malnutrition & Disability, indicator and age

I don't think the age-group it makes much difference except in terms of feasibility. You need to have a sufficient number of cases (malnourished) and a sufficient number of exposed (i.e. disabled) and you'd probably want to avoid zero cells in tables. Both malnutrition and significant disability are rare conditions. These considerations would favour a broader age-range. I would not extend the age-range much above ten years so as to avoid issues of sexual maturation.

You may want to go for a study that finds cases of disability by (e.g.) chain-referral sampling and use familial and neighbourhood controls. This a bit "backward" as selection would be on exposure (disability) rather than outcome (nutritional status) but that should not be a problem. It would, I think, be more cost-effective, easier to plan, and you would have some control over the eventual power of the study.

One concern that I have is that by "malnutrition" you probably mean below some threshold of an anthropometric index such as weight-for-height. The problem here is that some disabilities (e.g. curvatures of the spine) make measurement of height difficult. Height-for-age has the same problem. If you use MUAC then you would probably want to pick the favoured arm. Weight-for-age might be better but you'd need good age data. Severe disability (e.g. absence of limbs) might make this difficult. What are your thoughts on this?

Why have the two sets of controls? It seems to me that age / sex matched neighbourhood controls might be easier and you control for age, sex, and "neighbourhood" (usually a melange of class, tribe, family, livelihood zone, &c.).

I hope this is of some help.

Some answers / things to consider - probably not well thought through:

1. Anthropometry in adolescents is very tricky and needs to be adjusted for sexual maturation. I think this probably holds for the pre-adolescent growth spurt (Woody knows more about this than I do - he monitors these forums and may reply separately). I think that a maximum age of 9 or 10 years would be sensible. I am not convinced that you need a minimum age. I tend to avoid measuring the length of young (i.e. < 6 months old) children in the field. You need special equipment and I worry about rough handling in survey contexts.

2. OK. The "Key Informant Methodology" is a good start. I would have a snowball type sample as is used in CSAS / SQUEAC / SLEAC and use carers of disabled children to identify other disabled children.I was confused by "cross sectional survey" in your original post.

3. This is a common approach. It is covered in here on EN-NET. I think that it is simplest to use use in adults because, in children, limb to trunk length changes with age. The sitting to standing height ratio (SSR) declines steadily from birth to puberty. This probably accounts for a some of the estimation error seen in the reference you supply above. It seems to me that error will increase as the age range of the study widens. This means that you should probably include age (if possible to collect with reasonable accuracy) in the model used to estimate height. I'd probably include sex or build separate models for males and females. You could use your control group as the data source for finding the height estimating equation (you MUST do this with local data as SSR varies with ethnicity, climate, altitude and diet). There may be advantages of using estimated height in both the case and control group as this would means that any bias was the same for each group (ask a statistician about this).

4. There are gains in power that come with having multiple controls per case. These decrease with increasing additional controls so that the increase in power after five controls per case is marginal. I think that your use of active case-finding will mean that you will find sufficient cases. This would have been a problem if you were to rely on (e.g.) a SMART survey to provide disabled children. I would consider employing a matched case-control design and matching on age, sex, and neighbourhood.

Any help?

How will you treat the within household case-control pair? It is possible (likely?) that a lot of the antecedents for your “nutrition” outcome will be fixed at the household level. This suggests a matched analysis (i.e. on HH) would be appropriate. There is an issue with excluding singletons (no with-HH controls) and also with age (cases and controls may be different WRT in age and, with children, small differences in age can be important). I think that you should talk these through with a statistician or fellow epidemiologist. Maybe present as a lunchtime seminar and present the issues and ask for suggestions.