Menu ENN Search
Language: English Français

Calculating a sample size to detect associations between maternal nutrition, HH food security status and dietary diversity.

This question was posted the Assessment forum area and has 4 replies. You can also reply via email – be sure to leave the subject unchanged.

» Post a reply

Briony

Nutritionist

Normal user

6 Jul 2015, 07:26


Good morning.

Hoping you may be able to help us....

We wish to calculate the association between maternal undernutrition (defined as MUAC <23cm), HH food security status (using HFIAS 4 point scale from food secure to severely insecure) and dietary diversity (using WDDS). We are hoping to explore whether pregnant women from severely food insecure households, or that have low dietary diversity, are more likely to be undernourished. We also wish to explore whether food insecure households are more likely to have low dietary diversity.

My statistical skills are not strong, and hoping you may be able to show us how to calculate the required sample size.

Information we have includes:
1. Population size: 303,000 total population; 49.4 percent of total are women. We are estimating that around 3 percent of women are pregnant at any one time.
2. We hope to include approx 8-12 villages, which have a population of around 1,100 each.
3. Approx 34 percent of pregnant women are undernourished.
4. HH food insecurity (low, medium and severe) approx 60 percent.
And womens dietary diversity scores are approx 5 food groups (range 1-9).

Because we know that dietary diversity scores fluctuate with seasonal availability of food, we also hope to collect data over a one year period. Can you clarify the best study design? (its longitudinal, though we have no control group).

All thoughts are appreciated!

thank you!


Mark Myatt

Consultant Epideomiologist

Frequent user

6 Jul 2015, 14:48

I think you need to consider the design of your study. I think there are two designs that you may consider:

(1) A cross sectional design - Here you take a survey and look at associations between risks factors (e.g. in the lower quartile of WDDS) and outcome (MUAC < 230 mm). The problem with this is that the study may have low power because you may only find a few PLWs with MUAC < 230 mm in a survey sample.

(2) A case-control study - Here you actively seek cases (as many as you can) and controls (mush easier to find). You could take more controls than cases to increase power. Such a study will usually be cheaper and have more power than the cross-sectional design.

Some notes ...

The cross-sectional design (1) will also give you prevalence of risks and outcomes.

You can increase power somewhat by not collapsing variables into binary categories. You might (e.g.) look for the correlation between MUAC and WDDS. A significant positive correlation will be seen if lower WDDS and lower MUAC (and higher WDDS and higher MUAC) are associated with each other.

I think you should think about your design and the size of effect that you think is important to detect (e.g. PLWs is the lower quartile of WDDS are X times more likely to be malnourished than other PLWs) and post back and we can work through some sample size calculations.

You may want to review the relevant sections of this basic epideomiology text.

I hope this helps.

Briony

Nutritionist

Normal user

7 Jul 2015, 11:38

Hi Mark,

thank you for your response. Very helpful. However, I am struggling to identify an appropriate study design. A cross sectional would provide us with a snap-shot only (which doesnt factor in the limitations of DDS). Im assuming that the case-control design would be best. However, if we use this design, can we invite women to complete the survey over a one year period?

We wish to invite all pregnant women to participate in selected communities over a one year period. Women can form two groups, those undernourished, and those nourished (defined by MUAC) (we were initially considering one group, but this is possible). the purpose of inviting women over a one year period is so we can see how seasonality affects DDS and HFIAS. Each pregnant woman completes one questionnaire only (that said, we could do follow-up questionnaires..).

Your thoughts (and others!) are appreciated.

thank you!

Anonymous 425

Normal user

7 Jul 2015, 13:20

I think Mark is referring case-control as retrospective. But you are referring cohort study or prospective.

Mark Myatt

Consultant Epideomiologist

Frequent user

8 Jul 2015, 10:50

Sorry, I misunderstood your requirement.

A cohort design might work. This can be quite tricky to do as you have to keep track of each member of the cohort and make efforts to follow-up members who have moved away (i.e. because moving might be associated with low DDS or high HFIAS) and to distinguish losses due to deaths (which may also be associated with your risk factors). The cohort design will allow you to look at multiple outcomes (e.g. maternal survival, birth accident. SGS/LBW, &c.). Analysis can be complicated if you want to lag exposures (i.e. see if low DDS twelve, nine, or six months ago affects your outcomes of interest).

Another approach which will be simpler and (probably) cheaper would be to do a set of small cross-sectional surveys (e.g. three months apart). This will allow you to track DDS, HFIAS, &c. over time. Analysis could be per survey and (more difficult) over all surveys.

What do you think will suit your needs best?

Back to top

» Post a reply