Designing and implementing a nutrition program takes both a lot of thought and planning as well as coordination and effort. Causality for Program Design will help guide an analyst through some of the more probing routines to extract information from data that will greatly enhance the design of a nutrition program that addresses micronutrient malnutrition. Evidence for causality between an intervention and improved nutrition status lends some assurance that the intervention effort will succeed if continued or expanded. Often the answers to successful micronutrient programming lay buried in data sets that have only been gathering dust. This section should offer some practical steps toward unearthing a wealth of analysis skills to improve micronutrient program planning.
Due to the lack of micronutrient data (especially sub-clinical data) collected in this current day, it is first important to recap what type of data will likely be available for this type of analysis, and therefore what is presented in this section. As stated previously, the MN PANDA will cover analysis of VAD, IDA, and IDD only (for the reason that there is not adequate survey data for other deficiencies such as Calcium, Riboflavin, Selenium, Zinc etc.). Below is a listing of the outcome indicators used in this section:
| Nutrients | Primary Indicator(s) |
Comments |
|
Vitamin A |
Nightblindness and Signs of Xeropthalmia Serum retinol* |
Nightblindness and Signs of Xeropthalmia
are often assessed visually by a clinician but they are all later stages of deficiency and
are only useful in some populations (relatively rare events) and even nightblindness does
not detect all sub-clinical VAD. Serum Retinol in children is the suggested indicator because it is the best indicator of sub-clinical VAD, although facilities are often not currently available in developing countries. |
|
Iodine |
Goitre
by palpation Urinary Iodine*
|
Goitre assessed by palpation is
the most used population-wide outcome indicator in highly endemic areas initially, but is
inadequate as progress is made. Goitre in school age children is a good indicator until
the prevalence are quite low. Urinary Iodine (median value) in school age kids is likely the best measure for long-term monitoring of iodine status and adequate salt iodization. Iodized Salt Consumption can be used successfully to assess program coverage for salt iodization. |
|
Iron |
Hemoglobin
Serum ferritin* |
Hemoglobin is inadequate for
measuring iron reserves, although because anemia prevalence is initially very high in
developing countries, this measure is practical and adequate. Serum ferritin is likely the best measure of iron status (reserves as well), but it is still not often feasible in developing countries, especially on a large scale. |
* These indicators are listed because they are highly suggested for future use, although not often seen in the past on large-scale surveys.
Identify Associations
Continuous Outcomes
Initially, on the path to analysis, it is important to list out the basic variables of interest that have been collected on nutrition and nutrition related factors. Separate those that will be used as outcome indicators (at times) and those that will be used as a measure of risk or as an intervention. Look again at the conceptual framework to begin forming a mental picture of the possible relationships between these variables—this is the beginning of the PANDA method, forming SPIDERS. To revisit the UNICEF Conceptual Framework, click here. The same pathways exist for micronutrient malnutrition as for general malnutrition, except the foods and diseases are more specific to the nutrient(s) of interest. A spider might look like the following:
INSERT SPIDER using VAD-IDA example
Running some basic tabulations such as compare means and cross tabs is the next step in trying to identify associations and begin testing for evidence of causality. Using the compare means routine is very useful for looking at one and two-way analysis with continuous outcome variables, such as hemoglobin scores, serum retinol scores, or urinary iodine. For example, look at the following one-way analysis with hemoglobin and birth size:
Example of a compare means using hemoglobin as the outcome and birth size as the categorical independent variable
Birth Size |
Mean Hemoglobin |
N |
Standard Deviations |
Normal to large |
9.706 |
971 |
1.414 |
Less than average to very small |
9.510 |
400 |
1.450 |
Total |
9.649 |
1371 |
1.427 |
* Significant at the 0.05 level (significance= 0.020, F= 5.385) in an ANOVA test
An ANOVA shows a significant difference in hemoglobin values for those that are "low birth weight" indicated by recording birth size and those that are not low birth weight. This difference suggests there is an association between birth weight and anemia that will be worth exploring further. This is not only suggested by this analysis, but also in the scientific literature, which has shown that premature babies are more likely to become anemic because they have a fewer red blood cells at birth. Usually, babies are born with high hemoglobin levels and when many cells are hemolyzed in the first few weeks of life, the iron liberated is stored in the liver and spleen. This extra iron storage is accessed in the first few months as the baby grows to increase blood volume. Low birth weight is an identified risk factor for anemia; therefore it is a concern that could be addressed in an intervention program by providing iron supplements to the mother during prenatal care clinics.
This type of analysis can also be layered to show hemoglobin outcome by sex and birth size within each sex. This will give a better idea of possible interaction between the variables, as discussed in PANDA Analysis. Look at the following example of a layered compare means without interaction:
Birth Size |
SES |
Mean Hemoglobin |
N |
Average to large |
Not low |
9.825 |
422 |
Low |
9.615 |
549 |
|
Less than average to very small |
Not low |
9.573 |
147 |
Low |
9.473 |
253 |
|
Total |
Not low |
9.760 |
569 |
Low |
9.570 |
802 |
Graph of Mean Hemoglobin by birth size and SES
(no interaction)
A correlation matrix might also offer some insight into the linear relationships that exist in the data set. This will help to identify some of the variables that you will likely carry into regression modeling to test for evidence of causality.
From this point, it might be good to test in a regression model to see if the relationship between birth size and lower hemoglobin stays strong even when controlling for other possible influencing factors such as SES, vitamin A status, and sanitation access. Again, the difference in what the regression model can do is dramatic. It is the way to get at the heart of the research question. This will involve model building, just as it was shown in the Analysis section, but tailored to look at micronutrient deficiency as the outcome. Specific examples are shown within the Section 5: VAD, IDA, and IDD.
Categorical Outcomes
This same type of analysis using mean outcome with layers of independent variables could be used to explore and display differences in an outcome by multiple factors of interest, but when associations are identified then usually the next more rigorous test of the association is a regression model.
In some cases, the outcome variable will not be a continuous variable but a categorical, such as nightblindness or goitre. In these cases, the analytical techniques will vary slightly, although we are still going to begin by looking for associations and follow through with testing them using regression analysis. Initially, chi-square tests are used for testing the hypothesis that the row and column variables are independent or if there is a likely association. The next step would be to use logistic regression analysis, which is useful to predict the presence or absence of an outcome (e.g. goitre) based on values of a set of predictors (e.g. age, eco-zone, consumption of salt). Logistic regression is similar to linear regression (as was used previously in the Analysis section), but suited for models with a dichotomous outcome. Logistic regression coefficients can be used to estimate odds ratios for each of the independent variables in the model.
Presentation
There is a lot of thinking, testing, re-testing, and research that will go into making a conclusion that a relationship is likely causal, though it will be a fruitful pursuit when a program is successful because of discerning analysis and program planning that reflects this.
Using the data analysis results for report writing or presentation is a final, crucial step. This will allow you to bridge the information to many other groups that will likely influence the success of the suggested program design. The data can be presented so that it persuades policy makers, health workers, food production managers, sanitation specialists, etc. to look into the benefits it may provide and shows the role they might play in the success of the program. Here are some hints for effective presentation of data: