- Should survey weights be taken into account to come up with beta coefficients?
- What do I do with a negative or 0 value for any biomarker of nutrition or inflammation (e.g., ferritin, CRP)?
- Which β coefficients should we use? The ones from the internal regression correction (IRC) or those from the BRINDA regression correction (BRC)?
- How many decimal places should be used for estimates in the regression equation?
- When do you use the BRINDA external deciles for CRP and AGP vs individual survey deciles?
- When calculating the external decile, what was the rationale for excluding surveys?
- Can I add additional covariates in the BRINDA regression correction approach?
- Do I still apply the BRINDA regression correction if there is no correlation between CRP and/or AGP and the selected nutritional biomarker?
- What if I only have CRP or AGP?
- What do I do with the model intercept in the BRINDA regression correction for inflammation approach?
- What if I have a different inflammatory biomarker other than CRP or AGP (e.g., IL-1, calcitonin)?
- Can the BRINDA approach be used for populations other than preschool children, school-aged children, or women of reproductive age? For example, newborns, men pregnant women?
- Can BRINDA approach be used in clinical settings?
- Does BRINDA exclude outliers?
Either approach may be used. We suggest using unweighted regression estimates (e.g., using proc reg instead of proc surveyreg or any weighted linear regression model with design effects) since the intent of the regression correction is to account for the biological association between inflammation and given nutrition biomarker among the sampled individuals, not to produce a regression coefficient with an exact variance estimation and/or to be representative of an external population. Once the regression coefficient is calculated, we suggest accounting for complex survey design in later applications of data. back to top
What do I do with a negative or 0 value for any biomarker of nutrition or inflammation (e.g., ferritin, CRP)?
We recommend replacing the 0 with a nominal value (0.001) as logarithm of 0 is undefined. Thus, in order to Log transform (normalization) the variable, a very low number is added to the zero values to facilitate this numeric operation. The low number can be too low because it may automatically turn to zero in the software procedures. We recommend recoding negative biomarker values with missing. back to top
Which β coefficients should we use? The ones from the internal regression correction (IRC) or those from the BRINDA regression correction (BRC)?
We suggest using the IRC approach to calculate your own β coefficients. back to top
We recommend using 4 decimal places when calculating regression coefficients (i.e. nutrition biomarker-inflammation) as well as CRP and AGP reference decile values. Different statistical programs may automatically produce more or less decimal places, so this should be checked. *In SPSS: Download and double click “spss output decimals tool”. Click Utilities -> Set Decimal Places for Output Tables. Some Stata versions may need to install “Python essentials” first. However, if you are not cross-comparing with others, then the use of more or less than 4 decimal places would not hinder your work. back to top
If assay’s limit of detection for CRP is ≤ 0.1 mg/L and AGP and is ≤ 0.5 g/L, then use the BRINDA external decile; If not, use individual survey decile (10th percentile). back to top
Lab data quality (limit of detection). If assay’s limit of detection for CRP was > 0.1 mg/L or > 0.5 g/L for AGP, then that survey was excluded from the calculation of the BRINDA external decile. back to top
At this point we are NOT recommending this, as the intent is to adjust for the biological confounding effects of the acute phase response on nutrition biomarkers. We have also explored malaria as a confounder. The implications of other covariates in regression correction needs further exploration. back to top
Do I still apply the BRINDA regression correction if there is no correlation between CRP and/or AGP and the selected nutritional biomarker?
Generally speaking, regression correction should be done if both the biological relation between CRP and/or AGP and the selected nutritional biomarker has been warranted in previous literature, and there is also an evident correlation in the individual survey dataset. back to top
The BRINDA regression correction can be applied for CRP or AGP only or both. Single inflammatory biomarker macro is provided on the BRINDA website. back to top
What do I do with the model intercept in the BRINDA regression correction for inflammation approach?
Ignore the intercept, just use the beta coefficient. But do not specify a ”no intercept model”, as that would artificially influence your beta estimate. It is essential that a model with linear model with intercept is fitted. back to top
There is literature that have published ways to account for inflammation when assessing micronutrient not using CRP and AGP, specifically. For the BRINDA approach, however, the relationships between other inflammatory biomarkers and nutritional biomarkers has not been conducted. Therefore, we don’t have formal recommendations. CRP and AGP are most commonly measured inflammatory proteins in nutrition surveys, so BRINDA analyses have focused on these biomarkers. back to top
Can the BRINDA approach be used for populations other than preschool children, school-aged children, or women of reproductive age? For example, newborns, men pregnant women?
Unclear. Interrogation of these target groups is warranted in future studies. However, the BRINDA approach has been applied to other populations groups and that decision is at the discretion of the researchers. back to top
At the current time, the BRINDA approach cannot be used in clinical settings. It is intended to improve the accurate estimation of population micronutrient deficiencies. The validity of the BRINDA approach in clinical settings is warranted in future studies. back to top
No, we don’t exclude outliers. We have a flag variable for hemoglobin based on biological plausible values – “hbbivflag”. We don’t have any flag variables for other biomarkers. The rationale is that we don’t know the extreme value is due to error or due to inflammation. back to top