Hi Patrice,

I am not familiar with the data that you have, if you could elaborate then I may be able comment further. 

Regarding PCA, PyChem by Jarvis et al. is a opensource software package written in python that would be suitable for initial PCA analysis. http://pychem.sourceforge.net/

Adam

Thanks Adam. I measured a total of 17 cytokines and chemokines (pg/ml) after 6-day stimulation of whole blood with mycobacterial antigens. Cord blood, and infant samples collected at 7 different time points (weeks after birth) were cultured, and supernatants were assayed using Luminex. I would now like to analyse the results to look at longitudinal changes in responses with age, and also the effect of maternal factors (age, TB infection, BCG scar etc etc) and infant factors in these responses. I thought PCA would help, but I haven't used it for analysis of longitudinal responses. And I am only familiar with STATA not R.

Hi Patrice, 

I think being outside of your field I am having difficulty understanding what exactly your dataset includes, from my perspective I would require the data to be arranged within a matrix/table, where the variables (in this instance the 17 cytokine concentrations) occupy the columns and the re-measurement of these (over the 7 time points) would occupy the rows. I dont understand  how the maternal factors are variables in this dataset without the whole dataset being repeated for a number of patients etc. 

Many thanks 

Adam

Hi Patrice,

You have a unique dataset there to nail down what happen in early weeks of life on BCG response in your setting. I assume all your new-borns received BCG at bird if not this information must also be captured.

I am not sure PCA would be of much help here because what you want to see is the pattern of multiples cytokines/chemokines responses overtime and not just how the different group of new-borns cluster with regards to specific parameters at particular time points (what PCA is best at).

You can do a lot in stata, a statistician will best advice you. You can try some mixed-effect  regression model analysis, which helps you to analyse such complex longitudinal data accounting for missing value, multiple correlations over time and adjusting for  confounders. I guess you have defined groups and you are trying to understand how their responses vary overtime under the influence of the other co-morbidity and mother's parameters etc......

Hope this help.

BW,

Leopold