Study design -longitudinal or cross-sectional

Hi all, in the next few weeks we are going to start an experiment and we have a question about which design is better. The aim of the study is to analyse if the induction, in rats, of a gum disease affects the gut microbiome. For this, we have thought of the following experimental design:

  • Control group -> n=8
  • Group with the disease (induction for 4 weeks) -> n=8

and at the end of the experimental period, the fecal matter will be collected for microbiome analysis.

The question arises as to whether a longitudinal study is better. In this case, faeces would be collected before and after induction.

  • Which approach is better, longitudinal or cross-sectional?
  • 8 rats per group, is it a small sample size for differential abundance analysis (for example, for ANCOM-BC)?

Hello and welcome to QIIME2 Forum!

With 8 subjects per group, the study will have really low statistical power.

  • The longitudinal design looks more appealing, as it will show real change after the baseline, standardizing for more random effects (the same non-induced mice becoming a baseline for identical induced mice).
  • Yes, technically sample size is too small for standard frequentist statistical analysis. Take a look at that publication for the best small sample statistics practices or consult with a statistician about the possibility of applying Bayesian methods.



Hi @P_cc and @crusher083,

Hopefully I can add a few observation that may be helpful.

First, I agree wit @crusher083, having 2 groups of 8 will be hard and working with longitudinal data should help alleviate the issue, in theory, and that you're just at the edge of where you can apply models. Differential abundance, in particular, will be challenging because you will be paying a penalty for multiple hypothesis correction. Maybe do a back of the envelope calculation and estimate if you have 100 features how small you'd need to get a baseline p-value to see a difference.

On to a few other issues I think are important (and don't get discussed nearly enough.) First, murphy's law of microbiome says that some samples will always fail to amplify, and the more you need them, the more likely they are to fail. I would assume a10% drop out rate distributed randomly through out your population. So, in 32 samples, that's about 3. At worst, it means you don't have paired samples for 3 animals. So, you now only have 13 pairs instead of 16, or worst case, 5 vs 8. ...How does your power calculation work here?

The second issue is that rodents are coprophagic. I have caught rodents eating their own poop. Like, I walked in, he looked up and give me a shit eating grin, and then he went back to it. Because your animals are coprophagic, the cage becomes a unit of similarity for a cross sectional study. Typically you have to include that in your modeling. This creates additional levels of complication when you have to communicate your results.

There are, as @crusher083 mentioned, some decent baysian techniques that might work. You could try some of the semi-ensamble longitudinal techniques. (I'm currently low key in love with CTF for paired samples because its worked beautifully for me.) But, I think your best case scenario of 16 animals and more likely scenario of 12-13 is going to make it really hard to do the analysis no matter what you analyze



Hi @crusher083 and @jwdebelius,

Thank you both very much! I greatly appreciate your responses.

I thought so. Sample size is too small and this complicates everything...

@jwdebelius, you are absolutely right about coprophagy, and we should also take that into account. But, is there any way to prevent it? Rats would be offered food ad libitum...

Thank you in advance !

1 Like

Hi @P_cc,

I don’t think there is a humane way to prevent coprophagy in rodents. You can sometimes break the cage as a confounder in your outcome-exposure relationship by switching up your housing groups, but that depends on how your model gets induced. For instance, co-housing cre-lox littermate controls can sometimes help remove the confunding of cage effect.

However, I'm not a rodent specialist; I'm a gut microbiome generalist who kept rats for a few years because they're lovely. You might look into some of the experimental design articles specifically for rodents. I know Jerome Raes has done some work there that might be a jumping off point.