I am analysing human neuronal samples in culture after treatment with different concentrations of NGF. For each patient I am analysing around 30-40 neurons for each of 4 concentrations (including a negative control). I'm using Graphpad Prism for the analysis.
As expected, there is a lot of variability between human samples, so I am more interested in the changes between samples from the same patient. I can easily do an ANOVA per patient to analyse the results, but I'm looking for a way to then see if I get statistical significance overall, using data from all the patients. I could do a repeated measures ANOVA to keep the 'pairing' between samples from the same patient, but I can only input a mean for the neurons from each concentration, so I lose all the information about the variability of the neurons within each sample. Is there any test I can do which would keep this information, while also allowing me to keep the pairing for each patient?
Thanks!
Won't be doable in GraphPad, but a mixed model with a random effect for patient would be a better approach - since samples are clustered within an individual. And since you aren't interested in interpatient differences, the effects observed from the model estimates between concentrations will be interpreted for a "typical individual".
The fact that have several (20-40) neurons per patient is irrelevant for the statistical analysis of the treatment effect. Having more replicates per patient allows you to reduce the "technical variance". Knowing that variance and the biological variance is required to plan future studies, because it allows you to judge if it would be beneficial to increase or decrease the number of neurons per patient that you would have to measure. You can use a hierarchical (aka mixed) model, which will give you the these variances separated, as Joe Stanek mentioned. And I also think that Prism can't do that.
If you are just interested in the significance of the treatment effect, it's sufficient to average the values per patient first, and do a standard analysis (regression or ANOVA) using these averages. The result (for the treatment effect) will be identical to that from the hierachical model.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA