Best way to handle data


I am running a one-arm study in which a scale result (blood ketones) are reported across two consecutive 28 day periods (Control period=diet alone, Intervention=diet+supplement). This means I will have 56 data points per patient. I want to know if there is a significant change in both the total level of ketones and it would also be good to know if there is a difference in the daily variation of ketones. My plan was as follows:

Take a mean for each 28 day period per patient. Compare means across the two periods via paired t-test.

Calculate SD for each 28 day period, compare SDs via paired t-test as above.

Is this approach valid or should I use some form of ANOVA?

Actually, to further complicate matters, there is also a three day baseline period (no special diet) before the 56 day study starts in which patients report their blood ketones. Is there a good way to utilise all 59 data points across three different periods (3 day, 28 day, 28 day). We take more in depth ketone measures the for three days of baseline, control and intervention periods, so perhaps a better approach would be to just use 3 days across each period as then they are equal length.

I appreciate this sounds like an odd study design, but there were good reasons for it!


TS Contributor
Averaging each patient's measurements within each period would throw away statistial information. A multilevel model would use the available data mor appropriately. Maybe a repeated measures ANOVA with 2 factors ("period" with 2 levels and "measurement" with 28 levels) could be used, but I am not sure whether a factor with 28 levels could produce problems. Regarding variabilty, I think using SD as dependent variable is appropriate. Regarding baseline data, it is not clear from your description which role they play in your study. Maybe you want to use them as covariate(s) and analyse their interaction with the "period" effect?

With kind regards