Statistical Analysis for Blood Sugar Measurements

Question

I've conducted a lab aimed towards finding the impact of soluble fiber intake on the change in blood sugar levels. Following an 8 hour fast, I had test subjects consume a fixed amount of carbohydrates, with a varying amount of soluble fiber, and measured their blood sugar in 20 minute intervals. This was done for two hours each test, generally until blood sugar levels dropped to normal.

5 test subjects were used for the test, each with individual blood sugar responses - for instance, each had consistently different fasting blood sugar levels before food consumption, each had consistently different blood sugar peaks, and so on. Each test subject was tested 4 times, once with no soluble fiber, and 3 more with varying amounts of soluble fiber. The raw data I am now left with is simply a collection of scatter plot graphs displaying rises in blood sugar from normal condition, a peak, and a subsequent drop to normal condition.

I can visibly see from my graphs that more soluble fiber made blood sugar peaks smaller and longer to reach. Data did not always follow this trend, due to some source of error, but a relationship is nevertheless evident. I hope to make a conclusion from this data. What method of statistical analysis could I use to conclude a relationship from these scatter plots? I would like the method to take into account different individual responses, as well as the few data sets that did not confirm a relationship.

I appreciate your help!

Answer 1

It would help to see your plots. The way I would imagine to present this data is not scatter plots but line plots:

(Each line would have a confidence interval/standard deviation from your different test subjects. This is just a schematic representation. Color represents different experimental conditions. Let me know if I misunderstood your experiment.)

Now, what is your actual question? Your hypothesis? You want to know if the time or height of the peak of blood sugar levels is different between the different fiber levels, right?

So, why not measure this? At what time do you find the peak for each subject and experimental condition and what's the blood sugar level at that time point? Your null hypothesis would be that there is no difference between the test conditions.

Here some mock data show the data spread from your 5 test subjects of the peak height measurements in the 4 different test conditions. Now you reduced your data to a question, where you could use a statistical test to see if you can reject the null hypothesis.

One thing you could do is to perform an ANOVA with e.g. Dunnett's post hoc test to compare your fiber conditions to the control condition

A problem with that is that your conditions are not really independent. And your actual question is not if any of these conditions is different. Your hypothesis would rather be if there is a relationship between fiber content and peak height/delay (null hypothesis would be: there is no relationship).

Here is the same mock data in a scatter plot (color represents test subjects, they overlap, but let's assume there are 5 data points for each condition) with a trendline showing, that with increasing fiber content there is a reduction in peak height:

You could therefore try a correlation analysis and test the statistical significance of your correlation coefficient.

As you see, the statistical analysis very much depends on your hypothesis and how you analyse your raw measurements. I hope, I have correctly identified the issue and this helps you with your analysis.