I have data for 200 patients' blood sugar levels before and after an intervention. There are multiple readings both before and after the intervention.

In order to see if the intervention had a significant effect on the blood sugar, I plan on doing a paired t-test on the means of each persons levels before and after intervention.

My question is, is there a test that I can perform to analyze the trend in blood sugar readings, both before and after intervention? I get the idea that the paired t-test loses information about any trends.

The purpose of this is to see how peoples blood sugars trended (via absolute measure, or through categories such as prediabetic, hyperglycemic, etc) before and after intervention.

Thanks in advance for your help, it is greatly appreciated.


This data does not tell you much about the intervention due to the lack of a control group. Any causal interpretation is a serious mistake. You could try using a suitably matched historical control group, but a randomized trial design is the easiest approach that allows the type of interpretation you seem to want.

  • $\begingroup$ Even an observational study would permit recruiting controls who are matched to cross-over patients at baseline but, for whatever reason, they did not elect for the investigational treatment/intervention. $\endgroup$ – AdamO Jan 24 '18 at 15:39
  • $\begingroup$ Sure. That was not part of what was available as per the question. If it were, then compared to a RCT, it still makes things a lot more complicated though. One would then have to do some kind of approach that reflects that patients that do not elect to/get prescribed the treatment may differ systematically (e.g. propensity score matching). $\endgroup$ – Björn Jan 24 '18 at 20:17

EDIT: I have previousely adviced mixed-design ANOVA, because I somehow missed the fact, that You don't have a control group. Sorry.

In the regard of the experimental design, Bjorn is definietly right. Maybe any change after the treatment was a seasonal change and blood sugar of people who weren't treated would change in the exactly same way?

But as far as comparing several measurements for 1 group is concerned, Anova for repeated measures is the proper test.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.