I have blood sugar values for diabetic patients over the past few years from a clinical practice. Half way through the observation period, a treatment intervention was applied to all of the patients. My data consists of multiple blood sugar readings per person, both before and after the intervention.

I would like to compare the change in blood sugar values for each patient before and after the intervention, in order to determine if the intervention had a significant effect on blood sugar in each patient.

Which test(s) would be appropriate to achieve this? Should I be looking at the overall change in absolute blood sugar over time? Or should I be using categories (normoglycemic, prediabetic, hyperglycemic) to determine this?

I am rather new to statistics, and any help is greatly appreciated.


There are multiple ways of doing this, but the most obvious ones would be a paired t-test or a repeated measures ANOVA. These will compare mean blood sugar levels in the two groups (before / after the intervention) and determine whether there was a significant change. If that is all you need, categorization is unnecessary, just use the absolute values. These tests are available in most statistical software.

  • $\begingroup$ Thank you for your answer! Do you suggest I take the mean of each persons readings after intervention, take the mean of those values across all patients, and then use the tests you mentioned to compare it with the mean of means prior to the intervention? Thanks again. $\endgroup$ – Simran Parmar Jul 11 '17 at 18:07
  • $\begingroup$ @SimranParmar exactly, that's the way. you're welcome! $\endgroup$ – khaozavr Jul 11 '17 at 18:40
  • 1
    $\begingroup$ I'm not familiar with how to analyse this type of problem, but this suggested approach (taking the mean of the data before and after) would involve throwing away useful information if there is some sort of trend in blood sugar with time. I suggest you investigate ways to incorporate that information into your model. $\endgroup$ – mkt - Reinstate Monica Jul 11 '17 at 20:35
  • $\begingroup$ That's a very good point. I will keep looking around for how to incorporate this into the analysis. Any other input on this would be greatly appreciated! $\endgroup$ – Simran Parmar Jul 11 '17 at 22:39
  • $\begingroup$ There could very well be autocorrelation within the individual time series, which should be accounted for. $\endgroup$ – kjetil b halvorsen Aug 2 '17 at 13:47

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.