I would like some advice on what statistical model I can choose and how to implement in R. I have 2 groups of high blood pressure patients (each group consists of 40 subjects ), a treatment group (Group 1 ) that receives conventional treatment + new treatment and a control group (Group 2)that only receives the conventional treatment . The outcome that was measured is the systolic blood pressure in all subjects at 2 different time frames, after 4 weeks and after 8 weeks follow up. I would like to test the efficacy of the new treatment in lowering blood pressure. So my main outcome is the change in blood pressure from baseline.

My null hypothesis is that the new treatment does not have an effect on the level of blood pressure in hypertension patients

Null : Change in mean PB in treatment group = Change in mean PB in control group

Alternative: Change in mean PB in treatment group > Change in mean PB in control group

to test my hypothesis, I performed Welch's two-sample t-testI as well as 95% CI using R. But now I would like to fit the data into a model. I'm not sure what type of model would be appropriate for this data set and the goal of the study. I found only that Linear mixed-effects model might be appropriate but I'm not sure how to perform it in R. Below is a snippet of the first 5 rows of the data.

enter image description here

enter image description here

Thank you


1 Answer 1


A couple of points:

  • Instead of considering changes from baseline, it is better to run an analysis in which you include the baseline measurement as a covariate. Or still analyze all three measurements together.
  • Assuming that all patients were exactly measured at 4 and 8 weeks without too much variability in the actual timing of the measurements, you could consider the time as a categorical variable, and its interaction with treatment. Depending on what you do with the baseline measurements, you may need to exclude / not test the main effect of treatment.
  • For the correlations over time, again if time is categorical, you can consider an unstructured covariance matrix. Otherwise, you will need to select an appropriate structure using serial correlation models and/or random effects.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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