3

I am trying to study the difference in a biomarker (ANG7) levels pre and post bariatric surgery (a weight-loss surgery). I am using paired Wilcoxon signed-rank test for this analysis. The test shows there is a significant reduction in ANG7 after the surgery. But the ANG7 is also associated with BMI, so I want to control BMI to say that "there is a significant reduction in ANG7 after the bariatric surgery". My question is, how to control a covariate in a paired data design, what is the appropriate test to do this? Any help is highly appreciated.

arshad
  • 863
  • 1
  • 7
  • 13

1 Answers1

1

To account for covariates, use a regression model instead of a statistical test. It turns out that models generalize tests [1, 2]. In your case you would regress post-surgery ANG7 on pre-surgery ANG7 and BMI. Since you want to generalize the Wilcoxon signed-rank test, use ordinal regression (aka proportional odds logistic regression or POLR).

References

[1] Biostatistics for Biomedical Research course notes. Available online
[2] Inference of nonparametric tests as linear models.

More about the covarite BMI:

  • Analyze log(BMI) instead of BMI because BMI is a ratio of two quantities, weight and height (squared).
  • Even better, if you have those measurements, you can substitute log(BMI) with log(weight) and log(height).
dipetkov
  • 9,805