For some studies, e.g., in heart failure, it is traditional to normalize certain outcomes for diuretic use. For example in this paper the authors divided net fluid output by “furosemide 40mg equivalents”. In other studies the investigators might divide weight change by the same denominator. The problem as I see it is that diuretic use is adjusted over the course of observing the main response variable, so the ratio may be hard to interpret. This may be more of a situation of having a bivariate outcome: (main response variable, serial diuretic dosage). Is anyone comfortable with using the ratio as a single response variable? If we don’t do that how should concomitant therapy such as diuretic use be adjusted for?
A possibly useful background paper is here. This addresses the problem of whether linearity should be assumed for furosemide dose.
I’m not sure why someone would need to do this and what it represents. I would think the best way to mimic physiology would be model diuretic dose continuously with an interaction with renal function (Cr) and that would get at how UOP relates to dosing in HF.
Reminds me of Kronmal’s “fallacy of the ratio” and “spurious correlation” paper (198…6?). Highly recommend the read. I would raise concern on two counts: post-baseline diuretic has a high likelihood of being a mediating variable for monitoring other associations. Edema one of the “cardinal features” of HF which is managed by loop diuretic. Second, if does varies, which if any dose should be used? And what are the indicators of dose change?
Kronmal’s major conclusion is that if a variable is confounder, a ratio does not suffice to give unbiased estimates. The solution: adjust. Log transform, with coefficient adjustment for denominator is more flexible and removes spurious correlations. If dose varies with time, perhaps a marginal structural model is called for?
Question to clinicians: aside from cardiorenal syndrome, with HF, is edema primarily/to some extent explained by impaired kidney function d/t HF?
The relation between diuretic dose and renal function can be complex and depend on setting. Often need for higher doses in renal failure but sometimes dose is reduced if renal function deteriorates quickly.
I really appreciate the excellent thoughts. From what I am gathering, my prior bias of avoiding ratios as dependent variables is sustained, and that if a concomitant therapy such as diuretic dose is to be accounted for, it should be accounted for in a less assumption-laded way. What is still an open question is whether diuretic use should be accounted for at all. At present I can’t imagine a way that allows a clean interpretation of the main response variable (e.g., weight loss).