4
$\begingroup$

In a paper published in Ebiomedicine (part of the Lancet family journals), the researchers built a logistic model including a variety of predictors and then reported both "univariate" regressions (model with only one predictor) and "multivariate" regressions (logit model with all predictors), see the following table : univariate and multivariate regression

I am surprised by this methodology choice as a model with only one parameter (univariate regression) obviously fit the data in a sub-optimal manner, the estimated parameter is different than the estimated parameter in the corresponding multivariable logit model and so is its associated (asymptotic) confidence interval (same reasoning goes for its exponentiated transform, interpreted as an odd-ratio).

I have to admit that I'm not familiar with the publication standards in medicine journals, but is reporting univariate next to multivariate regression common practice ? If so, how is it supposed to be interpreted by the reader ?

$\endgroup$
1
  • 3
    $\begingroup$ The terminology in the table you display is "univariable" versus "multivariable" regressions, not "univariate" versus "multivariate" as you say in the question. That might have been intended to respect current best terminology, which is to reserve the word "multivariate" for models having multiple outcomes. $\endgroup$
    – EdM
    May 26 at 12:20

1 Answer 1

7
$\begingroup$

The univariate column shows the simple association between each predictor and the outcome. The multivariate column shows the association when controlling for all other predictors. Both of these things are important, and they're clearly distinct, so this is good practice. For example, the number of unique VPDs is a negative predictor on its own (OR <1), but is a positive predictor when controlling for other variables.

$\endgroup$
1
  • 3
    $\begingroup$ I would say it is a 'decent' practice, when readers will understand it correctly. This is certainly common, but it should be understood as exploratory. The data appear to be observational & none of these should be given a causal interpretation, eg. $\endgroup$ May 26 at 18:54

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.