I take it by multivariate you mean a regression with more than one predictor. If you strip your initial model of all the predictors except for the single one whose estimated regression coefficient turns out statistically different from zero in its 'univariate' version you are basically saying that all but one of the regressors originally thought to be important to predict your binary outcome make, jointly considered, a negligible contribution to the model and, therefore, there's no point in conditioning on them. This is a strong statement which I'd put to the test (e.g. a likelihood ratio test or AIC) before drawing that conclusion.

You are missing a lot of information and not proceeding in the best manner. See the attached papers for guidance. There is a complete protocol for doing this in these papers. Best, David

Hi Luca and David,

Thanks for the advice!

You may not get all the variables significant at 5 % LOS in univariate analysis. But to apply multiple logistic regression you can consider predictors significance at 20% LOS( P-0.20) in univariate analysis. Put all the predictors which have P-0.20 in multivariate logistic regression model and test for significance of associated predictors.

I know this my answer to this question may be too late, but it may be useful for others how may face the same situation. A very good explanation of the answer to this question could be found in this article:

Article Understanding logistic regression analysis

Ravishankar Suryanarayana Thank you for your comment. Do you have a reference for that?

Update:

The article in Khaled's comment suggested adding predictors to the multiple logistic regression model if they showed significance at %25 LOS in univariate analysis, citing this paper:

Bendel RB, Afifi AA. Comparison of Stopping Rules in Forward “Stepwise” Regression. J Am Stat Assoc 1977;72:46–53

Khaled Abuzeid has provided theceequired reference

I think our attached clinical research fits your question. We have clearly explained regression-correlation-univariate&multvariate logistic regressions.

Thanks for your sharing! Helpful! Izzettin Hatip-Al-khatib