I suppose that VD is a continuous independent variable (IV), and smoking status is binary IV. To asses the interaction between VD and smoking, you might try a logistic regression analysis, and put VD, smoking, and VD*smoking (interaction term) in the multivariate model. Certainly, the status of case or control is the dependent variable (DV) of this logistic regression model. The parameter estimate of the interaction term as well as its Wald statistic will provide you the information regarding whether the interaction is statistically significant or not.

thinks for your answer, i do binary regression analyse with status of case or control is the dependent variable (DV), and i put status of case or control is the dependent variable (DV) but  i dont find how i will enter VD*smoking (interaction term) in the multivariate model. 

Hi Ben, To assess interaction, you can create a new term. This new term will be called or labeled the interaction term. To generate this new interaction term, simply multiply your VD term by your smoking term to create a VD x Smoking term (VD*Term). When you do the logistic regression, you should have your dependent variable and then a multivariable model that includes the VD term, the Smoking term, and the VD*Smoking term. If the VD*smoking term produces a Wald Statistic with a p-value of < 0.05, then it is likely that interaction is present, even if the VD and Smoking terms are not significant in this new model. If interaction is present, interpreting your results becomes more difficult, because everything will now depend on multiple variables. Overall, If your sample size is small, this approach may not be appropriate. (If you are a Stata user, this page may be helpful for taking it a step further to assess for effect modification using likelihood ratio tests to compare models with and without the interaction term -- http://courses.umass.edu/biep640w/pdf/5.%20%20Logistic%20Regression%202014.pdf )

thinks jason . we have 250 cases and 250 controls in the regression test  smoking status is labeled as 1 ever smoker and 2 never smoker and vitamin d is continous 

and this is the result attached file 

as i understand i should look to the wild not to sig

The sig. is related to the Wald statistic, and  based upon what the significance tells us, it is likely that the interaction term is significant (p < 0.05). I'm not sure of the commands in SPSS, but the next step would be to see if the model deviance changes significantly between the two-term model and the model with the the interaction term. If the deviance of the models are significantly different, than interaction is presumed to be present. The test between the model deviances is called the likelihood ratio test. (Since the interaction term is significant, it is very possible that the likelihood ratio test comparing the deviance values of the two models (model without interaction versus model with the interaction term) will be significant.   *I did find a very helpful set of Powerpoint slides as a pdf online that provides instructions here. These methods from the slideshow appear to me as being the same as the recommended methods from Levy and Lemeshow's Applied Logistic Regression.   (Slides 42 to 49 should be quite helpful) http://core.ecu.edu/ofe/StatisticsResearch/logistic%20regression%20analysis.pdf

thinks a lot for your helpful advices and links

Cher ami kacem

Il n’est pas judicieux de procéder directement à l’analyse des données par une régression logistique  sans effectuer au préalable une analyse univariée (tableaux à une seule variable), une analyse bivariée (tableaux à deux variables) et une analyse de tableaux à trois variables.

 L’analyse de tableaux à trois variables permet notamment de déceler des interactions entre les variables avant d’introduire ces interactions dans le modèle, le nombre d’interactions introduites ne devrait pas dépasser deux. Au-delà de deux interactions, le modèle deviendrait vite inextricable.

Le modèle que vous avez joint ne contient que vos deux variables et l'interaction pour laquelle on a un test de Wald significatif. Je ne sais pas si vous avez introduit dans votre modèle votre variable VD sous une forme quantitative, les seuls tests que je connaisse pour évaluer une interaction entre deux variables qualitatives sont les tests de Zelen, de Woolf et l'indice de Breslow. 

A toutes fins utiles, ci-joint la signification d'une interaction avec deux exemples concrets (tirés de mon manuel à paraître "Précis d'épidémiologie") J'espère que cela sera utile.

merci beaucoup professeur  Abdeldjellil Bezzaoucha pour votre précieuse aide et enchante de vous connaitre sa fait plaisir de communique avec vous. un grand bonjour pour l Algérie