I have built a binomial logistic regression model for disease X with 20 predictors for a sample size = 400, for both prediction and inference purposes. The overall model has a great predictive power (McFadden pseudo R2 = 0.4), likelihood ratio is significant and overall model prediction accuracy is 97%. But only 4 predictors are significant. For inference purposes, I know it is not necessary to throw away the non significant coefficients. But I am struggling with how to interpret the model for prediction purposes. Do you report that it is a predictive model, with an interpration of the odds ratios of only the 4 significant predictors or can you interpret the odds ratios of all the predictors if you look at the model as a whole?