OBJECTIVES: The main goal of this research is to determine the factors which best predict the completion of a nurse substance abuse monitoring program.   METHODS: A retrospective cross sectional analysis was conducted in a state database (Florida) of 65,000 nurses enrolled in CMP. The entire dataset and the subset of 10,000 were used for analysis. The outcome variable was the status of CMP program. The predictor variables included demographics, treatment type and length, drug usage, healthcare setting and experience, status of treatment, aftercare treatment, and nursing specialty.  Missing data was not considered in the study. After checking for all the assumptions, univariate analysis using chi-square test was performed on the entire data as well as the subset. All features with significant relationships in the univariate analysis were entered in the forward, backward and stepwise multiple logistic models to predict the completion of contract by the nurses.  All tests were conducted at 5% level of significance. RESULTS: All independent variables had a significant relationship with the status of CMP. The model using the entire data did not converge. The forward logistic model of the subset data showed that drug usage, treatment type, status of treatment in contract, annual income and healthcare setting had a higher association with the completion of CMP (p-value=0.1188). The backward logistic model showed that aftercare treatment, type of treatment and nursing specialty had a significant association with CMP status (p-value =0.064). The stepwise model converged with aftercare treatment having a significant association with the CMP status.  CONCLUSIONS: In this study, the forward logistic model was preferred over the backward or the stepwise model to reduce the bias resulting from selection of variables and eliminate the resulting quasi-complete separation of data. Therefore, this CMP data for nurses was best fit using the forward logistic model.