OBJECTIVE: One of the disadvantages of propensity score matching is failure to apply on continuous variables. This paper proposes a method to control for selection bias when propensity score matching technique is not applicable. METHODS: The proposed method first uses continous variable rather than a binary variable in the first stage estimation. Since non-treated patients will have zero use and treated patient will have positive use of treatment. Tobit regression is proposed to estimate treatment use. Second, using Tobit residuals in the second stage equation to estimating health care cost or utility, we showed that selection bias due heterogeneity of patients are removed. RESULTS: Market Scan data were used to estimate total health care expenditures of migraine patients treated by triptan. Number of triptan scripts is used to do matching rather than binary variable. Mean value for triptan scripts for treated patients were 4.37. After certain inclusion and exclusion criteria 43,776 migraine patient with triptan used created our analytic samples. We used same number of control patients. After controlling for demographic and clinical factors, we add Tobit residuals as an additional variable in to our heath expenditure model. Significance of coefficient on residuals showed that selection bias exists and failure to account for that bias would yield spurious results. CONCLUSIONS: Propensity Score matching may not be applied in certain situations. This paper examined a case where selection was due to continuous variable and proposed and applied a technique under this circumstance.