OBJECTIVES: Censoring is common problem with medical count data. Estimation over only uncensored patients yields bias toward the patients who have shorter survival time since patients who have longer survival times are high likely to be censored. Standard survival analysis is not applicable since the number of visits during the study time and after censoring time is not independent. The objective of this paper is to propose a method, which can be applied to censored count data. METHODS: The proposed method first estimates the probability of censoring by using logit model. Then, second stage involves estimating weighted Poisson regression where weights are calculated as inverse of estimated probability of censoring. We show that the resulting estimators are consistent. Standard errors from second stage are not valid and should be adjusted for first stage estimation. We estimate the errors by using bootstrapping techniques. RESULTS: Medstat Market Scan data is used as an application of the method. Total hospitalization days after a year of initial diagnoses is estimated. Patients who are diagnosed less than a year before the end of study period are considered as censored. After using inverse probability weighted poisson regression, we also estimate the total hospitalization days by dropping the patients whose visits are censored. A test is proposed to compare the coefficients. We found that the difference in coefficients are significant (p<0.0004). CONCLUSION: This paper presents a method for testing and correcting for possible sample selection bias for cross sectional data. In our application we assessed the influence of explanatory variables, such as patient and clinical characteristics, on inpatient visits of asthma two years following diagnosis after accounting for possible selection bias due to censoring. We applied poisson and our proposed method and to show that failing to do the adjustments yield different estimators.