Estimating the economic costs of functional gastrointestinal tract disease using regression two-part parameters model

Accurate estimation of medical costs is one of the health care policymaker's goals. Regarding to the feature of health data and the complexity of their analysis conventional models are not suitable for them. To model these data, all these characteristics should be taken into consideration. This study has estimated the average cost of gastrointestinal tract diseases and some factors influencing this cost using two part regression model. During 2006 to 2007 information of total 1907 gastrointestinal tract patients were collected in the research center for gastroenterology and liver disease of Shahid Beheshti University of medical science of Tehran as a retrospective cross-sectional study. For the purpose of modeling gastrointestinal tract disease cost, the two-part model was employed. In the first part, a logistic regression was fitted to the dichotomous events of having zero or positive expense and in the second part a multiple linear regression was fitted to positive expense and the effect of demographic variables, work ability, number of times visiting a doctor, the number of diagnostic tests, insurance, number of days absent from work or reduced efficiency and hospitalization on expense were assessed. The average costs of gastrointestinal tract diseases and their standard deviations in parametric method were yielded as $75.93±122.29. Minimum and maximum of costs were respectively estimated $2.89 and $1394.32. While the actual average cost was $78.35±222.36, minimum and maximum were respectively zero and $5183.81. The results obtained from the two- part model revealed that "some demographical variables", "number of days absent from work or reduced efficiency" and "number of times visiting a doctor" have influenced having expense and the variables "number of days absent from work or reduced efficiency", "number of times visiting a doctor" and " hospitalization" " have influenced positive expenses. The estimated cost made by this method is close to the true value. According to the lower standard deviation of estimations made by model, compared to the actual values, estimations from this model have high precision. In addition, this model has performed well to maximize zero costs.