In order to analyze the patients' referral behavior pattern among different hospitals, the sequential pattern mining algorithm can be applied. Apriori-like algorithm is the classical algorithm in the sequential pattern mining, but there are some deficiencies, such as generating too many candidate sequences and scanning the database too often. To solve such problems, this paper has proposed some measures for improvement, including using a new pruning strategy and reducing the unnecessary scans of the database. The experiments prove that the improved algorithm performs more efficiently in the frequent referral sequence mining.