Intent classification and slot filling are critical tasks in Spoken Language Understanding (SLU). Currently, prototypical networks are used as the primary method to solve the problems of few-shot learning. However, directly applying prototypical networks to few-shot Chinese medical SLU tasks can lead to issues such as the prototypes being susceptible to noise interference in samples and non-medical entities introducing irrelevant features, which hinder the performance of prototypical networks. To address these problems, we propose a Chinese medical SLU model based on Prototypical Modification Network and Contrastive Learning (PMNCL). Its highlights are as follows: (1) To improve the efficiency of distance measurement, the distance weights between support set samples are calculated and used to modify the prototypes. (2) To obtain better class discriminative representations, the contrastive learning loss function is designed so that similar and dissimilar samples can be separated. (3) We conduct sufficient experiments on the benchmark Chinese medical intent classification datasets such as IMCS-V2, KUAKE-QIC, and CMID. Results demonstrate that our proposed model outperforms previous methods and achieves state-of-the-art performance.