Perioperative risk assessment (PRA) aims to evaluate the risk to patients from surgery, and knowing the risk can contribute to allocating scarce medical resources. Despite the clinical significance of PRA, to our best knowledge, there is currently no comprehensive perioperative risk (PR) dataset to facilitate the development of standard criteria and machine learning-based models for PRA. In this paper, we propose the first perioperative risk assessment dataset (PRAD) with multi-view data by applying online accelerated pairwise comparison (OAPC). Specifically, OAPC combines prior knowledge-based presorting and online probability insertion sorting to efficiently obtain robust pair comparisons. To obtain labels from doctors conveniently, we develop an online PRA system to enable doctors to label medical records anywhere and anytime. Our PRAD provides 300 medical records with multi-view data, including various types of preoperative and postoperative data, together with the corresponding comparisons and risk scores obtained from doctors with different experiences. Furthermore, we analyze the PRAD to investigate relationships between the patient's preoperative data and risk score, e.g., cardiovascular disease history is highly related to PR, providing a complementary view with current research on PRA. The labeling procedure is still ongoing, and additional records and analyses will be made available in the future. We believe our dataset and analysis provide new insights that will significantly facilitate the building of new PRA models.