The surgery department is the core division of hospitals. Reasonably formulating surgical plans is crucial for efficient allocation of medical resources and essential for improving the quality of healthcare services. The uncertainty of surgery duration presents challenges for surgery scheduling. To address the uncertainty of surgery duration, this paper utilizes historical surgical data and applies the variance inflation factors to reduce multicollinearity between patient features. Quantile regression is used to characterize the interval of surgery duration, and an uncertainty set of surgery duration driven by patient feature is then constructed. On this basis, considering the resource constraints of the beds in downstream intensive care units, a robust surgery allocation model is developed to decide on the allocation of operating rooms and the scheduling of patients' surgery date. The effectiveness of the proposed method is tested using real data, and the experimental results show that, by adjusting the model parameters, the proposed model outperforms the corresponding deterministic model in terms of decreasing the overall operating cost and alleviating the overtime hours; compared to the stochastic programming model, it reduces the overall operating cost and has advantages in terms of computational time with a sacrifice of fewer overtime hours.