Cancer staging is a crucial process that determines the severity of an individual’s cancer based on specific risk factors and clinical outcomes, such as time-to-event outcomes or the presence of a disease. This process involves classifying a heterogeneous set of cancer patients into several homogeneous groups. Accurately classifying the cancer stages helps doctors identify patients for clinical trials, understand the disease’s severity and prognosis, and facilitate clinical decision-making on therapy and surveillance. Tree methods have emerged as promising tools for cancer staging due to their ease of interpretation and ability to handle complex datasets with minimal assumptions [Bre+17; LWC13; Lin+16]. However, integrating multiple risk factors into cancer staging using tree methods presents several challenges. First, it is unclear how to leverage the ordering indicated by ordinal risk factors. Second, with a high number of categories defined by risk factors, it remains unknown whether patients in each category have a distinct prognosis. If not, it is unclear how to combine them into one stage. Finally, allowing a general grouping pattern is challenging, as most approaches have restrictions on the patterns of groupings. For instance, the classification and regression tree (CART) method only permits straight-line groupings on a partially ordered two-way grid. To address the limitations of tree methods with ordinal variables, we introduce a new method: Ordering Partially Ordered Set Elements by Recursive Amalgamation (OPERA)[Wan20]. This approach utilizes ordered information and accommodates general grouping patterns. OPERA, combined with pruning, demonstrates improved performance compared to traditional tree methods without pruning. This data-driven tool simplifies staging by condensing multiple groups into a single stratum based on a distinct prognosis, enhancing ease of use and interpretation. A well-trained tool can also accurately classify cancer stage and predict clinical outcomes. Beyond cancer staging, this method has implications for clustering in healthcare, aiding the identification of homogeneous patients for clinical trials and resource prioritization.