Decision Tree Model to Predict One-Year Survival in Ambulatory Patients with Advanced Cancer

Background. An accurate prognosticationis crucial for end-of-life decision-making in advanced cancer care. While existing prognostic tools focus on short-term survival (weeks/months), there is a paucity of studies that have examined the long-term prediction at one year. A one-year timeframe is regarded as a general indicator of palliative care referral; however, there are many uncertain issues. This study aimed to develop a one-year survival prediction model using objective parameters for patients with advanced cancer. Methods. This was a secondary analysis of data from a South Koreanprospective cohort study. Participants, with clinician-predicted survival of ≤1 year, were assessed using clinical data, performance status, laboratory data and chemotherapy response. Cox proportional hazards and recursive partitioning analyses (RPA) were used to identify the prognostic factors and build a prediction model. Results. Of the 200 advanced cancer patients (mean age 64.4, 36% female; 33.5% lung cancer), the median survival was 7.5 months. Multivariate analysis confirmed five variables: chemotherapy response, Karnofsky Performance Scale (KPS), edema, C-reactive protein/albumin ratio (CAR), and lactate dehydrogenase (LDH). RPA) using four variables (chemotherapy response, KPS, CAR, and LDH) helped create a 6-node survival tree, demonstrating significant survival differences (p <0.001) between the subgroups, with estimated ratios ranging from 0.2 to 3.4. Conclusion. We developed an RPA model to facilitate one-year survival prediction in patients with advanced cancer. The 6-leaf model incorporated only four readily available variables. Following external validation, this model may prove valuable in assisting clinicians with one-year survival prognostication.