Pirun Saelue, Jakrawadee Julamanee, Adisak Tantiworawit, Thanawat Rattanathammethee, Weerapat Owattanapanich, Smith Kungwankiattichai, Chantiya Chanswangphuwana, Chantana Polprasert, Wasithep Limvorapitak, Supawee Saengboon, Kannadit Prayongratana, Chantrapa Sriswasdi, Pimjai Niparuck, Teeraya Puavilai, Chajchawan Nakhakes, Chinadol Wanitpongpun
The THAI-LEDGE model performed well in predicting the survival of AML patients. However, external validation of the model in other ethnic populations and healthcare settings are necessary before widespread clinical implementation.
INTRODUCTION: Despite the development of advanced therapeutic approaches in the last two decades, acute myeloid leukemia (AML) has a poor prognosis, especially in older patients. The main causes of death are refractory/relapsed disease, fatal bleeding, or serious infection. A model to predict survival in patients with AML is necessary for clinicians to make decisions regarding appropriate treatment. In this study, we aimed to evaluate the predictive factors for death and generate a model to predict survival in patients with AML.
METHODS: We conducted a multicenter prospective cohort study across nine tertiary medical care institutes in Thailand, enrolling patients aged ≥ 18 years with newly diagnosed AML between January 1, 2014, and December 31, 2023. Patients with acute promyelocytic leukemia were excluded. Multivariable Cox proportional hazards regression analyses identified the predictors of mortality, and the final model was constructed using backward stepwise regression with Akaike Information Criterion selection. Model performance was assessed using Harrell's C-index and calibration plots, with internal validation performed via bootstrapping.
RESULTS: A total of 1,055 patients were included. The median overall survival was 10.9 months, with a 10-year survival rate of 22.4%. Eight variables were independently associated with survival outcomes: age > 55 years, Eastern Cooperative Oncology Group performance status, tumor lysis syndrome, leukostasis, disseminated intravascular coagulation, white blood cell count, genetic risk, and type of induction therapy. The THAI-LEDGE model demonstrated a good discriminatory ability (C-index = 0.743) and satisfactory calibration. The internal validation yielded a C-index of 0.737, confirming the robustness of the model.
CONCLUSIONS: The THAI-LEDGE model performed well in predicting the survival of AML patients. However, external validation of the model in other ethnic populations and healthcare settings are necessary before widespread clinical implementation.