BACKGROUND: AML exhibits substantial molecular and genetic heterogeneity. Therefore, identifying key biological processes and related genes involved in the pathogenesis, as well as contributing to therapeutic resistance, is imperative for enhancing clinical outcomes. However, the assessment of mitochondrial function in AML has gradually been acknowledged but not been widely emphasized. Hence, prioritizing the identification of mitochondrial-related biomarkers is crucial to enhance existing stratification methodologies and guide decisions on risk-adapted therapies. METHODS: We systematically integrated and analyzed data from nine online AML transcriptomics sequencing databases, screening the Human.MitoCarta3.0 mitochondrial gene database to identify AML-specific mitochondrial genes. A prognostic mitochondrial score was developed using LASSO regression analysis in HOVON database as training cohort (n = 618) and validated in another eight publicly available independent cohorts (n = 1,697). RESULTS: A 19-mitochondrial function gene AML score was further generated and exhibited high prognostic power in 2,315 AML patients, named as MitoScore. MitoScore was an independent survival prognosis biomarker (p <
0.001). The MitoScore effectively distinguishes several genetic abnormalities and significantly improves the ELN (European Leukemia Net) classification. Patients with a high MitoScore demonstrated a notably poor response to induction chemotherapy and related refractory AML (p <
0.001). In the favorable risk gene variant and cytogenetic abnormality group, MitoScore was significantly lower compared to patients without those variants. Conversely, in the adverse group, MitoScore was significantly higher compared to patients with favorable genetic abnormality. CONCLUSIONS: Our findings underscore the utility the MitoScore as a powerful tool for refined risk stratification and predicting chemotherapy resistance.