Cumulative incidence and prognostic factors for leukemic transformation in chronic myelomonocytic leukemia: a competing risk analysis

BACKGROUND: Leukemic transformation represents a pivotal event in the natural history of chronic myelomonocytic leukemia (CMML), yet conventional survival analysis may yield biased estimates by treating non-transformation deaths as censored observations. We hypothesised that competing risk methodology would provide more accurate transformation estimates and could alter the interpretation of treatment-associated effects compared with traditional approaches. METHODS: We retrospectively analysed 102 patients with CMML diagnosed and treated at Ganzhou People's Hospital between January 2014 and November 2025. Cumulative incidence functions (CIF) were estimated using the Aalen-Johansen method with death without transformation as the competing event. Fine-Gray subdistribution hazard models identified independent predictors, with parallel Cox regression for methodological comparison. Treatments were classified by primary modality at baseline (HMA, chemotherapy, best supportive care (BSC), allo-HSCT) and entered as a stratification covariate rather than as a randomised exposure. RESULTS: During follow-up, 54 patients (52.9%) developed acute myeloid leukemia and 7 (6.9%) died without transformation. The 12-, 24-, and 36-month CIF were 46.9%, 63.4%, and 73.8%, respectively, whereas 1 - Kaplan-Meier estimates were 49.2%, 67.2%, and 78.4% (4.9-6.2% relative overestimation). Baseline treatment strategy was associated with transformation risk: compared with HMA, BSC showed a higher subdistribution hazard (SHR 4.77, 95% CI 1.86-12.22, P = 0.001) and chemotherapy SHR 2.16 (1.04-4.48, P = 0.039), with allo-HSCT not significantly associated (SHR 0.52, P = 0.261); these associations should be interpreted as exploratory given non-random allocation and small subgroup sizes (allo-HSCT n = 13, BSC n = 22). Intermediate and high-risk karyotypes independently predicted transformation (SHR 3.09 and 2.98, both P < 0.05). Within the NGS-tested subgroup, SETBP1 (SHR 3.09, P = 0.009) and TP53 (SHR 3.66, P = 0.009) mutations marked small subsets at very high transformation risk. For allo-HSCT, Cox regression yielded a significant protective effect (HR 0.31, P = 0.026) that became non-significant in the Fine-Gray model, illustrating how cause-specific analysis can overstate the isolated anti-transformation effect of a treatment that simultaneously reduces mortality. CONCLUSIONS: This study shows how competing-risk analysis complements standard Kaplan-Meier estimates of overall and progression-free survival when the cumulative incidence of a specific event, leukaemic transformation, is the quantity of interest. Although the modest, single-centre sample and non-randomised treatment allocation preclude definitive treatment recommendations, cytogenetic risk was the strongest disease-related prognostic factor, and molecular profiling provided additional risk discrimination within the NGS-tested subgroup. Cumulative incidence functions should therefore be reported routinely alongside Kaplan-Meier estimates in CMML and other myeloid neoplasms with non-negligible competing mortality.