AUTHORS
Muhammad Ali Khan, MBBS, Kanishka Uttam Chandani, M.B.B.S.2, Cecilia Arana Yi, M.D.3, Zoey Harris, MD4, Ayalew Tefferi, MD5, Naseema Gangat, MBBS6, Animesh Pardanani, PhD, MBBS7, James Foran, FRCPC8, Candido E Rivera, MD9, Ziyi Huang, PhD10, Irbaz Riaz, MD, MS, MBI, PhD11, Jeanne Palmer, MD12
AFFILIATIONS
1 Mayo Clinic, Phoenix, Arizona, United States
2 Mayo Clinic, Phoenix, Arizona, United States
3 Mayo Clinic, Phoenix, Arizona, United States
4 Mayo Clinic, Phoenix, Arizona, United States
5 Mayo Clinic, Rochester, Minnesota, United States
6 Mayo Clinic, Rochester, Minnesota, United States
7 Mayo Clinic, Rochester, Minnesota, United States
8 Mayo Clinic, Jacksonville, Florida, United States
9 Mayo Clinic, Jacksonville, Florida, United States
10 Mayo Clinic, Phoenix, Arizona, United States
11 Mayo Clinic, Phoenix, Arizona, United States
12 Mayo Clinic, Phoenix, Arizona, United States
ABSTRACT
Background: Progression to myelofibrosis (MF) is associated with reduced survival and can influence treatment decisions in polycythemia vera (PV) and essential thrombocythemia (ET). However, available risk models are static and do not account for changes in clinical data over time. We evaluated whether routinely collected longitudinal data could support individualized, dynamically updated prediction of progression to MF.
Methods: Patients with PV or ET were identified by retrospective chart review; those with MF within 6 months of diagnosis were excluded as prevalent disease. Demographics, disease subtype and duration, serial laboratory data, vascular events, symptom data, and treatment exposure were recorded longitudinally from the date of PV or ET diagnosis until progression to MF, death, or last follow-up. Progression to MF was defined by pathology or hematologist confirmation; death before progression was treated as a competing event. Risk of MF progression was estimated at 3-month intervals throughout follow-up, with progression within 36 months considered as the primary outcome and within 12 and 60 months as secondary outcomes. Patients were partitioned into development (90%) and held-out test sets (10%) by a stratified split on disease subtype and progression status. Individual predictors, including treatment exposure (cytoreductive, disease modifying therapy, phlebotomy), were grouped into prespecified clinical categories and retained by permutation importance with stability selection within the development set. Separate cause-specific random survival forest algorithms were developed for PV and ET, and progression and death estimates were combined using the Aalen-Johansen estimator. Performance was assessed using time-dependent area under the curve (AUC). Progression to MF was compared across tertiles of important continuous features and across categories of ordinal features. Differences between the highest and lowest groups were evaluated using patient-level bootstrap resampling, with P<0.05 considered statistically significant.
Results: A total of 11,230 patients (5,926 PV, 5,304 ET) with 327,896 assessments were included, of which 1,812 progressed to myelofibrosis (1,054 PV, 758 ET) and 2,390 died before progression (1,315 PV, 1,075 ET). Median follow-up was 76.6 months; among progressors, median time from diagnosis to MF was 110.8 months. In a held-out test set of 1,117 patients, the 36-month AUC was 0.81 (95% CI 0.77-0.84) in PV and 0.75 (0.70-0.80) in ET, with comparable discrimination at 12 (PV: 0.82; ET: 0.78) and 60 months (PV: 0.81; ET: 0.74). Because risk was re-estimated every 3 months, individual predictions moved during follow-up: 31% of patients crossed a 10% risk threshold, and among those who progressed the predicted 36-month risk rose from a median of 8% to 14% in PV and 8% to 11% in ET as progression approached. In PV, the highest-contributing variable categories included red cell parameters (hemoglobin, hematocrit), red cell indices (RDW, MCV), and bone marrow morphology (cellularity, megakaryocyte proliferation), accounting for 53% of model performance. In ET, the highest-contributing variable categories included LDH, red cell parameters, and red cell indices, accounting for 52% of model performance. Within these categories, observed 36-month progression in PV rose with tertiles of RDW (1.0% to 15.2%) and marrow cellularity (5.0% to 19.0%) and with megakaryocyte proliferation (2.8% to 12.2%), and fell with hemoglobin (12.3% to 0.5%) and hematocrit (11.5% to 3.2%). In ET, progression rose with LDH (2.6% to 12.7%), RDW (3.8% to 8.4%), and MCV (3.5% to 7.8%), and fell with hematocrit (8.6% to 2.7%) and hemoglobin (7.0% to 3.1%). All observations were significant (P<0.05).
Conclusion: Routinely collected longitudinal data enable dynamic, individualized prediction of myelofibrotic progression in PV and ET, with discrimination maintained at 12, 36, and 60 months in the held-out test set. Unlike risk estimates fixed at diagnosis, predictions are updated at each visit. Even with treatment exposure included in the models, red cell parameters and indices were among the highest-contributing variables in both diseases, with marrow morphology additionally important in PV and LDH in ET. Notably, the models relied primarily on readily available hematologic and morphologic features without requiring additional specialized data. External validation is warranted before clinical implementation.
DISCLOSURES
Muhammad Ali Khan, MBBS: No relevant conflicts of interest to declare.