Leukemic transformation among 1306 patients with primary myelofibrosis: risk factors and development of a predictive model

Among 1306 patients with primary myelofibrosis (PMF), we sought to identify risk factors that predicted leukemic transformation (LT) in the first 5 years of disease and also over the course of the disease. 149 (11%) LT were documented; patients who subsequently developed LT (n = 149), compared to those who remained in chronic phase disease (n = 1,157), were more likely to be males (p = 0.02) and display higher circulating blasts (p = 0.03), ASXL1 (p = 0.01), SRSF2 (p = 0.001) and IDH1 (p = 0.02) mutations. Logistic regression analysis identified IDH1, ASXL1 and SRSF2 mutations, very high-risk karyotype, age > 70 years, male sex, circulating blasts ≥ 3%, presence of moderate or severe anemia and constitutional symptoms, as predictors of LT in the first 5 years of diagnosis. Time-to-event Cox analysis confirmed LT prediction for IDH1 mutation (HR 4.3), circulating blasts ≥ 3% (HR 3.3), SRSF2 mutation (HR 3.0), age > 70 years (HR 2.1), ASXL1 mutation (HR 2.0) and presence of moderate or severe anemia (HR 1.9). HR-based risk point allocation resulted in a three-tiered LT risk model: high-risk (LT incidence 57%; HR 39.3, 95% CI 10.8–114), intermediate-risk (LT incidence 17%; HR 4.1, 95% CI 2.4–7.3) and low-risk (LT incidence 8%). The current study provides a highly discriminating LT predictive model for PMF.


Introduction
Primary myelofibrosis (PMF) is an aggressive myeloid malignancy currently listed under the World Health Organization (WHO) category of myeloproliferative neoplasms (MPN) 1 . PMF represents a stem cell-derived clonal expansion of myeloid cells that often harbor one of three driver mutations, including JAK2, CALR and MPL. PMF is morphologically characterized by abnormal megakaryocyte proliferation that is often accompanied by reticulin fibrosis. Patients with PMF typically display severe anemia, marked hepatosplenomegaly and profound constitutional symptoms. Other complications of the disease include cachexia, thrombosis, bleeding and leukemic transformation (LT). Overall survival (OS) in PMF is estimated at 6 years and can range between a few months to over 20 years, depending on the presence or absence of specific clinical and genetic risk factors [2][3][4] . Current treatment in PMF includes allogeneic stem cell transplant (allo-SCT), which is the only treatment modality with the potential to cure the disease or prolong survival 5 . Other treatment approaches in PMF are mostly palliative and include drug therapy (e.g. JAK2 inhibitors), splenectomy and involved field radiation therapy 6 .
Taking the above into consideration, the primary objective in developing a treatment strategy for the individual patient with PMF is to establish the timing of allo-SCT. The particular task is often accomplished by considering risk level, according to previously established risk models for OS. At present, these include the geneticallyinspired prognostic scoring system (GIPSS) 3 and the mutation-and karyotype-enhanced prognostic scoring system (MIPSS70 + version 2.0) 4 . GIPSS relies on genetic risk factors only, including karyotype, driver mutations and other mutations, including ASXL1, SRSF2 and U2AF1 Q157. MIPSS70 + version 2.0 utilizes the same genetic risk factors used in GIPSS but also incorporates three specific clinical risk factors, including constitutional symptoms, presence of severe/moderate anemia and ≥ 2% circulating blasts. The main objective for the current study was to develop a robust LT predictive model that complements GIPSS and MIPSS70 + version 2.0 and thus further facilitates treatment decision-making in PMF; in this regard, it is to be recalled that, in the context of GIPSS/MIPSS70 + , leukemia-free survival (LFS) was previously shown to be affected by karyotype, SRSF2 and ASXL1 mutations, platelet count < 100 × 10 9 /l and circulating blasts ≥ 2% 3,7 .

Methods
The current study was approved by the institutional review board of the Mayo Clinic, Rochester, MN, USA. The study population consisted of consecutive patients with PMF seen at our institution between April 26, 1976 and November 21, 2017. Diagnoses of PMF and LT were confirmed by both clinical and bone marrow examinations, in line with the 2016 WHO criteria; specifically, LT required presence of ≥ 20% blasts in the peripheral blood (PB) or bone marrow (BM) 1 . Data was collected retrospectively corresponding to the time of first referral which in the majority of cases was at the time of or within the first year of diagnosis. All patients were followed until death or last follow-up as assessed by medical records or through direct contact with patients or their physicians. Data collection was updated as of April 2018. The determination of prognostically relevant mutations was made by next generation sequencing (NGS)-derived mutation information 8,9 . Cytogenetics data were analyzed using standard techniques and reported in conformity with the International System for Human Cytogenetic Nomenclature criteria 10 .
Statistical analyses considered clinical and laboratory data collected at the time of initial PMF diagnosis or Mayo Clinic referral point. Continuous variables are presented as median (range) and categorical variables as frequency (percentage). The differences in the distribution of continuous variables between categories were compared using the Mann-Whitney or Kruskal-Wallis test. Categorical variables were compared using the χ 2 test. Logistic regression statistics was employed in order to identify predictors of LT at 5 years (i.e., early events) from initial diagnosis/referral; in the particular method, patients with documented LT within 5 years were "uncensored" while those followed up for at least 5 years, without developing LT, were "censored"; the analysis excluded patients without LT and not followed for at least 5 years. In addition, Cox regression analysis was performed to identify risk factors for overall leukemia-free survival (LFS). The Kaplan-Meier method was used to construct time-to-leukemia curves, which were compared by the log-rank test. P values of < 0.05 were considered significant. In order to develop LT predictive model, HRbased risk point allocation was employed and predictive accuracy was compared to those of GIPSS and MIPSS70 + version 2.0, using Akaike Information Criterion (AIC) and receiver operating characteristic (ROC) curve-derived area under the curve (AUC) estimates. The JMP® Pro 13.0.0 software from SAS Institute, Cary, NC, USA, was used for all calculations.

Discussion
Leukemic transformation (LT) is a dreaded complication of myeloproliferative neoplasms (MPN); reported 10year estimates of LT incidence range from 0.7-3% for ET, 2.3-14.4% for PV and 10-20 % for PMF 2,[16][17][18] . In a recent communication of 410 patients with post-MPN LT, recruited from the Mayo Clinic (n = 248) and multiple centers in Italy (n = 162), median survival was only 3.6 months and post-LT survival was independently affected by unfavorable karyotype, platelet count < 100 × 10 9 /l, age > 65 years and transfusion need at time of LT 19 . In general, long-term survival after LT was unusual, despite the achievement of close to 60% rate of complete remission, with or without incomplete count recovery 19 . The particular study revealed treatment-specified 3-year/ 5-year survival rates of 32%/10% for patients receiving allo-SCT, 19%/13% for patients achieving remissions following intensive chemotherapy but were not subsequently transplanted, and 1%/1% in the absence of both allo-SCT and chemotherapy-induced remission 19 . In other words, the survival benefit of allo-SCT in MPN 20 might not extend to those with LT, which underscores the need to identify patients at risk, before they undergo LT.
The current study provides a highly discriminating LT predictive model for PMF, which was shown to be superior to both GIPSS and MIPSS70 + version 2.0 in its LT predictive accuracy (Fig. 2). However, it should be noted that almost all of the variables used in the new LT predictive model (i.e., IDH1, ASXL1, SRSF2 mutations, circulating blasts ≥ 3%, age > 70 years and moderate/ severe anemia) were previously associated with shortened LFS (see above). What is different in the current study was i) the much larger sample size of informative cases; ii) the distinction between early events (logistic analysis of LT risk in the first 5 years of diagnosis) and overall risk (assessed by Cox analysis of LFS); iii) the combined analysis of mutations, cytogenetic abnormalities and clinical variables, in order to decipher inter-independent risk factors; and iv) development of a novel LT predictive model that includes both genetic and clinical risk factors. From a practical standpoint, the new LT risk model for PMF complements GIPSS and MIPSS70 + version 2.0 and should provide additional layer of prognostic information to assist with treatment decision-making, especially in terms of patient selection for allo-SCT. The current study also confirms the prognostic importance of specific mutations, sex-adjusted anemia and excess circulating blasts, for both OS and LFS, in PMF. Our observations require further validation, which might not be easy to accomplish, considering the difficulty in securing adequate number of informative cases.