Validation and comparison of contemporary prognostic models in primary myelofibrosis

Analysis based on 334 patients from a single institution




Survival in primary myelofibrosis (PMF) is predicted by several prognostic scoring systems (PSSs); the most widely recognized is that of Dupriez. Two other PSSs, Cervantes and Mayo, were recently reported as being more useful in younger patients. The current study compares these 3 PSSs among all age groups.


The Mayo Clinic PMF database was queried to identify a consecutive series of patients in whom pretreatment bone marrow and complete blood count (CBC), obtained within 6 months of diagnosis, were available for review.


Among 334 study patients (median age, 57 years), median survival was 70 months. Multivariable analysis of all 6 adverse prognostic factors utilized in the aforementioned PSSs (ie, hemoglobin <10 g/dL, leukocyte count <4 or >30 × 109/L, constitutional symptoms, circulating blasts ≥1%, platelet count <100 × 109/L, absolute monocyte count ≥1 × 109/L) identified all but platelet count as being significant. The Mayo PSS, which is based on the 4 CBC-derived parameters (ie, hemoglobin, platelet, leukocyte, and monocyte counts), displayed a better hazard ratio profile compared with the other 2 PSSs in identifying long-lived patients as well as delineating intermediate-risk disease category. The latter effect was even more pronounced in patients younger than age 60 years.


The Mayo PSS for survival in PMF is an objective CBC-derived prognostic model that might be superior in its performance over that of either the Dupriez and Cervantes PSSs. Cancer 2007. © 2007 American Cancer Society.

Primary myelofibrosis (PMF), also known as chronic idiopathic myelofibrosis (CIMF) or agnogenic myeloid metaplasia (AMM), is characterized by stem cell-derived clonal myeloproliferation that is often associated with bone marrow fibrosis and extramedullary hematopoiesis (EMH).1 The primary clonogenic mutation(s) in PMF remains at large, although much attention has been given to 2 recently described gain-of-function mutations, JAK2V617F2 and MPLW515L/K,3, 4 which occur in approximately 45% and 5% of patients, respectively.5, 6 Typical clinical presentation in PMF includes anemia, marked splenomegaly, and debilitating constitutional symptoms.7 Laboratory features include leukoerythroblastosis, dacryocytosis, leukocytosis (or leukopenia), thrombocytosis (or thrombocytopenia), and increased serum level of lactate dehydrogenase (LDH). Over time, patients display progressive organomegaly, worsening cytopenia, cachexia, peripheral edema, symptomatic portal hypertension, thrombotic or bleeding complications, multiorgan EMH, and blast phase disease (PMF-BP).8

Neither conventional nor investigational drug therapy has been shown to improve survival in PMF. Such therapy is instead used to alleviate cytopenia and/or organomegaly.9–11 Both splenectomy and involved field irradiation therapy are also employed for palliative purposes.12, 13 The promise of ‘cure’ through the use of either myeloablative or reduced-intensity conditioning (RIC) allogeneic stem cell transplantation (ASCT) is undermined by the relatively high incidence of treatment-related mortality and morbidity.14 Therefore, many patients are managed by ‘watchful waiting,’ if asymptomatic. Conversely, recent information on disease pathogenesis has rejuvenated the focus on experimental targeted therapy as an alternative treatment strategy. In order to facilitate treatment decisions in PMF, a number of prognostic scoring systems (PSSs) have been constructed, based on the presence or absence of independent risk factors for survival.15–29 In the current, single institution-based retrospective study of 334 patients with PMF (ie, meeting the WHO criteria for CIMF), we compared the performance of the most widely used PSS (the Dupriez PSS)15 against 2 other recently published models (the Cervantes PSS and the Mayo PSS)17, 19 that were derived from prognostic factors in younger patients (Table 1).

Table 1. Frequently Utilized Prognostic Models in Primary Myelofibrosis
Prognostic scoring systemRisk categoryScore sumMedian survival, moScore for Hgb <10 g/dLScore for WBC <4 or >30 ×109/LScore for Plt <100 ×109/LScore for AMC ≥1 ×109/LScore for symptomsScore for circulating blasts ≥1%
  1. Hgb indicates hemoglobin; WBC, white blood cell count; Plt, platelet count; AMC, absolute monocyte count; NA, not applicable.

Mayo PSS19 (n=129) (Ages <60 y; median 52)Low01731111NANA
Dingli et al.36 (n=160) (Ages <60 y; median 52)Low0155111NANANA
Cervantes PSS17 (n=116) (Ages ≤55 y; median 46)Low0 or 11761NANANA11
Cervantes et al.16 (n=106) (All ages; median 64 y)Low0 or 1991NANANA11
Dupriez et al.15 (n=195) (All ages; median 65 y)Low09311NANANANA


Permission was obtained from the Mayo Clinic Institutional Review Board to review the medical records of all patients seen at the Mayo Clinic and diagnosed with any of the aforementioned different names for PMF, including CIMF, AMM, and myelofibrosis with myeloid metaplasia (MMM), during the period from 1976 through 2006, with subsequent follow-up information updated in November 2006. Diagnosis of PMF was made according to the WHO criteria after re-review of both clinical information and bone marrow histology in all cases.30 In this regard, patients who received their diagnosis before the formal publication of the WHO criteria were reevaluated retrospectively and included in the study only if they fulfilled the WHO criteria for PMF (ie, CIMF). Special care was undertaken to avoid inadvertent inclusion of patients with myelodysplastic syndrome associated with bone marrow fibrosis (MDSf), chronic myelomonocytic leukemia (CMML), or acute myelofibrosis. Diagnosis of PMF-BP was according to criteria discussed but not finalized by the International Working Group for Myelofibrosis Research and Treatment (IWG-MRT): the presence of either 20% blasts in the bone marrow or 30% in the peripheral blood.31

All bone marrow examinations, obtained at time of diagnosis, were either performed or reviewed at our institution to ensure accuracy of diagnosis. In addition, all study patients were required to have documented information regarding a complete blood count (CBC) that was obtained either at the time of diagnosis or within 6 months of diagnosis but before therapeutic intervention. In patients that were alive, every attempt was made to update follow-up information by means of a questionnaire/telephone call sent to both patients and their primary doctors and the ‘date of last contact’ reflected this timepoint and not the last time they were seen at the Mayo Clinic.

Conventional statistical procedures were followed in the presentation and analysis of clinical and laboratory information. All data were analyzed by using StatView (SAS Institute, Cary, NC). All P-values were 2-tailed and statistical significance was set at the level of P < .05. Continuous variables are summarized as medians and ranges. Categorical variables are described as count and relative frequency (%). Comparison between categorical variables was performed by chi-square statistics. Comparison between categorical and continuous variables was performed by either the Mann-Whitney U-test or Kruskal-Wallis test. Survival functions were estimated by Kaplan-Meier plots taking the interval from the date of diagnosis to death or last contact. The log-rank test was used to test the homogeneity of overall survival curves over different groups. The Cox proportional hazards model was utilized to determine the impact of various clinical and laboratory variables on survival.


A consecutive series of 334 patients with WHO-defined PMF (ie, CIMF) were included in the study and their age and gender distribution is outlined in Table 2. All patients fulfilled the study eligibility criteria; pretreatment CBC and bone marrow information was available for review either at or within 6 months of diagnosis. In addition, absolute monocyte count and the presence or absence of hypercatabolic symptoms was documented in 311 and 312 patients, respectively. For the purposes of the current study we focused on the 6 prognostic factors that are utilized in the Dupriez (hemoglobin <10 g/dL, leukocyte count <4 or >30 × 109/L), Cervantes (hemoglobin <10 g/dL, circulating blasts ≥1%, presence of constitutional symptoms), or Mayo (hemoglobin <10 g/dL, leukocyte count <4 or >30 × 109/L, absolute monocyte count ≥1 × 109/L, platelet count <100 × 109/L) PSSs (Table 1). This was not to undermine the value of other prognostic factors but to remain within the scope of the main objective for the current communication. As previously defined, constitutional symptoms included unexplained fever, excessive sweats, or weight loss of >10% of the baseline value in the year before diagnosis.

Table 2. Presenting Clinical and Laboratory Features in 334 Patients With Primary Myelofibrosis and Their Correlation With Survival
ParametersValuesUnivariate PMultivariate P
  • AMC indicates absolute monocyte count.

  • Unless otherwise indicated, n = 334 for all analyses.

  • *

    Multivariate analysis including laboratory covariates only: hemoglobin <10 g/dL, leukocyte count <4 or >30 × 109/L, platelet count <100 ×109/L, AMC ≥1 ×109/L, circulating blasts ≥1%.

Age, y (median and range)57 (15–88)  
No. of males (%)201 (60)  
Hemoglobin in g/dL (median and range)11 (4.5–16.1)  
No. of patients with hemoglobin <10 g/dL (%)104 (31)<.0001<.0001/<.0001*
Leukocyte count ×109/L (median and range)8.2 (1.5–157)  
No. of patients with leukocyte count <4 or >30 ×109/L (%)67 (20)<.0001.02/.09*
Platelet count ×109/L (median and range)271 (12–2200)  
No. of patients with platelet count <100 ×109/L (%)63 (19)<.0001.19/.05
Absolute monocyte count (AMC) ×109/L (median and range)0.34 (0–5.9)  
No.of patients with AMC ≥1 ×109/L (%)37/311 (12).002.0004/<.0001*
Circulating blast % (median and range)0 (0–25)  
No. of patients with circulating blasts ≥1% (%)111/312 (36).001.04/.02*
No. of patients with hypercatabolic symptoms (%)95 (28)<.0001.005
Follow-up in months (median and range)28 (0–282)  
No. of patients who developed blast phase disease (%)32 (10)  
No. of deaths (%)140 (42)  
Median survival in months among Dupriez prognostic categories (n=334)
 Low-risk100 (n=197)<.0001 
 Intermediate-risk44 (n=103).007 
 High-risk22 (n=34)  
Median survival in months among Cervantes prognostic categories (n=322)
 Low-risk81 (n=242)<.0001 
 High-risk32 (n=80)  
Median survival in months among Mayo prognostic categories (n=319)
 Low-risk134 (n=152)<.0001 
 Intermediate-risk50 (n=96).0001 
 High-risk29 (n=71)  

Table 2 outlines clinical and laboratory values of interest at presentation. After a median follow-up of 28 months (range, 0–282), 140 deaths (40%) were documented and the overall median survival was 70 months. In a univariate analysis, each 1 of the 6 aforementioned risk factors displayed significant correlation with shortened survival (Table 2). During multivariable analysis that included these 6 parameters as covariates, all but platelet count sustained their prognostic significance (Table 2). Because of the subjective nature of the evaluation for constitutional symptoms, a second multivariable analysis that included the 5 laboratory-derived objective parameters (hemoglobin <10 g/dL, leukocyte count <4 or >30 × 109/L, absolute monocyte count >1 × 109/L, platelet count <100 × 109/L, and circulating blasts ≥1%) was performed. The latter analysis again demonstrated the strong prognostic value of hemoglobin level and absolute monocyte count (Table 2). In addition, platelet and circulating blast count, but not leukocyte count, were found to carry additional prognostic value.

We next evaluated the performance of each of the 3 above-mentioned PSSs using the current dataset of 334 patients. All 3 PSSs performed adequately to distinguish long-lived from short-lived patients groups. However, as is evident from both Table 1 and Figure 1, the Mayo PSS appeared to outperform the other 2 PSSs in identifying a low-risk patient group with risk parameters that allowed median survival exceeding 10 years. Similarly, compared with the Dupriez PSS, the Mayo PSS displayed a better hazard ratio profile, including narrower confidence intervals, in delineating an intermediate-risk disease category (Fig. 1). The latter effect was even more pronounced in patients younger than age 60 years (ie, the subgroup of patients that are usually considered transplant-eligible; Fig. 2).

Figure 1.

Survival curves in primary myelofibrosis based on the (A) Dupriez (n = 334 evaluable), (B) Mayo (n = 319 evaluable), and (C) Cervantes (n = 322 evaluable) prognostic scoring systems (see Table 1 for details on prognostic models). CI, confidence interval.

Figure 2.

Survival curves in primary myelofibrosis based on the (A) Dupriez (n = 193 evaluable), (B) Mayo (n = 185 evaluable), and (C) Cervantes (n = 186 evaluable) prognostic scoring systems, in patients younger than age 60 years. CI, confidence interval.


In PMF, several studies have demonstrated the adverse prognostic relevance of advanced age,15, 16, 20, 22–25, 28 male sex,15, 22 anemia that is usually defined by a hemoglobin level of <10 g/dL,15–29 leukocytosis and/or leukopenia that is usually defined by a leukocyte count of <4 or >30 × 109/L,15, 18–21, 25 thrombocytopenia that is usually defined by a platelet count of <100 × 109/L,18–20, 25, 27 absolute monocyte count defined by a value of ≥1 × 109/L,19 circulating blast percentage often defined as either ≥1% or ≥3%,15–17, 22 circulating immature granulocyte percentage of ≥10%,21 constitutional symptoms,16, 17 and cytogenetic abnormalities either in the bone marrow18–20, 32, 33 or splenic tissue.34 With regard to the relation between cytogenetic findings and survival, the presence of solitary abnormalities involving either 13q- or 20q- were associated with significantly better prognosis compared with all other abnormalities.32, 33 Most recently, the prognostic relevance of JAK2V617F for survival in PMF was evaluated in a preliminary fashion by 2 groups of investigators whose findings were not consistent.5, 35 Regardless, some of the aforementioned risk factors have been utilized to construct several PSSs15–17, 22, 36 of which the Dupriez15 and Cervantes17 PSSs are the most widely used (Table 1). In addition, 2 recent studies from the Mayo Clinic18, 19 reported on a modified Dupriez PSS (ie, the Mayo PSS) that appeared to better delineate intermediate-risk category in transplant-eligible patients (Table 1).

The main objective of the current study, which is the largest of its kind, was not to come up with additional prognostic factors or PSSs. Instead, we wanted to validate the significance of prognostic variables used in selected PSSs whose overall performance was also compared. Considering the fact that ASCT is currently a major treatment option in PMF, we focused on age-exclusive contemporary PSSs: Dupriez, Cervantes, and Mayo (Table 1). In other words, an effective PSS in PMF should enable accurate identification of patients in whom the benefit of ASCT offsets its risk (ie, high-risk category) as well those in whom experimental treatment protocol participation is appropriate, regardless of age (ie, intermediate-risk category). As is evident from Table 2, most if not all of the 6 adverse factors used in at least 1 of the 3 aforementioned PSSs displayed their independent prognostic value. From a practical standpoint, evaluation of laboratory-derived parameters, as opposed to quantification of patient symptoms, is in general more objective. As such, a multivariable analysis was performed using parameters that are easily accessible from either a CBC or peripheral blood smear evaluation and identified hemoglobin <10 g/dL, circulating blasts ≥1%, absolute monocyte count ≥1 × 109/L, and platelet count <100 × 109/L, but not leukocyte count <4 or >30 × 109/L, as being independently predictive of shortened survival. This raises the possibility of further refinement to current PSSs by adding and/or subtracting certain scoring variables.

The 3 PSSs tested in the current study were all effective in distinguishing low-risk from high-risk patients (Fig. 1). However, the incorporation of both platelet and monocyte counts, by the Mayo PSS, to the original Dupriez prognostic model15 resulted in both a more robust selection of low-risk patients with a median survival that exceeded 10 years in all age groups as well as better delineation of an intermediate-risk category, especially in those younger than age 60 years (Fig. 2). It is to be noted that almost twice as many patients with poor risk features are identified by the Mayo PSS when compared with the Dupriez PSS. Furthermore, from a practical standpoint, a 3-tier (Mayo or Dupriez PSS) as opposed to a 2-tier (ie, Cervantes PSS) risk stratification system is preferred because it can be employed to identify patients who are most suitable for ‘watchful waiting’ (low-risk), conventional or experimental drug therapy (intermediate-risk), or ASCT if transplant-eligible (high-risk).

In conclusion, the current study supports the use of the Mayo PSS, instead of either the Dupriez or Cervantes PSS, for current prognostication of survival in PMF. However, we believe that there is room for improvement and 1 possibility would be to incorporate all 6 aforementioned risk factors, as well as information from cytogenetic studies, into 1 global prognostic model. Establishment of such a model would require a much larger group of study patients; the International Working Group for Myelofibrosis Research and Treatment (IWG-MRT) is currently looking into a multicenter collaborative effort to establish an International PSS (IPSS) for PMF.31