Fax: (507) 266 4972
Prognosis in transplant-eligible patients with agnogenic myeloid metaplasia
A simple CBC-based scoring system
Article first published online: 20 DEC 2005
Copyright © 2005 American Cancer Society
Volume 106, Issue 3, pages 623–630, 1 February 2006
How to Cite
Dingli, D., Schwager, S. M., Mesa, R. A., Li, C. Y. and Tefferi, A. (2006), Prognosis in transplant-eligible patients with agnogenic myeloid metaplasia. Cancer, 106: 623–630. doi: 10.1002/cncr.21644
- Issue published online: 20 JAN 2006
- Article first published online: 20 DEC 2005
- Manuscript Accepted: 17 AUG 2005
- Manuscript Revised: 30 JUN 2005
- Manuscript Received: 10 JUN 2005
- agnogenic myeloid metaplasia;
- prognostic models
Allogeneic hematopoietic stem cell transplantation is potentially curative in agnogenic myeloid metaplasia (AMM) but is associated with substantial mortality and morbidity that necessitates accurate identification of patients in whom benefit outweighs risk. The current study describes the natural history of AMM in transplant-eligible patients and proposes a new prognostic scoring system that favorably compares with other established models.
Patients diagnosed with AMM before the age of 60 years and seen at Mayo Clinic were identified and the diagnosis confirmed. Relevant demographic, clinical, and laboratory characteristics were abstracted, and the impact of various parameters on overall survival (OS) was evaluated with univariate and multivariate analyses.
A cohort of 160 patients with AMM is described. OS was 78 months. Multivariate analysis identified a hemoglobin level of <10 g/dL, white blood cell count of either <4 or >30 × 109/L, platelet count of <100 × 109/L, presence of constitutional symptoms, and hepatomegaly as independent predictors of inferior survival. The first 3 complete blood count-based parameters were combined into a new scoring system that resulted in median survivals of 155, 69, and 24 months in the presence of 0, 1, or ≥2 adverse features. The chi-square value for the new model was 80.6 compared with 51.4, 48.4, and 43.7 for the models by Dupriez, Cervantes, and Visani, respectively.
A new scoring system based on blood count at the time of diagnosis can adequately stratify by risk transplant-eligible patients with AMM and can accurately identify high-risk as well as intermediate-risk disease. The new system displayed a stronger discriminative value, between risk categories, compared with currently existing prognostic models. Cancer 2006. © 2005 American Cancer Society.
Agnogenic myeloid metaplasia (AMM) is a clonal stem cell disease currently classified as a myeloproliferative disorder. The disease is characterized by bone marrow fibrosis and myelophthisis that are accompanied by progressive anemia, splenomegaly, and extramedullary hematopoiesis.1, 2 The annual incidence of the disorder is 0.5 to 1.46/100,000 population with a median age at diagnosis of 65 years.3, 4 At present, drug therapy has not been shown to alter the natural history of the disease and allogeneic hematopoietic stem cell transplantation (SCT) (AHSCT) remains the only treatment modality with curative potential.5–7 Unfortunately, AHSCT is associated with substantial mortality and morbidity that currently restricts its use in selected patients, often under age 60 years. In this regard, therapeutic decision is facilitated by accurate information regarding age-specific natural history as well as objective identification of patients in whom the benefit of AHSCT outweighs both the short-term risk of death and long-term morbidity from graft-versus-host disease.
The natural history of AMM in transplant-eligible patients is limited to only two studies.8, 9 In one study, the number of patients was only 9 where the authors described a cohort of patients younger than 30 years.9 In the other, larger study, 121 patients younger than 55 years were described. These patients were recruited from several medical centers.8 Over the years, various groups have proposed risk-stratification models for patients with AMM.10–12 In our experience, these models have been inadequate to differentiate intermediate-risk from high-risk disease. This distinction is not only an academic question because patients with intermediate-risk disease may be good candidates for experimental therapies aimed at altering the natural history of the disease. Therefore, a risk-stratification model that reliably and objectively identifies patients with intermediate-risk should enhance selection of patients for clinical trials. Proper selection is also important given the relative rarity of this disease, and implementation of uniform criteria will help accrual and enhance the meaning of such studies.
In the current study, we report on a single institution's experience with a large cohort of patients with AMM who were eligible for AHSCT based on their age. The natural history of AMM in this group is described, and we propose a new and simple prognostic scoring system that stratifies patients better than previous risk stratification models.
MATERIALS AND METHODS
After approval from the Mayo Foundation Institutional Review Board (IRB) in accordance with federal regulations, we searched all medical records available at Mayo Clinic Rochester for patients who were diagnosed with AMM before the age of 60 years. The diagnosis of AMM was based on established criteria.13 The search strategy spanned the time period from January 1, 1976 through December 31, 2003. To ensure complete follow-up of the cohort, we contacted patients or their next of kin for information relating to vital status, general health, and any therapy for the disease. If patients permitted, their local physicians were also contacted for additional relevant information. Finally, local counties were contacted to obtain death certificates for accurate information about the time and cause of death. Our initial search identified 256 patients with a diagnosis of myelofibrosis. Each chart was evaluated to confirm the diagnosis of AMM, and patients with postthrombocythemic and postpolycythemic myelofibrosis were excluded from further analysis. In addition, patients with incomplete follow-up or who missed relevant clinical or laboratory parameters were excluded from further study. Thus, a cohort of 160 patients with AMM were identified and evaluated for the purposes of this study. The relevant demographic, clinical, and laboratory parameters were abstracted.
Demographic parameters included gender, age at diagnosis, vital status, and cause of death. Clinically relevant parameters that were abstracted included presence of constitutional symptoms (night sweats, unexplained fever >100 °F, and weight loss of greater than 10% of baseline body weight), any history of bleeding and thrombosis, transfusion requirements, organomegaly, and pharmacotherapy for the disease. Laboratory parameters studied included hemoglobin (Hb), white blood cell (WBC), and platelet count (PLT) at diagnosis; circulating immature cells (myelocytes and metamyelocytes) and blasts; circulating CD34+ cells, serum lactate dehydrogenase (LDH) level; and bone marrow histology and cytogenetics. All bone marrow biopsies were initially evaluated by Mayo Clinic hematopathologists and rereviewed by one of the authors (CYL) using uniform criteria for diagnosis and scoring of bone marrow fibrosis. Cytogenetic studies were available in 49 patients.
All statistical analyses were performed using StatView (SAS Institute, Cary, North Carolina). A P value <0.05 was considered statistically significant. Actuarial survival curves were plotted using the product-limit estimate according to Kaplan and Meier and compared using the Mantel–Cox test.14 Survival was defined as the time between diagnosis and death or last contact with the patient. Univariate analysis identified several clinical and laboratory parameters that were strongly associated with prognosis. These parameters were then combined in a step-wise fashion for a multivariate analysis by using a Cox proportional hazard model.15 Parameters that were statistically significant were retained, and the iterative addition of parameters was stopped when the P value for an additional parameter exceeded 0.05. Parameters where data was only available in <70% of the cohort were excluded from the multivariate analysis. A scoring system based on results of the multivariate analysis was developed.
Comparison with Other Scoring Systems
Patients were also stratified on the scoring systems developed by Visani et al.,10 Dupriez et al.,11 and Cervantes et al.,12 to evaluate their performance in our cohort of patients and to compare them to a new system developed with our multivariate analysis. The relative strength of each score was evaluated by Kaplan–Meier plots for the cohort after stratification based on each system and then compared by the chi square obtained for each model.14 In our analysis, the chi-square statistic evaluates significance of simultaneous exclusion of all covariates in the model and, thus, a higher chi-square value implies a stronger discriminant function between survival curves.
The study cohort comprised 160 patients with AMM in whom diagnosis was established before age 60 years. Complete follow-up information was available in all instances. The median age at diagnosis was 51.7 years (1–59.9 yrs), and 71 patients were females. Table 1 outlines patient demographics and presenting clinical features. Similarly, Table 2 lists relevant laboratory parameters at diagnosis. Bone marrow karyotype analysis was available in 49 patients. In 35 of these patients, results were either normal or had isolated abnormalities of chromosome 13 or 20 (13q− or 20q−) that have been associated with a good prognosis.16, 17 The other 14 patients had complex abnormalities that have been shown to be associated with an unfavorable prognosis.
|< 45 yrs||35 (22)|
|> 45 yrs||125 (78)|
|Hemoglobin (Hb; g/dL), median (range), n = 160||10.7 (5–14.9)|
|White blood cell (WBC; × 109/L), median (range), n = 160||8.2 (1–116)|
|Platelet count (PLT; × 109/L), median (range), n = 160||310 (1–1512)|
|No. of patients with Hb < 10 g/dL, n = 160||56 (35%)|
|No. of patients with WBC < 4 or > 30 × 109/L, n = 160||37 (23%)|
|No. of patients with PLT < 100 × 109/L, n = 160||38 (23%)|
|No. of patients with immature cells > 5%, n = 95||32 (34%)|
|No. of patients with circulating blasts > 2%, n = 58||26 (45%)|
|No. of patients with unfavorable karyotype, n = 49||14 (29%)|
Overall, 97 patients have required therapy for symptomatic anemia, control of symptomatic splenomegaly, clonal thrombocytosis, or constitutional symptoms. Hydroxyurea was given to 31 (32%) patients. Androgens were prescribed to 14 patients and interferon or anagrelide were given to 10 patients each. Alkylating agents or erythropoietin was prescribed to 6 patients each, whereas the remainder received colchicine, danazol, etanercept, imatinib mesylate, penicillamine, pirfenidone, prednisone, or thalidomide. Symptomatic anemia has been treated with regular transfusions in 41 (25%) patients. Splenectomy was performed in 49 patients, and another 6 patients had splenic irradiation either for symptomatic splenomegaly or control of cytopenias.
To date, 104 (65%) patients have died and 56 (35%) are alive. The median follow-up time for alive patients is 73 months (1–296 mos). Median overall survival (OS) for the entire cohort of 160 patients was 78 months (1–296 mos) (Fig. 1). Information regarding cause of death was available for 73 (70%) deceased patients and included progressive disease (PD) in 22 patients and leukemic transformation in another 22 patients. Other causes of death included sepsis (8 patients), arterial or venous thrombosis (9 patients), nonhematologic malignancies (7 patients), and gastrointestinal bleeding (4 patients). In addition, 1 patient developed a perforated peptic ulcer, 2 died of cardiorespiratory failure, and another died suddenly.
Univariate Analysis of Prognostic Parameters
We hypothesized that combining parameters from previously described prognostic models for AMM may lead us to an improved prognostic scoring system that better defines specific risk groups, especially intermediate-risk patients. Accordingly, we investigated the potential impact of age, gender, presence of splenomegaly or hepatomegaly, Hb, WBC, PLT, circulating blasts or immature cells, LDH, karyotype and bone marrow fibrosis on OS by the Kaplan–Meier method. Because Hb and WBC have been reported as independent predictors of prognosis in a previous large study, we opted to use the same limits as used in that study to simplify comparisons.11 Results are summarized in Table 3, and the impact of various parameters on survival are in agreement with previously published models.10–12 With few exceptions, survival in our cohort was somewhat better than that in published series, probably because of the younger age of our patients.
|< 45 yrs||35||134||7.244||0.0071|
|> 45 yrs||125||65|
|> 10 g/dL||105||135||42.1||< 0.0001|
|< 10 g/dL||55||29|
|> 4 but < 30 × 109/L||122||102.6||32.534||< 0.0001|
|< 4 or > 30 × 109/L||38||28.6|
|> 100 × 109/L||122||102.6||44.659||< 0.0001|
|< 100 × 109/L||38||21.8|
|≤ 300 U/L||24||28.6||0.31||0.5775|
|> 300 U/L||38||79|
|Grade ≤ 2||81||77||0.610||0.4347|
|Grade > 2||58||71|
On univariate analysis, we identified 9 parameters that were significantly associated with prognosis at the 0.01 level or lower (Table 3). The presence of splenomegaly did not have any impact on survival, confirming previous observations.10, 11 Similarly, serum LDH level and the degree of bone marrow fibrosis had no effect on prognosis. When analyzing the impact of PLT count, we studied both a limit of 150 as well as 100 × 109/L. Although both limits were significantly associated with an adverse survival, the lower limit was associated with a stronger chi-square (24.718 and 44.659), and this was the limit used in multivariate analysis. Although the percentage of circulating blasts was statistically significant, the chi square was relatively small, probably because of the small number of patients where this was reported (58 total).
Multivariate Analysis and Comparison with Other Scoring Systems
For the multivariate analysis, we considered only variables with a P < 0.001, and data was available for greater than 70% of patients in the current study. Accordingly, we performed a Cox proportional hazards regression model based on PLT, Hb, WBC, presence of constitutional symptoms, and hepatomegaly, in that order. Parameters were added serially to the model based on the strength of their chi square obtained in univariate analysis (Table 3). Results from multivariate analysis are presented in Table 4 and demonstrate that all 5 variables were independent of each other and had an impact on prognosis. The importance of PLT was striking and in contrast to previously published models.10–12
Given that all 3 hematologic parameters available from a complete blood count (CBC) displayed independent prognostic value, we devised a CBC-based scoring system that takes all 3 parameters into consideration. A score of 1 was attributed for each adverse prognostic factor (Hb <10g/dL, WBC <4 or >30x109/L, and PLT <100 × 109/L) and summed. Thus, the total score can range from 0 to 3. To simplify the model further and identify low, intermediate, and high-risk patients, we scored patients only with the range 0 to 2, i.e., no abnormalities (0), any 1 abnormality1 or any 2.2 When survival was analyzed by using this scoring system, there was a clear demarcation of patients into 3 groups (Fig. 2A). Patients with a score of 0 (good prognosis) had a median survival of 155 months; patients in the intermediate group1 had a median survival of 69 months; and the poor risk group (≥2) had a median survival of 23.5 months (P < 0.0001).
We also evaluated survival of our cohort based on the scoring systems previously described (Figs. 2B–D).10–12 It is clear that the new CBC-based scoring system is superior at identifying intermediate-risk patients compared with other systems. The scoring systems were then compared statistically by using a Kaplan–Meier estimate with each system as a nominal variable. The new CBC-based system is associated with a higher chi square, which suggests a more powerful predictive model (Table 5).
|System||No. (%)||Survival in mos||χ2||P|
The number of patients with available karyotypes was small. However, a multivariate analysis restricted to patients where all parameters were available showed that presence of an unfavorable karyotype, a low PLT, and a low Hb remained independent, and significant parameters (chi squares of 15.98, 21.144, and 4.199, respectively) and WBC lost prognostic significance. Thus, an unfavorable karyotype is associated with a poor prognosis, most likely because this karyotype portends leukemic transformation with associated dismal prognosis.18 A similar multivariate analysis showed that extremes of WBC and presence of constitutional symptoms at the time of diagnosis were the only independent risk factors with respect to leukemia-free survival.
AMM is biologically very heterogenous both in clinical presentation as well as prognosis, and patients can survive from months to decades.3, 4, 19, 20 The etiology of this disorder remains unclear,21 and the role of the recently described JAK2V617F activating mutation that is found in up to 50% of patients with AMM is yet to be defined.22–24 In the absence of a molecular target that is related to disease pathogenesis, the only curative option is ASHCT.5, 7, 25 This procedure is usually limited to younger patients with a good performance status. Unfortunately, ASHCT is associated with significant early mortality and delayed morbidity, and, therefore, clear risk stratification procedures must be available to correctly identify patients and to guide therapeutic decisions. Although it may be reasonable to propose ASHCT to patients with a high risk of death, this is not justifiable for patients who can survive for many years (low-risk patients). The optimal time for ASHCT in AMM is not clear, but data from patients with myelodysplastic syndrome suggest that high-dose therapy before leukemic transformation is optimal.26 However, regardless of the question of ASHCT, there is a third group of patients with intermediate-risk disease that have been difficult to capture with current prognostic models. These patients are ideal candidates to enroll in experimental therapies that are designed to alter the natural history of this disease. Therefore, a clear understanding of the natural history of patients with AMM who are potential ASHCT candidates because of age must be well described. In this report, we describe the largest single-institution experience of AMM in young patients who could be eligible for ASHCT, and we propose a more powerful prognostic, yet simple, scoring system to identify patients with intermediate-risk disease.
Because myelofibrosis may be primary (AMM) or due to progression of polycythemia rubra vera or essential thrombocythemia, we were careful to restrict our analysis to patients with de novo AMM.27 Our results show that AMM in patients younger than the age of 60 years is associated with a median survival that exceeds 6 years. This is somewhat longer than that reported for the general population of patients with AMM and is probably because our cohort was younger in age.4, 10–12 Most patients will require therapy for control of clonal cytosis, symptomatic splenomegaly, and anemia at some point during the course of their disease. Constitutional symptoms are common, and both thrombosis and bleeding occur with a significant frequency. The principal causes of death are PD with marrow failure, infection, and transformation to acute leukemia. In our cohort, AML developed in 30% of patients where the cause of death was established. This may appear to be higher than previously reported rates of transformation and is probably related to the longer follow-up of our cohort of patients.8, 10–12, 28
In our series, we confirmed the importance of established prognostic variables such as Hb, WBC, circulating immature cells and blasts, hepatomegaly, bone marrow karyotype, and presence of constitutional symptoms.4, 10–12, 29, 30 Perhaps the most striking observation was the impact of PLT on prognosis in multivariate analysis. To our knowledge, this is a novel observation because none of the established models include it in their final risk stratifications.10–12 Inclusion of PLT provides a more complete evaluation of hematopoiesis and how this is deranged in AMM and probably explains the additional strength of our model compared with prior ones. Although on multivariate analysis all 5 parameters selected remained independent predictors of mortality, we opted to develop a risk-stratification model based on 3 parameters available through CBC at diagnosis. The rationale for this approach is the reliability of the test that is both observer and patient independent compared with the more subjective reporting of constitutional symptoms and evaluation of hepatomegaly. In addition, the success of any prognostic scoring system is to a great extent based on its simplicity and widespread availability of the parameters it considers. In this respect, nothing can surpass the universality of the CBC in hematology. Although our system could in principle allow 4 subgroups, our studies show that an abnormality (as defined) in any hemopoietic cell lineage is associated with intermediate risk and any 2 abnormalities portend poor-risk disease. In this respect, the model extends the observations of Dupriez et al., where they only included Hb and WBC at diagnosis in their prognostic model.11
The presence of karyotype abnormalities has long been reported as an adverse prognostic factor.11, 31 However, not all cytogenetic abnormalities are associated with an adverse prognosis. In previous studies, we have shown that isolated 13q− or 20q− abnormalities are not associated with an adverse prognosis.16, 17 In our subgroup analysis of patients with available cytogenetics, the presence of a “favorable” karyotype (defined as either a normal or an isolated 13q− or 20q− abnormality) was associated with a significantly better outcome compared with survival in patients with an “unfavorable karyotype” (defined as any abnormality except as above) that often had complex cytogenetic abnormalities and was associated with a high risk of progression to AML. In the presence of an unfavorable karyotype, both Hb and WBC lost their prognostic relevance, and only PLT remained an independent variable. Informative karyotypes are available in <50% of patients with AMM, and, therefore, we elected to exclude this prognostic factor from our model. In the presence of an unfavorable karyotype (fortunately uncommon), the present scoring systems do not apply, and it can be assumed that the patient has high-risk disease.
In conclusion, AMM in young patients is associated with a median survival that exceeds 6 years. The CBC obtained at the time of diagnosis provides a powerful prognostic tool that can be used to stratify patients by risk and to help decide therapy. We described a new prognostic model that identifies an important group of patients with intermediate-risk disease who should be enrolled in studies aiming to modify the natural history of this disease. Allogeneic SCT should be considered only in patients with high-risk disease.
- 15Regression models with life tables (with discussion). J Royal Stat Soc. 1972; 34: 187–220..