|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
© 2001 American Society for Clinical Oncology Bone Marrow Features Improve Prognostic Efficiency in Multivariate Risk Classification of Chronic-Phase Ph1+ Chronic Myelogenous Leukemia: A Multicenter TrialFrom the Institutes of Pathology, Universities of Cologne, Freiburg, and Essen; Department of Oncology and Hematology, Klinikum Leverkusen, Leverkusen; and First Clinic of Medicine, University of Cologne, Cologne, Germany. Address reprint requests to Hans Michael Kvasnicka, MD, Institute of Pathology, University of Cologne, Joseph-Stelzmann-Str 9, D-50924 Cologne, Germany; email: hm.kvasnicka{at}uni-koeln.de
PURPOSE: Multivariate risk classifications for chronic (stable)-phase Ph1+ chronic myelogenous leukemia (CML) are generally focused on hematologic variables, and the putative prognostic property of bone morphology has been neglected or even contested so far. PATIENTS AND METHODS: A total of 510 consecutively recruited patients in first chronic phase Ph1+ CML and pretreatment bone marrow biopsy specimens were entered onto this multicenter observational trial to evaluate the effect of bone marrow histopathology. According to generally accepted criteria, patients with any signs of accelerated disease were excluded. Treatment modalities included administration of interferon alfa-2b (IFN) and chemotherapy with hydroxyurea (HU) or busulfan. Immunohistochemical and morphometric techniques were applied to identify marrow cells and to quantify fiber density. Patients were separated into learning and validation samples, and classification and regression tree (CART) analysis was performed to establish a prognostic decision tree. RESULTS: CART analysis of the validation sample (123 patients with HU therapy) revealed the amount of erythroid precursors in the bone marrow, myelofibrosis, and splenomegaly as the most important prognostic features. Three risk profiles with significantly different survival patterns were established, with median survival times ranging from 33 to 108 months (two-sided log-rank test, P = .0001). The new score was confirmed by application to the learning sample with IFN therapy (two-sided log-rank test, P = .0002). Furthermore, risk status defined by the new score was significantly correlated with the occurrence of blast transformation. CONCLUSION: Our data strongly implicate that prognostic classification of chronic-phase Ph1+ CML can be significantly improved by the inclusion of morphologic parameters. The variables of the presented scoring system may be easily assessed by routinely processed aspirates and bone marrow trephines.
MULTIVARIATE RISK classifications for chronic (stable)-phase Ph1+ chronic myelogenous leukemia (CML) based on pretreatment features have repeatedly been the subject of prognostic studies.1-8 Several models have been established that categorize patients into different groups with distinctive survival characteristics.3-5 However, the reproducibility and transportability of most of these staging systems have been repeatedly questioned because these systems were generally restricted to certain therapy regimens.9,10 Application of the various risk models to cohorts with other treatment modalities, in particular interferon alfa-2b (IFN) therapy, frequently disclosed a significant lack of prognostic efficiency with overlapping risk discrimination.3,9-12 Therefore, recently a new collaborate CML score was proposed that was especially adapted to IFN-treated patients.3 So far, all attempts to calculate prognostic staging systems and to assess more individually patients’ risk profiles have generally focused on hematologic variables, and thus, the putative prognostic property of bone marrow changes has been neglected in leading trials.8,13 In this context, there is still disagreement about whether and to what extent morphologic characteristics may predict differences in prognosis.14,15 However, in accordance with previous studies, it may be concluded that the significantly wide spectrum of hematologic findings on admission1 is also reflected by a corresponding variety of histomorphologic features.16 Consequently, there is growing evidence derived from former studies that a number of morphologic findings exert a significant effect on survival for Ph1+ CML. In particular, myelofibrosis was regarded as one of the most important morphologic features.8,13,17,18 From these investigations it was obvious that an increase in argyrophilic fibers generally implies a worsening of prognosis. However, because of semiquantitative grading systems in most trials,19 only manifest fibrosis was correlated with an unfavorable outcome.17,18 In more comprehensively conducted studies, including morphometry, confirmative data were produced that borderline to marginal changes in the fiber density indicate a slightly more advanced stage of disease and result in a reduced survival.8,13,16,20 In considering allogeneic bone marrow transplantation, these findings are of particular interest. Recently published data were in keeping with the assumption that pretransplant myelofibrosis might be responsible for a delayed or failing engraftment.21,22 Depending on the close biologic and functional relationship between fibers and megakaryocytes, this cell lineage was also reported to exert a substantial prognostic value.8,13 Of the other histologic bone marrow features that were regarded as possible indicators for survival, macrophages,13 but in particular so-called pseudo-Gaucher cells8,13,23 and the iron-laden subset of reticular histiocytes,24,25 were frequently studied. However, corresponding findings proved to be ambiguous, and therefore, their prognostic impact gave rise to discussion and controversy.26 Furthermore, medullary (nucleated) erythroid precursor cells have rarely been investigated with respect to their prospective property,13,27 although anemia and the presence of erythroblasts or normoblasts in the peripheral blood were repeatedly described as distinctive predictors for survival in CML.4,8,9,28,29 With respect to these inconsistencies surrounding certain aspects of prognostic evaluation of chronic (stable)-phase Ph1+ CML, the purpose of this multicenter observational study on a large number of patients with different treatment modalities was to investigate systematically, by applying morphometry, the histologic features that characterize pretreatment bone marrow biopsies and to evaluate relevant correlations with survival and occurrence of blast transformation. In the course of this analysis, we tried to extract and validate a new prognostic scoring system for chronic (stable)-phase Ph1+ CML, which is based on readily to determine morphologic and hematologic findings on admission. Finally, the prognostic performance of already published and generally accepted clinical scores was compared with the newly constructed risk model.
Patients The results of this multicenter observational trial are based on a consecutively recruited patient cohort with chronic (stable)-phase Ph1+ CML from three different institutions over a 15-year period (January 1982 to July 1997). Observation time ranged between 3 to 120 months, with a median of 32 months. Admission criteria included a strictly confirmed diagnosis of CML by clinical, cytogenetic (Ph1+, bcr/abl translocation), and morphologic findings; complete clinical and hospital records, including treatment modalities and follow-up status; and representative (pretreatment) bone marrow biopsy specimens, in addition to smears. According to generally accepted criteria,30 patients with any signs of accelerated disease, ie, more than 15% myeloblasts or more than 30% myeloblasts plus promyelocytes in the peripheral blood, thrombocyte counts less than 100 x 109/L, basophils greater than 20% in the blood, and a follow-up interval of less than 6 months, were excluded from further analysis. With these criteria, from a total of 611 collected patients, 510 cases with different therapeutic modalities were selected for subsequent analysis: 133 patients (79 men, 54 women; median age at diagnosis, 47 years; range, 17 to 81 years) with recombinant IFN therapy mostly received 9 x 106 U/m2/d as an initial dosage, which was followed by an appropriate maintenance treatment (2 to 5 x 106 U/m2/d), according to leukocyte counts and toxicity. Fifty (37.6%) of these 133 patients had been previously included as part of a prospective randomized single-center therapy study.31,32 One hundred twenty-three cases (71 men, 52 women; median age at diagnosis, 53 years; range, 19 to 88 years) were treated with hydroxyurea (HU), with daily doses of 20 to 40 mg/kg adapted to peripheral leukocyte counts; 154 patients (83 men, 71 women; median age at diagnosis, 52 years; range, 16 to 81 years) from a historical group with busulfan (BU) therapy in dosages between 4 and 8 mg/d discontinued therapy with BU when leukocyte levels dropped below 20 x 109/L and resumed therapy at a count of more than 50 x 109/L. One hundred patients (60 men, 40 women; median age at diagnosis, 57 years; range, 19 to 89 years) had varying therapeutic modalities, including cross-over regimens and different combinations of regimens. Mean time between diagnosis and treatment ranged between 0.5 and 4 months for the four groups. A bone marrow transplantation was performed in 86 of the 510 patients (IFN group, n = 37; HU group, n = 32; BU group, n = 6; and other treatment modalities, n = 11), with a median interval between diagnosis and transplantation of 18.5 months (range, 6 to 62 months). At the closure of this investigation (deadline August 1, 1998), 298 patients were dead (censoring of 39.8%), and no patient was lost from follow-up. During the follow-up period, 206 patients (40.4%) developed a blast transformation, and in 28.6% (n = 146) this progression of disease occurred within 3 years of diagnosis. Further details regarding important clinical features of the patients, including risk classification according to the Sokal score5 and the new collaborate CML score,3 are listed in Table 1 for the whole cohort. With the exception of age on admission, no significant differences could be calculated between the groups. Compared with the other three cohorts, the IFN-treated group was significantly younger at diagnosis. Values for spleen and liver size were based on ultrasound measurements in most of the patients; however, with regard to former prognostic studies, these values were transformed into centimeters below costal margin, with a normal reference maximum length of 11 cm. Only in a small subset of cases (n = 39; 7.6%) was spleen size assessed by palpation, but these patients generally presented with a marked splenomegaly and therefore were definitively classified as having a spleen size of more than 2 cm in further analysis. With the proposed rigid standard criteria,12 approximately 76% of patients treated with IFN therapy showed a complete hematologic remission within 6 months, and of these, 16 patients had a major or complete cytogenetic remission. Under HU therapy, a partial hematologic response was recognized in 54%, and in the group with varying therapeutic modalities, it was recognized only in 49% of the patients. Regarding patients with BU treatment, in approximately 60%, no relevant remission or an improvement that lasted for more than a few months could be achieved.
Bone Marrow Biopsies Representative trephine biopsies of the bone marrow were performed from the posterior iliac crest before any treatment as initial examinations. All samples were processed by standard techniques to ensure a homogeneity of handling and consequently also of histochemical results.33 After fixation of bone marrow biopsies specimens in an aldehyde solution for 12 to 48 hours (2 mL of 25% glutaraldehyde, 3 mL of 37% formaldehyde, 1.58 g of anhydrous calcium acetate, and distilled water per 100 mL), further processing included decalcification for 3 to 4 days in 10% buffered EDTA, pH 7.4, and paraffin embedding. Several routine stainings, such as Giemsa, periodic acid-Schiff (PAS) reagent, Perls’ reaction, and naphthol-AS-D-chloroacetate esterase,33 were used. Argyrophilic fibers (reticulin-collagen) were demonstrated by the silver impregnation method (Gomori’s stain) because this method is generally known to be highly reliable on paraffin-embedded bone marrow trephines.17-19,34 In contrast to other staining procedures for demonstration of myelofibrosis (Snook’s reticulin stain, Sweat’s stain), this technique is easily reproducible and enables a standardized assessment of argyrophilic fibers. For a specific staining of marrow cells, a set of appropriate monoclonal antibodies was selected: CD61 (antiplatelet glycoprotein IIIa) for the identification of megakaryocytes, including abnormal microforms and precursor cells, such as promegakaryoblasts and megakaryoblasts35; PG-M1 (CD68) for the staining of all mature, resident macrophages or phagocytic reticular cells36; GSA-I (alpha-D-galactosyl residues; Griffonia simplicifolia isotype I-B4) to identify the so-called activated subpopulation37; and finally, Ret40f (antiglycophorin C) to mark selectively nucleated erythropoietic precursor cells.35 This monoclonal antibody is directed against sialoglycophorin beta, commonly known as glycophorin C, which is one of four sialic acid–rich polypeptides found in the human erythrocyte membrane; it identifies exclusively red cells and their precursors, either in routinely fixed, paraffin-embedded bone marrow trephines or in bone marrow smears. The monoclonal antibodies and other reagents were purchased from Dako-Diagnostica GmbH (Hamburg, Germany). Details of staining procedures were reported in detail in previous articles.8 Iron-laden macrophages were determined by Perls’ reaction, and detection of pseudo-Gaucher cells was based on their characteristic fibrillar birefringence after polarization of Giemsa-stained slides and was also confirmed by their onion skin–like, finely striated cytoplasmic pattern visualized by the PAS reaction. After immunostaining, morphometric analysis was performed by two manual optic planimeters (VIDAS-Zeiss-Kontron; Carl Zeis, Oberkochen, Germany) with a standard program set on large trephine biopsies with an artifact-free mean marrow area of 10 ± 3 mm2. Morphometric measurements were made by five individuals and regularly checked by two others for accuracy. Frequencies of CD61+ megakaryocytes, including atypical microforms and precursors, CD68+ and GSA-I+ macrophages, the iron-laden subset, and Ret40f+ nucleated erythropoietic cells, were evaluated (per square millimeter) at x500 magnification by calculation of the total marrow and also the fat-cell–free (hematopoietic) area of the trephine biopsy. Argyrophilic (reticulin and collagen) fiber density was measured by using an ocular grid and expressed as numbers of intersections (i) per square millimeter of hematopoietic area (so-called point-intersection method).
Statistical Analysis
Multivariate regression methods were applied to assess the relative prognostic value of patient characteristics associated with survival by using Cox’s proportional hazards model.42,43 Clinical variables were categorized according to cutoff points derived from previous studies.8,44 The cutoff point for fiber density was defined by the doubling of the mean value derived from the normal bone marrow (mean, 16.1 ± 5.3 i x 102/mm2; range, 4.1 to 20.3); this implied an initial borderline (reticulin) myelofibrosis.8 With regard to the amount of CD61+ megakaryocytes, CD68+ and GSA-I+ macrophages, and Ret40f+ nucleated erythroid precursors, the median of the total patient cohort was chosen as the splitting value. Consequently, boundaries were set at a density of 32 i x 102/mm2 for argyrophilic fibers, at 300/mm2 for CD68+ and 170/mm2 for GSA-I+ macrophages, and at 360 Ret40f+ cells/mm2 for erythropoiesis. All variables were entered into the model by using a forward stepwise selection procedure43 with a statistical significance level of 2 alpha = 0.1. For each variable, the proportional hazards assumption was examined by plotting the logarithm of the cumulative hazards function. Variables were entered into the model or removed on the basis of the value of the Wald A second multivariate analysis was performed by using the classification and regression tree (CART) approach.46,47 The purpose of this analysis is to obtain a classification rule based on pretreatment parameters to identify subsets of the patient population with homogeneous prognosis. Consequently, the result of the final stratification can be represented as a binary decision tree. Initially, the entire population was partitioned into two subgroups according to the variable that produces the best split with respect to survival probability, ie, the largest two-sample test statistic over all splits.48,49 In each of the resulting strata, the process was repeated recursively until none of the candidate determinants showed statistically significant differences between the two compared survival curves or until the size of a given subgroup was too small. In a final amalgamation process, subgroups of patients who did not differ in survival were joined to homogeneous prognostic classes.48 Essentially, tree-structured survival analysis provides another way of understanding the predictive structure of the underlying data and is useful for detection of nonlinear interactions between baseline variables. Creation of the subgroups according to a tree structure enables an assessment of the relative prognostic importance of covarying factors. A major advantage of this technique is the ease of interpretation of the results,49 because classification of a new patient to a prognostic group is straightforward by means of simply following the sequence of binary questions (splits). Moreover, CART analysis is able to illustrate effects specific to subsets of the patient population, and no parametric assumption regarding the survival distribution of the determinants of interest is required.47,48 Thus, this technique has to be considered as complementary to the proportional hazards (Cox) model.49,50 Further details regarding this statistical methodology are available.46 Finally, the probability and relative risk of blast transformation (ie, onset of blast crisis within 3 years of diagnosis) between the compared scores and individual risk groups were assessed by logistic regression analysis. The prognostic efficiency of the developed multivariate risk models was compared by constructing a receiver operating curve (ROC) of sensitivity and specificity of predicting death within 3 and 5 years of diagnosis.51,52 Furthermore, we used a recently proposed measure of separation (SEP) to determine the performance of the derived classification schemes.52 This calculation describes the weighted geometric mean of relative risks within prognostic strata compared with the baseline risk of the entire sample, ie, the therapy group. Hence, this rationale incorporates weights of relative size for the subgroups. Because of an exponential transformation, the resulting SEP can be interpreted in terms of relative risk.52
Bone Marrow Features on Admission The wide ranges of clinical and hematologic data on admission (Table 1) were properly reflected by corresponding distinctive features in the pretreatment bone marrow trephines (Table 2). These were particularly expressed by applying immunohistochemistry and morphometry. A conspicuous variability was found for CD61+ and mature PAS+ megakaryocytes. Compared with normal values, 40.8% of our cases presented as a megakaryocyte-rich subtype, with megakaryocyte counts of more than 60/mm2, showing predominantly clusters and sheets of atypical small cells (micromegakaryocytes). Similarly, Ret40f+ nucleated erythropoietic precursors were characterized by significantly different frequencies, including 237 patients (46.5%) with marked reduction of erythroid islets (Fig 1, a to f). In keeping with the cellular components of the bone marrow, morphometric assessment of argyrophilic fiber density disclosed an extreme heterogeneity ranging from normal to a minimal and pronounced increase, consistent with advanced (collagen) myelofibrosis (Fig 1, g to i). With regard to a widely used semiquantitative scoring system,19 early to mild reticulin myelofibrosis with a more than three-fold increase in normal (mean) reticulin fiber density (50 i x 102/mm2) was determined in 26.5% of patients (Fig 1i), whereas an advanced myelofibrosis compatible with a more than four-fold density was found in 9.6% of patients at diagnosis. However, according to the more accurate morphometric evaluation, a marginal increase in fiber density compatible with a doubling of the normal value was already recognizable in approximately 21% of patients at onset (Fig 1h). Macrophages could be stained most specifically with CD68 and GSA-I to identify additionally the activated subset, including paratrabecular localized pseudo-Gaucher cells, which were found in 29% of our patients.
Several significant correlations between pretreatment bone marrow features and clinical baseline variables could be ascertained. Patients with marked decrease in Ret40f+ erythroid precursors (< 360/mm2; Fig 1, c and f) were particularly characterized by greater anemia, ie, reduced hemoglobin level (P  .005), increased spleen size (P .001), and occurrence of myeloblasts in the peripheral blood (P .001). Comparable findings were revealed for cases with moderate myelofibrosis (> three-fold increase in reticulin fibers). These patients generally presented with an increased reduction of Ret40f+ erythroid precursors on admission (mean difference, 90.2 Ret40f+ precursors/mm2; P = .003). Moreover, significant correlations could be established between nucleated erythroid precursors and number of macrophages (r = .27, P .001) and between megakaryocytes and fiber density (r = .30, P .001). All these results proved to be independent of the presenting risk profile, according to the Sokal score,5 and were observed in all therapy cohorts.
Survival According to Therapy
Validation of Already Published Risk Scores When applying the widely accepted Sokal score5 to our IFN cohort, only a clearly distinguished low-risk group could be determined. The intermediate- and high-risk groups revealed no distinctive survival pattern, because with this score, high-risk-classified cases were associated with a better prognosis than intermediate-risk cases. The corresponding survival curves are shown in Fig 3a. Similar findings could be obtained for the HU-treated cohort (Fig 3b). According to the Kantarjian staging system,4 patients were classifiable into four stages. However, depending on our stringent exclusion criteria, patients were not included in this trial who presented with stage IV features on admission. Constellation of these parameters indicated an advanced stage of disease and thus transformation into an accelerated phase characterized by thrombocytopenia with an excess of peripheral blasts and an increase in basophils. When this scoring system was applied, a crossing of survival curves was obtained for the IFN group (Fig 3c), and none of the correspondingly derived survival curves revealed a clear-cut separation of survival patterns. Regarding chemotherapy with HU, only stage I patients were characterized by a more favorable prognosis (Fig 3c), and comparable results were found for BU and combination regimens. According to the new collaborate CML score, which was proposed by Hasford et al,3 a conspicuous disproportionate risk classification for our IFN group was calculable. With this scoring system, only 13 (9.7%) of the 133 patients were regarded as high risk. However, low- and intermediate-risk cases showed distinctive survival patterns, with a median survival difference of 28 months (P = .0324). Concerning patients with HU or BU chemotherapy or combination and cross-over regimens, no distinctive risk classification could be established.
Prognostic Effect of Bone Marrow Features Analysis of parameters evaluated by morphometry disclosed a strong effect on survival for several histologic features that were uniformly expressed in all therapeutic regimens. Compared with normal ranges of bone marrow fiber content (Fig 1g), an already borderline to minimal increase in reticulin at onset (Fig 1h) was significantly associated with a worsening of prognosis (mean survival difference over therapy groups, 27 months; P = .0298). Regarding the proportion of patients with mild to advanced fibrosis on admission (ie, > three-fold increase in fibers; Fig 1i), subsequent analysis revealed that the IFN group did not exhibit any difference in survival compared with the cohort receiving chemotherapy with HU and combination regimens. However, these cases showed overall significantly higher survival rates than the BU-treated cohort. Furthermore, patients with an excessive reduction of Ret40f+ nucleated erythropoietic precursors (Fig 1, c and f) revealed significantly shorter survival times (mean survival difference over therapy groups, 12 months; P = .0094). From the other morphologic variables under study, only an increase in megakaryocytes was associated with a worsening of prognosis in all therapy cohorts (mean survival difference over therapy groups, 15 months; P = .0514). The remaining histologic features failed to disclose a relevant and therapy-independent effect on survival. We additionally assessed the distribution of pretreatment adverse histologic features among the risk groups defined by already published scoring systems, in particular the Sokal formula5 and the new collaborate CML score.3 Approximately one third of IFN and HU low-risk patients presented with a marginal increase in fibers or a marked reduction of erythropoietic precursors in the bone marrow. However, patients presenting with a high-risk profile revealed a significantly higher incidence of adverse morphologic variables. Survival analysis disclosed that these findings were also correlated with differences in median survival times ranging up to 36 months. Thus, even in low-risk patients, consideration of unfavorable bone marrow features yielded a striking reduction of survival probability. Comparing the overall median survival on the basis of the clinical scores (Fig 3), our data strongly suggest that the disproportionate risk classification with overlapping of survival times is related to the failure to enter distinctive morphologic predictors into prognostic evaluations.
Generation of a New Prognostic Score
Validation of the New Prognostic Score When our new score was applied to the other three independent therapy cohorts, which were chosen as validation samples, the clear-cut separation into three risk profiles could be confirmed (Table 5). By following the decision tree, equally sized prognostic groups were distinguishable, with a proportion of low-risk cases ranging from 30.0% to 36.1% and corresponding amounts of intermediate- and high-risk patients. Especially in the IFN cohort, three risk profiles with distinctive survival patterns could be obtained. Regarding the observed, as well as age- and sex-adjusted, relative survival rates clearly to distinguish survival, curves with no overlapping were estimated (Fig 5, c and d). According to our score, survival times ranged from 98 months for the low-risk group to only 43 months for the high-risk group. These differences were established as statistically significant (Table 5). When analysis was restricted to the 50 patients who were included as part of a prospective, randomized, single-center (IFN) therapy study31,32 and patients receiving bone marrow transplant were excluded, this result was confirmed. Even in the historical group with BU therapy, the new score was able to establish different risk profiles, mainly with regard to low-risk patients. Similar findings were observed for the sample with varying therapeutic modalities and cross-over treatment (Table 5). Again, the prognostic efficiency of our CART score proved to be significantly better than the risk classification based on the Cox model. In particular, for the IFN group, a higher measure of SEP could be observed (SEPCART = 1.79 v SEPCOX = 1.67). Finally, we tested the performance of the new CART score against the recently proposed collaborate CML score,3,11 which was primarily adapted to IFN-treated patient samples. Compared with our score, ROC analysis revealed a significantly lower sensitivity and specificity of predicting 3- and 5-year survival. Furthermore, this worse performance could be confirmed by a defective measure of SEP (SEP = 1.40).
Risk of Blast Transformation Probabilities and risks of blast transformation are listed in Table 6 for the learning sample with HU therapy and the validation sample with IFN treatment. According to our new score (Fig 4), significant differences in the median duration of the chronic phase could be observed. These ranged from only 29 months for the high-risk group to 75 months in low-risk-classified patients under HU therapy. Similar differences could be calculated in the independent validation sample of IFN-treated patients (Table 6) and the other cohorts receiving BU therapy and combination regimens. These findings were further confirmed in a subsequent logistic regression analysis to assess relative risk of blast transformation within 3 years of diagnosis. Patients with low-risk status showed a significantly lower probability of developing an unstable phase of disease in this period, independently of underlying therapeutic regimen. With our new score, relative risks of blast transformation ranged from 0.1 for low-risk status to 1.1 for the high-risk profile in the cohorts with HU or IFN therapy (Table 6). Compared with the new score, the clinically based risk stratification of patients revealed a lower predictive sensitivity in this context. However, low-risk cases generally showed a longer interval from diagnosis to blastic transformation in all scores. With the Sokal score,5 this risk group had a median length of chronic phase of 66 months under IFN treatment and 58 months under chemotherapy with HU, respectively. However, the intermediate- and high-risk patients disclosed no significant difference in their hazard ratios (IFN therapy: intermediate risk, 35 months; high risk, 51 months; HU therapy: intermediate risk, 32 months; high risk, 49 months). Consequently, the relative risks of blast transformation within 3 years of diagnosis were not significantly different in the logistic regression analysis. Similar results were obtained for the Kantarjian staging system.4 Regarding the Hasford score,3 in particular for the IFN-treated group, a correlation between risk status and duration of chronic phase could be determined (low risk, 73 months; intermediate risk, 51 months; and high risk, 37 months), but these differences were not significant in the log-rank test statistic. Accordingly, relative risks of blast transformation within 3 years were not different among the three risk groups. In contrast, all scores were able to predict evolution of blastic crisis in the historical cohort of BU-treated patients with a significantly different length of stable phase and associated higher relative risks.
Prognostic scoring systems that reflect the biologic behavior and progression of chronic-phase Ph1+ CML are generally intended to translate clinical experience into more precise knowledge of outcome.53 With regard to optimal treatment strategies and bone marrow transplantation, pretreatment risk stratification is therefore highly desirable.10 In this context, the most important result of our investigation is the extraction and validation of a new synoptic risk classification system. The presented prognostic scheme is built on easy-to-assess morphologic and clinical baseline variables and is not explicitly restricted to a certain treatment regimen. Our findings are based on a clinicopathologic multicenter observational trial that includes a large series of unselected patients receiving IFN and chemotherapy with HU or BU. In the concert of prognostic studies for chronic-phase Ph1+ CML, stimulating efforts regarding independent pretreatment disease features and therapy-related indicators predictive for differences in prognosis were frequently published.2-5,7,8,14,28,54 Several multivariate risk models have been proposed that categorize patients into different risk groups.2,4,5,8,28 Among these, the Sokal index5 and the Kantarjian staging system4 have gained the greatest acceptance. These risk scores were exclusively based on clinical or hematologic findings on admission. Older age, enlarged spleen, increased platelet counts, and high amounts of peripheral myeloblasts and basophils usually emerged as the most relevant predictors for survival.3-5,8,28 However, leading clinical trials repeatedly produced conflicting evidence with respect to the generalizability of these risk models.3,4,10 Especially for IFN-treated patients, a significant lack of prognostic efficiency was observed.3,11 This deficiency is confirmed by our results, because we were not able to obtain a clear-cut separation into distinctive risk groups for our unselected material (Fig 3, a to d). Taking this shortcoming into account, the proposed collaborative new CML score3 was reported to discriminate clearly defined risk patterns for IFN, even for patients treated by chemotherapy.11 According to our data and results from a recently published smaller series,32 this risk index revealed an unequal distribution between risk groups, with a low incidence of high-risk patients. The latter finding is in general agreement with the original data reported by Hasford et al,3 which showed a proportion of high-risk cases of approximately 12% to 14%. Because patients presenting with severe thrombocytopenia and excess of basophils as hallmarks of an advanced disease process30 were at least partially included in the original analysis, it has to be speculated whether this cohort represents cases that were not in the early chronic phase of disease. Furthermore, overlapping cases for which the complete set of stringent criteria indicating advanced stages was not available were not explicitly excluded from this series. When applying this score to our group of patients receiving chemotherapy, we found similar restrictions (Fig 3f) as for the Sokal index5 and Kantarjian staging system.4 However, in the cohort receiving IFN therapy, low- and intermediate-risk profiles were assigned to significantly different survival patterns (Fig 3e), which is in line with a recently published comparative evaluation of this staging system.11 With regard to risk of blast transformation, our new score further proved to exert a significantly higher predictive sensitivity than the other tested clinically based scores.3-5 In particular, patients who were classified as intermediate or high risk according to their clinical risk status revealed no difference in their hazards ratios for evolution of blast transformation. However, low-risk profile was generally associated with a less progressive course of disease, ie, a delayed occurrence of blastic crisis. Until now, prognostic evaluation of morphologic features has gained little attention, and development of staging systems was generally focused on hematologic parameters.4,7,55 Moreover, considerable controversy still persists regarding whether and to what extent bone marrow morphology is able to yield any improvement concerning pretreatment risk status.14,15 However, several lines of evidence suggest that histologic characteristics are important predictors for progression in chronic-phase Ph1+ CML.8,13,16,20 Because the biologic source of this disease is the neoplastic hematopoiesis, it might be postulated that deviations of the cellular and stroma components of the bone marrow are primarily responsible for the biologic behavior and progression of disease. Therefore, assessment of hematologic parameters reflects only secondary changes indicative for a gross expansion or generalization of the neoplastic cell clone. With this rationale, the new prognostic score extracted in this study is based on morphologic and hematologic features. Three easy-to-evaluate pretreatment variables were found to be the most important predictors for chronic-phase Ph1+ CML. According to our score, survival depends mainly on a decrease in medullary erythroid precursor cells, bone marrow fibrosis, and splenomegaly. In the process of our analysis, a decision tree was constructed and thoroughly validated in independent samples receiving different treatment modalities. This also includes a cohort of patients with IFN therapy derived from a randomized clinical trial.31,32 Compared with low-risk patients, intermediate- and high-risk cases were characterized by significant shorter median (observed) survival times with corresponding 3- and 5-year relative (age- and sex-adjusted) survival rates (Table 5). For the IFN-treated cohort, median survival ranged between only 43 months for high-risk profiles and 98 months for the low-risk cases. A similar discrimination was obtained for the group receiving chemotherapy with HU. It is noteworthy that, according to our score, low-risk patients did not obtain a significant benefit from IFN administration when this was compared with HU therapy (Table 5). This tendency of an only marginal survival difference between these two therapeutic regimens was already recognizable from our raw data (Table 3). With regard to IFN treatment, a median survival advantage of only 6 months was disclosed with failing separation of survival curves in the first 3 years of observation (Fig 2). However, with regard to relative survival, a 5-year probability of 57.2% and 45.6% could be calculated for both groups. These results are in line with survival data provided by ongoing prospective randomized studies.56-62 On balance, the accumulated evidence from these studies suggests that, opposed to HU or BU, IFN improves survival because of significantly higher hematologic and cytogenetic response rates in early chronic-phase patients with favorable hematologic features.53 Concerning intermediate- and high-risk cases, our results support the data of these study groups, because these patients did not profit from IFN administration. In keeping with these findings, adverse effects of IFN treatment have to be considered, because HU is effective for almost all Ph1+ CML patients and is tolerated well.53 A marked reduction of erythropoiesis, accompanied by a predominant growth of neutrophil granulopoiesis, is a commonly exhibited feature in Ph1+ CML.27,63 Applying a monoclonal antibody directed against glycophorin C and morphometry, we found a significant reduction of the red cell lineage in comparison with the normal bone marrow in approximately one third of our patients. With regard to semiquantitative grading of corresponding bone marrow smears (Fig 1a) and trephines (Fig 1, b to f), this finding is compatible with a shift of the granulocyte to erythroid ratio in excess of 10:1 to 12:1. However, opposed to erythroblasts and normoblasts of the peripheral blood,4,8,9,28,29 medullary erythroid precursor cells have been rarely studied in Ph1+ CML with respect to their predictive value.13,27 In extension of these preliminary data, including only a small series, a remarkable association with survival was confirmed in this multicenter trial. Our data imply that the amount of erythroid cells probably reflects the expansion of the Ph1+ granulocytic cell mass and, thus, the progression of disease. In this context, the naphthol-AS-D-chloroacetate esterase stain (Leder-stain; Fig 1d) is very helpful for assessment of the granulopoietic and erythropoietic cell lineages in bone marrow trephines and consequently is highly recommended as a routine method.33 Of the histomorphologic variables with questionable prognostic effect, myelofibrosis was most frequently assigned to an unfavorable outcome.8,17,18 In confirmation of other studies, approximately 27% of our patients presented with an early reticulin to advanced collagen fibrosis at diagnosis of Ph1+ CML.64 Survival analysis disclosed a significant worsening of prognosis for these patients, independent of therapeutic regimens. However, with regard to semiquantitative gradings,19 morphometric evaluation revealed that already a borderline to marginal increase in reticulin fibers, compatible with a doubling of the normal value (score 1+, Fig 1h), is indicative of a poor prognosis. Because cases with manifest myelofibrosis (score 2+, Fig 1i) disclosed no significant difference in survival rates under HU or IFN treatment, the considerable inconvenience, costs, and side effects of the latter regimen have to be considered for these patients. In this context, recent studies reported a stimulating effect of IFN on myelofibrosis.20,65 These results are in support of data derived from idiopathic (primary) myelofibrosis that showed no relevant improvement of bone marrow fibrosis by single-agent IFN therapy.66 With regard to allogeneic bone marrow transplantation, these findings are of particular interest, because recently published data provide confirmative evidence that pretransplant myelofibrosis might be responsible for a delayed engraftment and consequently unfavorable outcome.21,22 Similar to these findings, quantification of erythroiesis before bone marrow transplantation has proved to exert a predictive value regarding hematopoietic reconstitution, and thus outcome, in CML patients.67 Conflicting statements were also published regarding the prognostic effect of megakaryocytes.8 However, megakaryocyte-rich subtypes of Ph1+ CML are generally considered to herald transition into myelofibrosis and thus indicate an unfavorable course.68 Further prognostic implications have been attributed to the macrophage population, including their subsets in the bone marrow. In particular, the fraction of pseudo-Gaucher and the so-called sea-blue histiocytes was reported to indicate a more favorable survival.8,23 However, these data were based on historical patient cohorts mostly receiving chemotherapy with BU. In keeping with a smaller series of patients with IFN monotherapy, occurrence of these cells did not exhibit any significant effect on survival in our trial.13 Moreover, in confirmation with former studies, statistical analysis also revealed no influence on prognosis for the iron-laden subset of macrophages.69 As has been already pointed out, our findings emphasize the important correlation of subsequent hematologic variables and the underlying bone marrow pattern, which cannot be assessed by the widely accepted clinical scores.3-5 The extent of myelofibrosis and amount of erythroid precursors are in keeping with a more advanced stage of the disease process and therefore are related to an extramedullary expansion consistent with transition into myeloid metaplasia. In this context, enlargement of spleen size accompanied by an occurrence of myeloid and erythroid precursors in the peripheral blood has to be regarded as an important clinical parameter. In conclusion, the new risk model, extracted and thoroughly validated in this multicenter observational clinicopathologic study, strongly implies that prognostic classification of stable-phase Ph1+ CML, including evolution of blast transformation, can be significantly improved by regarding morphologic parameters. The variables of the presented scoring system may be readily assessed by routinely processed aspirates (nucleated erythroid precursors) and bone marrow trephines (fiber content). Because of an easy-to-follow decision tree, no complicated risk formula has to be calculated. Furthermore, the score has demonstrated a clear-cut discrimination of prognosis in representative patient samples with IFN and chemotherapy. Therefore, compelling evidence has been produced that an elaborate quantification of pretreatment risk profile should always be accompanied by a bone marrow biopsy to determine morphologic features, such as myelofibrosis, that cannot be evaluated in aspirates.
We are greatly indebted to B. Rosenbach, M. Wonschick, and G. Simons for their excellent technical assistance.
1. Cortes J, Kantarjian HM, Giralt S, et al: Natural history and staging of chronic myelogenous leukaemia. Baillières Clin Haematol 10: 277-290, 1997[Medline]
2.
Cervantes F, Rozman C: A multivariate analysis of prognostic factors in chronic myeloid leukemia. Blood 60: 1298-1304, 1982
3.
Hasford J, Pfirrmann M, Hehlmann R, et al: A new prognostic score for survival of patients with chronic myeloid leukemia treated with interferon alfa: Writing Committee for the Collaborative CML Prognostic Factors Project Group. J Natl Cancer Inst 90: 850-858, 1998 4. Kantarjian HM, Keating MJ, Smith TL, et al: Proposal for a simple synthesis prognostic staging system in chronic myelogenous leukemia. Am J Med 88: 1-8, 1990[Medline]
5.
Sokal JE, Cox EB, Baccarani M, et al: Prognostic discrimination in "good-risk" chronic granulocytic leukemia. Blood 63: 789-799, 1984
6.
Sokal JE, Baccarani M, Tura S, et al: Prognostic discrimination among younger patients with chronic granulocytic leukemia: Relevance to bone marrow transplantation. Blood 66: 1352-1357, 1985 7. The Italian Cooperative Study Group on Chronic Myeloid Leukaemia: Confirmation and improvement of Sokal’s prognostic classification of Ph+ chronic myeloid leukemia: The value of early evaluation of the course of the disease. Ann Hematol 63: 307-314, 1991[Medline] 8. Thiele J, Kvasnicka HM, Titius BR, et al: Histological features of prognostic significance in CML: An immunohistochemical and morphometric study (multivariate regression analysis) on trephine biopsies of the bone marrow. Ann Hematol 66: 291-302, 1993[Medline] 9. Hehlmann R, Ansari H, Hasford J, et al: Comparative analysis of the impact of risk profile and of drug therapy on survival in CML using Sokal’s index and a new score: German Chronic Myeloid Leukaemia (CML) Study Group. Br J Haematol 97: 76-85, 1997[Medline] 10. Hasford J, Ansari H, Pfirrmann M, et al: Analysis and validation of prognostic factors for CML: German CML Study Group. Bone Marrow Transplant 17: S49-S54, 1996 11. Bonifazi F, De Vivo A, Rosti G, et al: Testing Sokal’s and the new prognostic score for chronic myeloid leukaemia treated with alpha-interferon. Br J Haematol 111: 587-595, 2000[Medline]
12.
Kantarjian HM, Smith TL, O’Brien S, et al: Prolonged survival in chronic myelogenous leukemia after cytogenetic response to interferon-alpha therapy: The Leukemia Service. Ann Intern Med 122: 254-261, 1995 13. Thiele J, Kvasnicka HM, Niederle N, et al: Clinical and histological features retain their prognostic impact under interferon therapy of CML: A pilot study. Am J Hematol 50: 30-39, 1995[Medline] 14. Cervantes F, Rozman C, Feliu E: Prognostic evaluation of initial bone marrow histopathological features in chronic granulocytic leukemia. Acta Haematol 82: 12-15, 1989[Medline] 15. Rozman C, Cervantes F, Feliu E: Is the histological classification of chronic granulocytic leukaemia justified from the clinical point of view? Eur J Haematol 42: 150-154, 1989[Medline] 16. Thiele J, Kvasnicka HM, Schmitt-Graeff A, et al: Bone marrow features and clinical findings in chronic myeloid leukemia: A comparative, multicenter, immunohistological and morphometric study on 614 patients. Leuk Lymphoma 36: 295-308, 2000[Medline] 17. Dekmezian R, Kantarjian HM, Keating MJ, et al: The relevance of reticulin stain-measured fibrosis at diagnosis in chronic myelogenous leukemia. Cancer 59: 1739-1743, 1987[Medline] 18. Lazzarino M, Morra E, Castello A, et al: Myelofibrosis in chronic granulocytic leukaemia: Clinicopathologic correlations and prognostic significance. Br J Haematol 64: 227-240, 1986[Medline] 19. Bauermeister DE: Quantitation of bone marrow reticulin: A normal range. Am J Clin Pathol 56: 24-31, 1971[Medline] 20. Thiele J, Kvasnicka HM, Schmitt-Graeff A, et al: Effects of interferon and hydroxyurea on bone marrow fibrosis in chronic myelogenous leukaemia: A comparative retrospective multicentre histological and clinical study. Br J Haematol 108: 64-71, 2000[Medline] 21. Thiele J, Kvasnicka HM, Beelen DW, et al: Relevance and dynamics of myelofibrosis regarding hematopoietic reconstitution after allogeneic bone marrow transplantation in chronic myelogenous leukemia: A single center experience on 160 patients. Bone Marrow Transplant 26: 275-281, 2000[Medline]
22.
Rajantie J, Sale GE, Deeg HJ, et al: Adverse effect of severe marrow fibrosis on hematologic recovery after chemoradiotherapy and allogeneic bone marrow transplantation. Blood 67: 1693-1697, 1986
23.
Kelsey PR, Geary CG: Sea-blue histiocytes and Gaucher cells in bone marrow of patients with chronic myeloid leukaemia. J Clin Pathol 41: 960-962, 1988 24. Cervantes F, Rozman C, Brugues R, et al: Iron stores in chronic granulocytic leukaemia at presentation. Scand J Haematol 32: 469-474, 1984[Medline] 25. Welborn JL, Lewis JP: Correlation of marrow iron patterns with disease status of chronic myelogenous leukemia. Leuk Lymphoma 10: 469-475, 1993[Medline] 26. Buesche G, Majewski H, Schlue J, et al: Frequency of pseudo-Gaucher cells in diagnostic bone marrow biopsies from patients with Ph-positive chronic myeloid leukaemia. Virchows Arch 430: 139-148, 1997[Medline] 27. Thiele J, Hoefer M, Kvasnicka HM, et al: Erythropoiesis in CML: Immunomorphometric quantification, PCNA-reactivity, and influence on survival. Hematol Pathol 7: 239-249, 1993[Medline]
28.
Kantarjian HM, Smith TL, McCredie KB, et al: Chronic myelogenous leukemia: A multivariate analysis of the associations of patient characteristics and therapy with survival. Blood 66: 1326-1335, 1985 29. Levis A, Marmont F, Ciocca Vasino MA, et al: Peripheral erythroid precursor and prognosis in chronic granulocytic leukemia. Cancer 58: 229-233, 1986[Medline] 30. Kantarjian HM, Dixon D, Keating MJ, et al: Characteristics of accelerated disease in chronic myelogenous leukemia. Cancer 61: 1441-1446, 1988[Medline] 31. Niederle N, Kloke O, Wandl UB, et al: Long-term treatment of chronic myelogenous leukemia with different interferons: Results from three studies. Leuk Lymphoma 9: 111-119, 1993[Medline] 32. Kloke O, Opalka B, Niederle N: Interferon alfa as primary treatment of chronic myeloid leukemia: Long-term follow-up of 71 patients observed in a single center. Leukemia 14: 389-392, 2000[Medline] 33. Schaefer H: How to fix, decalcify and stain paraffin embedded bone marrow biopsies, in Lennert K, Hübner K (eds): Pathology of the Bone Marrow. Stuttgart, Germany: Fischer, 1984, pp 6-9 34. Thiele J, Wickenhauser C, Neuwirth C, et al: Effect of IFN-alpha on normal human hematopoiesis: An immunohistochemical and morphometric study on trephine biopsy specimens. J Interferon Cytokine Res 18: 247-253, 1998[Medline] 35. Gatter KC, Cordell JL, Turley H, et al: The immunohistological detection of platelets, megakaryocytes and thrombi in routinely processed specimens. Histopathology 13: 257-267, 1988[Medline] 36. Falini B, Flenghi L, Pileri S, et al: PG-M1: A new monoclonal antibody directed against a fixative-resistant epitope on the macrophage-restricted form of the CD68 molecule. Am J Pathol 142: 1359-1372, 1993[Abstract]
|
TOP
ABSTRACT
INTRODUCTION
2 statistic. For the investigation of alternative models, highly correlated covariates were exchanged. The thus-derived prognostic model was used to calculate hazards ratios for each patient and to divide the population into three subgroups with different risk status. Observed survival probabilities for each risk group were plotted and tested with the log-rank test for homogeneity.
