Professional Documents
Culture Documents
Introduction
The myelodysplastic syndromes (MDS) are a heterogeneous group of disorders, mostly appearing in older people,
which are characterized by cytopenias, dysplastic hematopoiesis, and an inherent propensity to clonal evolution into
acute myeloid leukemia (AML) [1]. Most patients with MDS
have symptomatic anemia at presentation or develop it during the clinical course. Although recombinant erythropoietin
and, more recently, thalidomide and lenalidomide can ameliorate symptomatic anemia in some cases, frequent red
blood cell (RBC) transfusion remains cornerstone in the
treatment of many patients with MDS [2].
A history of heavy RBC transfusion has recently been
identied as a predictor for shorter survival, increased risk
of progression into AML, and poorer outcomes after allogeneic stem cell transplantation [36]. By extrapolating the
well known deleterious effects of transfusional hemosiderosis in patients with thalassemia major and other congenital
anemias, several authors have postulated iron overload as
the causative factor linking heavy transfusion burden to
poor outcomes in the MDS [5,79]. The association found
in several studies between high serum ferritin levels and
poor prognosis adds further weight to that purported causal
link [3,10,11], and has driven many experts to recommend
that MDS patients for whom chronic transfusion support is
likely should be started on iron chelation therapy [9,1215].
On the contrary, other authors are sceptical about that explanation and believe instead that heavy transfusion
requirements are merely a surrogate for more severe disease in patients with MDS [16,17].
Conicting hypotheses on the causal pathway between
heavy transfusion burden and prognosis in the MDS will
probably not be resolved in the short-term because of the
complexities involved in conducting the necessary clinical
studies. Meanwhile, statistical modeling of competing
hypotheses using available patient datasets can provide
some clues that may prove useful in the management of
transfusion-dependent patients with MDS.
245
http://wileyonlinelibrary.com/cgi-bin/jhome/35105
research article
factors that are more probable to be causally related with
shorter survival.
Methods
Study population and measures. Clinical, hematologic, and followup data of patients with MDS seen at the Hospital Clnic of Barcelona
have prospectively been recorded in an electronic database from the
mid 1990s onwards. We queried this database and the Transfusion
Service database in order to identify transfusion-dependent MDS
patients. Transfusion-dependency was dened according to the WPSS
criteria [4] as requiring at least one RBC transfusion every 8 weeks
over a period of 4 months. Patients who were seen for second opinion
consultation or at irregular intervals and whose transfusion management was carried out at other hospitals were excluded, as were those
transfused only in the context of bone marrow transplantation or after
transformation into AML. Patients seen before 2003 had been diagnosed following the FAB criteria and were reclassied according to the
2001 WHO classication [18]. Patients with MDS unclassied or with
20% blast in the bone marrow were excluded so that the following
categories were considered: refractory anemia (RA), RA with ringed
sideroblasts (RARS), refractory cytopenia with multilineage dysplasia
(RCMD), RCMD with ringed sideroblasts (RCMD-RS), RA with excess
of blasts (RAEB-1 and -2), and MDS with isolate del(5q). For the purpose of this study we also considered patients diagnosed with chronic
myelomonocytic leukemia (CMML). The IPSS criteria were used to
derive the karyotype-based risk classication [19].
Clinical and hematologic variables analyzed for prognostic signicance included sex, age at the time of rst transfusion, WHO category
(RA, RARS, del(5q) vs. RCMD, RCMD-RS vs. RAEB-1, RAEB-2,
CMML), the karyotype-based risk group, and the IPSS. Transfusionrelated variables included the cumulative transfusion burden (in number
of RBC units) and the transfusion intensity, both being measured at the
time of each transfusion episode. Transfusion intensity was calculated
using the following formula:
Number of RBC units transfused at the episode
3 365 days=Number of days from the previous transfusion episode:
It represents the number of RBC units the patient would have
received over a full year had he/she kept being transfused at the current rate over a complete year. For instance, a patient who receives 2
RBC units 40 days after the previous transfusion episode has a transfusion intensity of [2 units 3 365 days/40 days] 5 18.25 RBC units per
year. If he/she is transfused with 3 RBC units 30 days later, the transfusion intensity at that time would have increased to [3 units 3 365 days/
30 days] 5 36.5 RBC units per year.
End-points and statistical analysis. The primary end-point was
AML-free survival from the time of rst transfusion. To test the
study hypotheses, we conducted several nested Cox models in
which the covariate transfusion burden 25 RBC units was successively adjusted for transfusion intensity and the median follow-up
at the time patients had received 25 RBC units, both covariates
being analyzed as time-dependent. Initial patient features signicantly inuencing on prognosis were included in every model. The
underlying assumption was that transfusion burden would be an
spurious prognostic factor (mediated or confounded by other variables) if its association with survival was abrogated by the inclusion
of any of the above covariates in the regression, and the resulting
Cox model had signicantly better specication and goodness of t
statistics [20].
We cut-off the cumulative transfusion burden at 25 RBC units
because this is the threshold for pathologic iron deposition in the liver
as measured by T2* MRI [21] and the threshold at which many practice
guidelines recommend starting on iron chelation therapy [12.14,15]. In
the multi-state model, we set an additional cut-off at 100 RBC units
because it has previously been identied as the critical threshold for
increased risk of myocardial iron overload [21,22].
To gain further insight into the prognostic signicance of transfusiondependency, we built a multi-state model representing the most relevant events of the disease course and the patient transfusion history.
Transitions from one state to each of the next states were modeled as
competing-risks, taking the occurrence of AML/death as an absorbing
state. The effect of covariates on transition hazards was analyzed by
Cox proportional hazards regression adapted to the competing-risks
framework [23].
Survival curves were drawn by the Kaplan and Meier method.
Nested Cox models were evaluated for goodness of t and specication
246
At rst
transfusion
25
RBC units
191 (100%)
74 (1699)
112 (59%)
75 (1899)
100 RBC
units
29 (15%)
73 (2886)
97 (51%)/94
63 (56%)/49
17 (59%)/12
24 (13%)
91(48%)
54 (28%)
22 (11%)
14
53
32
13
8 (28%)
15 (51%)
6 (21%)
0
141 (85%)
25 (15%)
(12%)
(47%)
(29%)
(12%)
83 (84%)
16 (16%)
9.4 (169)
months
21 (84%)
4 (16%)
29 (1262)
months
At the time the surviving patients reached the cumulative transfusion burden.
Available in 166 patients.
by the likelihood ratio test and the Akaikes information criterion (AIC)
[24]. The AIC is a measure of the goodness of t of an estimated statistical model; lower AIC indicates better likelihood. In nested Cox models, the mediation or confounding effect on the association between the
cumulative transfusion burden and survival was measured by the percentage reduction of its regression coefcient after adjustment for the
mediating or confounding covariate [25]. All the statistical analyses
were done with Stata, version 10 (www.stata.com). Cox regression
models and cumulative incidence curves were calculated within the
competing-risks framework with the stcompadj module [26]. The timespan splitting method [27] was used to t the Cox models with time-dependent covariates. By using this method, the 191 patient records were
split at the time of each transfusion episode into a total of 5,306 observations. This study was approved by the Research Ethics Committee of
the Hospital Clinic of Barcelona, Spain.
Results
Patient characteristics and survival
Among the 466 patients recorded in the MDS database
up to December 31, 2009, 191 met the eligibility criteria
and were included in this study. Their main clinical and hematologic features at the time of rst transfusion are summarized in Table I. At the study closing date, median follow-up from rst transfusion was 1.5 (0.410.5) years, 128
(67%) patients had died, 8 (4%) were lost to follow-up, and
55 (29%) remained alive. Forty-six patients had progressed
to AML after a median of 8 (range: 453) months from the
rst transfusion. Median survival after progression to AML
was 2.5 months, with 44 (95%) patients having died within
the rst year following leukemic transformation.
Overall median survival from rst transfusion was 2 years
for the 191 patients, and the median AML-free survival was
1.96 years (Fig. 1). Male sex, WHO category, and high risk
karyotype were independent predictors for overall and
AML-free survival (Table S1, Fig. S1).
Transfusion data
A total of 5,306 transfusion episodes were recorded.
They involved 10,684 RBC units, 1,022 (9.5%) of which
were transfused after the recipient had progressed to AML.
Since prognosis after leukemic transformation was nearly
invariable fatal in the short term and presumably not
affected by transfusions given once the patient had progressed to AML, further analyses of the prognostic signicance of RBC transfusion were restricted to the AML-free
phase of MDS. The median number of RBC units transfused per patient before leukemic transformation was 30
research article
Figure 1. AML-free survival from rst transfusion in 191 MDS patients with transfusion dependency.
Figure 2. Transfusion intensity during the AML-free survival in 191 MDS patients
with transfusion dependency. According to follow-up from rst transfusion (A), year of
transfusion (median, quartiles, and 95% CI (B), and cumulative transfusion burden (C).
247
research article
tions. The WHO category and high-risk karyotype had its
main prognostic impact on the transition from state 0 (rst
transfusion) or state 3 (25 RBC units) to state 1 (AML or
TABLE II. Nested Cox Models Showing the Prognostic Signicance of
Transfusion Burden 25 RBC Units After Adjustment for Selected Covariates
Covariate
HR
Model A
AIC 5 842
Male sex
WHO type
Poor prognosis karyotype
Transfusion burden 25 RBC units
Model B
Male sex
WHO type
Poor prognosis karyotype
Transfusion burden 25 RBC units
Time from rst transfusion >9 months
Model C
Male sex
WHO type
Poor prognosis karyotype
Transfusion burden 25 RBC units
Transfusion intensity
Model D
Male sex
WHO type
Poor prognosis karyotype
Transfusion burden 25 RBC units
Time from rst transfusion 9 months
Transfusion intensity
1.6
2.5
3.4
2.7
AIC 5 824
1.6
3.1
3.7
2.3
10.8
AIC 5 831
1.4
2.6
2.5
1.5
1.4
AIC 5 819
1.4
2.9
2.9
1.5
8.4
1.3
95% CI
Log-likelyhood:
2417
1.12.4
1.83.5
2.05.9
1.74.3
0.011
<0.001
<0.001
<0.001
Log-likelyhood:
2407
1.12.4
2.24.4
2.16.5
1.43.7
2.546
0.02
<0.001
<0.001
<0.001
0.001
Log-likelyhood:
2410
0.92.1
1.83.6
1.44.5
0.82.6
1.21.7
0.10
<0.001
0.001
0.17
<0.001
Log-likelyhood:
2403
0.92.1
2.14.2
1.65.2
0.82.6
1.936
1.11.5
0.08
<0.001
<0.001
0.20
0.004
0.008
Figure 3. Multi-state model representing the disease progression and transfusion history in 191 MDS patients with transfusion dependency. Panel A shows the outline
of the model. Risks of transition from state S0 or state S2 are shown in panels B and C, respectively. Panel D represents the risk of transition from state S3.
248
research article
TABLE III. Factors Associated With Progression Along the Multi-State Model
From cumulative burden 25 RBC units to
AML or death
Covariate
HR (95% CI)
Male sex
WHO categorya
High risk karyotypeb
Transfusion intensity
1.1
2.8
3.6
1.5
a
b
(0.61.9)
(1.65.1)
(1.58.7)
(1.12.1)
P
0.13
<0.001
0.005
0.026
HR (95% CI)
1.0
1.1
2.4
3.5
(0.71.5)
(0.81.5)
(1.34.5)
(2.84.5)
AML or death
P
HR (95% CI)
HR (95% CI)
0.9
0.4
0.005
<0.001
1.9 (1.043.6)
2.1 (1.33.6)
2.9 (1.46.2)
1.8 (1.22.5)
0.04
0.004
0.005
0.001
2.7 (1.017.3)
0.6 (0.31.4)
4.4 (0.922)
2.3 (1.33.9)
0.05
0.23
0.07
0.002
249
research article
References
1. Tefferi A, Vardigan JW. Mechanisms of diseaseMyelodysplastic syndromes.
N Engl J Med 2009;361:18721885.
2. Stone RM. How I treat patients with myelodysplastic sndromes. Blood
2009;113:62966303.
3. Malcovati L, Della Porta MG, Pascutto C, et al. Prognostic factors and life expectancy in myelodysplastic syndromes classied according to WHO criteria:
A basis for clinical decision making. J Clin Oncol 2005;23:75947603.
4. Malcovati L, Germing U, Kuendgen A, et al. Time-dependent prognostic scoring system for predicting survival and leukemic evolution in myelodysplastic
sindromes. J Clin Oncol 2007;25:35033510.
5. Cermak J, Kacirkova P, Mikulenkova D, Michalova K. Impact of transfusion
dependency on survival in patients with early myelodysplastic syndrome without excess of blasts. Leuk Res 2009;33:14691474.
6. Alessandrino EP, Della Porta MG, Bacigalupo A, et al. Prognostic impact of
pre-transplantation transfusion history and secondary iron overload in patients
with myelodysplastic syndrome undergoing allogeneic stem cell transplantation: A study from the Gruppo Italiano Trapianto di Midollo Osseo (GITMO).
Haematologica 2010;95:476484.
7. Dreyfus F. The deleterious effects of iron overload in patients with myelodysplastic syndromes. Blood Rev 2008;22 (Suppl 2):S29S34.
8. Remacha AF, Arrizabalaga B, Del Canizo C et al. Iron overload and chelation
therapy in patients with low-risk myelodysplastic syndromes with transfusion
requirements. Ann Hematol 2010;89:147154.
9. Pullarkat V. Objectives of iron chelation therapy in myelodysplastic syndromes: More than meets the eye? Blood 2009;114:52515255.
10. Takatoku M, Uchiyama T, Okamoto S, et al. Retrospective nationwide survey
of Japanese patients with transfusion-dependent MDS and aplastic anemia
highlights the negative impact of iron overload on morbidity/mortality. Eur J
Haematol 2007;78:487494.
11. Armand P, Kim H, Cutler C, et al. Prognostic impact of elevated pretransplantation serum ferritin in patients undergoing myeloablative stem cell transplantation. Blood 2007;109:45864588.
12. Valent P, Horny H-P, Bennett JM, et al. Denitions and standards in the diagnosis and treatment of the myelodysplastic syndromes: Consensus statements and report from a working conference. Leuk Res 2007;31:727736.
13. Suzuki T, Tomonaga M, Miyazaki Y, et al. Japanese epidemiological survey
with consensus statement on Japanese guidelines for treatment of iron overload in bone marrow failure sndromes. Int J Hematol 2008;88:3035.
14. Mittelman M, Lugassy G, Merkel D, et al. Iron chelation therapy in patients
with myelodysplastic syndromes: Consensus conference guidelines. IMAJ
2008;10:374376.
250
15. Malcovati L. Red bood cell transfusion therapy and iron chelation in patients With
myelodysplastic syndromes. Clin Lymph Myeloma 2009;9 (Suppl 3):S305S311.
16. Steensma DP. Myelodysplasia paranoia: Iron as the new radon. Leuk Res
2009;33:11581163.
17. Tefferi A, Stone RM. Iron chelation therapy in myelodysplastic syndromeCui
bono? Leukemia 2009;23:1373.
18. Vardiman JW, Harris NL, Brunning RD. The World Health Organization
(WHO) classication of the myeloid neoplasms. Blood 2002;100:22922302.
19. Greenberg P, Cox C, LeBeau MM, et al. International scoring system for evaluating prognosis in myelodysplastic syndromes. Blood 1997;89:20792088.
20. MacKinnon DP, Krull JL, Lockwood CM. Equivalence of the mediation, confounding and suppression effect. Prev Sci 2000;1:173181.
21. Di Tucci AA, Matta G, Deplano S, et al. Myocardial iron overload assessment
by T2* magnetic resonance imaging in adult transfusion dependent patients
with acquired anemias. Hematologica 2008;93:13851388.
22. Buja LM, Roberts WC. Iron in the heart. Etiology and clinical signicance. Am
J Med 1971;51:209221.
23. Putter H, Fiocco M, Geskus RB. Tutorial in biostatistics: Competing risks and
multi-state models. Stat Med 2007;26:23892430.
24. Akaike H. A new look at the statistical model identication. IEEE Trans Autom
Control 1974;19:716723.
25. Judd CM, Kenny DA. Process Analysis: Estimating mediation in treatment
evaluations. Evaluation Rev 1981;5:602619.
26. Coviello VBN. Cumulative incidence estimation in the presence of competing
risks. Stata J 2004;4:103112.
27. Jann B. Splitting time-span records with categorical time-varying covariates.
Stata J 2004;4:221222.
28. Oliva EN, Dimitrov BD, Benedetto F et al. Hemoglobin level threshold for cardiac remodeling and quality of life in myelodysplastic syndrome. Leuk Res
2005;29:12171219.
29. Goldberg SL, Chen E, Corral M, et al. Incidence and clinical complications of
myelodysplastic syndromes among United States Medicare beneciaries. J
Clin Oncol 2010;28:28472852.
30. Malcovati L, Della Porta MG, Cazzola M. Predicting survival and leukemic evolution
in patients with myelodysplastic syndrome. Haematologica 2006;91:15881590.
31. Chee CE, Steensma DP, Wu W et al. Neither serum ferritin nor the number of
red blood cell transfusions affect overall survival in refractory anemia with
ringed sideroblasts. Am J Hematol 2008;83:611613.
32. Vamvakas EC, Blachman MA. Deleterious clinical effects of transfusion-associated immunomodulation: Fact or ction? Blood 2001;97:11801195.
33. Tinmouth A, Fergusson D, Yee IC, Hebert PC. Clinical consequences of red
cell storage in the critically ill. Transfusion 2006;46:20142027.