The present study aimed to directly compare the efficacy and safety of azacitidine and decitabine in patients with myelodysplastic syndromes (MDS). We compared the overall response rate (ORR) (complete responses, partial responses, marrow complete responses, and haematological improvements), overall survival (OS), event-free survival (EFS), time to leukaemic transformation, and adverse outcomes between azacitidine and decitabine. To minimize the effects of treatment selection bias in this observational study, adjustments were made using the propensity-score matching method. Among 300 patients, 203 were treated with azacitidine and 97 with decitabine. Propensity-score matching yielded 97 patient pairs. In the propensity-matched cohort, there were no significant differences between the azacitidine and decitabine groups regarding ORR (44% vs. 52%), OS (26 vs. 22·9 months), EFS (7·7 vs. 7·0 months), and rate of leukaemic transformation (16% vs. 22% at 1 year). In patients ≥65 years of age, survival was significantly better in the azacitidine group (P = 0·017). Patients who received decitabine experienced more frequent episodes of grade 3 or 4 cytopenia and infectious episodes. We found that azacitidine and decitabine showed comparable efficacy. Among patients ≥65 years of age, survival was significantly better in the azacitidine group (ClinicalTrials.gov Identifier: NCT01409070).
Myelodysplastic syndromes (MDS) are clonal disorders of haematopoietic stem cells characterized by ineffective haematopoiesis in the bone marrow; these disorders lead to peripheral cytopenia and are susceptible to leukaemic transformation (Tefferi & Vardiman, 2009). MDS probably arise from genetically transformed haematopoietic stem cells in which the subsequent aberrant DNA hypermethylation is believed to silence tumour-suppressor genes, contributing both to clonal variation and to the aforementioned tendency for leukaemic transformation (Jiang et al, 2009; Tefferi & Vardiman, 2009). The application of epigenetic modifiers that reverse hypermethylation and allow the re-expression of the silenced tumour-suppressor gene might therefore be a rational strategy in the treatment of MDS.
Two hypomethylating agents (HMAs), azacitidine and decitabine, were approved by the US Food and Drug Administration for the treatment of MDS in 2004 and 2006, respectively (Kaminskas et al, 2005; http://www.cancer.gov/cancertopics/druginfo/fda-decitabine). However, several phase III trials comparing azacitidine or decitabine with conventional treatment including best supportive care have shown discrepant efficacy in the 2 medications (Silverman et al, 2002; Kantarjian et al, 2006; Fenaux et al, 2009; Lubbert et al, 2011). The AZA-001 study in high-risk MDS patients reported that patients treated with azacitidine achieved a survival benefit of 9·4 months over patients given conventional treatment (Fenaux et al, 2009). In contrast, 2 phase III trials, including the European Organization for Research and Treatment of Cancer (EORTC) trial, compared decitabine with best supportive care and failed to demonstrate a survival advantage for this medication (Kantarjian et al, 2006; Lubbert et al, 2011). Subsequent meta-analysis comparing both HMAs with conventional treatment showed a significant overall survival benefit only with azacitidine (Gurion et al, 2010; Kumar et al, 2010).
The inferior outcome of decitabine should be interpreted with caution, however. This result may come from the differences in the number of delivered HMA cycles between studies, the high heterogeneity of the study populations, and/or confusion resulting from different studies defining different study end-points (Silverman et al, 2002; Kantarjian et al, 2006; Fenaux et al, 2009; Lubbert et al, 2011).
Given the absence of a prospective, randomized, head-to-head study, most haematologists choose a HMA according to their own experience and other experts' opinions. The present study aimed to directly compare the efficacy and safety of azacitidine and decitabine in patients with MDS.
The Korea MDS registry, sponsored by the Korean Society of Haematology, holds data from major centres in Korea that diagnosed and treated patients with MDS between January 2004 and December 2011. We conducted a retrospective review of these data, including patients 18 years of age and older who had received either azacitidine or decitabine for MDS. For study inclusion, patients needed to have an International Prognostic Scoring System (IPSS) lower risk score (IPSS low or intermediate-1) with significant cytopenia, or a higher risk score (IPSS intermediate-2 or high) (Greenberg et al, 1997). To reflect recent changes in diagnostic criteria, we excluded patients who had refractory anaemia with excess blasts in transformation, and those with chronic myelomonocytic leukaemia by French-American-British (FAB) criteria (Bennett et al, 1982). Patients who were classified according to FAB criteria or the 2001 World Health Organization (WHO) criteria were reclassified by independent pathologists using the 2008 WHO criteria (Vardiman et al, 2002, 2009). The IPSS included a classification of karyotype findings according to their prognostic impact. Patients with normal marrow karyotype, del(5q) alone, del(20q) alone, and –Y alone were in the good karyotype group. Patients with complex abnormalities, constituting 3 or more chromosomal anomalies, or chromosome 7 abnormalities were in the poor karyotype group. The remaining patients were classified in the intermediate karyotype group. This study was approved by the individual institutional review boards of all participating institutions and was carried out according to the principles of the Declaration of Helsinki. This trial is registered at www.clinicaltrials.gov as NCT01409070.
Subcutaneous azacitidine at 75 mg/m2 was given for 7 consecutive days at 4-week intervals. Intravenous decitabine at 20 mg/m2 was given over 1 h, for 5 consecutive days, at 4-week intervals. Dose reductions or treatment delays were implemented at the judgment of the treating physicians. HMA treatment was continued until the patient experienced disease relapse, unacceptable medication toxicity, disease progression, or death. Bone marrow examination was performed after the initial 2–4 HMA courses and was repeated when either further clinical improvement or disease progression was noted. Prophylactic antimicrobials were given at the discretion of the treating physicians.
Assessment of efficacy and safety
All patients who completed at least 1 cycle of HMA were included in the efficacy and safety analyses. The efficacy end points were overall response rate (ORR), overall survival (OS), event-free survival (EFS), and time to acute myeloid leukaemia (AML) transformation. The ORR included rates for complete response (CR), marrow complete response (mCR), partial response (PR), and haematological improvement (HI). Treatment response was assessed with the 2006 modified International Working Group response criteria (Cheson et al, 2006). OS was defined as the time from initiation of HMA treatment to death from any cause, and EFS was defined as the time from medication initiation to treatment failure, death from any cause, or to the date of allogeneic stem cell transplantation (ASCT). The time to AML transformation was measured from initiation of HMA to development of 20% or greater bone-marrow blasts. The safety end points comprised the number of infectious episodes requiring intravenous antimicrobials and the occurrence of adverse events. Infections requiring intravenous antimicrobials were counted from the initiation of HMA treatment to the last date of the final course, and calculated per 100 treatment cycles. Adverse events were assessed with the Common Toxicity Criteria of the National Cancer Institute, version 3.0(http://ctep.cancer.gov/protocolDevelopment/electronic_applications/docs/ctcaev3.pdf).
Clinical, pathological, and outcome data were collected with the use of a dedicated internet-based reporting system. All outcomes of interest were confirmed at each individual hospital and then adjudicated by a central reviewer (Y-G. L.) at Seoul National University Hospital.
Categorical variables were compared using of the chi squared test or Fisher's exact test, and continuous variables were compared using Student's unpaired t-test. Time-to-event outcomes were estimated using the Kaplan–Meier method and compared by log-rank tests. The Cox proportional-hazards regression model was used to estimate hazard ratios (HR) and associated 95% confidence intervals (95% CI). Interaction tests were used for variables that may have affected the impact of each HMA in multivariate analysis. Proportional hazard assumptions were tested statistically and graphically.
To minimize the effects of treatment selection bias and observed confounding bias in this observational study, adjustments were made using the propensity score matching method for the following baseline characteristics: sex; age at HMA initiation; disease aetiology; WHO classification; bone marrow blast percentage; karyotype risk; IPSS score; and WHO classification-based Prognostic Scoring System (WPSS) (D'Agostino, 1998). To develop propensity score-matched pairs for azacitidine and decitabine, the nearestneighbour-matching algorithm was used without replacement, giving a 1:1 match (Leuven & Sianesi, 2010).
After propensity score matching was performed, we compared the pretreatment characteristics between the azacitidine and decitabine groups. Categorical variables were compared with McNemar's test, and continuous variables were compared with the paired t-test. Stratified log-rank tests were used to compare time-to-event outcomes between the groups. To compare the HR of outcomes between groups, Cox proportional-hazards regression models were used with robust standard errors that accounted for the clustering of matched pairs. All reported P-values were 2-sided, and P-values of <0·05 were considered to indicate statistical significance. All analyses were performed using Stata 12.0 software (Stata Corp LP, College Station, TX, USA).
Between January 2004 and December 2011, a total of 300 patients at nine centres in Korea met the criteria for inclusion. Of these, 203 patients were treated with azacitidine and 97 with decitabine, according to the preference of their physicians. The median age at treatment initiation was 64 years (range, 18–83 years) in the azacitidine group and 65 years (range, 28–85 years) in the decitabine group. The baseline characteristics were well balanced between the groups, with the exception of the karyotype risk (Table 1). Propensity score matching yielded 97 patient pairs; there were no significant differences in variables between groups in this cohort of 194 patients (Table 1).
Table 1. Baseline characteristics in the overall and the propensity-matched cohorta
Data are median (range) or number (%), RCUD, refractory cytopenia with unilineage dysplasia; RARS, refractory anaemia with ring sideroblasts; RCMD, refractory cytopenia with multilineage dysplasia; RAEB, refractory anaemia with excess blasts; MDS-U, myelodysplastic syndrome, unclassified; IPSS, International prognostic scoring system; WPSS, WHO-based prognostic scoring system.
* MDS denotes myelodysplastic syndrome.
P-values are based on the paired t-test for continuous variables and on the McNemar's test or marginal homogeneity test for categorical variables.
MDS with del(5q)
Marrow blast (%)
IPSS risk category
WPSS risk category
Duration of MDS
Comparison of treatments
Both azacitidine and decitabine were given for a median of 5 cycles (interquartile range, 4 to 8 cycles). In the azacitidine group, 35 patients (17%) received subsequent ASCT and 14 patients (8%) received decitabine treatment following the conclusion of azacitidine therapy. In the decitabine group, 16 patients (16%) received subsequent ASCT but no patients received azacitidine following the conclusion of decitabine therapy. Oral antimicrobialand antifungal prophylaxis was used in 32 patients (16%) in the azacitidine group and 75 patients (77%) in the decitabine group.
Comparison of outcomes
A response to treatment was seen in 93 (46%) of 203 patients in the azacitidine group: 23 CR (11%); 13 PR (6%); 23 mCR (11%); and 34 HI (17%). In the decitabine group, 50 (52%) of 97 patients showed a treatment response: 9 CR (9%); 4 PR (4%); 23 mCR (24%); and 14 HI (14%) (Table 2). There was no significant difference in the ORR between the groups (P = 0·35). The proportion of patients with mCR was significantly higher in the decitabine group than in the azacitidine group (P = 0·006). The median time to best response was 4·2 months for patients in the azacitidine group and 4·0 months for patients in the decitabine group. In the propensity score-matched cohort, the ORR in the azacitidine group was not significantly different than that in the decitabine group (P = 0·30) (Table 2). The response of unmatched patients is presented in Table SI.
Table 2. Treatment response by hypomethylating agents
P-values are based on the McNemar test for categorical variables.
ORR (CR + PR + mCR + HI)
After a median follow-up period of 29·6 months, the median OS was 23·2 months in the azacitidine group and 22·9 months in the decitabine group (Fig 1A); there was no significant difference (HR 1·01;95% CI 0·71–1·43; P = 0·97). The median EFS was 8·4 months in the azacitidine group versus 7·0 months in the decitabine group (HR 0·98; 95% CI 0·74–1·29; P = 0·87) (Fig 1C). The cumulative incidence of AML at 1 year was similar between groups (21% vs. 22%; P = 0·70) (Fig 1E, Table 3). In the cohort of 97 propensity score-matched pairs, there was no significant difference in OS between the azacitidine and decitabine groups (26·0 vs. 22·9 months; HR 1·36; 95% CI 0·94–1·97, P = 0·10) (Fig 1B). Likewise, there was no difference in EFS (7·7 vs. 7·0 months; HR 0·98; 95% CI 0·71–1·33, P = 0·87) (Fig 1D) or cumulative incidence of AML (16% vs. 22% at 1 year; P = 0·34) (Fig 1F, Table 3). In the full propensity score adjusted analyses with inclusion of unmatched individuals, there were no significant differences between the azacitidine and decitabine groups regarding OS (adjusted HR 0·97; 95% CI 0·68–1·38; P = 0·85), EFS (adjusted HR 0·99; 95% CI 0·75–1·32; P = 0·96) and rate of leukaemic transformation (adjusted HR 1·10; 95% CI 0·65–1·85; P = 0·73).
Table 3. Hazard ratios for clinical outcomes in the overall and the propensity-matched cohort
Data are median (interquartile range) or percentage. Hazard ratios calculated with Cox proportional hazards model.
HR, hazard ratio. NR, not reached.
P-values are based on weighted Cox proportional hazards regression analysis with robust standard errors.
Overall survival (months)
Event-free survival (months)
1-year cumulative incidence of AML (%)
In the univariate analyses, shorter survival was associated with age ≥ 65 years at HMA initiation (vs. < 65 years) (HR 1·78; 95% CI 1·30–2·46; P < 0·001), poor karyotype (vs. good karyotype) (HR 2·31; 95% CI 1·59–3·35; P < 0·001), IPSS higher-risk status (vs. IPSS lower-risk) (HR 1·89; 95% CI 1·38–2·60; P < 0·001), and increasing WPSS risk score (from one unit to the next) (common HR 1·67; 95% CI 1·41–1·97; P < 0·001). In multivariate analyses, the effects of HMAs on survival were modified by age group. Therefore, we calculated stratum-specific HR by performing stratified analyses for each age group (Table SII). Increasing WPSS risk category was the only independent prognostic factor for shorter OS both in patients < 65 years of age (common HR 1·74; 95% CI 1·35–2·24; P < 0·001) and patients ≥ 65 years of age (common HR 1·55; 95% CI 1·25–1·94; P < 0·001). With regard to the effects of HMAs, decitabine showed a trend of better survival than azacitidine in patients < 65 years of age (HR 0·63; 95% CI 0·35–1·14; P = 0·13) and a trend of worse survival than azacitidine in patients ≥ 65 years of age (HR 1·21; 95% CI 0·76–1·90; P = 0·42), however, these were not statistically significant.
In the propensity score-matched cohort, we also stratified analyses according to age at HMA initiation. In patients <65 years of age, survival was not statistically different between the azacitidine and decitabine groups (HR 1·17; 95% CI, 0·54–2·54; P = 0·70) (Fig 2A). When patients ≥ 65 years of age were analysed, survival was significantly better in the azacitidine group (HR 3; 95% CI 1·22–7·37; P = 0·017) (Fig 2B). In addition, the effect of azacitidine over decitabine on survival was marginally significant for patients with higher risk IPSS scores (24·9 vs. 15·3 months; HR 3·5; 95% CI 0·97–12·58; P = 0·055).
Table 4 shows the early mortality and safety profiles, broken down by treatment group, in both the overall and the propensity score-matched cohort. During the first 3 months of treatment, deaths occurred in 10 (5%) of the 203 patients in the azacitidine group and 8 (8%) of the 97 patients in the decitabine group. These deaths were attributable to sepsis after 1 or 2 cycles of HMA, with the exception of 1 patient with a life-threatening brain haemorrhage. Grade 3 or 4 neutropenia was more common in the decitabine group (87%) than in the azacitidine group (67%). The number of infectious episodes treated with intravenous antimicrobials was significantly lower in the azacitidine group (11·8 per 100 cycles) than in the decitabine group (15·7 per 100 cycles) (relative risk 0·75; 95% CI 0·58–0·96; P = 0·02). In the propensity score-matched cohort, the higher incidence of grade 3 or 4 cytopeniain the decitabine group was unchanged. Susceptibility to infection was shown to be more prominent in the decitabine group (relative risk 0·55; 95% CI 0·40–0·76; P < 0·001) (Table 4). The rates of infectious episodes were not significantly different according to the use of infection prophylaxis in both treatment groups.
Table 4. Deaths and safety profile in the overall and the propensity-matched cohort
Overall cohort (n = 300)
Propensity-matched cohort (n = 194)
Azacitidine (n = 203)
Decitabine (n = 97)
Azacitidine (n = 97)
Decitabine (n = 97)
Data are number (%).
National Cancer Institute's Common Toxicity Criteria based on laboratory data.
Infectious episodes requiring antimicrobial per 100 cycles (episodes/cycles)
The present study aimed to compare the treatment efficacy of azacitidine with that of decitabine, and to suggest the optimal therapeutic option for patients with MDS. To accomplish these aims, we compared data on patients who received HMA under the same criteria and treatment schedule, after using propensity score matching to make the treatment groups more similar. In both the overall and the propensity-matched cohorts, there were no significant differences between the azacitidine and decitabine groups regarding ORR, OS, EFS, or rate of leukaemic transformation; this means that both HMAs showed comparable treatment efficacy.
With regard to the safety profile, however, patients receiving decitabine experienced more frequent grade 3 or 4 cytopenia and more frequent infectious episodes requiring antimicrobials than patients receiving azacitidine in both the overall and the propensity-matched cohorts. Though about 77% of patients in the decitabine group received broad application of prophylactic antimicrobials, they appeared to be more vulnerable to infection than patients in the azacitidine group; this vulnerability became quite apparent in the more elderly study participants. If the inferior survival in elderly patients who received decitabine is a result of their high vulnerability to infection, we would suggest azacitidine as the preferred treatment option in these patients.
In the absence of a randomized comparison study, a recent meta-analysis attempted to compare the efficacy of the 2 HMAs (Gurion et al, 2010); survival benefits were shown for azacitidine but not for decitabine. The authors noted that reduced efficacy from a shorter administration period may account for the lack of survival improvement seen with decitabine. Another meta-analysis included the same randomized trials and performed similar analyses using a different method (Kumar et al, 2010), obtaining practically the same results. They also presented an indirect comparison of azacitidine versus decitabine, showing a significant survival benefit for azacitidine. In our study, patients who received either medication for the same timeframe of administration did not demonstrate a significant difference in survival. However, when we confined our target population to patients ≥ 65 years of age in the propensity-matched cohort, we found inferior survival with decitabine. As the patients represented in the meta-analysis were older than those in our study (median age, 67–70 years vs. 64 years), our findings seem to be consistent with those of the meta-analyses.
In the subgroup analysis of patients in the AZA-001 trial who were ≥ 75 years of age, azacitidine was found to significantly improve survival; it also demonstrated good tolerability compared with the conventional care regimen (HR 0·48; 95% CI 0·26 to 0·89; P = 0·019) (Fenaux et al, 2009; Seymour et al, 2010). However, in the subgroup analysis of patients in the EORTC trial who received decitabine versus best supportive care, the benefit of decitabine seemed to disappear in patients ≥ 75 years of age (HR 1·29; 95% CI 0·64–2·59) (Lubbert et al, 2011). Taken together, these results indirectly suggest the advantage of azacitidine over decitabine in the elderly population and are in agreement with our results.
The major limitation of our study is that we evaluated observational, retrospective data, meaning that the patients were not randomized and were subject to treatment selection bias. As decitabine was approved in Korea 2 years later than azacitidine, the choice of HMA was influenced by the drug approval status as well as by the preferences of the treating physicians. Although the benefit of subsequent decitabine in patients pretreated with azacitidine is not established (Borthakur et al, 2008), this regimen was used in our study patients; this contamination may cause bias and reduce our statistical power.
Our findings are also limited by selection and confounding bias with respect to the pretreatment risk differences in patients who received the 2 medications. To minimize these inherent biases, we performed propensity score matching and adjusted for the possible confounders which were likely to have influenced HMA selection in each patient (D'Agostino, 1998; Leuven & Sianesi, 2010). Nevertheless, our methods cannot control for unmeasured potential confounders. Considering these issues and the results of our study, a randomized direct comparison study between azacitidine and decitabine in patients with MDS is warranted.
In conclusion, we found that azacitidine and decitabine showed comparable efficacy as demonstrated by the ORR, OS, EFS, and the rate of leukaemic transformation in MDS patients with IPSS lower risk status and significant cytopenia, as well as in those with higher risk IPSS status. Among propensity score-matched patients ≥ 65 years of age, survival was significantly better in the azacitidine group than in the decitabine group. Patients who received decitabine experienced more frequent episodes of grade 3 and 4 cytopenia and more frequent infectious episodes, giving azacitidine a better safety profile. When both azacitidine and decitabine are available, the safety profile as well as the treatment efficacy needs to be considered.
We are grateful to EUN JU KWAK (Clinical Research Coordinator) at AML/MDS Working Party in the Korean Society of Haematology for the data collection.
Y-G.L. and H-J.K. contributed to the study design, data collection/analysis/interpretation and writing the manuscript. I.K. treated many of the patients and contributed to the study design, data analysis/interpretation and writing of the manuscript. S-S. Y., S.P., J.W.C., Y.H.M., J-O.L., S-M.B., H.G.Y., C.S.K., Y.P., B-S.K., Y-C.M., C-M.S., J.P., J.H.L., S-Y.K., H.G.L. and Y-K.K. treated many of the patients in this study. All authors checked the final version of the manuscript.
The authors state that they have no conflict of interest.