Predicting infections in high-risk patients with myelodysplastic syndrome/acute myeloid leukemia treated with azacitidine: Aretrospective multicenter study


  • Conflict of interest: Nothing to report

Correspondence to: Yishai Ofran, Deptartment of Hematology and Bone Marrow Transplantation, Rambam Health Care Campus, Haifa 31096, Israel. E-mail:


Hypomethylating agents have become the standard therapy for patients with high-risk myelodysplastic syndrome (MDS). In Israel, azacitidine (AZA) is routinely used. Yet, infectious complications are common during AZA therapy. The current study was aimed to evaluate the incidence and predisposing risk factors for infections in AZA-treated patients. This retrospective study included patients treated with AZA in 18 Israeli medical institutions between 2008 and 2011. Data on 184 patients [157 high-risk MDS and 27 acute myeloid leukemia (AML)], with a median age of 71.6 (range 29–92) were recorded. Overall, 153 infectious events were reported during 928 treatment cycles (16.5%) administered to 100 patients. One hundred fourteen, 114/153 (75%) events required hospitalization and 30 (19.6%) were fatal. In a univariate analysis, unfavorable cytogenetics, low neutrophil, hemoglobin (Hb) and platelet (PLT) counts were found to be associated with infections (24.4% vs. 12.9%, P < 0.0001; 27% vs. 13.5%, P < 0.0001; 20.4% vs. 11%, P < 0.0001 and 29.2% vs. 14.2%, P < 0.0001, respectively). In multivariate analysis, only low Hb level, low PLT count, and unfavorable cytogenetics remained significant. Prior to therapy, poor cytogenetics, PLT count below 20 × 109/L and neutrophil count below 0.5 × 109/L were predictive of the risk of infection during the first two cycles of therapy. In conclusion, patients with unfavorable cytogenetics, presenting with low neutrophil and PLT counts, are susceptible to infections. Evaluation of infection risk should be repeated prior to each cycle. Patients with poor cytogenetics in whom AZA is prescribed despite low PLT count are particularly at high risk for infections and infection prophylaxis may be considered. Am. J. Hematol. 88:130–134, 2013. © 2012 Wiley Periodicals, Inc.


Infection is a common potentially fatal event among patients with myelodysplastic syndrome (MDS) [1]. In particular, the risk of infection is elevated among high-risk MDS patients who often present with neutropenia [2-4]. Neutrophil count may further decrease following hypomethylation agents therapy [5, 6] which has become the standard of care for these patients. Indeed, the risk of infection during hypomethylation agent treatment was demonstrated to be substantial [7-9]. However, there are no established indications for primary or secondary prophylactic antibiotics or for the use of granulocyte colony-stimulating factor (G-CSF) during therapy with hypomethylation agents for patients with MDS [10]. In Israel, azacitidine (AZA) is the most popular hypomethylation agent, therefore, we conducted a national retrospective analysis aiming to evaluate risk factors for infection among high risk MDS patients treated with AZA.


The survey was designed to include all high-risk MDS or acute myeloid leukemia (AML) patients treated with AZA in 18 Israeli hospitals between the years 2008–2011. Eighteen of 19 Israeli hospitals where AZA was available during the study period participated in the survey. The study was approved by Institutional Review Boards as required. All adult patients diagnosed with high-risk MDS or AML who received AZA were included in the analysis. After reviewing patients' medical charts, those found to have IPSS score of less than 1.5 or those who received AZA doses lower than 75 mg/m2 for 5 days in all cycles were excluded. Clinical and demographic data, complete blood count (CBC) and serum chemistry values on the first day of each AZA cycle were collected. Parameters selected for assessment in univariate analysis comprised: age, sex, serum creatinine level, cytogenetics, percentage of blasts in the bone marrow (BM) aspirate and transfusion dependence. In addition, AZA doses, neutrophil, platelet (PLT) and hemoglobin (Hb) levels measured prior to each cycle, were also included (Table 1).

Table 1. Characteristics of Patients and Cycles
 No of patients, N = 184No of cycles, N = 928
  1. Hb: hemoglobin; PLT: platelets.

Age (median, range)  
>70118 (64.1%)611 (65.8%)
Male117 (63.5%)606 (65.3%)
Good7 (4%)32 (3.6%)
Intermediate108 (64%)607 (68.9%)
Poor55 (32%)242 (27.5%)
0–5%26 (14%)140 (15.2%)
6–10%48 (26%)207 (22.4%)
11–20%82 (45%)420 (45.5%)
>20%27 (15%)156 (16.9%)
mean ± STD1.05 ± 0.431.06 ± 0.44
Azacitidine dose  
<75 m2 × 5d9 (5%)65 (7%)
75 m2 × 5d67 (37%)303 (33%)
75 m2 × 7d106 (58%)558 (60%)
<1060 (33%)382 (41%)
<20,00041 (22%)144 (15.5%)
<50049 (27%)213 (23%)

Data collection was limited to only include the first 12 cycles of therapy for each patient, as infection rate during subsequent cycles is known to be very low [6]. Including the latter cycles may bias the analysis due to over-representation of good responders. Findings on 39 patients from three medical centers were obtained for the first six cycles only.

Data analysis

Statistical analysis was performed using SPSS program, version 18. Normal distributions of the quantitative variables were tested by the Kolmogorov–Smirnov tests. Univariate analysis (Fisher exact test, Pearson chi-square T-test, and Munn–Whitney U-test) was used for assessment of infection incidence. Pearson correlation was used to calculate linear correlation between quantitative variables (Hb level, neutrophil count, serum creatinine level, WBC and PLTs count). Multinomial logistic regression analysis was employed to identify any relation between infection and studied parameters. P < 0.05 was considered statistically significant.


Description of patients and cycles

Seventeen patients who were found to have an IPSS score below 1.5 and 15 patients who received very low AZA doses in all reported cycles were excluded from the analysis. The characteristics of remaining 184 patients are presented in Table 1. The median age of patients was 71.6 years and ranged between 29 and 92 years. The median percentage of BM blasts was 13%. In 27 out of 184 (14.6%) patients, a BM blast level of above 20% (which is consistent with AML) was reported. Numbers of cycles and doses of AZA varied. The median number of cycles prescribed to a single patient was 3 (1–27). Doses of AZA were grouped into three categories: a standard dose of 75 mg/m2 or higher for 7 days (a total dose of 525 mg/m2), a low dose of 75 mg/m2 for 5 days (a total dose of 375 mg/m2) and a very low dose of less than 350 mg/m2 for a cycle. Overall, data on 928 cycles were collected, including 558 (60.1%), 303 (32.7%), and 65 (7%) cycles for each dose group, respectively. For patients, who received very low doses in some of the cycles while getting higher doses in others, all treatment cycles were included in the analysis. Although it is sometimes difficult to obtain information from medical records, it appeared that in a significant number of cycles the reduced doses were attributed to interruption of a planned conventional course due to various reasons, including infections.

Prevalence and severity of infections

Infection was defined as any event diagnosed as such by the treating physician. Results of complete workups for infectious pathogens were available in 124/153 of events (81%). Empiric anti-microbiology therapy was prescribed in 69/153 (45.1%) events, including the 29 events where data regarding microbiology workup were not available and 40 events where despite a comprehensive workup no pathogen had been identified. Out of the 124 cases which underwent a complete microbiology workup, bacterial, viral and fungal diseases were diagnosed in 73 (59%), 5 (4%), and 6 (4.8%) events respectively. Three of the five viral cases were diagnosed as H1N1.

Overall, out of 184 patients, 100 (54.3%) developed 153 infectious events (16.5% of cycles). One hundred fourteen infectious events (74.5%) required hospitalization, and 30 of them (19.6%) were fatal. Infections were more common during the first two cycles of therapy and their incidence in correlation with number of a particular cycle is presented in Figure 1. The rate of infection events declines gradually along with the progression of therapy. Prophylactic antibiotics were prescribed during only 39/928 (4.2%) cycles administrated to 20 patients. Infections developed after 26 and 23% of the first and second AZA cycles, respectively, but only after 7% of cycles given as fifth or later cycles. At the end of 2011, 80 (43.5%) patients succumbed to either infections or MDS/AML progression.

Figure 1.

Incidence of infection events as a function of the sequential number of azacitidine cycle.

Factors predicting infections prior to each cycle

Patients with unfavorable cytogenetics (55/184; 30%) had a significantly higher incidence of infections compared to those with normal or favorable cytogenetics (24.4% vs. 12.9% of prescribed cycles, P < 0.0001). Transfusion dependency, defined as receiving of at least four units of blood within a 2-month period prior to the first AZA cycle, increased patients' susceptibility to infection. Infections were diagnosed in 117/625 (18.7%) cycles of therapy given to transfusion-dependent patients compared to only 19/157 (12.1%) cycles prescribed to nontransfusion dependent individuals (P = 0.059). No significant difference was found in the infection risk between patients receiving standard versus low-dose regimens [100/556 (18%) and 46/303 (15.2%), respectively; P = 0.34]. In univariate analysis, patients' sex, age, and creatinine level were not found to be associated with the rate of infectious events.

As to the BM blast percentage, infections occurred in 17/140 (12.1%), 44/207 (21.3%), 62/418 (14.8%), and 30/156 (19.2%) cycles in patients with a blast count of <5%, 5–10%, 11–20%, and >20%, respectively (Table 2). The difference between infection rates in patients with 11–20% and >20% of BM blasts was not statistically significant (P= 0.2) and for further analysis those groups were combined into one group defined as containing >10% of BM blasts.

Table 2. Univariate Analysis of Risk Factors for Infections in MDS Patients Treated with Azacitidine
ParameterWith infectionWithout infectionP-value
  1. a

    Significant differences (P < 0.0001) between intermediate versus poor cytogenetics.

  2. b

    Significant differences (P = 0.03) between 6 and 10% blasts versus 11–20% blasts.

<7060 (19.0%)256 (81.0%)Ns
>7093 (15.2%)517 (84.8%) 
Male94 (15.5%)511 (84.5%)Ns
Female59 (18.4%)262 (81.6%) 
Good4 (12.5%)28 (87.5%)P < 0.001 (b)
Intermediate78 (12.9%)527 (87.1%) 
Poor59 (24.4%)183 (75.6%) 
0–5%17 (12.1%)123 (87.9%)P = 0.03 (a)
6–10%44 (21.3%)163 (78.7%) 
11–20%62 (14.8%)356 (85.2%) 
21+%30 (19.2%)126 (80.8%) 
Transfusion dependency   
Yes117 (18.7%)508 (81.3%)P = 0.059
No19 (12.1%)138 (87.9%) 
Azacitidine dose   
7 days100 (18.0%)456 (82%)Ns
5 days46 (15.2%)257 (84.8%) 
Creatinine1.057 ± 0.441.069 ± 0.44Ns
<500 cells/µL57 (27%)154 (73%)P < 0.0001
>500 cells/µL95 (13.5%)608 (86.5%) 
<20,000 cells/µL42 (29.2%)102 (70.8%)P < 0.0001
>20,000 cells/µL111 (14.2%)676 (85.8%) 
<10111 (20.4%)434 (79.6%)P < 0.0001
<1042 (11.0%)339 (89.0%) 

Prior to each AZA cycle, CBC and routine serum biochemistry were performed. The mean Hb level was significantly lower (9.35 ± 1.26 vs. 10.16 ± 1.8, P < 0.0001) prior to cycles complicated with infections. When AZA prescription was anteceded by a Hb level lower than 10 g/dL, infection rate was as high as 20.4% compared to only 11% following cycles prescribed to patients with higher Hb levels (P < 0.0001). AZA was given in 211 (23.1%) cycles when the neutrophil count was lower than 0.5 × 109/L. Indeed, infection rate in those cycles was twice as high as in cycles with a higher neutrophil count (27% vs. 13.5%, P < 0.0001). Similarly, infection rate was doubled after AZA cycles prescribed to patients with a PLT count lower than 20 × 109/L compared to cycles prescribed to individuals with higher PLT counts (29.2% vs. 14.2%, P < 0.0001).

Multivariate analysis

Multivariate analysis of infection risk following AZA treatment included the following parameters: age, sex, cytogenetics, BM blast percentage and transfusion dependence prior to the first AZA cycle. Additionally, AZA dose, as well as Hb, neutrophil and PLT counts on the first day of therapy were analyzed in each cycle. PLT count lower than 20 × 109/L, Hb level lower than 10 g/dL, and poor cytogenetics, were found to be the only statistically significant risk factors for infection (Table 3). A low PLT count appeared to be the most important risk factor, resulting in a 2.26-fold increase in infection risk, while poor cytogenetics and low Hb were associated with a 1.77- and 1.75-fold rise in infection rate, respectively.

Table 3. Multivariate Analysis of Risk Factors for Infections in MDS Patients Treated with Azacitidine
 Odds ratio95% C.I.Sig
  1. Sig: significance.

PLT < 20,0002.2651.4103.637.001
Poor cytogenetics1.7701.1712.674.007
Hb < 101.7551.1012.798.018

Surprisingly, low neutrophil count did not come up as one of the significant factors. Remarkably, when serum creatinine level (insignificant as a univariate parameter) was included in the multivariate analysis, low PLT count and poor cytogenetics remained predictive factors for infection; however, Hb lower than 10 g/dL was replaced with low neutrophil count, probably due to a hidden correlation between serum creatinine and Hb levels (r = −0.1253, P< 0.0001).

Factors identifying patients prone to develop infections

Sixty five patients experienced one infectious event, 23 patients developed two events, seven patients had three events, four patients had four events, and one patient experienced five events. Eighty four patients (45.6%) had no infection during the study period. Patients who developed infections during the first two cycles of therapy (73 patients, 39.7%) were defined as “prone to infection.” Patients who developed infections only during advanced cycles were analyzed as a separate group.

Poor cytogenetics, PLT count below 20 × 109/L, and neutrophil count below 0.5 × 109/L recorded prior to first AZA cycle identified “prone to infections” patients. (52.7% vs. 33.9%, 56.1% vs. 35%, 53.1% vs. 35.6%, respectively; P < 0.05 for all comparisons). Infections during the first two cycles of therapy occurred in 48.1% of patients receiving standard AZA dose as opposed to 26.9% of those received a lower dose (P < 0.05). Age, sex, serum creatinine level, Hb level, BM blast percentage and the need for blood transfusions were similar among all groups.

In multivariate analysis, a PLT count below 20 × 109/L, neutrophil count below 0.5 × 109/L and a 7-day course of AZA were predictive of patient's risk of infection during the first two cycles. PLT and neutrophil counts are rapidly changing with disease dynamics during therapy. Therefore, their values prior to the first cycle of therapy are not expected to predict infection developing during the third or subsequent cycles. Indeed, no parameter identified patients who developed infections during advanced cycles only (27, 14.7%).


The introduction of hypomethylating agents in general, and AZA in particular, has become a major breakthrough in the management of high-risk MDS patients [11-13]. AZA delays leukemic transformation, can induce remission in AML patients and has been shown to prolong overall survival in high-risk disease [14-17]. However, this therapy is associated with an elevated risk for infections, resulting in a substantial mortality rate [9]. Upon achievement of good hematological response, the risk for infection declines. However, the favorable effect of AZA evolves over a time period that may last 4–6 months. Therefore, prevention of life-threatening infections during this period is crucial for successful management of such patients. Routine antibiotics are still debatable. Currently, in cancer patients who are neutropenic after chemotherapy, the pendulum is swinging towards the use of fluoroquinolone prophylaxis for individuals with high-risk disease, which is defined by the expected duration and profoundness of neutropenia [18, 19]. In a small retrospective study of 28 patients receiving decitabine, prophylaxis with antibiotics was reported to decrease infection incidence during hypomethylation therapy [20]. The high rate of documented bacterial infections observed in this series suggests that fluoroquinolone prophylaxis during AZA therapy may prevent some infection events. Identifying patients at high risk for infection, and determining treatment cycles when this risk is maximal, may contribute to the development of a personalized and more efficient infection prevention protocol for AZA-treated MDS patients [21]. It is noteworthy that the use of G-CSF has also been suggested to prevent infection [22, 23]. One may also consider reducing the AZA dose, but this could hamper the effect of the drug.

High national coverage rate of different medical centers ensures that the results obtained do not reflect specific practice of a single institution and are likely to be applicable to other environments around the world.

Intuitively, neutrophil count is regarded a major risk factor for infection. Indeed, low neutrophil count at presentation did correlate with patient's risk for infection during the first two cycles of therapy. However, when measured prior to each cycle, neutrophil count failed to be the single most significant risk factor for infection in univariate analysis. Surprisingly, in a multivariate model its significance was lost. During AZA therapy, neutropenia may be disease- or therapy-related. Thus, it is suggested that the risk for infection depends on the BM status. At presentation, low neutrophil count is a marker for the severity of the disease, and therefore, predicts infection in the first cycles of therapy. As treatment progresses, some patients respond but still may be neutropenic due to the AZA-related adverse effect. In such cases, factors other than neutropenia measured prior to each cycle may better correlate with BM reserves. Indeed, low PLT and Hb levels, along with malignant clone carrying poor cytogenetic aberrations are more accurate predictors for infection risk than neutropenia. This is consistent with the recent report suggesting that nearly all BM cells in patients with MDS and secondary AML are clonally derived [24].

Surprisingly, BM blast percentage did not correlate with infectious events. Infections were more prevalent among AML patients (blasts > 20%) than in MDS patients with blast counts lower than 5%; yet, this difference did not reach statistical significance (19.2% vs. 12.1%, P = 0.12). In addition, higher rates of infection were observed among patients with 5–10% of blasts compared to those presenting with 10–20% of blasts (21.3% vs. 14.8%, P = 0.03). In our analysis, blast count recorded prior to therapy was applied to all consecutive cycles prescribed to a patient. As the number of blasts in the BM has not been followed during therapy, real data about the actual blast number during each cycle are missing.

Recently, a revised version of the IPSS score taking into account patient's transfusion dependency has been proposed [25]. Transfusion dependency is defined as administration of more than two red pack-cells a month for two consecutive months. In the current study, 145 cycles of therapy were given to the patients who did not have a 2-month delay between MDS diagnosis and the first AZA dose. With the increased availability of AZA, smaller numbers of patients will have their therapy delayed and hence, stratifying patients upon the present definition of transfusion dependency will become less practical.

A recent Dutch retrospective survey of 90 MDS/AML patients, who were treated with AZA, identified the absence of circulating blasts, standard and good cytogenetics and an increase in PLT count following the first AZA dose of more than twofolds as independent predictors for response and survival [26]. This is the other side of the same coin, cytogenetics and PLT counts represent BM potential for better (response) or for worse (infection).

This is a large retrospective series, and some practical conclusions may be drawn from it. The presence of at least one of the three identified risk factors (unfavorable cytogenetics, low Hb or PLT) prior to AZA administration, indicates cycles with higher infection risk. Such stratification allowed coverage of 82% (126/153) infectious events; however, this high sensitivity came at a price of a need to classify 659/928 (71%) cycles as high risk for infection (specificity of 31%).

Five hundred forty five AZA cycles (58.8%) were administered while Hb was <10 g/dL. However, since the mean Hb level prior to cycles not complicated with infection was 10.16 g/dL, the Hb <10 g/dL is not a practical alarm sign for imminent infection. Additionally, Hb <10 g/dL overlapped with a PLT count of <20 × 109/L. Of 144 cycles prescribed while the Hb level was <10 g/dL, a PLT count of <20 × 109/L was recorded in 119 (82.6%). Low Hb level was observed in 149/242 (61.6%) cycles prescribed to patients with poor cytogenetics. Narrowing down the infection risk group to include only patients with adverse cytogenetics or low PLT count will provide a sensitivity of 53% (81/153 infectious events) and specificity of 66% (341/918 cycles). Remarkably, 70% (21/30) of all fatalities occurred in patients of this risk group.

Based on this study, during AZA therapy, infection prophylaxis may be considered in a limited group of patients with adverse cytogenetics or during AZA cycles prescribed when PLT counts are low. Efficacy of such method should be further examined in prospective trials.

Author Contributions

DM: designed the study, performed research, analyzed data, wrote the paper, KF: performed research; AG-G: performed research, wrote the paper, LV: performed research, wrote the paper, AA, MEG, IS, YH; AA, TT, ND, AN, OR, AR, KH-T, LA, AB, IH, SY, AN: performed research, RL: analyzed data, MM: performed research, wrote the paper, and YO: designed the study, performed research, analyzed data, wrote the paper.