Validation of the MD Anderson Prognostic Risk Model for patients with myelodysplastic syndrome


  • Data were presented at an oral session at the American Society of Hematology 52nd Annual Meeting, December 2010, Orlando, Florida (Abstract # 444).



The proposed MD Anderson Risk Model Score (MDAS) for myelodysplastic syndromes (MDS) refines outcome discrimination compared with the International Prognostic Scoring System (IPSS). We applied the MDAS to Moffitt Cancer Center (MCC) MDS patients to validate its prognostic utility.


This was a retrospective analysis of MDS cases, as defined by World Health Organization criteria, from the Moffitt database, with confirmatory chart review.


A total of 775 patients evaluated between January 2001 and December 2009 were included. Patients were reclassified by MDAS as low (20.6%), intermediate-1 (31%), intermediate-2 (21%), high risk (16.1%), and unknown (11.2%). Median overall survival (OS) from diagnosis was 92, 49, 27, and 14 months for low, intermediate-1, intermediate-2, and high-risk MDAS groups, respectively (P < .005). Median OS from referral was 61, 28, 15, and 8 months (P < .005), respectively. Among 484 patients classified by IPSS as low or intermediate-1 risk, 25% were up-staged as intermediate-2 or high risk by MDAS; 4 prognostically distinct subgroups were identified among lower risk IPSS categories with median OS of 93, 53, 31, and 18 months (P < .005). Among 201 intermediate-2 or high-risk IPSS patients, 15.4% were down-staged to intermediate-1 by MDAS, with 3 groups identified by MDAS having median OS of 33, 23, and 14 months. Acute myeloid leukemia transformation rate was highest among high-risk MDAS (50.4%). In Cox regression analysis, higher risk MDAS predicted inferior OS (hazard ratio 1.42; P < .005) independent of IPSS.


Our data validated MDAS' prognostic value. MDAS is complementary to IPSS and refines prognostic precision. Cancer 2011;. © 2011 American Cancer Society.


Myelodysplastic syndromes (MDS) include a spectrum of hematopoietic stem cell malignancies with variable natural history and risk for progression to acute myeloid leukemia (AML).1 Prognostic risk models have proved to be valuable tools to gauge disease behavior in order to properly tailor disease management.2 The International Prognostic Scoring System (IPSS) is the most widely used model in clinical practice for risk stratification.3 Patients are stratified based on bone marrow myeloblast percentage, number of cytopenias, and karyotype at the time of original diagnosis into 4 categories: low risk, intermediate-1 (int-1), intermediate-2 (int-2), and high risk to segregate patients according to expectations for overall survival (OS) and AML transformation. More importantly, treatment currently is tailored based on IPSS risk category, in patients with “lower risk” MDS (low or int-1) goal of treatment is to improve cytopenias and restore effective hematopoiesis, whereas in “higher risk” disease the goal is to alter the natural history of disease, extend OS, and delay AML transformation.4

The MD Anderson Risk Model is a newly proposed prognostic model that addresses recognized shortcomings of the IPSS such as severity of cytopenia and transfusion dependence.5 Patients are risk stratified based on weighted summed score of age, performance status, hemoglobin, platelet count, bone marrow myeloblast percentage, karyotype, leukocyte count, and transfusion history. Patients are classified according to the MD Anderson Score (MDAS) into 4 categories similar to IPSS: low, int-1, int-2, and high-risk disease.

The primary objective of this study was to externally validate the new risk model calculated at time of initial MDS patients' referral to the Moffitt Cancer Center (MCC).


Data were collected retrospectively from the MCC Total Cancer Care (TCC) database and verified by chart review after institutional internal review board approval. All patients with a diagnosis according to World Health Organization (WHO) 2000 criteria were included. MDS/myeloproliferative neoplasms (MDS/MPN) were excluded from this analysis. The MDAS was calculated as described based on age, performance status, bone marrow myeloblast percent, degree of thrombocytopenia, cytogenetics, white blood cell count, and prior history of transfusion. Patients were divided into 4 risk groups according to total score: low (0-4 points), int-1 (5-6 points), int-2 (7-8 points), and high risk (≥9 points).5 IPSS score was calculated as previously described.3

Analyses were conducted using SPSS version 15.0 (SPSS Inc, Chicago, Ill). The Kaplan-Meier method was used to estimate median OS. Log-rank test was used to compare Kaplan-Meier survival estimates between groups, and Cox regression was used for multivariable analysis.


Between January 2001 and December 2009, 775 MDS patients were captured in the MCC MDS database. Median duration of follow-up from date of diagnosis was 55 months and 37.5 months from date of MCC referral. The mean age at diagnosis was 66.7 years, with 74% of patients above age of 60 years. Table 1 summarizes baseline characteristics. Approximately two-thirds of the patients were characterized as low or int-1 risk by IPSS. Majority of patients were primary MDS (>90%), MDS/ MPN group including chronic myelomonocytic leukemia (CMML) were not included. The median OS for all patients from diagnosis was 37 months (95% confidence interval [CI] 31.8-42.2).

Table 1. Cohort Baseline Characteristics
 N = 775 Patients
  1. Abbreviations: Int, intermediate; IPSS, International Prognostic Scoring System; MDS, Myelodysplastic syndromes;

Age, y 
  Means ± SE66.7 ± 0.45
  ≥60 years (%)575 (74%)
IPSS score, n (%) 
  Low146 (18.8%)
  Int-1338 (43.6%)
  Int-2160 (20.6%)
  High41 (5.3%)
  Unknown90 (11.6%)
Myelodysplastic syndrome subtype 
  RA175 (21.9%)
  RARS94 (12.1%)
  RCMD144 (18.6%)
  Del 5 q20 (2.4%)
  RAEB324 (41.8%)
  t-MDS23 (3%)
  MDS, unclassified20 (2.6%)

The median OS from diagnosis by IPSS risk group was 84 months, 43 months (95% CI 33.7-52.3), 20 months (95% CI 16.4-23.6), and 14 months (95% CI of 11.1-16.9), respectively, for low, int-1, int-2, and high-risk groups (P < .005, log-rank test) (Fig. 1A). Median OS from time of referral to MCC was 55 months (95% CI 46.2-64.6), 26 month (95% CI 21.8-29.4), 12 months (95% CI 9.9-13.4), and 8 months (95% CI 4.8-11.3), respectively, for low, int-1, int-2, and high-risk IPSS groups (P < .005, log-rank test) (Fig. 1B).

Figure 1.

Shown are Kaplan-Meier estimates of overall survival by the International Prognostic Scoring System (IPSS) from time of diagnosis (A) and from time of referral (B) to the Moffitt Cancer Center. Int, intermediate.

Patients were reclassified according to MDAS into 4 risk categories: low risk (20.6%; n = 160), int-1 (31%; n = 240), int-2 (21.0%; n = 163), high risk (16.1%; n = 125), and unknown (11.2%; n = 87). The median OS from date of diagnosis was 92 months (95% CI 72.5-111.5), 49 months (95% CI 40.4-57.6), 27 months (95% CI 21.5-32.5), and 14 months (95% CI 12.0-16.0) for low, int-1, int-2, and high-risk MDAS groups, respectively (P < .005, log-rank test) (Fig. 2A). The median OS from date of referral to MCC was 61 months (95% CI 48.1-74.7), 28 months (95% CI 23.0-33.5), 15 months (95% CI 11.7-19.3), and 8 months (95% CI 4.7-11.8) for low, int-1, int-2, and high-risk MDAS groups, respectively (P < .005, log-rank test) (Fig. 2B).

Figure 2.

Shown are Kaplan-Meier estimates of overall survival (OS) by The MD Anderson Risk Model Score (MDAS) from time of diagnosis (A) and from time of referral (B) to the Moffitt Cancer Center.

Among 484 patients with low or int-1 risk according to IPSS, the MDAS permitted reclassification of patients into 4 prognostically distinct groups; 154 patients (31.8%) who were low risk, 207 patients (42.8%) who were int-1, 98 patients (20.2%) who were int-2, and 21 patients (4.3%) who were high risk. Only 4 patients (<1%) were not reclassified due to missing information. Overall, 119 patients (25%) classified by IPSS as low or int-1 risk were up-staged to int-2 or high risk by MDAS. The median OS from time of diagnosis according to the MDAS among patients with low/int-1 IPSS risk group was 93 months (95% CI 70.3-115.7), 53 months (95% CI 43.7-62.3), 31 months (95% CI 21.8-40.2), and 18 months (95% CI 13.1-22.9) for low, int-1, int-2, and high-risk MDAS (P < .005, log-rank test) (Fig. 3A).

Figure 3.

Shown are Kaplan-Meier estimates of overall survival by The MD Anderson Risk Model Score (MDAS) among the International Prognostic Scoring System (IPSS) low/int-1 risk groups by IPSS (A) and int-2/high-risk groups (B) by IPSS.

According to the IPSS risk model, 201 patients were int-2 or high risk. We were able to reclassify these patients into 3 groups by MDAS: 31 patients (15.4%) were int-1, 65 patients (32.3%) were int-2, and 101 patients (50.2%) were high risk. The median OS from date of diagnosis was 33 months (95% CI 0-66), 23 months (95% CI 18.6-27.4), and 14 months (95% CI 11.5-16.5), for int-1, int-2, and high-risk MDAS groups, respectively (P < .005, log-rank test) (Fig. 3B). Of the 201 patients categorized as int-2 or high risk by IPSS, 31(15.4%) were down-staged to int-1 risk by MDAS and only 4 patients were low risk.

The rate for AML transformation based on MDAS was 5.6% (9 patients) for low-risk, 15.4% (37 patients) for int-1 risk group, 38% (62 patients) for int-2 risk group, and 49.6% (62 patients) for the high-risk group (P < .005, Pearson chi-square test).

In multivariable analysis, adjusted for age >60 years, treatment with azacitidine/decitabine, red blood cell transfusion dependency (RBC-TD), and serum ferritin level. IPSS and MDAS were statistically significant independent prognostic factors for outcome hazard ratio (HR) for IPSS of 1.47 (95% CI 1.2-1.8; P < .005) and 1.42 (95% CI 1.2-1.7); P < .005) for MDAS. Detailed treatment records were not available in the database, 375 patients 48.4% received azacitidine or decitabine, the only treatment with survival data in higher risk MDS.

Finally, we evaluated outcome (median OS from diagnosis) by known prognostic factors, including age, RBC transfusion dependence, and serum ferritin level. All of these prognostic factors correlated with median OS (summarized in Table 2).

Table 2. Overall Survival by Age, RBC Transfusion Dependency, and Serum Ferritin Levels
 Median Survival, Months (95% CI)P
  1. Abbreviations: CI, confidence interval. RBC-TD, red blood cell transfusion dependency.

  2. aUnivariable analysis.

  ≥60 years old (n = 575)35 (29.3-40.7).02
  <60 years old (n=200)45 (32.7-57.3)
  Yes (n = 358)27 (21.7-32.3).005
  No (n = 379)47 (40.7-53.3)
  Unknown (n = 32)53 (1.5-104.5)
Serum ferritin levels  
  ≥1000 ng/ml (n = 177)33 (23.4-42.3).01
  <1000 ng/mL (n = 340)45 (37.7-52.3) 


The IPSS is currently the most widely used prognostic tool for classification and risk stratification of MDS patients.3 IPSS effectively distinguishes outcome for 4 risk groups and has been established as a useful tool in tailoring patient treatment. However, IPSS has several “shortcomings” that limit precision. IPSS was developed from a pooled cohort of patients managed solely by supportive care or hematopoietic growth factor therapy calculated at time of diagnosis, and therefore was not intended to serve as a dynamic model.6 IPSS does not account for severity of cytopenia, namely, thrombocytopenia or magnitude of RBC transfusion burden, and excluded cases of therapy-related MDS and proliferative CMML.7,8 Finally, IPSS does not account for host factors such as age or performance status. The MDAS is a recently proposed model that was intended to address many of the IPSS limitations. The most appealing features of the model are its flexibility, applicability to CMML and therapy-related MDS, its ability to adjust to severity of cytopenias, and the incorporation of patient age and performance status.

In this study, we aimed to validate the prognostic value of the MD Anderson Risk Model in a large external cohort of MDS patients. We found that the MDAS is complementary to the IPSS and refines its prognostic precision. The new risk model identified a subset of patients with IPSS “lower risk” disease with “higher risk” disease behavior and inferior outcome; MDAS also identified a subset with IPSS “higher risk” disease with “lower risk” or indolent behavior.

The MDAS appears reproducible among externally distinct patient data sets, demonstrating comparable OS from time of referral to specialized cancer centers in 3 separate institutional analyses. Table 3 summarizes median OS by MDAS at The MD Anderson Cancer Center, 5 Mayo Clinic, 9 and the Moffitt Cancer Center.

Table 3. Overall Survival from Time of Referral by MDAS
 Overall Survival (Mo)
 MD Anderson Cancer CenterMayo ClinicMoffitt Cancer Center

The greater prognostic precision of the MDAS and applicability at any time point in the disease course should lead to improved management decisions, which could potentially impact an individual's disease outcome. Given its flexibility and precision, the MDAS should be incorporated into clinical trials and management decisions. An international effort is underway to revise the IPSS and incorporate variables from the MDAS and identify new cytogenetic risk groups.10 The WHO-based Prognostic Scoring System is another model that has been validated and currently is incorporated in treatment decisions.6 A unified, objective, and reproducible tool that can be generalized is needed for all academic and community practices. In the near future, prognostic tools will incorporate molecular variables with introduction of newer technologies, permitting treatment algorithms to be based on biological profiles identified.


No specific funding was disclosed.


The authors made no disclosures.