Abstract

The International Prognostic Scoring System (IPSS) provides relatively homogeneous prognoses for untreated patients with MDS in terms of survival and transformation to AML. In many cases, cytogenetic results are not available and prevent use of the IPSS. Our objective was to create and validate a dynamic prognostic model that would allow identification of high-risk MDS patients when cytogenetics are unavailable. We followed the methodological approach used by Greenberg et al (

Blood
1997
;
89
:
2079
) and used our new model to re-analyze the FAB-based CALGB data (
JCO
2002
;
20
:
2429
). The variables used in the prognostic model were identified as statistically significant independent variables by Greenberg and included % marrow blasts, number of cytopenias, the multivariate risk factors (age and gender), the univariate risk factor (FAB classification), and time since diagnosis because CALGB data were collected at randomization and IPSS at diagnosis. Both blasts and FAB were included in our model since there is considerable outcome heterogeneity within FAB subtypes. Predicted survival for each of the 191 CALGB patients was determined from a stratified proportional hazards regression model fitted to patients’ survival times. The 191 patients were stratified using their randomized treatment assignment (azacitidine or supportive care). All patients with predicted survival ≤ 1.2 years (median survival of IPSS INT-2 risk group) comprised a high-risk subgroup; those with predicted survival > 1.2 years formed the low-risk subgroup. To assess its predictive accuracy, our model was validated by demonstrating significant segregation of risk groups in an independent dataset of 2,318 untreated primary MDS patients from the MDS Registry, Düsseldorf. In all, 964 (42%) registry cases and 72 (38%) CALGB cases fit into the high-risk subgroup. Log-rank analyses of overall survival and time to AML transformation demonstrated clear separation of the low- and high-risk subgroups (both p-values < 0.0001).

Figure 1:

Validation of Pregnostic Model with MDS Registry, Düsselderf, Overall Survival

Figure 1:

Validation of Pregnostic Model with MDS Registry, Düsselderf, Overall Survival

There was fair agreement between the low and high subgroups from our model with the low (low+INT-1) and high (INT-2+high) subgroups from the IPSS, Cohen’s Kappa = 0.30 (95% CI: 0.21 to 0.39) in 547 patients from the MDS Registry, Düsseldorf. A sensitivity analysis of the CALGB efficacy results was conducted by varying the ≤ 1.2 years survival criterion from our model. High-risk subgroups, defined by using predicted survival of 1.0 to 2.8 years in increments of 0.1 years, were investigated. In general, the sensitivity analyses revealed that clinically meaningful benefit was statistically demonstrated in the smaller subgroups but lost significance as the subgroups approached the full data set. The smaller subgroups were more homogeneous in predicted survival. In conclusion, our model is a valid prognostic model that bridges reasonably well to the IPSS high-risk group and identified a high-risk homogeneous subgroup of CALGB patients to further investigate the effects of azacitidine.

Author notes

Corresponding author