Page 86 - 2019_05-HaematologicaMondo-web

N. Gagelmann et al.
58.4-78.4) for the low-risk group, 43.3% (95% CI: 37.2- 49.4) for the intermediate-risk group, 26.0% (95% CI: 20.7-31.3) for the high-risk group, and 11.5% (95% CI: 5.4-17.6) for the very high-risk group (P<0.001) (Figure 2B). The cumulative incidence of relapse and non-relapse mortality at 5 years according to risk group were, respec- tively: 16.9% and 12.8% (low-risk group), 28.2% and 27.7% (intermediate-risk group), 36.0% and 36.9% (high- risk group), and 46.0% and 44.3% (very high-risk group) (Figure 2C,D).
Nomogram
We aimed to refine the transplant-specific risk score fur- ther by using age and platelet count at transplantation as continuous variables while the other variables remained in their categorical structure. We provide a discrete/continu- ous nomogram to interpolate the final score and assess an individual patient’s risk in an easy manner (Figure 2). For each of the seven prognostic factors, individual points are assigned, which are subsequently summed to a total point scale. The final score is then translated into predicted sur- vival rates at different time points for each patient. For instance, 55-year old patients presenting with a platelet count of 150 x 109/L, blasts ≤1%, good cytogenetics, a Karnofsky Performance Status of 90% who are seroposi- tive for CMV and for whom an identical sibling donor is available show estimated survival rates at 3 and 5 years of 57% (95% CI: 51-63) and 47% (95% CI: 41-54), respec- tively.
Validation of existing systems
The overall score from the CIBMTR ranged from zero to seven and was used to define four risk categories: low risk (0-1, n=46), intermediate risk (2-3, n=434), high risk (4-5, n=365), and very high risk (>5, n=31). The 5-year overall survival rates were 55.8% (95% CI: 39.9-71.7) for the low-risk group, 42.3% (95% CI: 37.2-47.4) for the intermediate-risk group, and 27.2% (95% CI: 21.9-32.5) for the high-risk group. Rates were not estimable in the very high-risk group because the median follow-up was 42.6 months in this group while median overall survival was 11.8 months (95% CI: 3.1-20.4) (Figure 3A). The GITMO score values ranged from zero to eight. The actu- al score delineated four risk-groups: low (0-1, n=81), inter- mediate (2-3, n=200), high (4, n=123), and very high (>4, n=115). The overall survival rates at 5 years were 62.7% (95% CI: 51.7-73.7) for the low-risk group, 41.4% (95% CI: 34.0-48.8) for the intermediate-risk group, 24.1% (95% CI: 15.1-33.1) for the high-risk group, and 15.8% (95% CI: 8.2-23.4) for the very high-risk group (Figure 3B). Overall, both scores could be validated (P<0.001).
Comparison of prognostic systems
Both existing scores from GITMO and CIBMTR were then compared with the developed EBMT transplant-spe- cific risk score with respect to their performance regard- ing overall survival. The CIBMTR and GITMO scores showed modest performance, with C-statistics after cross-validation being 0.555 (95% CI: 0.524-0.586) and 0.579 (95% CI: 0.570-0.588), whereas the IPSS-R resulted in C-statistics of 0.551 (95% CI: 0.530-0.566). The devel- oped categorized EBMT transplant-specific risk score showed C-statistics of 0.609 (95% CI: 0.588-0.629) indi- cating an improvement in prognostic performance, which was further improved using age and platelet count as con-
tinuous variables as indicated by the C-statistics of 0.628 (95% CI: 0.616-0.637).
Discussion
The major findings of this analysis can be summarized as follows. First, seven independent risk factors (age, cyto- genetics, thrombocytopenia and increasing blood blasts at transplantation, Karnofsky Performance Status, donor relation, and CMV status of the recipient) could be suc- cessfully incorporated into a transplant-specific risk score. Second, this EBMT transplant-specific risk score enabled significantly improved prediction of outcome in compari- son with currently existing systems.
The EBMT transplant-specific risk score presented here considered conventional clinically derived risk factors at transplantation and can thus be readily calculated. The pri- mary objective in developing this score was to improve our ability to predict survival of MDS patients after trans- plantation; furthermore, the score was predictive of relapse-free survival, relapse and non-relapse mortality. Moreover, sub-analyses of all MDS patients in the EBMT registry (n=6181), of whom 5122 were not included in the present study because of missing data, revealed a better overall survival rate at 5 years for excluded patients (48%) compared with that of the patients used to develop the score (39%) while non-relapse mortality was 29%, respec- tively. Most patients had to be excluded because of miss- ing information regarding cytogenetics. Multiple imputa- tion of these patients revealed no difference in score per- formance. Collectively, the prognostic ability could be improved using the proposed EBMT transplant-specific
Table 2. Multivariate analysis on overall survival showing seven inde- pendent risk factors after stepwise selection using a Cox proportional hazards model.
Factor
Age at transplant, years < 50
HR (95% CI)
reference 1.71 (1.39-2.09)
reference
1.39 (1.03-1.86)
reference 1.46 (1.17-1.82)
reference
1.39 (1.13-1.71)
reference 1.71 (1.43-2.06)
reference
1.39 (1.16-1.65) reference
1.44 (1.20-1.72)
P Score value
≥ 50
<0.001
0.03
0.001
<0.001
<0.001
<0.001
<0.001
2
1
1
1
2
1
1
Blood blasts at transplant, % ≤ 1
>1
Platelets at transplant, x109/L >50
≤50
Donor type HLA-identical sibling matched unrelated
Cytogenetic risk
very good to poor
very poor/monosomal karyotype
Cytomegalovirus serostatus of recipient
negative
positive
Karnofsky index, % 90 to 100
<90
HR: hazard ratio; CI: confidence interval.
932
haematologica | 2019; 104(5)