Comparison of prognostic scoring systems in primary myelofibrosis

To the editor:

We read with interest the description by Cervantes and the International Working Group for Myelofibrosis Research and Treatment (IWG-MRT) of a new prognostic index for patients with primary myelofibrosis.¹  This system, based on 5 adverse characteristics, identified 4 prognostic subgroups. Using the c index of Harrell, the authors stated that the IWG-MRT scores have higher discriminating power than previous scores. However, the use of this index on the sample for which the classification score has been tailor-made is seldom of interest because it estimates probability of concordance between predicted and observed responses.²  The estimation of separation parameter D would better assess whether increasing complexity results in a useful increase in discriminating power.³ 

Using the bootstrap resampling method, we compared the separation parameters^3,4  of IWG-MRT¹  and Lille scores⁵  in a series of 229 patients (163 and 171 previously reported by Dupriez and Cervantes, respectively) diagnosed between 1963 and 2006 (median age, 64 years; median overall survival, 52 months; 10% of patients lost to follow-up). There was no difference in the distribution of characteristics of the present series and previous reports,^1,5  except a low frequency of deaths in the IWG-MRT series (78% vs 49%).

IWG-MRT low- (14%), intermediate-low– (32%), intermediate-high– (32%), and high-risk patients (22%) had significantly different median survival (136, 86, 45 and 21 months, respectively; P < .001). Median survival of Lille low- (49%), intermediate- (42%), and high-risk patients (9%) were 109, 30, and 15.5 months, respectively, without significant difference between the latter 2 subgroups in the present analysis, as previously reported.¹  Thus, combining intermediate- and high-risk patients may improve the interest of Lille score by identifying a large subset of patients eligible for innovative approaches with a median survival of 26 months (95% confidence interval [CI], 22-36 months). In addition, we confirmed that Lille low-risk patients aged less than 55 years (14%) had a median survival estimated at 172 months (95% CI, 100-204 months),⁶  similar to that of IWG-MRT low-risk patients.

Estimate of the separation parameter was D_Lille3 = 1.28 (95% CI, 0.96-1.65) and D_IWGMRT = 1.29 (95% CI, 0.95-1.66) for the Lille and IWG-MRT scoring systems, respectively. This difference was not statistically significant. Combining Lille intermediate- and high-risk patients, a similar lack of difference was observed (D_Lille2 = 1.26; 95% CI, 0.94-1.62).

The IWG-MRT system split low- and intermediate-risk subgroups of the Lille system (P = .002 in each subgroup) but not the high-risk subgroup. Conversely, log-rank test was significant in IWG-MRT intermediate-low–, intermediate-high–, and high-risk patients when the Lille system with 3 categories was assessed (P = .04 in each subgroup). Merging Lille intermediate- and high-risk subgroups, similar results were observed except in the 48 IWG-MRT high-risk patients (P = .11).

Although the present study, as previous reports, does not meet criteria for external validation, the results observed with a prolonged follow-up do not support a higher discriminating power of the IWG-MRT score. Moreover, they support the use of a simplified Lille scoring system, combining intermediate- and high-risk patients in a single category. Given the simplicity of Lille system, further evaluations are required to compare the usefulness of these tools for clinical practice.