Response: Capturing variables with prognostic relevance in development of a new scoring system for primary myelofibrosis

Morel and Duhamel question the statistical methodology used in the validation of the new prognostic scoring system for primary myelofibrosis (PMF) recently proposed by the International Working Group for Myelofibrosis Research and Treatment (IWG-MRT)¹  and the need for a new prognostic score. Instead, they suggest combining the intermediate- and high-risk groups of the Lille score²  into a new high-risk group, resulting in 2 groups: low-risk (median survival, 109 months) and high-risk (median survival, 26 months).

External validity and discriminating power are the main attributes of any prognostic classification. Unfortunately, both remain unsolved,^3,4  probably because statistical validity does not always mean clinical value. For this reason, we assembled an unprecedented large number of PMF patients and worked only with variables with previously demonstrated prognostic value. It should be noted that the proportional hazards assumption was checked in all Cox models and the discriminating power of the classification tested not only by the C index of Harrell but also by calculating the model's accuracy to predict actual survival (supplemental Figure 2 in our article). Obviously, because no PMF series as large as the one of the IWG-MRT was available to test the prognostic classification, we relied on resampling, which is a convenient approach for external validation of prognostic models of diseases with low prevalence,⁵  as recently acknowledged by Morel and Duhamel.⁶  Moreover, we tested the discriminating power of the new classification to predict relative survival, a method less sensitive to the changes in baseline life expectancy occurring over long time periods.

The Lille score has been an important tool in PMF prognostic stratification. However, it was unsatisfactory due to the high proportion of patients in the low-risk group and the poor separation between the intermediate- and high-risk categories, a fact actually acknowledged by Morel and Duhamel when proposing combination of the latter 2 categories. The weakness of the Lille score lies on the fact that it is based on only 2 prognostic factors: hemoglobin less than 10 g/dL and leukocyte count. Because leukocyte values greater than 30 × 10⁹/L are rare in PMF (7.6% of patients in the IWG-MRT series), whereas leukopenia is also infrequent (11% of patients), the score relies mostly on hemoglobin. PMF is not a disease of black or white, certainly not with regard to prognosis, where survival is not a matter of long and short survivors and the spectrum of possibilities is wide. In identifying prognostic variables, the chances of capturing all variables with prognostic relevance substantially increase with larger sample sizes. This was the case of the IWG-MRT study, in which we could demonstrate that other variables also easily available at presentation are important to assess PMF prognosis.

In recent years, considerable advances have been achieved in the molecular characterization of the classic myeloproliferative neoplasms. At clinical level, progress will likely come from collaboration between different investigators. This seemed to be the opinion of other members of the Lille group, who helped to make possible the new PMF prognostic score by contributing to the study.