Response: Statistical power and validation of the MIPI

With great interest we read the letter of Shah et al evaluating the Mantle Cell Lymphoma International Prognostic Index (MIPI) in a cohort of 93 of 97 previously untreated patients with aggressive advanced stage mantle cell lymphoma (MCL) treated with alternating cycles of rituximab plus hyperfractionated cyclophosphamide, vincristine, adriamycin, and dexamethasone (HyperCVAD) and rituximab plus high-dose methotrexate–cytarabine (MTX-AraC) within a monocentric phase 2 study.

In the published analyses, overall response rate was remarkable with 97% responding patients and no patients with either stable disease or progressive disease during induction.¹  Overall survival (OS) was also very favorable with two-thirds of patients alive after a median follow-up of 4.8 years, presumably, and 5-year OS rates of 65%.² 

Based on such a favorable outcome one might expect a loss of predictive power of any prognostic factor. However, external validation of a prognostic index requires careful planning, especially with regard to statistical power. Based on the observed hazard ratio of 0.5 between the low and intermediate risk groups in our original cohort, the power of the reported analysis is approximately 40% for this comparison. Accordingly, its statistical power is not sufficient to either confirm or reject the prognostic value of the MIPI.

As mentioned by Shah et al, treatment in our patient cohort (n = 455) varied in the different trials, but treatment selection was randomized and highly standardized within the study protocols. In fact, because a broad range of patient characteristics was also included in our analysis, we believe that it especially reflects the variability of standard care in advanced stage MCL patients. Within this line, the MIPI has been meanwhile also confirmed in other patient cohorts.³ 

In contrast to the statement of Shah et al, the MIPI was generated applying the commonly accepted standard procedure, the Cox regression model. In addition, the simplified prognostic index was developed to provide a simple risk assessment as bedside application. Although concordance to the quantitative MIPI was very high and median OS was almost identical according to the simplified prognostic index (not reached vs 53 months vs 27 months; not reached vs 51 months vs 29 months), we recommend the original quantitative MIPI to be used whenever possible.

In conclusion, we agree that a validation analysis is necessary for confirmation of the prognostic value of the MIPI before recommending a broad application of this tool as stated in our manuscript. However, external validation of a prognostic index requires careful planning, including selection of an adequate cohort with regard to statistical power. Currently, we and others are planning such an appropriately powered external validation applying the MIPI in recent trials of other study groups as well as the current study generation of the European MCL Network.