A new prognostic index (MIPI) for patients with advanced-stage mantle cell lymphoma

Mantle cell lymphoma (MCL) is a relatively rare lymphoma entity accounting for approximately 3% to 6% of all non-Hodgkin lymphoma (NHL) cases.^1,–3  It has a poor prognosis with reported median overall survival (OS) of only 30 to 43 months.^1,2,4  In the late 1980s, it was controversially debated whether centrocytic lymphoma, as defined by the Kiel classification,⁵  represented a separate lymphoma entity and whether it corresponded to the intermediate lymphocytic lymphoma or lymphocytic lymphoma of intermediate differentiation of the Working Formulation.⁶  After detection of the characteristic translocation t(11;14) associated with those lymphoma subtypes and based on further morphologic analyses, Banks et al⁷  proposed in 1992 the term “mantle cell lymphoma” for this now widely accepted entity of malignant lymphomas.

Since the international acceptance of MCL in the revised European-American lymphoma (REAL) classification⁸  in 1994 many reports on clinicopathologic features, treatment, clinical course, and prognostic factors for patients with MCL have been published.^{1,2,4,9,,,,,,,–17}  Treatment results have overall been unsatisfactory, although a substantial variation in outcome was noted among individual cases with a small fraction of patients even achieving long lasting remissions. Unlike the International Prognostic Index¹⁸  (IPI) for diffuse large B-cell lymphomas or the Follicular Lymphoma International Prognostic Index¹⁹  (FLIPI) for follicular lymphomas, no broadly accepted prognostic index has been defined for MCL so far. Previous reports revealed serious limitations applying the IPI and FLIPI to MCL patients,^{1,2,4,9,,,,–14,16,17,20}  and the development of a prognostic index specific to MCL patients has been postulated.^12,20  In addition, controversial results have been published concerning the prognostic value of several candidate prognostic factors, but results were mostly based on limited samples of fewer than 130 patients.

We analyzed the pooled data of the most recent randomized trials of the German Low Grade Lymphoma Study Group (GLSG) and the European MCL Network to validate the IPI, FLIPI, and other prognostic factors, and to develop, if needed, a new prognostic index that is particularly suited for patients with advanced stage MCL.

Data from 3 consecutive randomized trials, GLSG1996,²¹  GLSG2000,²²  and European MCL Trial 1,²³  which were performed between 1996 and 2004, were included in this analysis. Local ethics committees of the participating centers approved the study protocol, and written informed consent was obtained from all patients in accordance with the Declaration of Helsinki. Patient entry criteria of all trials were identical with the exception of age, which was limited up to 65 years in the European MCL trial. Otherwise all 3 trials recruited treatment-naive patients with advanced Ann Arbor stage III or IV MCL. Central pathology review was mandatory to ensure a high quality of histologic diagnosis. Exclusion criteria comprised serious concomitant diseases, bad performance status, or significant impairment of organ function.^21,–23  The initial diagnostic work-up comprised a bone marrow biopsy, ultrasound examination of the abdomen, and computed tomography (CT) scans of chest and abdomen. Normal organ function was assured by the respective laboratory tests, as well as by echocardiograms and electrocardiograms.

The GLSG1996 trial²¹  compared the response rates of initial chemotherapy with CHOP (cyclophosphamide, hydroxy-daunorubicin, vincristine and prednisone) and MCP (mitoxantrone, chlorambucil, prednisone). In 2000, the GLSG2000 trial²²  was initiated to assess the additional efficacy of incorporating the monoclonal CD20-antibody rituximab to primary CHOP in terms of time-to-treatment failure. In both GLSG trials, patients older than 65 years and responding to initial chemotherapy received interferon-α (IFNα) maintenance therapy. Younger patients were randomized between consolidating myeloablative radiochemotherapy followed by autologous stem cell transplantation (ASCT) and IFNα maintenance within the European MCL Trial 1.²³  Detailed study design and results of the respective trials have been published previously.^21,–23 

All baseline variables were considered as candidate prognostic factors, namely age, sex, Eastern Cooperative Oncology Group (ECOG) performance status,²⁴  Ann Arbor stage, presence of B-symptoms, bone marrow involvement, spleen involvement, number of involved nodal areas, number of extranodal sites, size of largest involved lymph node, and laboratory parameters such as white blood cell (WBC), lymphocyte, granulocyte, monocyte, and platelet counts, hemoglobin (Hb), lactate dehydrogenase (LDH), albumin, and serum β₂-microglobulin levels. Because upper limits of normal (ULN) differed considerably between laboratories for LDH, albumin, and β₂-microglobulin, ratios to ULN were calculated. In addition, the known prognostic indices established in other lymphomas, IPI,¹⁸  and FLIPI¹⁹  were assessed. The outcome parameter was OS calculated from the day of recruitment to death from any cause or to the latest follow-up date.

In line with IPI and FLIPI, our aim was to establish a clinical prognostic index (mantle-cell lymphoma international prognostic index, MIPI) easily available in clinical practice. Besides the clinical parameters, the proliferation markers Ki-67 and number of mitoses, possibly the most important prognostic markers in MCL,^{1,2,4,11,14,,–17}  have been assessed centrally and in a standardized way. Thus, we analyzed the additional prognostic relevance of cell proliferation and exploratively developed a combined biologic index (MIPI_b) incorporating the effect of cell proliferation.

To validate IPI and FLIPI, Kaplan-Meier OS curves stratified for risk group were calculated and compared by the log rank test. In addition, the individual risk factors constituting IPI and FLIPI, respectively, were checked for independent prognostic relevance by multiple Cox regression analyses.

The prognostic relevance of the candidate prognostic factors was evaluated using univariate Cox regression for OS. Subsequently, multiple Cox regression with backward variable selection was performed to identify independent prognostic factors. Size of largest involved lymph node, spleen involvement, lymphocyte, granulocyte, and monocyte counts, albumin, and serum β₂-microglobulin levels were not included in multiple regression because of a high number of multivariately missing values. Bone marrow involvement was excluded because of high congruence with stage IV. Continuous parameters were not categorized a priori because this would have negatively affected the power of the analysis.²⁵  ECOG performance status was classified into 3 groups: asymptomatic (ECOG 0), symptomatic but ambulatory and able to carry out light work (ECOG 1), and unable to work or bedridden (ECOG 2-4). WBC count, LDH, and β₂-microglobulin were analyzed on the log-scale because of highly skewed distributions. The proportional hazards assumption and the linearity assumption for the final model were checked using scaled Schoenfeld residuals^26,27  and martingale residuals,²⁸  respectively.

Prognostic groups were defined by categorizing the prognostic score (linear predictor) of the final Cox regression model. 2 optimal cutpoints were found maximizing the log rank statistic according to the “minimal P value approach,”²⁹  and P values for the log rank statistic were adjusted for multiple testing by the Bonferroni method. For practical reasons referring to clinical application of the prognostic index, the goal was to define 3 risk groups, none of which comprised more than 50% of the patients. An explorative simplification of the new classification rule was developed categorizing the prognostic factors according to standard cutpoints. The simplification with the best concordance to the original index measured by Cohen weighted kappa³⁰  was chosen.

Internal validation was performed applying the refined bootstrap described by Efron.³¹  Random data splitting in training and validation sample was not performed because this internal validation procedure reduces the power for both model development and validation and is known to be inferior to bootstrap validation.^31,32  Bootstrap validation used the likelihood ratio or log rank test statistic, the calibration slope (regression coefficient of the prognostic score), Harrell concordance index c³³  for censored outcome variables, and the measure of separation SEP³⁴  for survival curves.

Statistical analyses were performed using SAS version 9.1 (SAS Institute, Cary, NC) and R version 2.2.1 (www.r-project.org). Significance level was 5% for all analyses.

Between May 1996 and October 2004, 455 patients with advanced stage MCL entered the trials. The median age was 60 years (34-86 years) and 76% of the patients were male. Eighty-four percent of the patients were in Ann Arbor stage IV, 79% had bone marrow involvement. Forty-three percent presented with B-symptoms and 9% were unable to work or bedridden (ECOC 2-4). Further patient characteristics are depicted in Table 1.

Initial cytoreductive chemotherapy comprised CHOP in 56% of the patients, 31% of the patients received R-CHOP, 11% MCP, and 2% other chemotherapy regimens. Of the 438 patients evaluable for treatment response, 351 (80%) achieved a complete or partial remission, and 80 (18%) a complete remission. Treatment in remission was ASCT in 80 patients, IFNα maintenance in 199 patients, and 72 patients obtained no therapy in remission. Of the 455 patients, 159 had died and the median OS was 57 months with a median follow-up of surviving patients of 32 months.

According to the IPI, 99 patients (23%) were classified as low risk (LR), 173 patients (40%) as low intermediate risk (LIR), 119 patients (28%) as high intermediate risk (HIR), and 41 patients (9%) as high risk (HR); in 23 patients the IPI could not be evaluated due to missing data. The median OS was not yet reached in the LR group with a 5-year OS rate of 59%, and 61 months in the LIR group, 45 months in the HIR group and 20 months in the HR group (P < .001, Figure 1). LIR and HIR groups comprised together more than two-thirds of the patients and were not well separated. In multiple Cox regression age, ECOG performance status and LDH were prognostic for OS, whereas the number of extranodal sites was not (Table 2).

The FLIPI could be determined in 418 patients of whom 26 (6%) were classified as LR, 124 patients (30%) as intermediate risk (IR), and 268 patients (64%) as HR. The median OS was not yet reached for LR and IR patients with a 5-year OS rate of 61% and 57%, respectively, and 48 months in the HR group (P< .001, Figure 2). LR and IR groups were not separated and the large high risk group had relatively good outcome. In the multiple Cox regression, age and LDH were prognostic for OS, whereas neither hemoglobin nor the number of involved nodal areas were of prognostic relevance (Table 3).

Neither sex, Ann Arbor stage (III vs IV), bone marrow involvement, number of extranodal sites or number of nodal areas, nor platelet count or albumin showed prognostic relevance for OS in univariate analyses. In contrast, age, ECOG performance status, B-symptoms, spleen involvement, maximal lymph node size, WBC, lymphocyte, granulocyte, and monocyte counts, LDH, Hb, and β₂-microglobulin had a significant impact on OS (Table 4).

Sex, stage, B-symptoms, number of extranodal sites, number of nodal areas, platelet count, and hemoglobin lost their prognostic relevance in multiple Cox regression with backward variable selection on the dataset of 409 complete cases. Age, ECOG performance status, LDH, and WBC counts were identified as the 4 independent prognostic factors for OS. Relative risk (RR) of increased age was 1.04 per year (95% CI 1.02-1.06, P<.001), RR of patients with poor performance (ECOG > 1) was 2.01 (95% CI 1.19-3.39, P = .009), and patients with a tenfold elevated LDH or WBC count had a RR of 3.92 (95% CI 1.48-10.37, P = .006) and 2.56 (95% CI 1.66-3.95, P < .001), respectively (Table 5). Accordingly, the MIPI prognostic score was calculated as

• MIPI score = [0.03535 × age (years)]

• × age (years)] + 0.6978 (if ECOG > 1)

• + [1.367 × log10(LDH/ULN)]

• + [0.9393 × log10(WBC count)]

The median value of the prognostic score in 409 patients was 5.78 (4.31-9.18), 10% of the values were below 5.16 and 90% of the values below 6.63. Potential cutpoints between 5.15 and 6.65 in steps of 0.05 were assessed. On this basis, the value 6.2 showed the best discrimination (log rank statistic 53.06, 1 degree of freedom, df) and defined a high-risk group of 84 patients (21%, score ≥ 6.2). For the definition of an intermediate-risk group, the value of 5.7 was identified as best discriminator (log rank statistic 61.11, 2 df). In this way 145 patients were classified as intermediate risk (35%, 5.7 ≤ score < 6.2). The remaining 180 patients belonged to the low-risk group (44%, score < 5.7). After Bonferroni correction for the P values calculated from the log rank statistics, the separation of these risk groups remained statistically significant. Median OS was not reached in the low risk group with a 5-year OS of 60%, and it was 51 months and 29 months in the intermediate-risk group and the high-risk group, respectively (Figure 3).

The internal validation procedure correcting for overoptimism by bootstrap showed stability of statistical significance and prognostic separation (Table 6). In addition, the bootstrap corrected performance measures for the new prognostic classification system appeared to be better than the results of the external validation of the IPI and FLIPI.

The classification according to the new prognostic index involves some elementary mathematical operations best performed using an electronic calculator. To make the index most practicable, we exploratively simplified it categorizing the individual prognostic factors maintaining high concordance to the original index. Dichotomizing the prognostic factors, as is commonly done, at standard cutpoints (60 years for age, ULN for LDH, 10 × 10⁹/L for WBC count) yielded low concordance to the MIPI (weighted kappa 0.56), and suboptimal separation of survival curves (Table 6, SEP). Instead, weighting age, ECOG performance status, LDH, and WBC with 0 to 3 points using 3 cutpoints each (Table 7) and classifying patients with a total of at most 3 points as LR, patients with 4 to 5 points as IR and patients with more than 5 points as HR, yielded high concordance (weighted kappa 0.79) and good separation of OS curves (Table 6, SEP).

The percentage of Ki-67 positive cells was available in 236 lymph node biopsies of the 455 MCL patients with a median of 14.5% and range of 1.2% to 91.0%. The number of mitoses per square millimeter was counted in 162 cases with a median of 4, range 0 to 72 and one extreme value of 1039, which was excluded from further analysis. Both proliferation parameters showed strong univariate prognostic relevance for OS with an RR of 1.29 (95% CI 1.16-1.44, P < .001) for Ki-67 elevated by 10% and 1.27 (95% CI 1.09-1.48, P = .003) for the number of mitoses elevated by 10/mm². In bivariable regression including Ki-67 and number of mitoses, only Ki-67 remained independently significant, whereas the number of mitoses lost its prognostic relevance (P = .68, n = 126), and was thus excluded from further analyses. Patients with Ki-67 assessed showed significantly less spleen and bone marrow involvement, better performance status, more involved nodal areas, and larger lymph nodes than patients without available Ki-67. However, there was no significant difference in OS of patients with or without data on Ki-67 (median OS 58 and 57 months, P = .39).

Ki-67 remained independently significant from the MIPI clinical score (regression coefficient 0.02142, P < .001) and inclusion of Ki-67 did not substantially change the regression coefficient of the MIPI score (0.9554, P < .001). Thus, we calculated the combined biologic score by addition of 0.02142 times Ki-67 (%) to the clinical score. Optimal cutpoints were 5.7 and 6.5 for the log rank statistics with respect to OS, and Figure 4 shows OS according to the so defined combined biologic index (MIPI_b).

Based on data of 455 patients with advanced stage MCL from 3 randomized trials of GLSG and European MCL Network, a new prognostic index (MIPI, mantle cell lymphoma international prognostic index) has been developed. The MIPI is the first prognostic index specific to MCL patients and, more importantly, allows a clear separation of 3 well-balanced groups of patients with significantly different prognoses. The low-risk group comprised 44% of the patients with a median OS not yet reached after a median follow-up of 32 months, and a 5-year OS rate of 60%. The intermediate risk group consisted of 35% of the patients and revealed a median OS of 51 months, while the high-risk group (21%) had a poor outcome with a median survival of only 29 months. The MIPI is based on the 4 independent prognostic factors age, ECOG performance status, LDH, and WBC counts, which are clinical variables easily and routinely available in clinical practice. A simplification of the MIPI was developed allowing for bedside application reproducing the original index remarkably well.

The application of the IPI¹⁸  for diffuse large B-cell lymphoma to our MCL patient cohort revealed a lack of separation of the 2 intermediate-risk groups, which comprised more than two-thirds of the patients. The application of the FLIPI¹⁹  for follicular lymphoma to our MCL patients showed even worse results, with no separation of low- and intermediate-risk groups and a large high-risk group with relatively good outcome. In addition, of the individual IPI and FLIPI risk factors, the number of extranodal sites and the number of involved nodal areas showed no prognostic relevance and hemoglobin no independent prognostic relevance in our MCL patient cohort. ECOG performance status, which showed high prognostic relevance in our analyses and many previous reports, was not included in the development of the FLIPI.

Previous attempts to apply the IPI in particular to patients with MCL proved rather unsatisfactory^{1,12,–14,16,20}  (Table 8). The IPI showed either no prognostic relevance for OS,^4,10,17  or the original 4 risk groups were combined to 3^12,14  or 2 risk groups^1,11,16  or different risk groups had very similar outcome.^2,9,13,20  In addition, of the 5 IPI risk factors, the number of extranodal sites showed mostly no, and Ann Arbor stage inconsistent, prognostic relevance in previous univariate analyses (Table 8).

In contrast to our results, a recent report by Møller et al²⁰  stated superiority of the FLIPI over the IPI in terms of size and number of risk groups, separation of survival curves, and prognostic value of individual risk factors. However, the number of patients analyzed, 93, was rather small, low- and intermediate-risk groups were not well separated, and the high-risk group included more than half of the patients. Interpretation of multiple Cox regression results was hampered by inclusion of the interdependent factors IPI, FLIPI and the individual prognostic factors.

Of the independent prognostic factors of the MIPI, consistent prognostic relevance on OS had been shown for age, performance status, and LDH in previous univariate analyses (Table 8). Hematologic parameters were mostly not considered as candidate prognostic factors. Three studies,^1,11,12  however, included the WBC count and described a prognostic relevance in univariate analysis. In several studies peripheral blood involvement was included^4,10,,–13  and showed significance in 3^4,12,13  of 5 univariate analyses and in 2^12,13  multiple regression analyses. The methods for the detection of peripheral blood involvement, however, were mostly not adequately described and sensitivity of different methods may vary substantially. Although peripheral blood involvement may occur without leukocytosis, a high WBC count probably reflects peripheral blood involvement, and our results thus may confirm previous observations.^12,13 

Despite these encouraging results, it must be emphasized that the current analyses did not include patients with limited stage I or II MCL because for these patients no comparable clinical trial is available. However, the prognostic relevance of stage is not consistently seen in the literature, the proportion of patients with stages I or II is rather low in MCL (Table 8) and they also require a different therapeutic approach. Thus, as recommended by Altman,³⁵  to avoid any treatment bias, we limited our investigation to the advanced-stage MCL patients with standardized treatments in randomized trials. The present data are also limited to patients who can tolerate moderately intensive chemotherapy. This limitation, however, also applies to the IPI¹⁸  and FLIPI¹⁹  and does not hamper the relevance of the MIPI within these borders.

Spleen involvement, maximal tumor size, serum β₂-microglobulin levels, and lymphocyte, granulocyte, and monocyte counts were excluded from multiple Cox regression due to missing data, but had shown univariate prognostic relevance on OS. Nevertheless, further exploratory analyses revealed no additional prognostic relevance of these variables to the 4 MIPI prognostic factors. For β₂-microglobulin this stands in contrast to the results of Khouri et al.³⁶  Two MIPI adverse prognostic factors, higher age and worse ECOG performance status, are also contraindications against more aggressive experimental therapies, for which the MIPI could be applied to select candidate patients. However, excluding age and ECOG performance status as candidate prognostic factors, only LDH and WBC count remained independent prognostic factors still allowing for patient stratification. In addition, in the cohort of patients younger than 65 years, the same 4 independent prognostic factors were identified, thus, there was no “age-adjusted” prognostic index as developed for the IPI. Finally, our findings are independent of the choice of initial cytoreductive therapy, because inclusion of treatment variables did not change the relevance of the 4 independent prognostic factors. In particular, the inclusion of primary rituximab treatment as covariate did not change the final prognostic model. In addition, a significant improvement in OS has not been shown for ASCT so far for MCL patients.

In an attempt to develop a combined biologic index as postulated by Räty et al¹⁵  and Tiemann et al,¹⁷  we included the proliferation marker Ki-67 in our analyses and showed high prognostic relevance independent from the MIPI prognostic score. Including Ki-67 we exploratively defined a combined biologic index (MIPI_b) which revealed a low-risk group with relatively good outcome. However, proliferation data were available in only approximately half of the patients, and the selection of patients with Ki-67 appeared to be nonrandom with less spleen and bone marrow involvement and more and larger lymph node involvement. As it is assessed on lymph node biopsies, proliferation is not available in all MCL patients. In addition, standardization and reproducibility of proliferation assessment still requires improvement as essential prerequisite to be used as molecular prognostic marker.³⁷  Hence, a combined biologic prognostic index including Ki-67 is currently not applicable except for research studies.

We performed internal validation by bootstrap, the method that achieves the highest possible power from the available data,^31,32  and could confirm a high stability of the developed prognostic model. Nevertheless, an external validation on an independent data set is still warranted to allow a broad application of this prognostic tool.

In conclusion, our new prognostic classification tool, MIPI, might be helpful to allow individualized, risk-adapted treatment decisions in patients with advanced stage MCL, to hopefully optimize treatment and to improve outcomes of this aggressive disease. In addition, our results will allow stratification in clinical trials, interstudy-comparisons of clinical trial results according to patient risk profiles, and provide a basis for establishing future novel biologic prognostic markers.

We thank the German Low Grade Lymphoma Study Group and the European MCL Network for participation in this study. Many thanks to Olaf Determann from the Lymph Node Registry Kiel, who also worked intensively in providing the proliferation data.

This work was supported in part by grants from the Deutsche Krebshilfe (T14/96/Hi 1, project No. 70-2208-Hi 2), the Bundesministerium für Bildung und Forschung, Kompetenznetz Maligne Lymphome (no. 01 GI 9994), the European Commission (no. LSHC-CT-2004-503351), and the Lymphoma Research Foundation (no. MCLI-04-016).

E.H. analyzed and interpreted data, performed statistical analysis, and drafted the manuscript; MD, W.H., and M.U. designed the trials and revised the manuscript; W.K. provided proliferation data; J.H. provided statistical expertise and revised the manuscript; C.G., A.v.H., H.C.K.-N., M.P., M.R., B.M., H. Einsele, N.P., W.J., B.W., W.-D.L., U.D., H. Eimermacher, and H.W. recruited, treated, and documented patients.

E.H. is a candidate at the Faculty of Medicine of the Ludwig-Maximilians-University Munich for her doctoral degree and part of this work has been developed in partial fulfillment of the requirements for her degree.

A complete list of the members of the German Low Grade Lymphoma Study Group and the European MCL Network who contributed to this analysis appears in Document S1, available on the Blood website; see the Supplemental Materials link at the top of the online article.

Conflict-of-interest disclosure: M.P. is a member of the Advisory Board of Roche, Genentech, and Lilly. M.D. received support for clinical studies and speaker honoraria from Roche. W.H. is member of the Advisory Board of Roche and received support for clinical studies and speaker honoraria from Roche. The remaining authors declare no competing financial interests.

Correspondence: Dr. Michael Unterhalt, Medizinische Klinik III, Klinikum Groβhadern, Marchioninistr. 15, D-81377 München, Germany; e-mail: Michael.Unterhalt@med.uni-muenchen.de.