Comparison of the NCCN‐IPI, the IPI and PIT scores as prognostic tools in peripheral T‐cell lymphomas

Abstract

Survival in peripheral T-cell lymphomas (PTCL) is poor compared to aggressive B-cell lymphomas. The majority of patients will progress or relapse after front-line therapy and long-term overall survival (OS) rates are only 20–40% in most population-based studies (Vose et al, 2008; Ellin et al, 2014). The most commonly used score to prognosticate survival in PTCL is the International Prognostic Index (IPI), which was developed in aggressive lymphomas, and has been validated in smaller studies in PTCL (Ansell et al, 1997; Vose et al, 2008). Although associated with survival, the IPI has limited ability to identify those PTCL patients at truly low risk of treatment failure. A different prognostic model, the Prognostic Index for T-cell lymphoma (PIT) score, was initially reported as superior to the IPI in a cohort of PTCL not otherwise specified (NOS) (previously termed PTCL-unspecified) (Gallamini et al, 2004). However, subsequent validation in PTCL NOS and application in different PTCL subtypes has shown a performance similar to the IPI (Weisenburger et al, 2011; Ellin et al, 2014). In the most common aggressive B-cell lymphoma subtype, diffuse large B-cell lymphoma (DLBCL), a novel model, the National Comprehensive Cancer Network (NCCN)-IPI, has been shown to have a superior prognostic ability compared to the IPI in patients treated with immunochemotherapy. The NCCN-IPI includes the same factors as the IPI, but introduces weighted categorization with respect to age and lactate dehydrogenase in serum, and includes specific extranodal sites (central nervous system, bone marrow, liver/gastrointestinal tract, lung) rather than the absolute number of involved extranodal sites (Zhou et al, 2014). We evaluated the prognostic capacity of the NCCN-IPI compared to the IPI and PIT in a pooled analysis of three international cohorts of nodal PTCL, including anaplastic lymphoma kinase-negative anaplastic large cell lymphoma (ALK ALCL), angioimmunoblastic T-cell lymphoma (AITL) and PTCL NOS. Details regarding the methods can be found in Data S1. In total, 660 patients met the selection criteria with respect to diagnosis and treatment, and 603 (91%) had complete data for calculation of prognostic indices (ALK ALCL n = 152; AITL n = 145; PTCL NOS n = 306), and were included in the analysis. The 57 patients with missing data did not differ significantly from the study population by baseline characteristics and OS (data not shown). Median follow-up was 81 months (range 1–185 months), during which 449 patients had a documented relapse/progression and 400 patients died. Basic clinical variables are listed in Table I. Initial treatment in the population consisted of anthracycline-based chemotherapy in 577 (96%), with 26 patients receiving other curative intent treatment. Prognostic scores according to NCCN-IPI, IPI and PIT were calculated and grouped into low, low-intermediate, high-intermediate and high scores (Table SI). Looking at single scores, this grouping could potentially be made differently, but was beyond the scope of this study (Figure S1). Distribution of prognostic groups is presented in Table I together with hazard ratios (HRs) for OS by individual factors. Of note, in AITL, age was not significantly associated with OS or event-free survival (EFS) (e.g., age > 60 years HR = 0 97; 95% confidence interval [CI] 0 66–1 41, P = 0 86) by any of the cut-offs (Table I). Kaplan–Meier curves for OS (Fig 1) were calculated according to the three indices for the PTCL subtypes separately, and were compared visually and by the c-statistic (Table I). The NCCN-IPI seemed to separate prognostic groups slightly better by visual judgement of Kaplan–Meier curves in ALK ALCL and PTCL NOS (Fig 1), although this difference did not result in markedly better c-statistics, with values similar and below 0 7 in all subtypes and indices (Table I). All indices performed distinctly worse in AITL than in ALK ALCL and PTCL NOS. Results were comparable using the EFS endpoint (Figure S2). Analyses stratified according to gender and up-front autologous stem cell transplantation (n = 114) showed similar results (data not shown). The recent improvement in prognostication in DLBCL with the introduction of the NCCN-IPI was hypothesized to improve prognostication in PTCL. In contrast to DLBCL, the NCCN-IPI performed similarly to the IPI, and also the PIT score, in the three nodal PTCL entities included in our study. In ALK ALCL and PTCL NOS, the NCCN-IPI seemed to be slightly better in identifying groups with the lowest and highest risk of failure. However, the percentages of patients in these risk groups were smaller than with the IPI, limiting the clinical utility of these models and resulting in similar cstatistics for the NCCN-IPI and the IPI. Based on our results, alternative groupings of the scores would probably only result in a marginal improvement at best, and would need independent validation.