Improving risk-stratification of patients with chronic lymphocytic leukemia using multivariate patient similarity networks

Abstract

Background: Better risk-stratification of patients with chronic lymphocytic leukemia (CLL) and identification of subsets of ultra-high-risk (HR)-CLL patients are crucial in the contemporary era of an expanded therapeutic armamentarium for CLL.

Methods: A multivariate patient similarity network and clustering was applied to assess the prognostic values of routine genetic, laboratory, and clinical factors and to identify subsets of ultra-HR-CLL patients. The study cohort consisted of 116 HR-CLL patients (F/M 36/80, median age 63 yrs) carrying del(11q), del(17p)/TP53 mutations and/or complex karyotype (CK) at the time of diagnosis.

Results: Three major subsets based on the presence of key prognostic variables as genetic aberrations, bulky lymphadenopathy, splenomegaly, and gender: profile (P)-I (n = 34, men/women with CK + no del(17p)/TP53 mutations), P-II (n = 47, predominantly men with del(11q) + no CK + no del(17p)/TP53 mutations), and P-III (n = 35, men/women with del(17p)/TP53 mutations, with/without del(11q) and CK) were revealed. Subanalysis of major subsets identified three ultra-HR-CLL groups: men with TP53 disruption with/without CK, women with TP53 disruption with CK and men/women with CK + del(11q) with poor short-term outcomes (25% deaths/12 mo). Besides confirming the combinations of known risk-factors, the used patient similarity network added further refinement of subsets of HR-CLL patients who may profit from different targeted drugs.

Conclusions: This study showed for the first time in hemato-oncology the usefulness of the multivariate patient similarity networks for stratification of HR-CLL patients. This approach shows the potential for clinical implementation of precision medicine, which is especially important in view of an armamentarium of novel targeted drugs.

Keywords: CLL; Multivariate and network-based approaches; Precision medicine; Prognostication; Risk patient subsets.