Low serum cholesterol levels predict inferior prognosis and improve NCCN‐IPI scoring in diffuse large B cell lymphoma

Abstract

Low circulating cholesterol concentration is associated with elevated cancer incidence and mortality. However, the association between cholesterol levels and diffuse large B cell lymphoma (DLBCL) remains unknown. The aim of our study was to evaluate the prognostic value of serum lipid profile in DLBCL. Five hundred and fifty enrolled subjects with detailed serum lipid levels at diagnosis of DLBCL were randomly divided into a training set (n = 367) and a validation set (n = 183) (ratio, 2:1). Multivariate Cox regression analyses screened the prognostic factors associated with progression-free survival (PFS) and overall survival (OS). Performances of models were compared using C-index and area under the curve in internal and external validation. The results showed that decreased levels of total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C) and low-density lipoprotein cholesterol (LDL-C) were associated with unfavorable PFS and OS in the rituximab era, and concurrently low HDL-C together with low LDL-C was an independent prognostic indicator for both PFS and OS. Patients achieving complete remission or partial remission after 6-8 circles of chemotherapies had significantly increased cholesterol levels compared to the levels at DLBCL diagnosis, and HDL-C or LDL-C elevations were correlated with better survival. Furthermore, the predictive and discriminatory capacity of the National Comprehensive Cancer Network (NCCN)-International Prognostic Index (IPI) together with low cholesterol levels was superior to NCCN-IPI alone both in the training and validation set. In conclusion, serum cholesterol levels are simple and routinely tested parameters, which may be good candidates for predicting prognosis in the future clinical practice of DLBCL.

<p>Statistical Challenges in Development of Prognostic Models in Diffuse Large B-Cell Lymphoma: Comparison Between Existing Models – A Systematic Review</p>

“…24 Seven studies of 46 used training and validation sets to develop a prognostic index and to validate it. 1,2,16,31,[34][35][36] Among the studies using nomograms, three used both internal and external validation, [26][27][28] one used internal validation only, 25 and two did not assess the performance through validation. 29,30 Model performance was evaluated using different methods, which are discussed in the following subsections.…”

“…As a measure of discrimination, the area under the receiver operating characteristics (ROC) curve for survival outcomes (the c-index) was used in seven of 46 studies. 7,16,32,34,[37][38][39] Higher values of the c-index indicated better discrimination. The value of the c-index for the novel prognostic models ranged from 0.708 for the comorbidity National Comprehensive Cancer Network-IPI (cNCCN-IPI) to 0.830 for the lipoprotein prognostic index (Lipo-PI).…”

“…The c-index was also calculated for the indices that were previously reported, such as IPI (range 0.67-0.76), revised IPI (c-index 0.642), and NCCN-IPI (range 0.6950.791). Two studies pointed to the use of Harrell's c-index/modified method as a discrimination method, 16,32 while one reported the use of the c-index according to Uno et al 22…”

“…9,33,[49][50][51] However, the majority of studies used institutional upper/lower limits of normal (ULN/LLN) for the continuously distributed measurements, including hemoglobin levels, platelet counts, 52 absolute monocyte count, absolute neutrophil count, B2M, LDH, 37 B2M, 36,45,53,54 and lipid levels. 16 Three studies used other statistical techniques (eg percentile value, medians) for testing different cut-off points of continuous measurements. 8,33,55…”

“…Despite the standard treatment with chemoimmunotherapy, approximately 30% of patients experience disease relapse or are refractory to therapy. 16 Therefore, the utilization of selected group of patients' characteristics through a statistical model is important to identify high-risk patients who could potentially benefit from more intensive immunochemotherapy and/or molecular-targeted agents. 17 Prognostic risk models, which are a mathematically precise way to summarize properties of measurements and their associations, are created to predict events.…”