Investigation into prognostic factors of early recurrence and progression of previously untreated diffuse large B-cell lymphoma and statistical prediction model of POD12

Ke Lian¹Zhu Wenyao²Zhihui Hu¹Fang Su³Caixia Xu⁴Hui Wang^5*

• ¹Shanxi Bethune Hospital, Shanxi Medical University, Taiyuan, Shanxi Province, China
• ²The Affiliated Yantai Yuhuangding Hospital of Qingdao University, Yantai, China
• ³Shanxi Provincial Cancer Hospital, Taiyuan, Shanxi Province, China
• ⁴Houma People’s Hospital, Houma, Shaanxi Province, China
• ⁵Zhejiang University of Finance and Economics, Hangzhou, China

Objective: To evaluate the incidence, prognostic value and risk factors of progression of disease within 12 months (POD12) in the patients with diffuse large B-cell lymphoma (DLBCL). Methods: Retrospective analysis of the clinical, pathological and follow-up data was carried out on 69 DLBCL cases in Shanxi Bethune hospital from January 2016 to June 2020. One-way ANOVA and multivariate Cox regression analysis were used to explore the correlation between POD12 and prognosis, and the logistic regression analysis was used to explore the risk factors of POD12, accompanied by prediction model based on the convolutional neural networks and long short-term memory (CNN-LSTM) and particle swarm optimization and general regression neural network (PSO-GRNN) models. Results: (1) POD12 is significantly correlated with PFS (P<0.001) and OS (P=0.008). (2) From the univariate logistic regression analysis corrected by the first-line chemotherapy regimen that LDH, β2-MG, Stage, ECOG, NLR and SII are as the risk factors for POD12 (P<0.1), while β2-MG and ECOG are as the independent risk factors for POD12 from multivariate logistic regression analysis (p<0.05). (3) The prediction model of POD12 is established based on LDH, β2-MG, Stage, ECOG, NLR, SII indicators. AUC is 0.846 (95% CI: 0.749~0.944, P<0.001), suggesting that the prediction model is reasonable. A prediction method of the characteristic variables of the risk of POD12 by the CNN-LSTM deep learning model based on the chaotic time series is proposed. By comparison, the CNN-LSTM and PSO-GRNN models are the most suitable to predict the risk level of the POD12 in the future.