Immune cell abundance model to predict survival prognosis of patients with sepsis：a deep learning study

doi:10.3969/j.issn.1009-976X.2022.02.007

Abstract

Abstract: Objective To development of a deep learning immune cell abundance model to predict 28-day survival in patients with sepsis. Methods In this study, a total of 479 sepsis patients were included and patients were randomly divided into training and validation groups in a 9：1 ratio. We built the DeepSurv in TensorFlow, a deep learning survival neural network-based model on 431 sepsis patients′ data with 28 immune cells in training cohort from prospective cohort study （MARS study）. The algorithm was internally validated on 48 sepsis patients from validation group. The primary end point was 28-day survival, and the area under curve （AUC） were used for model evaluation. Results In the training group, we established a deep learning survival neural network model showed promising results to predict 28-day survival in sepsis patients, patients with low vs high risk score had statistically significantly longer 28-day survival （HR=0.022, 95%CI： 0.013~0.038, P<0.005）. The immune cell abundance risk score was associated with 28-day survival （AUCs for 14- and 28-day survival were 0.912 and 0.936, respectively）. Similarly, in the test group, patients with low vs high risk score had statistically significantly longer 28-day survival （HR=0.07, 95%CI： 0.008~0.63, P<0.005）, the AUCs for 14- and 28-day survival were 0.822 and 0.777 respectively. Further, this study identified that model obviously related to immune microenvironment characteristics. Conclusion This study developed and validated novel deep learning survival neural network model showed reliable individual 28-day survival information in prognostic evaluation and treatment recommendation in patients with sepsis.

Key words: sepsis, deep learning, prognostic model, immune microenvironment

摘要： 目的构建一种基于深度学习算法的免疫细胞丰度模型预测脓毒血症患者生存预后。方法本研究共纳入479例脓毒症患者,将患者按9：1的比例随机分为训练队列和验证队列,在TensorFlow中构建了一个基于深度学习生存神经网络的seDNT模型,根据前瞻性研究队列MARS研究的431名脓毒症患者的数据组成训练组。此外,该算法在验证组的48例脓毒症患者中进行内部验证,研究的主要终点为28天生存期,模型的评价使用曲线下面积（AUC）。结果在训练组中,深度学习生存神经网络模型对脓毒症患者28天生存期的预测效果良好,低风险评分患者与高风险评分患者28天生存期的预后差异具有统计学意义（HR=0.022,95%CI=0.013-0.038,P<0.005）。免疫细胞丰度风险评分与28天生存率相关（14和28天生存率的AUC分别为0.912和0.936）。同样,验证组中低风险评分患者与高风险评分患者28天生存预后较好,差异有统计学意义（HR=0.07,95%CI=0.008~0.63,P<0.005）,14天和28天生存期的AUC分别为0.822和0.777。此外,本研究还显示该风险评分与免疫微环境有明显的相关性。结论本研究构建并验证了一种新的深度学习生存神经网络模型,该模型可以准确预测脓毒症患者可靠的28天生存率,提供了预后评估和治疗建议。

关键词: 脓毒症, 深度学习, 预后模型, 免疫微环境

CLC Number:

R631

GU Yang, LIU Xun, OU Qi-yun, ZHANG Na, LI Han, QIN Wei-qiang, YU Tao, LI Li. Immune cell abundance model to predict survival prognosis of patients with sepsis：a deep learning study[J]. Lingnan Modern Clinics In Surgery, 2022, 22(02): 154-162.

顾杨, 刘勋, 区绮云, 张娜, 李涵, 秦伟强, 李莉. 基于深度学习算法的免疫细胞丰度模型预测脓毒症患者的生存预后[J]. 岭南现代临床外科, 2022, 22(02): 154-162.

References

[1] Islam MM, Nasrin T, Walther BA, et al.Prediction of sepsis patients using machine learning approach: a meta-analysis[J]. Comput Methods Programs Biomedi, 2019, 170: 1-9.
[2] Cheng Z, Abrams ST, Toh J, et al.The critical roles and mechanisms of immune cell death in sepsis[J]. Front Immunol, 2020, 11: 1918.
[3] Schrijver IT, Théroude C, Roger T.Myeloid-derived suppressor cells in sepsis[J]. Front Immunol, 2019, 10: 327.
[4] Chen X, Wei Q, Hu Y, et al.Role of Fractalkine in promoting inflammation in sepsis-induced multiple organ dysfunction[J]. Infect Genet and Evol, 2020, 85: 104569.
[5] van der Poll T, Shankar-Hari M, Wiersinga WJ. The immunology of sepsis[J]. Immunity, 2021, 54(11): 2450-2464.
[6] Chaudhary K, Vaid A, Duffy Á, et al.Utilization of deep learning for subphenotype identification in sepsis-associated acute kidney injury[J]. Clin J Am Soc Nephrol, 2020, 15(11): 1557-1565.
[7] Peters-Sengers H, Butler JM, Uhel F, et al.Source-specific host response and outcomes in critically ill patients with sepsis: a prospective cohort study[J]. Intensive Care Med, 2022, 48(1): 92-102.
[8] Finotello F, Trajanoski Z.Quantifying tumor-infiltrating immune cells from transcriptomics data[J]. Cancer Immunol Immunother, 2018, 67(7): 1031-1040.
[9] Katzman JL, Shaham U, Cloninger A, et al.DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network[J]. BMC Med Res Methodol, 2018, 18(1): 1-12.
[10] Sperk M, Domselaar RV, Neogi U.Immune checkpoints as the immune system regulators and potential biomarkers in HIV-1 infection[J]. Int J Mol Sci, 2018, 19(7): 2000.
[11] Letourneur D, Danlos FX, Marabelle A.Chemokine biology on immune checkpoint-targeted therapies[J]. Eur J Cancer, 2020, 137: 260-271.
[12] Jiang W, Li X, Ding H, et al.PD-1 in Tregs predicts the survival in sepsis patients using sepsis-3 criteria: A prospective, two-stage study[J]. Int Immunopharmacol, 2020, 89(Pt A):107175.
[13] Maiza H,Leca G,Mansur IG,et al.A novel 80-kD cell surface structure identifies human circulating lymphocytes with natural killer activity[J]. J Exp Med,1993,178(3):1121-1126.
[14] Inoue S, Bo L, Bian J, et al.Dose dependent effect of anti-CTLA-4 on survival in sepsis[J]. Shock, 2011, 36(1): 38-44.
[15] Karbian N, Abutbul A, El-Amore R, et al.Apoptotic cell therapy for cytokine storm associated with acute severe sepsis[J]. Cell Death Dis, 2020, 11(7): 535.
[16] Chousterman BG, Boissonnas A, Poupel L, et al.Ly6Chigh monocytes protect against kidney damage during sepsis via a CX3CR1-dependent adhesion mechanism[J]. J Am Soc Nephrol, 2016, 27(3): 792-803.
[17] Hoogendijk AJ, Wiewel MA, van Vught LA, et al. Plasma fractalkine is a sustained marker of disease severity and outcome in sepsis patients[J]. Crit Care, 2015, 19: 412.