论文题名(中文): | 重点呼吸道病毒感染入院患者低淋巴细胞血症与肺外器官损伤的关联与动态预警研究 |
姓名: | |
论文语种: | chi |
学位: | 博士 |
学位类型: | 学术学位 |
学校: | 北京协和医学院 |
院系: | |
专业: | |
指导教师姓名: | |
论文完成日期: | 2025-05-09 |
论文题名(外文): | Research on the Dynamic Prediction and the Association Between Lymphopenia and Extrapulmonary Organ Injuries Among Patients Hospitalized with Major Respiratory Virus Infections |
关键词(中文): | |
关键词(外文): | Major respiratory virus infection Extrapulmonary organ damage Dynamic lymphocyte count change Dynamic model for early warning |
论文文摘(中文): |
研究背景 流感病毒和严重急性呼吸综合征冠状病毒2 (Severe acute respiratory syndrome coronavirus 2,SARS-CoV-2,简称新冠病毒),是下呼吸道感染病毒感染中的主要病原体,主要侵害肺脏,亦可引发除肺脏以外的其他器官损害,常见的包括静脉血栓栓塞症(Venous thromboembolism,VTE)和急性肾损伤(Acute kidney injury,AKI)。淋巴细胞在清除病毒、调节炎症反应和控制病情中至关重要。在病情较为严重时,病毒感染和细胞因子风暴会诱导淋巴细胞大量凋亡和耗竭,临床表现为低淋巴细胞血症(Lymphopenia)。低淋巴细胞血症已被确定与疾病严重程度和死亡密切相关,然而低淋巴细胞血症与VTE、AKI等肺外器官损伤是否相关的研究较少,证据有限。此外,患者住院过程中产生的动态数据亦未被充分利用,通过探索淋巴细胞的动态变化模式、选择合适的统计学模型来全面描绘疾病变化规律可能是肺外器官损伤和院内死亡结局预警、进而指导临床治疗的关键一步,但目前该项内容的探索尚属空白。 鉴于当前研究进展存在的不足,有必要对低淋巴细胞血症与以VTE、AKI为代表的肺外器官损伤和院内死亡结局的关联进行分析,并探索可充分利用住院期间淋巴细胞纵向时序数据对肺外器官损伤和院内死亡结局进行动态预警的新模式,为新冠/流感病毒感染患者救治提供科学依据。 研究目的 1. 描绘新冠/流感病毒感染入院患者的肺外器官损伤特点,探索入院低淋巴细胞血症和相关外周血细胞计数指标与以VTE、AKI为代表的肺外器官损伤和死亡结局的关联; 2. 探索住院期间淋巴细胞的动态变化模式,从低淋巴细胞血症持续时间角度分析其与各个结局之间的关联; 3. 探讨不同动态模型对淋巴细胞纵向变化的拟合效果,分析比较不同模型对各类结局的预警价值。 研究方法 1. 采用回顾性队列研究,纳入2016年1月1日至2023年12月31日期间中日友好医院由于呼吸道病毒感染入院的患者,对不同结局患者群体采用与之相对应的纳排标准。将感染分为2类:新冠病毒感染或流感病毒感染。通过查阅文献结合既往临床经验支持,对不同的研究结局设置了不同的协变量调整:对VTE结局,调整年龄、性别、激素使用、免疫抑制、Charlson共病指数、预防性抗凝和入院D-二聚体水平;对于AKI结局,调整了年龄、机械通气和慢性肾脏病;对于死亡结局,调整了年龄,激素使用,CKD,慢性呼吸病和心血管病和入院低淋巴细胞血症等协变量。采用多因素Logistic回归模型、多因素Cox比例风险回归模型分别估计调整协变量之后的入院基线低淋巴细胞血症与住院期间发生VTE、AKI和全因死亡之间的统计学关联,并进一步分别探索了入院低淋巴细胞血症与早期AKI、晚期AKI和AKI恢复之间的关系。此外,由于住院期间的病死率较高,将院内死亡作为VTE和AKI结局的竞争风险,进行了竞争风险的多变量Fine-Gray模型分析。淋巴细胞相关外周血计数比率也同时纳入了分析。 2.采用广义线性混合模型,分析淋巴细胞绝对数与凝血指标/相关炎症因子的动态变化之间的关联。定义低淋巴细胞血症最长持续时间,并通过多变量logistic回归,定量估计住院期间低淋巴细胞血症最长持续天数分别为1 - 3天、4 - 7天和7天以上时的各结局关联强度。 3. 分析比较界标模型、联合模型和深度学习的长短时记忆网络(LSTM)模型在拟合纵向淋巴细胞计数的连续变化预警不同结局时的表现,并从区分度和校准度两方面对其进行评估;开发霍克斯过程联合生存结局的联合模型,探索低淋巴细胞血症发生发展过程中不同的激发-衰减模式及对不同结局的预警价值。 研究结果 1. VTE结局的分析共纳入3484例患者,其中新冠病毒感染患者2640例,流感病毒感染患者844例,2组VTE发生率分别为5.04%和2.61%。调整多因素后,各病毒感染入院患者入院淋巴细胞计数、低淋巴细胞血症与院内VTE无统计学关联。AKI结局的分析纳入2542例患者,其中新冠病毒1945例,流感病毒597例,AKI发生率分别为18.0%和23.3%。新冠病毒和流感病毒的入院淋巴细胞计数减少亦可预警AKI发生。死亡结局的分析纳入3554例患者,其中新冠病毒2699人,流感病毒855人,死亡率分别为11.7%和10.1%。新冠病毒、流感病毒的感染入院患者入院低淋巴细胞血症与AKI和死亡的相关性均具有统计学意义。 2. 住院期间的淋巴细胞计数动态变化与各炎症指标总体上均呈负相关。对新冠病毒感染,持续低淋巴细胞血症是VTE(持续4 - 7天时,比值比【Odds ratio,OR】2.06,95% 置信区间【Confidence interval,CI】1.07 - 3.96,P = 0.031)、AKI(1 - 3 天, OR 4.55,95% CI 2.57 – 8.06, P < 0.001;4 - 7 天, OR 3.92,95% CI 2.17 – 7.10, P < 0.001;> 7 天, OR 2.20,95% CI 1.20 – 4.03, P = 0.010)和死亡(1 - 3 天, OR 1.96,95% CI 1.05 – 3.64, P = 0.033;4 - 7 天, OR 3.55,95% CI 1.86 – 6.79, P < 0.001;> 7 天, OR 4.05,95% CI 2.12 – 7.74, P < 0.001)的危险因素;对于流感病毒感染,持续低淋巴细胞血症是AKI(1 - 3 天, OR 17.77,95% CI 6.74 – 46.87, P < 0.001;4 - 7 天, OR 10.81,95% CI 3.83 – 30.53, P < 0.001;> 7 天, OR 4.28,95% CI 1.38 – 13.32, P = 0.012)和死亡(1 - 3 天, OR 9.65,95% CI 2.59 – 35.88, P = 0.001;4 - 7 天, OR 27.41,95% CI 7.16 – 104.96, P < 0.001;> 7 天, OR 32.42,95% CI 8.29 – 126.81, P < 0.001)的危险因素,且可见流感病毒的关联强度高于新冠病毒。 3. 综合界标模型、线性混合效应-生存结局联合模型、霍克斯过程-生存结局联合模型和LSTM的拟合结果可见,各模型对住院期间淋巴细胞计数的动态变化与各结局之间的关联分析结果接近,即新冠病毒感染后淋巴细胞计数的动态增加可减少院内VTE的发生风险,且提示该效应随时间增加而增强,而流感病毒感染后则无上述关联和变化;新冠和流感感染患者的淋巴细胞计数动态增加可降低AKI和死亡风险,且强度类似。模型表现来看,界标模型的运算效率最快,校准度最好(Brier分值均低于0.10),然而区分度在不同结局之间差异较大(曲线下面积【AUC】在0.60 ~ 0.81之间波动);LSTM的区分度最好(AUC在0.76 ~ 0.92波动),但可解释性最差;线性混合效应-Cox联合模型的计算速度较慢,区分度和校准度均为中等(AUC在0.62 ~ 0.77之间,Brier分值最大超过0.15),但可以给出纵向趋势变化与结局的参数解释;霍克斯过程-Cox联合模型计算速度最慢,参数信息最为丰富,但是区分度和校准度均较差(AUC在0.48 ~ 0.78;Brier分值均大于0.25)。 研究结论 1. 新冠/流感病毒感染患者的入院低淋巴细胞血症与AKI和死亡结局相关。 2. 入院基线的淋巴细胞相关外周血细胞计数比率,包括NLR,MLR、NPR和PLR等均不同程度与VTE、AKI和死亡结局相关,且在不同病毒感染中时是否具有统计学差异和关联强度大小有所不同。 3. 新冠/流感病毒感染住院患者的淋巴细胞动态变化在预警不同的肺外器官损伤和死亡结局时各具不同特点,宜重视低淋巴细胞血症的早期发生和持续。 4. 动态预警方面,界标模型表现较好,而联合模型灵活性较强,均有进一步研究应用的潜力。 |
论文文摘(外文): |
Background Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and influenza virus are major etiological agents of lower respiratory tract viral infections. Both viruses primarily attack the lungs, while extrapulmonary organ injuries, such as acute kidney injury (AKI) and venous thromboembolism (VTE), are also commonly observed in severe cases. Lymphocytes play a pivotal role in viral clearance and inflammatory modulation. However, during severe infections, viral invasion and cytokine storms induce extensive lymphocyte apoptosis and exhaustion, manifesting as lymphopenia. While lymphopenia has been strongly associated with disease severity and mortality, limited evidence exists regarding its correlation with extrapulmonary organ injuries like AKI and VTE. Furthermore, dynamic longitudinal data on lymphocyte counts during hospitalization remains underutilized. Comprehensive characterization of lymphocyte dynamic patterns and fitting suitable statistical models could provide critical insights for early warning of extrapulmonary complications and mortality, yet this area remains unexplored. Considering the limitations of the current research progress, it is essential to investigate the association between lymphopenia and extrapulmonary organ injury as well as mortality outcomes such as VTE and AKI. Additionally, there is a need to develop a novel model capable of leveraging the longitudinal time-series data of lymphocytes during hospitalization for dynamic prediction of extrapulmonary organ injury and mortality outcomes. This will provide a robust scientific foundation for the prevention and treatment of respiratory viral infections in China. Objectives 1. To describe the characteristics of extrapulmonary organ injury in hospitalized patients with SARS-CoV-2 / influenza virus infections and evaluate the associations between baseline lymphopenia/peripheral blood indices and outcomes (VTE, AKI, mortality). 2. To analyze dynamic lymphocyte changes and assess correlations between prolonged lymphopenia and clinical outcomes. 3. To compare the predictive performance of landmark model, joint models (linear mixed-effects sub-model and survival sub-model, and Hawkes sub-model and survival sub-model), and deep learning model (long-short time memory, LSTM) for the early warning of worse outcomes. Methods 1. A retrospective cohort study was conducted, in which hospitalized patients admitted to China-Japan Friendship Hospital between January 1, 2016, and December 31, 2023, were enrolled. Inclusion and exclusion criteria were tailored to distinct patient subgroups based on clinical outcomes. Infections were categorized into two groups: SARS-CoV-2 infection or influenza virus infection. Covariate adjustments were implemented according to different study outcomes, supported by literature review and prior clinical expertise. For VTE outcomes, adjustments were made for age, sex, glucocorticoid use, immunosuppression, Charlson Comorbidity Index, prophylactic anticoagulation, and admission D-dimer levels. For AKI outcome, covariates including age, mechanical ventilation, and chronic kidney disease were adjusted. In mortality analyses, adjustments included age, glucocorticoid use, chronic kidney disease, chronic respiratory disease, cardiovascular disease, and admission lymphopenia. Multivariable logistic regression models and Cox proportional hazards regression models were utilized to estimate the adjusted associations between admission lymphopenia and the risks of VTE, AKI, and all-cause mortality during hospitalization. Furthermore, the relationships between admission lymphopenia and early AKI, late AKI, and AKI recovery were explored. Due to the high mortality rate during hospitalization, death was treated as a competing risk for VTE and AKI outcomes, and multivariable Fine-Gray sub-distribution hazard models were employed for competing risk analyses. Lymphocyte-related peripheral blood count ratios were also analyzed to assess their potential associations with clinical outcomes. 2. A generalized linear mixed model was used to analyze the association between absolute lymphocyte count and the dynamic changes of coagulation indicators/inflammatory factors. The longest duration of lymphopenia was defined, and the association strength of each outcome when the longest duration of lymphopenia during hospitalization was 1-3 days, 4-7 days, and more than 7 days was quantitatively estimated through multivariate logistic regression. 3. The performance of landmark models, joint models, and LSTM model in fitting the continuous changes of longitudinal lymphocyte counts to predict different outcomes was analyzed and compared, and they were evaluated from the aspects of discrimination and calibration. A joint model of Hawkes process and survival outcome was developed to explore different excitation-decay patterns during the occurrence and development of lymphopenia and their predictive value for different outcomes Results 1. A total of 3,484 patients were included in the VTE analysis, among which 2,640 were infected with the SARS-CoV-2 and 844 with influenza virus. The incidence rates of VTE in the three groups were 5.04% and 2.61%, respectively. After adjusting for multiple factors, the lymphocyte counts at admission and lymphopenia in patients admitted with each virus infection were not significantly associated with in-hospital VTE. A total of 2,542 patients were included in the AKI analysis, including 1,945 with SARS-CoV-2 infection and 597 with influenza virus infection. The incidence rates of AKI were 18.0% and 23.3%, respectively. Lymphopenia at admission was significantly associated with AKI in patients admitted with each virus infection, except for other respiratory viruses. A reduction in the lymphocyte count at admission could also predict AKI in patients with SARS-CoV-2 and influenza virus infections. A total of 3,554 patients were included in the analysis of the death outcome, including 2,699 with SARS-CoV-2 infection and 855 with influenza virus infection. The mortality rates were 11.7% and 10.1%, respectively. Lymphopenia at admission was significantly associated with death in patients admitted with SARS-CoV-2 and influenza virus infections. 2. The dynamic changes in lymphocyte count during hospitalization were generally negatively correlated with various inflammatory indicators. For SARS-CoV-2 infection, persistent lymphopenia was a risk factor for VTE (lasting for 4 – 7 days, OR 2.06, 95% CI 1.07 - 3.96,P = 0.031), AKI (1 - 3 days, OR 4.55,95% CI 2.57 – 8.06, P < 0.001; 4 - 7 days, OR 3.92,95% CI 2.17 – 7.10, P < 0.001; > 7 days, OR 2.20,95% CI 1.20 – 4.03, P = 0.010), and death (1 - 3 days, OR 1.96,95% CI 1.05 – 3.64, P = 0.033; 4 - 7 days, OR 3.55,95% CI 1.86 – 6.79, P < 0.001; > 7 days, OR 4.05,95% CI 2.12 – 7.74, P < 0.001); for influenza, persistent lymphopenia was a risk factor for AKI (1 - 3 days, OR 17.77,95% CI 6.74 – 46.87, P < 0.001; 4 - 7 days, OR 10.81,95% CI 3.83 – 30.53, P < 0.001; > 7 days, OR 4.28,95% CI 1.38 – 13.32, P = 0.012) and death (1 - 3 days, OR 9.65,95% CI 2.59 – 35.88, P = 0.001; 4 - 7 days, OR 27.41,95% CI 7.16 – 104.96, P < 0.001; > 7 days, OR 32.42,95% CI 8.29 – 126.81, P < 0.001), and the association was stronger for influenza virus than that for SARS-CoV-2. 3. A comprehensive comparison of landmark models, linear mixed-effects-survival joint models, Hawkes process-survival joint models, and LSTM fitting results revealed that the associations between dynamic changes in lymphocyte counts during hospitalization and clinical outcomes were similar across models. Specifically, dynamic increases in lymphocyte counts after SARS-CoV-2 infection were associated with reduced in-hospital VTE risk, with strengthening over time, whereas no such associations or temporal trends were observed in influenza virus infection. In both COVID-19 and influenza patients, dynamic increases in lymphocyte counts were linked to reduced risks of AKI and mortality, with comparable effect magnitudes. Regarding model performance: Landmark models demonstrated the fastest computational speed and best calibration (all Brier score < 0.10), but discrimination varied significantly across outcomes (varying from 0.60 ~ 0.81). LSTM achieved the highest discrimination but had the poorest interpretability. Linear mixed-effects-Cox joint models showed intermediate discrimination (AUC varying from 0.62 ~ 0.77) and calibration (max Brier score > 0.15), slower computational speed, but provided parametric interpretations of longitudinal trends and outcomes. Hawkes process-Cox joint models had the slowest computational speed and the richest parametric information, yet exhibited the poorest discrimination and calibration (AUC varying from 0.48 ~ 0.78; all Brier > 0.25). Conclusions 1. Lymphopenia on admission in patients with respiratory virus infection was associated with AKI and mortality outcomes. 2. Lymphocyte-related peripheral blood count ratios, including NLR, MLR, NPR, and PLR on admission, were associated with VTE, AKI, and mortality outcomes, while the significances and magnitude of the association differed among different respiratory virus infections. 3. The dynamic changes of lymphopenia in hospitalized patients with respiratory virus infection may predict extrapulmonary organ damage and mortality outcomes, and the occurrence and persistence of lymphopenia should be given more attention. 4. In dynamic early warning, superior performance of landmark models was observed, while joint models demonstrated stronger flexibility, and both were identified to hold potential for further research and application. |
开放日期: | 2025-06-30 |