Morales Chacón Lilia María, Galán García Lídice, Cruz Hernández Tania Margarita, Pavón Fuentes Nancy, Maragoto Rizo Carlos, Morales Suarez Ileana, Morales Chacón Odalys, Abad Molina Elianne, Rocha Arrieta Luisa
International Center for Neurological Restoration, Havana 11300, Cuba.
Cuban Neurosciences Center, Havana 11300, Cuba.
Behav Sci (Basel). 2022 Jul 15;12(7):234. doi: 10.3390/bs12070234.
To identify clinical phenotypes and biomarkers for best mortality prediction considering age, symptoms and comorbidities in COVID-19 patients with chronic neurological diseases in intensive care units (ICUs). Data included 1252 COVID-19 patients admitted to ICUs in Cuba between January and August 2021. A k-means algorithm based on unsupervised learning was used to identify clinical patterns related to symptoms, comorbidities and age. The Stable Sparse Classifiers procedure (SSC) was employed for predicting mortality. The classification performance was assessed using the area under the receiver operating curve (AUC). Six phenotypes using a modified v-fold cross validation for the k-means algorithm were identified: phenotype class 1, mean age 72.3 years (ys)-hypertension and coronary artery disease, alongside typical COVID-19 symptoms; class 2, mean age 63 ys-asthma, cough and fever; class 3, mean age 74.5 ys-hypertension, diabetes and cough; class 4, mean age 67.8 ys-hypertension and no symptoms; class 5, mean age 53 ys-cough and no comorbidities; class 6, mean age 60 ys-without symptoms or comorbidities. The chronic neurological disease (CND) percentage was distributed in the six phenotypes, predominantly in phenotypes of classes 3 (24.72%) and 4 (35,39%); χ² (5) 11.0129 = 0.051134. The cerebrovascular disease was concentrated in classes 3 and 4; χ² (5) = 36.63, = 0.000001. The mortality rate totaled 325 (25.79%), of which 56 (17.23%) had chronic neurological diseases. The highest in-hospital mortality rates were found in phenotypes 1 (37.22%) and 3 (33.98%). The SSC revealed that a neurological symptom (ageusia), together with two neurological diseases (cerebrovascular disease and Parkinson's disease), and in addition to ICU days, age and specific symptoms (fever, cough, dyspnea and chilliness) as well as particular comorbidities (hypertension, diabetes and asthma) indicated the best prediction performance (AUC = 0.67). : The identification of clinical phenotypes and mortality biomarkers using practical variables and robust statistical methodologies make several noteworthy contributions to basic and experimental investigations for distinguishing the COVID-19 clinical spectrum and predicting mortality.
为了确定在重症监护病房（ICU）中患有慢性神经系统疾病的COVID-19患者的临床表型和生物标志物，以实现基于年龄、症状和合并症的最佳死亡率预测。数据包括2021年1月至8月期间古巴ICU收治的1252例COVID-19患者。基于无监督学习的k均值算法用于识别与症状、合并症和年龄相关的临床模式。采用稳定稀疏分类器程序（SSC）预测死亡率。使用受试者操作特征曲线下面积（AUC）评估分类性能。通过对k均值算法使用改进的v折交叉验证，确定了六种表型：表型1，平均年龄72.3岁，患有高血压和冠状动脉疾病，伴有典型的COVID-19症状；表型2，平均年龄63岁，患有哮喘、咳嗽和发热；表型3，平均年龄74.5岁，患有高血压、糖尿病和咳嗽；表型4，平均年龄67.8岁，患有高血压且无症状；表型5，平均年龄53岁，有咳嗽且无合并症；表型6，平均年龄60岁，无任何症状或合并症。慢性神经系统疾病（CND）的比例分布在这六种表型中，主要集中在表型3（24.72%）和表型4（35.39%）；χ²（5）=11.0129，P = 0.051134。脑血管疾病集中在表型3和表型4中；χ²（5）=36.63，P = 0.000001。死亡率总计325例（25.79%），其中56例（17.23%）患有慢性神经系统疾病。住院死亡率最高的是表型1（37.22%）和表型3（33.98%）。SSC显示，一种神经症状（味觉丧失）、两种神经疾病（脑血管疾病和帕金森病），以及除ICU住院天数、年龄和特定症状（发热、咳嗽、呼吸困难和寒战）之外，还有特定的合并症（高血压、糖尿病和哮喘）显示出最佳的预测性能（AUC = 0.67）。使用实际变量和稳健的统计方法识别临床表型和死亡率生物标志物，为区分COVID-19临床谱和预测死亡率的基础研究和实验研究做出了一些值得注意的贡献。
