Subjects
Patients with COVID-19 in Bangladesh
Patients with clinically suspected SARS-CoV-2 infection who visited Evercare Hospital Dhaka between December 25, 2021 and September 21, 2022 were considered for this study. Most of these cases coincide with the sixth and part of the seventh wave in Japan. Patients were admitted immediately after the onset of symptoms. Outcomes after hospitalization were classified as Mild, Moderate, or Severe (Table 1). From those considered, 129 untreated COVID-19 patients who visited Evercare Hospital within 7 days of onset were included. All patients presented to the hospital in the Mild stage; 64 subsequently improved to Mild, 46 in Moderate, and 19 in Severe. The mean onset dates of patients were Mild: 2.3±0.12, Moderate: 2.3±0.14, and Severe: 3.0±0.37(mean±SE) days.
The mean date of admission was Mild: 3.08±0.32, Moderate: 3.24±0.45, and Severe: 3.71±0.21 days after the onset of symptoms. To facilitate comparison with Japanese data, sera from 23 healthy Japanese were prepared (66.9±3.3 years). These were measured simultaneously with the Bangladeshi samples.
Followings the hospital COVID-19 taskforce management guideline, disease was categorized as Mild, Moderate, or Severe. Disease severity was not determined on the patients’ first visit, rather it was mentioned on their case sheets at the they were discharged from the hospital. Laboratory confirmed Mild COVID-19 cases were those with one or more signs and symptoms of COVID-19 (e.g., fever, cough, runny nose, fatigue, headache, nausea, vomiting, diarrhea, chest pain, abdominal pain, and loss of taste or smell), but without shortness of breath, dyspnea on exertion, and abnormal radiological findings. Laboratory confirmed Moderate COVID-19 cases were those with pneumonia, blood oxygen saturation >93% on room air, and patients that may have required low oxygen support. Severe cases developed COVID-19 pneumonia and required hospitalization, dyspnea, respiratory frequency ≥30 breaths /min, blood oxygen saturation ≤93% on room air, lung infiltrates in >50%, and may have required mechanical ventilation and/or ICU support.
Patients with COVID-19 in Japan
This sample consisted of patients with clinical suspicion of SARS-CoV-2 infection who were admitted to Habikino hospital, and Tokushukai Hospital from the end of June 2020 to the middle of June 2022. All patients provided written informed consent and the study was approved by the Ethics Committee of Osaka Habikino Medical Center (Approved ID: 150-7), Tokushukai Hospital (TGE01547) and Louis Pasteur Center for Medical Research (LPC.29). This study followed the principles of the Declaration of Helsinki, and was approved by the institutional review board of Osaka University Hospital (No- 885). Data for healthy subjects were obtained from Louis Pasteur Center for Medical Research (LPC.8 and LPC.25).
In Japan, the disease severity of patients was determined at hospital admission according to The Guideline for Medical Treatment of COVID-19(https://www-mhlw-go-jp/content/ 000785119-pdf). Briefly, patients with “Mild” illness showed one or some of the signs and symptoms of COVID-19 (e.g., fever, cough, sore throat, malaise, headache, muscle pain, nausea, vomiting, diarrhea, and loss of taste and smell), but lacked shortness of breath, dyspnea on exertion, and abnormal imaging findings. “Moderate I” cases showed evidence of lower respiratory disease with a percutaneous oxygen saturation (SpO2) of >93% on room air, and were compatible with “moderate illness” described in the Coronavirus Disease 2019 (COVID-19) Treatment Guidelines of the National Institutes of Health (NIH; https://www-covid19treatmentguidelines-nih-gov). “Moderate II” cases were supported with non-invasive mechanical ventilation or supplemental oxygen (including high-flow oxygen devices), and were compatible with “severe illness” described in the NIH guidelines. “Severe” cases were admitted into the intensive care unit or supported with invasive mechanical ventilation or extracorporeal membrane oxygenation, and were compatible with “critical illness” described in the NIH guidelines. In Japan, invasive ventilation was not used for some terminal cases.
As indicated in Table 1, 197 untreated COVID-19 patients who visited Habikino and Tokushukai Hospitals within 7 days of onset were included. The mean onset dates of patients were Mild: 3.08±0.32, Moderate: 3.24±0.45, and Severe: 3.71±0.21 days.
In Japan, COVID-19 is classified as Mild, Moderate I, Moderate II, and Severe, but Moderate II and Severe were combined into Severe so that they could be compared with Bangladesh. The severity of illness here refers to the final outcome of the patient and not their condition at the point of admission to hospital. The breakdown of patients, the severity classification and age distribution are shown in Table 1. Data from 91 healthy Japanese subjects with an average age of 63.6±1.9 years is also shown.
Cytokines/chemokines/soluble receptors assay
Cytokines, chemokines, and soluble receptors were quantified using Bio-Plex 200 system, a multiplex cytokine array system (Bio-Rad Laboratories, CA, USA) according to the manufacturer's instructions. Blood sera from healthy subjects and COVID-19 patients were collected and centrifuged at 1600g for 10 min. Serum samples were frozen at −80 oC until they were analyzed. We simultaneously quantified cytokines, chemokines, and soluble receptors. The Bio-Plex Human Cytokine 48-Plex Panel and Inflammation Panel (Bio-Rad Laboratories, CA, USA) was used to simultaneously quantify 78 items: CTACK, Eotaxin, FGF basic, G-CSF, GM-CSF, GRO-α, HGF, IFN-α2, IFN-γ, IL-1α, IL-1β, IL-1ra, IL-2, IL-2Rα, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IL-10, IL-12(p40), IL-12(p70), IL-13, IL-15, IL-16, IL-17, IL-18, IP-10, LIF, MCP-1(MCAF), MCP-3, M-CSF, MIF, MIG, MIP-1α, MIP-1β, β-NGF, PDGF-BB, RANTES, SCF, SCGF-β, SDF-1α, TNF-α, TNF-β, TRAIL, VEGF) and inflammation panel (37 plex: APRIL, BAFF, CD30, CD163, Chitinase, sgp130, IFN-α 2, IFN-β, IFN-γ, sIL-6Ra, IL-10, IL-11, IL-12(p40), IL-12 (p70), IL-19, IL-20, IL-22, IL-26, IL-27, IL-28A, IL-29, IL-32, IL-34, IL-35, LIGHT, MMP-1, MMP-2, MMP-3, Osteocalcin, Osteopontin, Pentraxin-3, sTNF-R1, sTNF-R2, TSLP, TWEAK. The following were excluded from the inflammation panel for data analysis: IFN-α2, IFN-γ, IL-2, IL-8, IL-10, IL-12(p40), and IL-12(p70). Patients’ samples were measured several times over a two-year period. Since there were lot-to-lot and measurement-to-measurement errors, these data were corrected based on the values of healthy subjects. The following items, IL-19, IL-20, IL-26, IL-28A, IL-29, IL-35, LIGHT, were further excluded from the analysis because they had large inter-kit errors and low measurement sensitivity.
Statistical analysis
Cytokine, chemokine, and soluble receptor values were analyzed to determine whether the raw values or log-transformed values were more normally distributed. Based on the results of this analysis, the log-transformed values were used and ANOVA was performed. Quantitative data were presented as means ± SEM and the significance of the difference between the groups was evaluated using Dunnett’s test with a value of P < 0.05 considered significant. All statistical analyses were carried out with JMP 20.0.
The aim of the study was to use cytokines/chemokines/soluble receptors data collected within 7 days of COVID onset, to predict whether a patient would subsequently need hospitalization or deteriorate to a moderate disease state or worse. To achieve this, we employed a binary logistic regression model. To refine the predictors, we utilized the Least Absolute Shrinkage and Selection Operator (LASSO) regression to select relevant cytokines as candidate markers22). The selection of the optimal number of variables was guided by Leave-One-Out Cross Validation (LOO CV). LOO-CV is usually known to be over-trained. However, in this case, the correct generalization performance can be evaluated using validation data. This method does not need to consider variations due to partitioning (such as k-fold). Due to the disproportionate distribution of disease severities in the Bangladeshi COVID-19 patients, we applied weights to minimize potential bias in the data. The performance of the logistic regression model was assessed using both training and validation datasets, including determining the Area Under the Curve (AUC) from the Receiver Operating Characteristic (ROC) curve and calculating performance metrics such as specificity and sensitivity. All analyses were conducted using the R language v4.2 (https://www.r-project.org/), with the glmnet 4.1-7 package supporting variable selection and logistic regression analysis. Furthermore, the pROC package version 1.18.4 was utilized to evaluate the model's performance through the ROC curve and AUC.