Study design

This was an open, multicenter, non-interventional, prospective cohort study conducted in Japan. We collected information on study subjects who visited our institute for the diagnosis and treatment of pneumonia after obtaining the written informed consent. During routine medical care, biomarkers were measured using blood samples, and causative pathogens were identified.

This study was registered in the UMIN Clinical Trials Registry (ID: UMIN000034673) before study initiation (November 29, 2018), was approved by the Institutional Review Board or Ethics Committee for each site, and was performed in accordance with the Declaration of Helsinki revised in October 2013 and the Japanese Ethical Guidelines for Medical and Biological Research Involving Human Subjects partially amended on February 28, 2017. The current study enrolled patients who visited 14 institutions throughout Kyushu, Japan between December 2018 and December 31, 2020.

Study subjects

Eligible patients were adults aged ≥ 20 years who were diagnosed with CAP in accordance with the ATS/IDSA guideline or Japanese guidelines for the treatment of adult pneumonia and had satisfied the following two conditions:

(1)

Two or more of the following findings/symptoms associated with pneumonia: cough (productive or dry); purulent sputum; abnormal findings on auscultation or percussion (e.g., moist rales, diminished breath sounds, abnormal turbidity on percussion); dyspnea or tachypnea; fever (axillary body temperature ≥ 37 °C); increased white blood cells (WBCs) (> 10,000/mm3), increased stab cells (> 15%), or decreased WBCs (< 4,500/mm3); elevated CRP levels (> upper limit of normal at each institution); and hypoxemia (PaO2 < 60 Torr or SpO2 < 90%).

(2)

Findings suspicious for pneumonia (e.g., alveolar infiltration shadows on air bronchograms, pleural effusion, or other new increased lung shadows suspicious for infection) on chest radiographs or chest computed tomography images obtained within 48 h of subject registration.

Exclusion criteria included pneumonia that occurred after 48 h of hospitalization; patients who had been enrolled in this study; those who had already started antimicrobial therapy for this episode (pneumonia) and had shown improvement (patients who had received antimicrobial therapy, in principle, for a minimum of 3 days and had not shown improvement were allowed entry; however, this rule was applied only if the investigators determined that the patient was appropriate as a study subject); those with respiratory infections caused by Mycoplasma pneumoniae, Chlamydophila pneumoniae, Bordetella pertussis, Pneumocystis jirovecii, or mycobacterial species (including suspected cases); those who had used azithromycin within 7 days before study treatment initiation, excluding low-dose macrolide regimens (low-dose macrolide therapies before study treatment initiation were allowed to continue without changes in the dosage); those with an underlying disease that can significantly impact the diagnosis of CAP in this study, such as advanced cancer, primary lung cancer, lung metastasis of malignant tumors, severe heart failure, cystic fibrosis, and acquired immunodeficiency syndrome; and those determined to be inappropriate as study subjects by the investigators.

Data collection

Data obtained from patients and medical records included age, sex, height, weight, inpatient/outpatient status, underlying diseases, symptoms (temperature, cough, and sputum), disease severity [Acute Physiology and Chronic Health Evaluation II (APACHE-II) score, Confusion, Urea, Respiratory Rate, Blood Pressure and Age ≥ 65 Years (CURB-65) score, and Pneumonia Severity Index (PSI)], blood biomarkers, treatment drug, and safety information.

If possible, sputum was collected before antimicrobial administration (Day 0), and Gram staining and culture tests were performed. Sputum volume score [5 steps: 0 (none), 1 + (< 10 mL/day), 2 + (10 to < 50 mL/day), 3 + (50 to < 100 mL/day), 4 + (≥ 100 mL/day)] and sputum purulence [purulent (P), purulent-mucous (PM), and mucous (M)] [19] were also recorded. The causative microorganisms were identified based on data from samples obtained on Day 0 and subsequent days if necessary using sputum culture, urine antigen tests (Streptococcus pneumoniae and Legionella pneumophila), sputum antigen tests (S. pneumoniae), and FilmArray assay of nasopharyngeal swabs. The FilmArray respiratory panel can detect 20 pathogens including viruses [Adenovirus, Coronavirus (229E, HKU1, OC43, NL63), Human Metapneumovirus, Human Rhinovirus/Enterovirus, Influenza A (A/H1, A/H1-2009, A/H3), Influenza B, Parainfluenza 1–4, Respiratory Syncytial Virus] and atypical organisms (M. pneumoniae, C. pneumoniae, and B. pertussis). Assays were conducted by laboratory technicians at Department of Laboratory Medicine, Nagasaki University Hospital (Nagasaki, Japan) who were blinded to the patient information.

Blood samples were obtained on Day 0 for routine blood laboratory tests and biomarker measurements. Routine blood laboratory tests including blood cell counts, random plasma glucose, hemoglobin A1c (HbA1c), aspartate aminotransferase, alanine aminotransferase, alkaline phosphatase (ALP), γ-glutamyl transpeptidase, lactate dehydrogenase, total bilirubin, blood urea nitrogen, creatinine, and CRP were performed at participating hospitals. For additional biomarker measurements, serum (for IL-6, IL-8, IL-10, AAT, procalcitonin, and presepsin) and plasma (for pentraxin 3) were separated and cryopreserved immediately after blood collection. These biomarkers were measured at LSI Medience Corporation (Tokyo, Japan).

Data analyses

All cases were first classified into the following four categories based on the identified causative microorganism according to the flow diagram in Fig. S1: bacteria alone (purely bacterial pneumonia), virus alone (purely viral pneumonia), bacterial and viral coinfection (mixed bacterial–viral pneumonia), and none (no organism). Background data were presented as number (%) or median (interquartile).

The normality of continuous variables, including age, disease severity (PSI and CURB-65 score), sputum volume score, blood cell numbers (neutrophils, lymphocytes, and platelets), and biomarkers, was assessed using the Kolmogorov–Smirnov test, and variables were normalized using the Box-Cox transformation. Tests for outliers were performed with the Smirnov–Grubbs test. However, all outliers excluding abnormal data were used for analysis. Given that three sputum types were available for selection (P, PM, and M), some investigators selected M or blank in cases with no sputum or low sputum volume scores, which can indicate saliva. Therefore, sputum specimens were categorized into two types, namely P/PM and M/blank.

To identify items specific to bacterial pneumonia and unpredictable background effect on diagnosis, two data sets were prepared, one for the group of patients with purely bacterial pneumonia or mixed bacterial–viral pneumonia (BP group) and another for the group of those with purely viral pneumonia (VP group). Variables, including age, sex, disease severity, biomarkers, blood cell count, and sputum type, were then compared between the two groups. Continuous variables were compared using Mann–Whitney U test, whereas binary variables were compared using Fisher’s exact test. In addition, multivariate logistic regression analysis was performed with diagnosis (BP or VP) as the objective variable and age, sex, biomarkers, blood cell numbers, and sputum type and volume as explanatory variable, adjusting for age and sex as confounding factors. Furthermore, multivariate logistic regression analysis was repeated with the diagnosis (BP or VP) as the objective variable and items with a p value of < 0.05 in the previous univariate or multivariate logistic regression analysis, as well as age and sex, as explanatory variables while excluding items that were not significant (p ≥ 0.05). However, age and sex were retained as confounding factors until the end of the analysis.

The cutoff value for the identified items was determined using the receiver operating characteristic (ROC) curve and Youden’s index, after which the area under the ROC curve (AUC) and its 95% confidence interval (CI) were calculated. After dichotomizing the identified items at the cutoff values, odds ratios (95% CI) were obtained using multivariate logistic regression, and a diagnosis decision tree was constructed.

Finally, associations between each inflammatory biomarker and APACHE-II were examined using Pearson’s product moment correlation coefficient. All statistical analyses were performed using EZR software (version 4.1.2).