Differentiating influenza from COVID-19 in patients presenting with suspected sepsis

A prospective observational cohort study, as part of the Fast Assay for Pathogen Identification and Characterization (FAPIC) project, was performed between February 2019 and April 2020 at Jessa Hospital, Hasselt (clinicaltrial.​gov identifier NCT03841162). Adult patients presenting at the ED with suspected sepsis, defined as patients with symptoms of systemic infection for whom blood cultures were drawn, were asked to participate in the study. Patients who tested positive for influenza by PCR were included in this analysis.

During the COVID-19 pandemic, a retrospective cohort study was performed. All adult patients with SARS-Cov-2 PCR-positive test that were hospitalized for at least 24 h were included in the cohort. Patients that were suspected of sepsis and for whom blood cultures were drawn at admission (< 24 h) were included for this analysis.

In Jessa Hospital, for all patients presenting at the ED with suspected sepsis, blood cultures are performed to exclude bacteraemia. During the yearly influenza season (December–April), nasopharyngeal swabs for the detection of influenza were taken based on clinical suspicion. Since the emergence of COVID-19, SARS-Cov-2 PCR was performed according to the national case definition which was based on the WHO case definition guidance [14‐16]. After the first case of COVID-19 was admitted at Jessa Hospital on the 9th of March, the FAPIC study included 16 patients until the 17th of April. The last patient in whom influenza was detected (end of the 2020 season) was included on the 5th of March 2020. The clinical laboratory of Jessa Hospital started SARS-Cov-2 PCR testing on the 5th of March 2020.

Written informed consent was obtained from all participants with influenza. No informed consent was needed for the retrospective cohort study of COVID-19 patients, according to Belgian legislation. Documented approval for both studies was obtained from the Ethics committees of Hasselt University and Jessa Hospital.

Nasopharyngeal swabs for influenza detection were analysed with real-time PCR, either with the ARIES® Flu A/B & RSV assay on the ARIES instrument (Luminex Corporation) or with an in-house multiplex PCR on Quantstudie 7 flex (ThermoFisher) for the simultaneous detection of 23 respiratory pathogens. If available, lower respiratory tract samples, such as sputum, bronchial aspirates or broncho-alveolar lavage fluid, were preferred.

Nasopharyngeal swabs for the detection of SARS-CoV-2 were analysed with an in-house developed reverse-transcriptase PCR for the E-gene on the ARIES analyser (Luminex Corporation). In patients with a clinical suspicion based on history, laboratory and/or chest X-ray results, and a negative initial PCR test, a second nasopharyngeal swab or, if possible, a lower respiratory tract sample such as sputum, bronchial aspirates or broncho-alveolar lavage fluid, were collected after 24–48 h [17].

For all suspected pneumonia patients with chest X-ray abnormalities, urinary antigen tests for Streptococcus pneumoniae and Legionella pneumophila were performed.

Co-infections were defined according to the ECDC definitions on the surveillance of health care associated infections [18].

Parameters were chosen based on clinical relevance for suspected sepsis and based on literature. Relevant data were extracted from patients’ electronic medical files for all patients in both cohorts. Age, gender and comorbidities were registered. Clinical parameters at admission in the ED were collected. These parameters included body temperature, heart rate, blood pressure (systolic, diastolic and mean), oxygen saturation (SaO₂), partial oxygen pressure (PaO₂), Glasgow coma scale (GCS), vasopressor use and ventilation requirements. Laboratory testing at admission for suspected sepsis included white blood cell count (WBC), platelets, haemoglobin, red blood cell distribution width (RDW), C-reactive protein (CRP), creatinine, urea, lactate dehydrogenase (LDH), bilirubin, alanine aminotransferase (ALT), aspartate aminotransferase (AST), ferritin and serum lactate. To define whether patients eventually had sepsis according to sepsis-3 definitions, the SOFA score was calculated for all patients [9]. Microbiological diagnostics (pathogen identification and susceptibility) of blood cultures, of all other clinical cultures that were performed, and of virologic PCRs were collected. Data on co-infections within the first 24 h were extracted from ED discharge letters or clinical follow-up notes in the patient’s electronic medical file, and pathogens were collected from microbiology reports. Early empiric antibiotic treatment was recorded from discharge letters, from the ED and the ward or ICU, and from clinical notes made in the patient’s electronic medical file. Antibiotic use targeted at an isolated pathogen was excluded. Outcomes of patients that were collected included ICU admission, length of stay, in-hospital mortality and destination at discharge.

Descriptive statistics were used to analyse the patient’s characteristics. Continuous data are shown as median (interquartile range). Categorical data are reported as frequency (percentages). Differences between patients were analysed with the χ² test or Fisher’s exact test for categorical variables and with the Mann-Whitney U test for continuous variables. p values < .05 were considered statistically significant. A multivariable logistic regression model was built, where the viral infection during hospitalization was used as the outcome (influenza or COVID-19). Parameters that were statistically significant in the univariate analysis were inserted in the starting model. A backward selection based on significance level p < .05 was used. Odds ratios were calculated to define independent risk factors. Receiver operating curve (ROC) analysis was done to evaluate the performance of the logistic regression model to differentiate between influenza and COVID-19. Lastly, the χ² test or Fisher’s exact test was used to analyse the difference in patient outcomes (hospital LOS, ICU admission, mortality and destination at discharge) between patients with and without co-infection. All analyses were done using SPSS version 25 (IBM, Chicago, Illinois, USA).

In total, between the 12th of February 2019 and the 5th of March 2020, 103 patients from the FAPIC cohort with suspected sepsis on admission tested positive for influenza, 98 for influenza A and 5 for influenza B. From the 12th of March 2020 until the 12th of April, 110 patients admitted with suspected sepsis were PCR confirmed with SARS-CoV-2. No patients had an influenza and COVID-19 co-infection. Characteristics of patients, disease severity and outcomes are shown in Table 1. There were no differences in median age or gender between the two groups. Total comorbidities evaluated by the Charlson comorbidity index (CCI) were not different. However, there were significantly more COVID-19 patients with cardiac comorbidities (65.1% vs. 14.6%, p = .048) and hypertension (56.4% vs. 23.3%, p = .000). SOFA score was not different between patient groups. Eventually, 73.6% and 69.9% of patients had sepsis, respectively. Additionally, findings on chest X-ray differed, with more abnormalities for patients with COVID-19 (81.5% vs. 37.8%, p = .000). In the subgroup of patients with chest X-ray abnormalities, unilateral abnormalities were more frequently present for patients with influenza (31.8% vs. 59.5%, p = .004) and more bilateral abnormalities for patients with COVID-19 (68.2% vs. 40.5%, p = .004). Significantly, more patients with COVID-19 required oxygen therapy (90.9%) compared to patients with influenza (32%, p = .000). Lastly, all assessed outcomes were significantly different between the two groups. There were more ICU admissions (26.4% vs. 4.9%, p = .000), less discharges to home (84.2% vs. 89.6%, p = .044), more discharges to a rehabilitation centre (6.6% vs. 0%, p = .044), more deaths (30.9% vs. 6.8%, p = .000) and a longer hospital length of stay for patients with COVID-19 (8 days vs. 5 days, p = .000).

Clinical and laboratory characteristics of patients with suspected sepsis at clinical presentation are shown in Table 2. Patients with COVID-19 had a lower temperature at clinical presentation (37.5 °C) than patients with influenza (38.4 °C, p = .000). Furthermore, median SaO₂ and PaO₂/FiO₂ for patients with COVID-19 was 91.5% and 309.52 respectively, which were both significantly lower than patients with influenza (93%, p = .005 and 328.57, p = .027, respectively). More patients with COVID-19 had hypoxemia (SaO₂ < 90% or PaO₂/FiO₂ < 300) compared to patients with influenza (SaO₂: 34.6% vs. 16.7%, p = .003 and PaO₂/FiO₂: 46.5% vs. 32.7%, p = .046, respectively). Median absolute leukocyte count was 6.29 × 10^9 cells/L in patients with COVID-19 and 8.44 × 10^9 cells/L in patients with influenza, which was significantly lower (p = .012). This difference could also be seen for an absolute neutrophil count (5.17 × 10^9 cells/L vs. 6.35 × 10^9 cells/L, p = .049) but not for absolute lymphocyte count (0.63 × 10^9 cells/L vs. 0.63 × 10^9 cells/L, p = .960). Overall, 58.2% of patients with COVID-19 and 67.7% of patients with influenza had lymphopenia (p = .354). Other biochemical parameters that were significantly higher in patients with COVID-19 than in patients with influenza were LDH, ALT, AST, bilirubin and ferritin (370 U/L vs. 230 U/L, 29 U/L vs. 20 U/L, 40 U/L vs. 26 U/L, 0.52 mg/dL vs. 0.42 mg/dL and 1000 ng/mL vs. 290 ng/mL, respectively; p = .000 for each and p = .021 for bilirubin).

In Table 3, the multivariable logistic regression analysis of the association of clinical, laboratory and imaging characteristics with influenza or COVID-19 is presented. All variables that were significant in the univariate analysis were included in the multivariable regression model, and a backward selection was performed. In total, 14 cases were not inserted in the model since they had one or more missing values. The presence of hypertension was predictive for COVID-19 (OR 6.550 [95% CI, 2.590–16.565]). Abnormalities on chest X-ray, both unilateral and bilateral, were also predictive for COVID-19 (OR 4.764 [95% CI, 1.743–13.023] and OR 7.916 [95% CI, 3.858–21.925] respectively). Other factors that could differentiate influenza from COVID-19 were temperature, absolute leukocyte count, LDH, ALT, AST and ferritin. Patients with COVID-19 had lower temperature (OR 0.535 [95% CI, 0.343–0.833]), lower absolute leukocyte count (OR 0.857 [95% CI, 0.761–0.976]) and lower AST levels (OR 0.946 [95% CI, 0.917–0.976]). On the other hand, higher levels of LDH (OR 1.008 [95% CI, 1.004–1.012]), ALT (OR 1.044 [95% CI, 1.012–1.077]) and ferritin (OR 1.001 [95% CI, 1.000–1.002]) were predictive of COVID-19. The ROC of the model is shown in Fig. 1. ROC analysis of the logistic regression model showed an area under the curve (AUC) of 0.915 (p = .000; [95% CI, 0.877–0.953]). At a cutoff value of 0.49, the sensitivity was 83.5% and the specificity was 81.6%. The PPV and NPV of the logistic regression model were 82.6% and 82.5% respectively.

In Table 4, the type of co-infections and causative pathogens is shown. There were more patients with influenza diagnosed with bacterial co-infections during the first 24 h after the clinical presentation (10.7%) compared to patients with COVID-19 (2.7%, p = .011). Urinary antigen test for Streptococcus pneumoniae and Legionella pneumophila was performed in 59 patients with COVID-19 (53.6%) and in 81 patients with influenza (78.6%). Legionella pneumophila antigen was negative in all patients. Streptococcus pneumoniae antigen was not detected in any patient with COVID-19 but was positive in six patients with influenza. In Table 5, the univariate analysis evaluating the differences in clinical outcomes between all patients with early bacterial co-infections and those without is presented. There were no differences in clinical outcomes in patients with bacterial co-infection compared to patients without bacterial co-infection. Empiric antibiotic therapy was frequently started both in influenza and COVID-19 patients (76.7% vs. 84.5%, p = .147).