TY - JOUR
T1 - Clinical prediction model for bacterial co-infection in hospitalized COVID-19 patients during four waves of the pandemic
AU - Elbaz, Meital
AU - Moshkovits, Itay
AU - Bar-On, Tali
AU - Goder, Noam
AU - Lichter, Yael
AU - Ben-Ami, Ronen
N1 - Publisher Copyright:
© 2024 Elbaz et al.
PY - 2024/11
Y1 - 2024/11
N2 - The reported estimates of bacterial co-infection in COVID-19 patients are highly variable. We aimed to determine the rates and risk factors of bacterial co-infection and develop a clinical prediction model to support early decision-making on antibiotic use. This is a retrospective cohort study conducted in a tertiary-level academic hospital in Israel between March 2020 and May 2022. All adult patients with severe COVID-19 who had a blood or lower respiratory specimen sent for microbiological analyses within 48 h of admission were included. The primary study endpoint was the prevalence of bacterial co-infection at the time of hospital admission. We created a prediction model using the R XGBoost package. The study cohort included 1,050 patients admitted with severe or critical COVID-19. Sixty-two patients (5.9%) had a microbiologically proven bacterial infection on admission. The variables with the greatest impact on the prediction model were age, comorbidities, functional capacity, and laboratory parameters. The model achieved perfect prediction on the training set (area under the curve = 1.0). When applied to the test dataset, the model achieved 56% and 78% specificity with the area under the receiver operating curve of 0.784. The negative and positive predictive values were 0.975 and 0.105, respectively. Applying the prediction model would have resulted in a 2.5-fold increase in appropriate antibiotic use and an 18% reduction in inappropriate use in patients with severe and critical COVID-19. The use of a clinical prediction model can support decisions to withhold empiric antimicrobial treatment at the time of hospital admission without adversely affecting patient outcomes.
AB - The reported estimates of bacterial co-infection in COVID-19 patients are highly variable. We aimed to determine the rates and risk factors of bacterial co-infection and develop a clinical prediction model to support early decision-making on antibiotic use. This is a retrospective cohort study conducted in a tertiary-level academic hospital in Israel between March 2020 and May 2022. All adult patients with severe COVID-19 who had a blood or lower respiratory specimen sent for microbiological analyses within 48 h of admission were included. The primary study endpoint was the prevalence of bacterial co-infection at the time of hospital admission. We created a prediction model using the R XGBoost package. The study cohort included 1,050 patients admitted with severe or critical COVID-19. Sixty-two patients (5.9%) had a microbiologically proven bacterial infection on admission. The variables with the greatest impact on the prediction model were age, comorbidities, functional capacity, and laboratory parameters. The model achieved perfect prediction on the training set (area under the curve = 1.0). When applied to the test dataset, the model achieved 56% and 78% specificity with the area under the receiver operating curve of 0.784. The negative and positive predictive values were 0.975 and 0.105, respectively. Applying the prediction model would have resulted in a 2.5-fold increase in appropriate antibiotic use and an 18% reduction in inappropriate use in patients with severe and critical COVID-19. The use of a clinical prediction model can support decisions to withhold empiric antimicrobial treatment at the time of hospital admission without adversely affecting patient outcomes.
KW - COVID-19
KW - antibiotic stewardship
KW - bacterial coinfection
KW - prediction model
UR - http://www.scopus.com/inward/record.url?scp=85208772364&partnerID=8YFLogxK
U2 - 10.1128/spectrum.00251-24
DO - 10.1128/spectrum.00251-24
M3 - ???researchoutput.researchoutputtypes.contributiontojournal.article???
C2 - 39315820
AN - SCOPUS:85208772364
SN - 2165-0497
VL - 12
JO - Microbiology spectrum
JF - Microbiology spectrum
IS - 11
M1 - e00251-24
ER -