Utility of Empiric Antibiotics for Non-intubated Novel Coronavirus Diseases 2019 Patients

This study will examine the impact of empiric antibiotic therapy on patients who present to
hospital with an acute lower respiratory illness and a diagnosis of COVID-19
present-on-admission.

The Premier Healthcare Database will be used as the data source for administrative data. In
addition, the subset of hospitals reporting microbiology and laboratory data will be used for
subset analyses and validation purposes. The primary population to be studied will be
non-intubated patients diagnosed with COVID-19 on admission (identified by diagnosis coding
and/or polymerase chain reaction result present-on-admission) who have diagnosis codes
supportive of acute lung illness (e.g. pneumonia). Patients with extra-pulmonary infections
present-on-admission for which antibiotics would be generally administered and/or those
requiring vasopressors and/or mechanical ventilation on the day of admission or day after
will be excluded.

Patients will be analyzed according to their antibiotic treatment status, using an overlap
weight matching strategy. Patients will be matched on age, gender, ethnicity, Elixhauser
comorbidity index and month of admission as well as severity of acute illness (need for
intensive care unit and acute organ failure score present-on-admission), performance of rapid
diagnostic testing for bacterial respiratory pathogens, and receipt of concomitant putative
COVID-19 directed therapy (remdesivir, tocilizumab, systemic corticosteroids,
hydroxychloroquine) initiated on the day of or day after admission respectively. Logistic
regression will be performed downstream to matching to mitigate the impact of residual
confounding.The primary outcome and secondary outcomes are reported separately below.

Effect modification of the relationship between empiric antibiotics and outcomes will be
examined across clinically relevant subgroups based on antibiotic regimens (separately
comparing community and hospital acquired type coverage to no empiric antibiotics
respectively), and those with or without need for non-invasive ventilation on admission as
well as quartiles of hospital's frequency of empiric antibiotic use and admission
procalcitonin level (when available) respectively among patients admitted with COVID-19.

Sensitivity analyses will be performed to examine outcomes with vs without coding for
conditions that may or may not suggest a definite indication for antibiotic on admission
(e.g. chronic obstructive lung disease exacerbation) and/or explicit diagnosis for "sepsis"
(as it remains unclear in whom this code was indicated to represent confirmed viral sepsis).
Sensitivity analyses will also be performed to include patients without diagnosis codes for
acute lower respiratory illness present-on-admission to include patients with COVID-19
pneumonia who may not have been coded for pneumonia per se.