Nosocomial infections amongst critically ill COVID-19 patients in Australia

Purpose To determine the frequency of nosocomial infections including hospital-acquired pneumonia (HAP) and bloodstream infection (BSI), amongst critically ill patients with COVID-19 infection in Australian ICUs and to evaluate associations with mortality and length of stay (LOS). Methods The effect of nosocomial infections on hospital mortality was evaluated using hierarchical logistic regression models to adjust for illness severity and mechanical ventilation. Results There were 490 patients admitted to 55 ICUs during the study period. Adjusted odds ratio (OR) for hospital mortality was 1.61 (95% confidence interval (CI) 0.61–4.27, p = 0.3) when considering BSI, and 1.76 (95% CI 0.73–4.21, p = 0.2) for HAP. The average adjusted ICU LOS was significantly longer for patients with BSI (geometric mean 9.0 days vs 6.3 days, p = 0.04) and HAP (geometric mean 13.9 days vs 6.0 days p<0.001). Conclusion Nosocomial infection rates amongst patients with COVID-19 were low and their development was associated with a significantly longer ICU LOS.

Secondary infections in patients with influenza viral pneumonitis, including during the2009 H1N1 pandemic [6] , have been well characterised and are known to be associated with higher illness severity, usage of healthcare resources, morbidity and mortality [7][8][9][10][11] . These infections, most commonly Streptococcus pneumoniae and Staphylococcus aureus , occurred in 23% of patients with H1N1 [8] . Viral co-infections have been reported in 10-60% of these patients [7] .
However, there is paucity of such data for COVID-19 [12] . A singlecentre study demonstrated that mechanically ventilated COVID-19 patients were twice as likely to develop a ventilator-associated pneumonia (VAP) than patients without COVID-19, but both groups had similar pulmonary microbiomes [13] . Another single-centre study conducted in a repurposed COVID-19 ICU reported that 67% of COVID-19 patients in their cohort had bloodstream infections (BSI) [4] . A large cohort study of mostly non-critically ill hospitalized COVID-19 patients reported that nosocomial infections, mostly respiratory and bloodstream, were rare in a cohort where most patients (82.3%) did not have microbiological investigations [14] .
Further studies evaluating the characteristics and outcomes of COVID-19 patients with nosocomial infections are required to enable clinicians to risk stratify patients and guide therapy.
The Short Period Incidence Study of Severe Acute Respiratory Infections (SPRINT-SARI Australia) study [15] has been prospectively collecting comprehensive data on critically ill patients with COVID-19 admitted to Australian intensive care units (ICU) from February 2020.
We used the SPRINT-SARI Australia database to determine clinical characteristics and outcomes, including mortality and lengths of stay, of COVID- 19  comial infections compared to patients who did not develop such infections. Our primary hypothesis was that clinical outcomes would be inferior in those patients who developed a nosocomial infection, independent of admission illness severity.

Methods
This multicentre cohort study was performed following the recommendations of the STROBE Statement [ 16 ]. Ethics approval with full consent waiver was granted under the National Mutual Acceptance scheme by the Alfred Health Human Research Ethics Committee (HREC/16/Alfred/59) or by specific applications at individual sites.

Study design
The methodology for SPRINT-SARI Australia has been described in detail elsewhere [15] . In brief, the SPRINT-SARI Australia case report form, developed together with partners from the International Severe Acute Respiratory and Emerging Infection Consortium [17] , prospectively collected data on all COVID-19 admissions to participating ICUs. Patients had to have a positive polymerase chain reaction (PCR) test for COVID-19 and require ICU admission for COVID-19 related indications.
There were 79 participating ICUs across Australia. Decisions to admit and discharge patients were made by treating clinicians, mostly specialist intensive care physicians, based on local protocols at individual sites.

Data
Data pertaining to baseline demographics, clinical characteristics, treatments, microbiology and clinical outcomes were extracted from the SPRINT-SARI Australia database for patients admitted from inception until 30 September 2020. Data for patients without a complete outcome (i.e., they were still alive in ICU or hospital) were collected for descriptive purposes, but not used for analysing mortality.
Hospital-acquired pneumonia (HAP) was defined as an acute infectious process of the lungs with clinical and, if available, radiological evidence of focal or diffuse lung infiltrates that the treating clinician believed to be due to pneumonia occurring after 48 h of hospital admission [18] . When this occurred within 48 h of hospital admission, we termed this co-infection. We considered HAP with a microbiological diagnosis (i.e., evidence of an infectious organism from bronchoalveolar lavage, endotracheal aspirates or sputum samples) and/or BSI as constituting a nosocomial infection. Patients who had evidence of an infectious organism from bronchoalveolar lavage, endotracheal aspirates or sputum samples without meeting criteria for HAP were defined as having a colonising organism.
BSI was defined as presence of bacteria in blood detected on blood cultures. BSIs with organisms known to cause contamination such as coagulase negative Staphylococci were considered to be contaminants and not included in the BSI group unless specified in the database as a true BSI. Nosocomial infection was defined as the presence of either BSI and/or HAP.

Statistical analyses
Continuous variables were assessed for normality and summarised using mean and standard deviation or median and interquartile range (IQR) according to data type and distribution. Categorical variables were summarised using counts and proportions.
The primary outcome measure was hospital mortality. The primary exposure variables were the development of BSI or HAP during ICU admission. Univariable analysis was performed using logistic regression to determine the association between BSI, HAP and hospital mortality. Multivariable analysis was performed using hierarchical logistic regression adjusting for Acute Physiology And Chronic Health Evaluation 2 (APACHE-2) score and receipt of mechanical ventilation with patients nested within sites and sites treated as random effects. Sensitivity analyses were performed by excluding patients who did not receive mechanical ventilation throughout their ICU admission. A further sensitivity analysis was conducted by using nosocomial infection as a composite exposure variable that included both BSI and HAP. The secondary outcomes, ICU and hospital lengths of stay were analysed using generalised linear mixed modelling adjusting for APACHE-2 and mechanical ventilation with patients nested within sites and sites treated as random effects. The lengths of stay were analysed as log-transformed continuous variables due to positively skewed distributions. Multivariable competingrisks regression was used to identify predictors and quantify cumulative incidence of nosocomial infection with death as a competing risk. Patients with missing data for the outcomes being analysed or the covariates were excluded from the respective analyses. We did not perform any imputation for missing data.
Results are reported as odds ratios (OR) for hospital mortality, geometric means for lengths of stay and hazard ratios (HR) for nosocomial infection with corresponding 95% confidence intervals (95% CI). A twosided p value of 0.05 was chosen to indicate statistical significance in all analyses.
Microbiological characteristics were summarised using descriptive statistics.
All analyses were performed with SAS version 9.4 (SAS Institute, Cary, NC, USA) and Stata version 15 (StataCorp, Texas, USA).

Funding
SPRINT-SARI Australia is supported by funding from the Australian Department of Health (Standing Deed SON60002733). This post-hoc analysis did not receive any specific funding.

Patient characteristics
There were 490 patients with confirmed COVID-19 admitted to 55 Australian ICUs during the study period. Of these, 36%(176/490) were female and the overall median age was 61 years(IQR 50-70). Other characteristics of the cohort are presented in Table 1 .

Nosocomial infections
There were 30 out of 490 patients (6%) who developed BSI and 36(6%) who developed HAP during their ICU stay. There were 6 patients (1%) who developed both BSI and HAP leaving 430 patients (88%) who did not develop a nosocomial infection.
In the univariable competing risks models, we found that temperature (highest in first 24 h), body-mass index (BMI) and mechanical ventilation were all significantly associated with development of nosocomial infection. Mechanical ventilation was strongly and independently associated with development of nosocomial infection after adjustment for age, sex, APACHE-2 score, temperature and BMI in a competing-risks regression model (HR 6.62, 95% CI 2.29-19.10, p = 0.0005). BMI was also independently associated with nosocomial infection in this model with a HR of 1.04 per 1 unit of BMI (95% CI 1.02-1.06, p = 0.001). The other variables in the model were not statistically significant.
The cumulative incidence function for development of nosocomial infection is displayed in Fig. 1 . The incidence of nosocomial infection increased over time from Day 0 to approximately Day 60 of hospitalization.

Outcomes
Hospital mortality for patients with BSI and HAP were 27%(8/30) and 23%(8/35) compared to 11%(46/422) for patients without nosocomial infection. Lengths of stay and readmission rates as presented in Table 3 .   After adjustment for APACHE-2 score and mechanical ventilation in the multivariable hierarchical logistic regression model, the adjusted OR for hospital mortality was 1.61(95% CI 0.61-4.27, p = 0.3) with BSI. With HAP, the adjusted OR was 1.76(95% CI 0.73-4.21, p = 0.2). Sensitivity analyses for both these models were conducted by removing patients who did not receive mechanical ventilation. The respective ORs for the BSI and HAP models changed minimally to 1.43(95% CI 0.51-4.04, p = 0.5) and 1.40(95% CI 0.57-3.44, p = 0.5). There were similarly minimal changes to the results when use of antibiotics was added as a fixed effect to both models. When nosocomial infection was the exposure variable, the adjusted OR for hospital mortality was 2.30(95% CI 1.12-4.73, p = 0.02).

Nosocomial infection
Six out of 30 patients (20%) with BSI had gram positive cocci isolated from their blood. One of these was a methicillin-resistant Staphylococcus aureus (MRSA), two were Enterococcus faecalis and three were Enterococcus faecium . There were 12 patients (40%) with gram negative organisms including 3 with Enterobacter cloacae , 2 each with Pseudomonas aeruginosa and Escherichia coli , 1 each with Klebsiella pneumoniae and Stenotrophomonas maltophilia , and a further 3 patients with multiple organisms on blood culture. There were a further 12 patients (40%) with missing data concerning the BSI organism.
Twelve out of 36 patients (33%) had HAP with gram positive cocci isolated from cultures of respiratory samples including 1 patient with MRSA and 11 with other Staphylococcal species. There were 21 patients   (58%) with gram negative bacilli of which 8 were mixed growths and 4 were Pseudomonas aeruginosa . One patient had a fungal pathogen ( Aspergillus spp ) identified as the source of their HAP, and 2 had viruses isolated. Details of HAP and BSI organisms are presented in Table 2 .

Colonising organisms
In addition to the patients who developed HAP, there were 28 patients (6%) who had positive microbiological findings on respiratory specimen culture but did not meet criteria for diagnosis of HAP. Amongst these patients with colonising organisms in their respiratory system, there were ten gram-positive cocci including 5 Staphylococcus aureus (of which 2 were MRSA). There were eleven gram-negative bacilli, 4 Candida species and 2 Aspergillus species. Details are presented in Table 4 .

Nurse:patient ratios
Nurse:patient ratios in Australian ICUs are strictly mandated at 1.0 for mechanically ventilated patients, though the ratio can be 0.5 for certain selected non-ventilated patients. Over 90% of our cohort received a nurse:patient ratio of at least 1.0, with a further 8% of patients receiving a ratio of 0.5 and only 1% receiving a ratio lower than 0.5. All the patients with nosocomial infections had a ratio of 1.0 or 2.0.

Key findings
Nosocomial infections, BSI and HAP, occurred in 6% and 7% of patients with COVID-19 admitted to Australian ICUs respectively. The development of either BSI or HAP, was not independently associated with an increase in the risk of hospital mortality, after adjustment for illness severity and requirement for mechanical ventilation. However, both BSI and HAP were associated with significantly longer ICU lengths of stay.

Comparisons to literature
In the pre-COVID-19 ICU literature, it has been reported that 5-7% of ICU patients develop BSI, with up to 40% of patients in septic shock trials developing BSI [19] . Amongst ICU patients, BSI is known to be associated with increased risk of mortality [20] and both HAP and nosocomial infections with increased mortality and length of stay in ICU and hospital [ 21 , 22 ].
An Italian cohort study [4] found that 67% of patients with COVID-19 admitted to an ICU during the "first wave " of COVID-19 infections in Italy (February 2020 to April 2020) developed BSI, with most isolated organisms (80%) being gram positive cocci. The proportion of mechanically ventilated was high (93%) and the overall mortality rate was 49%. Another Italian study [5] during the "first wave " reported 40% of patients developed BSI also with a preponderance of gram positive cocci and ICU mortality was reported as 26%.
There are notable differences between our study and this prior work, which may account for the observed differences in BSI rates and other outcomes. We reported on cases from a longer recruitment period which incorporated both the first and second waves of COVID-19 in Australia. Our database provided near-complete coverage of Australian ICUs which admitted COVID-19 patients compared to these single or two centre studies from Italy. Australia has had less community transmission of COVID-19 and has thus far experienced fewer cases than Italy (121 versus 7075 cases per 100,000 population) [1] . Thus, Australian ICUs were not required to operate beyond their usual capacity [15] . At the height of the first wave in the repurposed Italian ICU [4] , the authors reported that nurse:patient ratios were as low as 0.2. Comparatively, such measures were never required in Australia with maintenance of usual nurse:patient ratios.
Reports of secondary infections from Wuhan, China, early in the pandemic suggested low rates (1-10%) but it is notable that follow-up periods were short and incomplete in these studies [ 23 , 24 ]. A study from New York, USA, reported that 6% of their cohort developed BSI [ 25 ]. For similar reasons discussed above, these results may not be directly comparable to our study.
A single-centre UK study reported that 48% of patients with COVID-19 admitted to their ICU developed microbiologically confirmed VAP [ 13 ] with ICU mortality rate of 38%. They have reported maintenance of usual nurse:patient ratios, though the nurses were not always critical care trained. Despite differences in patient characteristics, rates of VAP and outcomes, there were similarities with our study in the organisms isolated from respiratory samples with a preponderance of gram-negative bacilli. This is in contrast to nosocomial infections seen in influenza, where high proportions of gram-positive HAP are found.

Implications of findings
Several important implications arise from our findings. The rates of nosocomial infections in critically ill patients with COVID-19 in Australia, such as BSI and HAP, were low. This is surprising, given the prolonged lengths of stays in ICU and hospital. One explanation could be that patients who developed nosocomial infections on hospital wards, prior to or after ICU admission, were not identified in this study. We have demonstrated that risk of nosocomial infection in this cohort does increase as a function of hospital length of stay. The other implication of this low rate of nosocomial infection (HAP particularly), in combination with the gram-negative predominance, is that COVID-19 may not be a significant driver of nosocomial infection. Rather, the observed rate of nosocomial infection may be related to ICU stay in general, as would apply to any patient requiring ICU admission.
This study also highlights the role of resourcing issues on the outcomes of patients with COVID-19. Australian ICUs, faced with relatively low numbers of COVID-19 patients, were not required to repurpose non-ICU areas for ICU-level care, were able to maintain usual nurse:patient ratios and were able to operate within usual capacity [ 15 ].
Nonetheless, nosocomial infections in our cohort were associated with significantly longer ICU and hospital lengths of stay, and by extension, resource utilization. Therefore, vigilant monitoring for nosocomial infection, and attention to preventative strategies (infection control practices such as hand washing, evidence-based use of personal protective equipment and environmental controls such as negative pressure rooms) remain important.

Strengths and limitations
This study was performed using data from a database with nearnational coverage of ICUs that admitted critically ill patients with COVID-19. Data collection was performed by experienced research staff using standardized case report forms. The follow-up rate was high, with near-complete data for the primary outcome. Competing-risks analyses were performed to account for death as a competing risk for development of nosocomial infection.
There were, however, important limitations to consider. This was an observational study with confounding from numerous sources that may affect the rate of nosocomial infections and mortality. There was no routine screening for HAP or BSI. We only reported on patients who developed clinically apparent nosocomial infections. Only patients who were admitted to an ICU were included, and ICU admission criteria were at the discretion of clinicians at individual sites. Additional microbiological data, such as antimicrobial sensitivity and minimum inhibitory concentrations, were unavailable. Genome sequencing data to identify COVID-19 variants of concern were unavailable. There were missing data for some patients with positive blood cultures. The generalizability of our findings to other countries with different healthcare systems and higher rates of COVID-19 transmission is also uncertain. The sample size and event rates were low and hence this study was under-powered to detect small effect sizes. The point estimates suggested that nosocomial infections were associated with increased mortality, but the confidence intervals were wide and included the possibility of both harm and benefit.

Conclusion
In a healthcare system operating within capacity, the proportions of COVID-19 patients admitted to ICU who developed serious nosocomial infections, were low. In these patients, significant increases in both ICU and hospital length of stay were observed. The microbiology of noso-