The Incidence and Outcomes of Healthcare-associated Respiratory Tract Infections in Nonventilated Neurocritical Care Patients: Results of a 10-year Prospective Cohort Study

Background. The incidence of healthcare-associated respiratory tract infections in non-ventilated patients (NVA-HARTI) in neurosurgical intensive care units (ICU) is unknown. The impact of NVA-HARTI on patient outcomes and differences between NVA-HARTI and ventilator-associated healthcare-associated respiratory tract infections (VA-HARTI) are poorly understood. Our objectives were to report the incidence, hospital length of stay (LOS), ICU LOS, and mortality in neurosurgical ICU and compare these characteristics between NVA- and VA-HARTI. Methods. This prospective cohort study was conducted in a neurosurgical ICU in Moscow from 2011 to 2020. All patients with ICU LOS >48h were included. Time trends were analyzed for all outcomes. A competing risk model was used for survival and risk analysis. Results. A total of 3,937 ICU admissions were analyzed. NVA-HARTI vs VA-HARTI results were: cumulative incidence 7.2 (6.4-8.0) vs 15.4 (14.2-16.5) per 100 ICU admissions, incidence rate 4.2±2.0 vs 9.5±3.0 per 1000 patient-days in the ICU, median LOS 32 [21; 48.5] vs 46 [28; 76.5] days, median ICU LOS 15 [10; 28.75] vs 26 [17; 43] days, and mortality rates 12.3% (7.9-16.8) vs 16.7% (13.6-19.7). The incidence of VA-HARTI decreased in ten years while NVA-HARTI incidence did not change. VA-HARTI was found to be an independent risk factor of death, odds ratio 1.54 (1.11-2.14), p-value=0.009 while NVA-HARTI was not. Conclusion. Our ndings suggest that NVA-HARTI in neurosurgical ICU patients represents a signicant healthcare burden with relatively high incidence and associated poor outcomes. NVA-HARTI appeared to be different from VA-HARTI and persisted despite preventive measures; therefore, extrapolating VA-HARTI research ndings to NVA-HARTI

unclear whether VA-and NVA-HARTI are similar or different clinical categories (8) and whether it is warranted to extrapolate results from VAP research to NVA-HARTI. Therefore, our primary objective was to evaluate the incidence, prevalence, risks, and outcomes of NVA-HARTI in neurosurgical ICU patients. Our secondary objective was to compare these characteristics between NVA-and VA-HARTI within the same patient population and timeframe.

Study Design and Settings
This prospective cohort study was conducted at the national referral neurosurgical hospital, Burdenko National Medical Research Center of Neurosurgery (NSI) in Moscow, from 2011 to 2020. The hospital has an annual patient ow of approximately 9000 patients, with 98% undergoing neurosurgical procedures.
The ICU at NSI is combined with the post-operative care unit and has a total of 38 beds. On average, 4000 patients are admitted every year to the ICU, 10-13% stay for > 48 hours.

Patients and HARTI De nitions
Patients of all ages admitted to the ICU between January 1, 2011, and December 31, 2020, with a length of ICU stay (ICU LOS) greater than 48 hours were included in the study, excluding those with HARTI present on admission (POA) and tested positive for COVID-19 at any point during their hospital stay. HARTI POA was identi ed based on the 2019 CDC/NHSN de nition (9).
The 2008 CDC/NHSN de nition (10) was used in our study to de ne healthcare-associated infections (HAI); HARTI was de ned as a subset of HAI that included all respiratory tract infections: upper respiratory tract infections (pharyngitis, laryngitis, epiglottitis), sinusitis, pneumonia, and lower respiratory tract infections other than pneumonia (bronchitis, tracheobronchitis, tracheitis, other infections of the lower respiratory tract). VA-HARTI was de ned per the 2008 CDC/NHSN guidelines (10). The presence of HARTI was assessed daily; standard resolution criteria were used to de ne resolved cases. For more on de nitions, see Supplementary Methods.

Data Collection and Processing
Data was collected as a part of the ICU infection surveillance protocol and stored in the electronic medical record. ICU nursing staff performed surveillance and reported to the infection control team. Study team members (GD, ES, OE) veri ed data and performed data entry daily. For each patient, we collected 47 characteristics daily, and eight at the end of the hospital stay. Characteristics included demographics, hospital course and outcomes (Table S1).
The data from patients with ICU LOS exceeding 365 days was censored at ICU day 365. In case of readmission, if it took place within 24 hours from the discharge, it was considered a continuation of the previous ICU admission. If after 24 hours, then this constituted a new ICU admission.

Outcome Measures
The outcomes of the study included daily prevalence, cumulative incidence, incidence rates, time-risk assessment, and patient outcomes (hospital LOS, ICU LOS, and crude in-hospital mortality). The daily prevalence of HARTI was calculated by dividing the number of patients with active HARTI symptoms in the morning of each day by the total number of patients in the ICU who were enrolled in the study on the same morning. The daily prevalence was reported as a percentage. The cumulative incidence was reported as the number of HARTI cases per 100 ICU admissions. Incidence rates were de ned as the number of HARTI cases per 1000 patient-days in the ICU. For VA-HARTI, the incidence rate was also adjusted to days on mechanical ventilation and reported as the number of VA-HARTI cases per 1000 ventilator-days. The results of the time-risk assessment were reported as a cumulative probability and a daily risk (instantaneous hazard).
The hospital LOS and ICU LOS were reported in days. The crude in-hospital mortality was de ned as death from any cause occurring during the hospital stay. The hospital LOS and mortality were attributed to a particular group of patients based on a group hierarchy: 'Dual HARTI' > 'VA-HARTI' > 'NVA-HARTI' > 'Other HAI' > 'No HAI'. A patient was excluded from other groups where he/she might be counted based on his/her other ICU admissions.

Statistical analysis
The database was extracted and de-identi ed. The percentage of missing values was calculated for each variable; missing values were treated using manual ll, forward-ll or backward-ll where applicable (Table S2). Data cleaning did not include any data transformation or outlier removal. According to the rst objective of the study, we calculated the mean incidence, prevalence, risks, and outcomes of NVA-HARTI and their dynamics over the study period. Binary measurements were reported as a number of events with percentage and a 95% con dence interval (CI) for binomial distribution. Continuous and ordinal measurements were reported as a median with rst and third quartiles (Q1; Q3) or as a mean with standard deviation. The linear regression and Augmented Dickey-Fuller test (ADF) test were used to evaluate time trends. Monthly, quarterly, or yearly data was used depending on the model. For each variable, the model was checked for relevance based on the distribution of residuals and QQ plot. In the ADF test for stationarity, the variable's dynamic was considered non-stationary if mean, standard deviation, or frequency changed over time; non-stationarity was assumed as a null hypothesis. To extract a trend from prevalence data, we used the additive decomposition method de ned as a centered moving average of the data within the nearest n periods, assuming that the seasonal component is constant from year to year.
Included ICU admissions were divided in ve groups depending on the presence and type of infection: VA-HARTI, NVA-HARTI, Other HAI, No HAI, and the group of patients who had both types of HARTI during a single ICU admission ('Dual HARTI') ( Fig. 1, groups shown in blue). We excluded the Dual HARTI group from outcome analysis as an outlier with an insu cient sample size. The null hypothesis of the second objective of the study was set as a lack of difference between VA-and NVA-HARTI groups for all outcomes. The Kruskal-Wallis test was used to compare continuous variables among three or more groups with post-hoc pairwise comparison by Conover test. To compare the distribution of categorical variables among two or more groups, we used a chi-square test with a post-hoc pairwise chi-square test when needed. P-values were adjusted for multiple comparisons by the Bonferroni-Holm method if > 5 hypotheses were tested.
To compare mortality between groups, we used a competing risk model of a sub-distribution function (11) implemented in (12) for survival analysis; a discharge from the hospital was considered a competing event. For additional considerations regarding the methodology of survival analysis, see Supplementary

Methods.
The immortal-time bias was present; we utilized two approaches to account for it. First, we used the Cox time varying proportional hazard model to account for time-independent variables (age, sex, Charlson Comorbidity Index and admitting diagnosis). To account for all confounding variables (time-dependent and time-independent), we used logistic regression. Both regressions were applied separately for VA-and NVA-HARTI data. The procedure for the logistic regression was as follows. First, the univariate analysis was carried out comparing all factors between patients with and without an infection. Factors with a pvalue < 0.05 after adjustment for multiple comparisons were entered into the rst logistic regression.
Variables that demonstrated p-value > 0.05 (were not independent predictors of infection) and a binary infection variable were used in the second logistic regression to evaluate whether they can independently predict mortality. Model performance was assessed in ve-fold cross-validation with a receiving-operating curve, an area under the curve (ROC-AUC) score used as a performance metric.
The HARTI time-risk assessment was performed using a competing risk model; discharge from the hospital and death were considered competing events. Kernel-smoothed instantaneous (daily) hazard of infection was obtained using a Nelson-Aalen estimator.
The signi cance level was set at 0.05. Statistical analysis was performed in Python 3.7 using StatsModels (13) and Scipy (14) libraries.

Study population
A total of 3,842 unique patients accounting for 4,258 ICU admissions were admitted for greater than 48 hours; 310 ICU admissions were excluded due to HARTI present on admission, and 11 ICU admissions were excluded due to positive COVID-19 test. The nal data sample of 3,937 ICU admissions was analyzed (Fig. 1).

Characteristics of NVA and VA-HARTI
The univariate analysis showed that NVA-and VA-HARTI patients were similar in terms of patients' demographics, admitting diagnosis, surgeries, and complications, although their ICU courses and severity were different (Table 1). When comparing ve groups, there were signi cant differences among them, most notably in their ICU course and complications (Table S3). We also found that patient characteristics remained mostly stable during ten years while multiple aspects of medical and surgical practice have changed. Most notably, the use of mechanical ventilation, benzodiazepines, urinary catheters, and antibiotics decreased signi cantly (Table S4). Table 1 Baseline clinical characteristics of NVA-HARTI and VA-HARTI patients; for baseline characteristics of other patient groups refer to Among patients who had a single HARTI event during their hospital stay, NVA-HARTI occurred on median ICU day 4, with the onset of the half of the cases between ICU days 3 and 7 ( Figure S1A).  Figure S1C).

Prevalence of HARTI
The mean daily prevalence of NVA-HARTI was 5.6% and there was no longitudinal time trend observed, pvalue = 0.85 ( Figure S2A). The mean prevalence of VA-HARTI was 13.7% and it signi cantly decreased from 2011 to 2020, p-value = 0.00011 ( Figure S2B). For both infections, the prevalence uctuated signi cantly from day to day, and the extracted annual trend had irregular periodic patterns. 14.6) in 2020, p-value = 0.008 ( Fig. 2A).

Incidence of HARTI
The mean incidence rate of NVA-HARTI was 4.2 ± 2.0 per 1000 patient-days in the ICU with no time trend, p-value = 0.15. The mean incidence rate of VA-HARTI was 9.5 ± 3.0 per 1000 patient-days, also with no time trend, p-value = 0.09, Fig. 2B.
When normalized to ventilator-days, VA-HARTI incidence rate didn't change throughout the study period (pvalue = 0.79) and averaged at 16.4 ± 4.7 cases per 1000 ventilator-days (Fig. 2C). At the same time, the total number of ventilator-days per year decreased by 30%, from 3,690 in 2011 to 2,565 in 2020.  Figure S3B).
In the NVA-HARTI group, the mortality rate was 12.3% (7.9-16.8), and no different from other groups. In VA-HARTI patients, the mortality rate was 16.7% (13.6-19.7), twice that of the No HAI group, 8.2% (7.0-9.4), p-value < 0.0001, Fig. 4C. From 2011 to 2020, the mortality rate dropped signi cantly in all groups except for the NVA-HARTI, Figure S3C.  Figure S4A. This difference represented the immortal time bias. To account for it, we adjusted for time-independent variables (age, sex, diagnosis, and CCI) in a Cox time varying model. Neither infection was found to be independently associated with mortality, NVA-HARTI pvalue = 0.13 (Table S5A), VA-HARTI p-value = 0.82 (Table S5B).
Survival curves were obtained from the competing risk model while accounting for discharge as a competing event, Fig. 3D. The Kaplan-Meier model underestimated the risk of death and was deemed inappropriate ( Figure S4B).  3D).
In stepwise logistic regression, VA-HARTI was an independent risk factor of death with an odds ratio of 1.54 (1.11-2.14), p-value = 0.009 (Table 2A, all covariates in Table S6A). NVA-HARTI was not a predictor of mortality, odds ratio 1.43 (0.99-2.06), p-value = 0.057 (Table 2B, all covariates in Table S6B). The performance testing in ve-fold cross-validation yielded a ROC-AUC score of 0.63 for NVA-HARTI and 0.79 for VA-HARTI models.

Risk of HARTI
The risk of both HARTI types increased rapidly in the beginning of ICU stay. The NVA-HARTI probability grew faster and plateaued earlier than VA-HARTI, around day 8 in the ICU with the highest daily risk of 1.0% at that time. Whereas the risk of VA-HARTI continued to increase up to ICU day 20 before reaching a plateau phase (Fig. 4A). Also, a secondary peak of daily infection risk occurred earlier for NVA-HARTI, Fig. 4B. During the rst week in the ICU, the risk of infection doubled every 18.4 hours for NVA-HARTI and every 13.6 hours for VA-HARTI.
The cumulative probability of VA-HARTI depending on the number of days on mechanical ventilation increased rapidly during the rst 8 ventilation days, Fig. 4C. The probability of VA-HARTI doubled daily during the rst week on ventilation. A secondary peak became apparent after ventilation day 36 (Fig. 4D).

Discussion
A prospective observational study was performed at a large tertiary referral neurosurgical center to examine the incidence and outcomes associated with NVA-HARTI and to compare it to VA-HARTI. The main ndings of this study are: 1) both VA-and NVA-HARTI are prevalent in the neurosurgical ICU however they are different in terms of incidence, prevalence, and risks, 2) while the VA-HARTI incidence reduced over time with preventive measures in place, NVA-HARTI incidence did not change, 3) NVA-HARTI was associated with higher hospital and ICU LOS, but not with higher mortality compared to patients without HAI, and 4) VA-HARTI was an independent predictor of mortality while NVA-HARTI was not. We demonstrated that NVA-HARTI is a distinct clinical entity that persists despite preventive measures and is associated with poor outcomes.
The Rate of Occurrence of NVA and VA-HARTI To our knowledge, the incidence of HARTI in non-ventilated neurosurgical ICU patients is unknown and comprehensive epidemiological data on NVA-HARTI in ICU is limited (4). The most comparable study was done in Europe on mixed ICU patients who stayed > 48h in the ICU; it reported 1% incidence of nonventilator pneumonia (7). In our study, the cumulative incidence of NVA-HARTI was signi cantly higher, 7.2%. The difference can be explained by higher prevalence of factors predisposing neurosurgical patients to respiratory infections. The incidence rate in the mixed hospital population was 3.63 per 1000 patient-days in the USA (15) which is similar to our result, 4.2 per 1000 patient-days.
Notably, the cumulative incidence of NVA-HARTI did not change over the ten-year study period. As previously reported, the incidence of pneumonia remained stable for 5 years in acute-care non-ventilated patients (16).
In comparison, the cumulative incidence of VA-HARTI in our study was comparable to previous reports on VAP in neurosurgical ICU patients, 15.4% (17). However, in literature, the incidence varies signi cantly (2.3-50.7%) depending on the speci c neurosurgical procedure and patient characteristics (18, 19). For example, in patients with traumatic brain injuries, VAP incidence was found to be 36% (20).
The incidence rate of VA-HARTI in our study was 16.4 per 1000 ventilator-days which is within the range of previous European reports of 19.0 per 1000 ventilator-days (21) and 11.0 per 1000 ventilator-days (22).

Outcomes in Patients with NVA and VA-HARTI
There is an emphasis in literature that the resulting burden of NVA-HARTI on patient outcomes in neurocritical care is not fully known (20), and our primary aim was to report hospital and ICU LOS and mortality in this patient population. Our results are consistent with previously reported outcomes from mixed ICU in Spain (23). In our study, the ICU LOS was signi cantly higher in NVA-HARTI patients than in patients with no HAI. Such a difference was also reported before (23). Of note, the differences in LOS between patients with and without HARTI should be interpreted with caution as a number of baseline characteristics were different between groups (Table S3).
Whereas no available publications examine the mortality in neurosurgical ICU patients with NVA-HARTI, in mixed ICU population, mortality rate in non-ventilated patients with pneumonia was reported to be 23% within the 30 days of pneumonia onset (7). A 2010 mixed ICU study reported 36% hospital mortality in non-ventilator ICU-acquired pneumonia patients (23). Both rates are higher than those observed in our study. These studies included only patients with pneumonia, the most severe type of HARTI, while we included all respiratory infections that may ultimately skew the results toward lower mortality.
Additionally, the overall mortality in our study population was lower (10.8%) than previously reported in similar neurosurgical ICUs, 19.0% (22). Thus, lower NVA-HARTI mortality could be a re ection of lower overall mortality. Death rate in VA-HARTI patients was 26% higher than in NVA-HARTI patients. However, it was lower than reported previously for neurosurgical ICU patients with VAP, 28.4% (7). Of note, the mortality in patients with VAP in general ICU populations varies considerably across studies (4), and the latest European data suggests a rate of 37.7% (24).
In this study, NVA-HARTI was not associated with increased mortality while VA-HARTI independently increased chances of death by 54%. There is an ongoing discussion about the impact of VAP on mortality with controversial evidence (25). A 2017 study on ICU patients with intracerebral hemorrhage found that VAP independently increased in-hospital mortality, odds ratio 2.68 (2.58-2.77) (17). Similarly, a 2019 study in French mixed ICUs found that VAP signi cantly increased mortality by 38% (7). Conversely, a meta-analysis in patients with traumatic brain injury did not nd an association between VAP and mortality (20). Another study on patients in neurologic ICU also did not show that VAP in uenced mortality, but in that study the absolute number of patients who died in the VAP group and in relation to VAP was very low (26). When obtained from randomized prevention studies, a meta-analysis estimated an attributable VAP mortality to be 13% (27). Accounting for preexisting differences and evolution of the risk of death over time yielded an attributable mortality of 2.3% per day with VAP (28). The controversy around VAP impact on mortality can be due to a complex time-dependent balance of multiple factors contributing to overall risk.

Risks of NVA-HARTI and Comparison with VA-HARTI
We found that the risk of NVA-HARTI increased with increasing ICU stay in a non-linear manner. The highest daily risk was in the beginning of the ICU course followed by a smaller peak later on. A similar result was demonstrated previously for VAP (29). The fact that the risk is uneven throughout the ICU stay can be a basis for developing a exible prevention strategy with time-dependent preventive measures.

Strengths, Limitations and Further Directions
The strength of our study is the carefully curated and prospectively collected database that included a large cohort of neurosurgical ICU patients and many years of observation. Long period of observation allowed for a reliable estimate of infection and mortality rates. We utilized the method of survival analysis that was the most suitable for the speci cs of HAI in neurosurgical ICU and accounted for immortal time bias. This study also has some limitations including its observational nature and single-center settings.
Additionally, the de nition of HARTI was based on clinical parameters and subjective complicating comparisons to other studies. Accordingly, HARTI events may have also been tracheobronchitis or laryngitis episodes that may have otherwise overestimated infection rates and underestimated mortality rates. The metric of crude in-hospital mortality and lack of follow-up after the discharge also limited the ability to completely assess the impact of HARTI on mortality. Future multicenter studies may provide better estimates of the effect of HARTI on patient outcomes.

Conclusion
Our ndings suggest that NVA-HARTI in neurosurgical ICU patients represents a signi cant healthcare burden with relatively high incidence and associated poor outcomes. NVA-HARTI appears to be different from VA-HARTI and persisted despite preventive measures; therefore, extrapolating VA-HARTI research ndings to NVA-HARTI should be avoided. Declarations Ethics approval and consent to participate

List Of Abbreviations
The NSI Institutional Review Board approved this study. An informed consent waiver was obtained from the Institutional Review Board based on the study's observational nature. All data was anonymized before data analysis.  Incidence of HARTI in neurosurgical ICU patients. A, yearly cumulative incidence of HARTI per 100 ICU admissions with 95% con dence interval for binomial distribution. The mean incidence during the study period shown as a dotted line; B, incidence rates of VA-and NVA-HARTI per 1000 patient-days in the ICU; C, incidence rate of VA-HARTI per 1000 ventilator-days. Star (*) in the legend indicates groups with statistically signi cant time trend. In B and C we used incidence rate per quarter; regression line and pvalues obtained from the robust linear regression. Abbreviations: ICU, intensive care unit; NVA-HARTI, nonventilator-associated healthcare-associated respiratory tract infections; VA-HARTI, ventilator-associated healthcare-associated respiratory tract infections.