Limitations of care and comorbidities are associated with increased mortality in patients treated with non-invasive ventilation: A retrospective observational study in a single- center ICU. [version 1; peer review: 1 approved with reservations]

Background: Non-invasive ventilation (NIV) is a common treatment for acute respiratory failure in intensive care units (ICU). While there is increasing data on outcomes after NIV treatment, there are large variations in staffing and monitoring where NIV is provided, making results hard to generalize. The aim of this study was to characterize patients treated with NIV, describe outcomes, and identify factors associated with outcome in an ICU at a Swedish county hospital. Methods: A single-centre retrospective observational study during 2018 of patients treated with NIV in a six-bed ICU at a Swedish county hospital. Patient characteristics, including comorbidities, details of ICU stay, simplified acute physiology score (SAPS-3), details of NIV treatment and 30-day mortality were collected, and the Charlson comorbidity index (CCI) was calculated. Primary outcomes were 30-day mortality and associated factors. Results: 92 patients with mean age (71,3, SD 12,1) were treated with NIV during the study period. 42 (46%) were women. Median CCI was 3 (25th-75th percentiles 1.4)) and median SAPS-3 score was 66 (25th-75th percentiles 58). The 30-day mortality was 37% and in the univariate analysis, SAPS-3 score >66, Charlson comorbidity index, CCI>=3, pCO2 <5.5 and limitation of care were factors associated with increased 30day mortality. pH <7.35 and pO2<8 at admission showed no associations with 30-day mortality. Open Peer Review


Introduction
Non-invasive ventilation (NIV), i.e. respiratory support through a non-invasive interface such as a facemask or nasal prongs, is an evolving therapy with a broad application in intensive care units (ICU) [1][2][3] . NIV reduces the risks related to invasive mechanical ventilation in patients with acute respiratory failure in e.g. chronic obstructive pulmonary disease (COPD) and pneumonia 4 . Further, NIV improves survival after pulmonary oedema and respiratory failure of different aetiologies 5-8 .
A delayed start of NIV increases mortality 9 . Improper selection of patients and continuing treatment when NIV is not sufficient leads to unnecessary harm to the patients and delayed intubation and invasive mechanical ventilation, which is associated with increased mortality 10 .
The selection of patients who would benefit from an ICU-admission with NIV-treatment is often difficult, as the patients often present with a complex clinical picture 11 . For decision support, there are several models that predict ICU outcome, such as the Simplified Acute Physiology Score (SAPS-3) 12 and the Acute Physiology and Chronic Health Evaluations (APACHE II) 13 . Further, the Charlson comorbidity index (CCI) 14 , a frequently used measure of comorbidity, is also able to predict mortality 15,16 Patients treated with NIV frequently have "do not intubate" (DNI) orders 17,18 . It has been shown that the patients with DNI orders treated with NIV for acute respiratory failure who survive have similar life quality after three months compared to patients without limitations of care 19 .
The settings in which NIV is utilized differs between hospitals and countries where NIV is being increasingly used outside of the ICU 4,20 . The outcome for patients treated with NIV in the acute care setting for specific diagnoses has been extensively studied 7 . However, there is limited knowledge of the outcomes in a mixed population receiving non-invasive ventilation at an ICU.
The aim of this study was to describe mortality after NIV at a Swedish county hospital and to identify factors associated with mortality.

Patient population and study design
A retrospective observational single-centre study conducted in a six-bed ICU at a county hospital (250-bed, Sundsvall, Sweden). Patients >18 years who received NIV in the ICU from January 2018 to December 2018 were included in the study. The study was approved by the Swedish Ethical Review Authority, Uppsala, Sweden (2019-04564, 2019-09-27) and patient consent was waived. The study protocol was performed in accordance with relevant guidelines and the study was registered in clinicaltrials.gov (NCT04115969).

Data collection
Out of the 547 ICU-admissions during 2018, eligible patients were identified in the local ICU-registry. If a patient had more than one ICU-admission during 2018, only the first admission was included. Data sources were medical records, laboratory records, the local ICU-registry and ICU-charts. Variables collected were patient characteristics, admission status (diagnosis, laboratory findings and risk-scoring), details of ventilatory support, arterial blood-gases, limitation-of-care decisions, and 30-day mortality. Further, overall 30-day mortality and the number of patients with limitations of care was retrieved for all patients admitted to the ICU during 2018.

ICU-scoring
The SAPS-3 and APACHE-II-scores were retrieved from the local ICU-registry. The scores were validated for each patient by a review of the records.

Comorbidities
Charlson comorbidity index (CCI) was used to quantify comorbidities and the CCI was calculated with weighting (according to the original study), with a maximum score of 33 14 .

Statistical analysis
All variables were registered in an anonymized spreadsheet (Excel, version 1902; Microsoft Excel, RRID:SCR_016137. A free alternative software is Google sheets). Data are presented as numbers (rates), mean (standard deviation, SD) or medians (25 th -75 th percentiles) as appropriate.
Univariate analysis was performed using the Chi-2 test to analyse factors associated with 30-day mortality and presented as unadjusted odds-ratios with a corresponding 95% confidence interval. P-values lower than 0.05 were considered statistically significant.
The statistical analyses were performed using SPSS version 25.0 (IBM SPSS Statistics, RRID:SCR_019096 ). An open-source alternative software is R, which can perform equivalent functions (R Project for Statistical Computing, RRID:SCR_001905).

Patient characteristics
The inclusion of patients is presented in Figure 1. 92 patients were treated with NIV during the study period. The mean age was 71 years (SD 12) and 42 (46%) patients were females. Reason for admission to the ICU included respiratory (n=82; 87%) and/or cardiac causes (n=36; 38%). 30 patients (32%) were admitted from the emergency department, 56 (60%) from a hospital ward, 6 (6%) from the operating theatre and 2 (2%) from other ICUs. Further patient characteristics are presented in Table 1.

Characteristics of ICU stay
The median time spent in the ICU was 1.8 (0.8-5.0) days and the median duration of NIV-treatment was 17 (6-34) hours. 24 patients (26%) were also treated with invasive ventilation during the ICU stay. A total of 38 patients (41%) had limitations of care. 34 (36%) with 'no intubation' and 37 (39%) 'no CPR'. Further details of the ICU stay are presented in Table 2.
Outcome and predictors of 30-day mortality The 30-day mortality in the study cohort was 37% (n=34). In the univariate analysis, SAPS-3 score >66, CCI >=3, pCO2 <5.5 kPa and limitations of care were factors associated with an increased 30-day mortality, see Table 3.
Characteristics of the ICU 30-day mortality for all patients admitted to the ICU during the study period was 21% (n=547). Further, 12% of all patients had limitations of care.

Discussion
Our retrospective study included all patients treated with NIV at a county hospital ICU during one year. We found that one third of these patients had died within 30 days, and the mortality was almost double compared to all patients admitted to the ICU. Further, 30-day mortality was associated with the presence of co-morbidities and limitations of care.
A recent meta-analysis evaluating noninvasive ventilation and survival in acute care settings showed a hospital mortality of 11% 7 , a much lower mortality than in our cohort. However, there were great variabilities in the studies included in the meta-analysis, regarding the severity and type of conditions treated. A French study evaluating outcome after acute respiratory failure in 14 different ICUs showed a hospital mortality of 30% in patients treated with NIV, which is comparable to our results 4 . The high mortality in our cohort might also be explained by advanced age, severity of disease, and a high number of patients with limitations of care.
NIV was used in every fifth patient in our ICU, a higher amount number compared to a study of more than thirteen   We found that a high SAPS3-score, a high Charlson Comorbidity Index (CCI), limitation of care and a low pCO2 on admission were associated with an increased mortality. SAPS-3 is a validated score to predict ICU mortality, but when used for patients undergoing NIV in intermediate care units, the prediction accuracy could be lower, but by combining SAPS-3 with patient-risk factors, the mortality prediction can be improved 22 . This is in line with our results, as we found an association between mortality and co-morbidities.
Further, several studies identify CCI as a predictor of ICU mortality 15,16,23 . A study of 347 patients outside an ICU undergoing NIV due to pneumonia concluded that in-hospital mortality was mainly associated with patients' co-morbidities and limitations of care, rather than the degree of acute respiratory failure 24 . The evidence is conflicting however, as others have not found CCI to be a good predictor for short and medium-term outcomes in patients submitted for NIV due to acute respiratory failure 25 . However, our data indicates that CCI might be considered as a predictor of mortality in NIV patients in an ICU setting.
pCO2 on admission is not included in SAPS3 or APACHE II scoring systems. pCO2 has been studied as a prognostic indicator for patients admitted with pneumonia, but no significant association has been found with mortality 26 . We found that patients with a high pCO2 had a lower mortality. This might either be explained by a suggested mortality benefit with NIV treatment in hypercapnic COPD patients 6 or that patients with normal or low pCO2 had conditions causing respiratory failure with a higher mortality.
41% of our patients had limitations of care, in contrast to all our ICU patients where only 12% hade limitations of care. The observations are in line with a recent meta-analysis which concluded that 27% of patients with acute respiratory failure had a do-not-intubate order 27 .
It is a difficult clinical task to identify patients that would benefit from NIV treatment, and several factors might influence the selection of patients admitted to an ICU 28 . Since limitations of care are common among these patients, it is important to set clear goals with the NIV treatment 29 . An older age, ARDS or pneumonia, or failure to improve after one hour of treatment is associated with higher risk of NIV failure in hypoxemic patients 30 . Better prediction tools that can be used clinically, before the ICU admission, are needed 31 .
Except for details of comorbidities, pCO2 and NIV-parameters, our data included the same variables that is sent to/included in the Swedish intensive care registry and this makes it possible to perform similar studies on a national level. However, variables are usually collected during the ICU-stay and many of the variables are not known to the clinician when the decision to start NIV treatment is made. Therefore, a strength with the CCI is that it can be calculated at the time of admission to the ICU, in contrast to physiology-based scores that are dependent on laboratory and bedside clinical data. CCI as a predictor of outcome in the population treated with NIV in ICU has not previously been studied and our study indicates that it might be a usable tool.
Our study has limitations and strengths. Patients treated with NIV are a heterogenous group and NIV was used in many different clinical conditions making it hard to generalize results and our single study setting further decreases external validity. However, even though our study was retrospective, our review of patient records added more variable than included in national intensive care registries. Further, we included almost all patients admitted to our ICU for one year. This is a strength as it reflects a general cohort of patients in the need of NIV treatment.

Conclusions
We conclude that patients treated with NIV in ICU are a diverse population where comorbidities and presence of limitations of care might be considered as better predictors of 30-day mortality, rather than physiological parameters.