Prognosis after Intensive Care for COPD Exacerbation in Relation to Long-Term Oxygen Therapy: A Nationwide Cohort Study

Abstract Decisions to admit or refuse admission to intensive care for acute exacerbations of COPD (AECOPD) can be difficult, due to an uncertainty about prognosis. Few studies have evaluated outcomes after intensive care for AECOPD in patients with chronic respiratory failure requiring long-term oxygen therapy (LTOT). In this nationwide observational cohort study, we investigated survival after first-time admission for AECOPD in all patients aged ≥40 years admitted to Swedish intensive care units between January 2008 and December 2015, comparing patients with and without LTOT. Among the 4,648 patients enrolled in the study, 450 were on LTOT prior to inclusion. Respiratory support data was available for 2,631 patients; 73% of these were treated with noninvasive ventilation (NIV) only, 17% were treated with immediate invasive ventilation, and 10% were intubated after failed attempt with NIV. Compared to patients without LTOT, patients with LTOT had higher 30-day mortality (38% vs. 25%; p < 0.001) and one-year mortality (70% vs. 43%; p < 0.001). Multivariable logistic and Cox regression models adjusted for age, sex and SAPS3 score confirmed higher mortality in LTOT, odds ratio for 30-day mortality was 1.8 ([95% confidence interval] 1.5–2.3) and hazard ratio for one-year mortality was 1.8 (1.6–2.0). In summary, although need for LTOT is a negative prognostic marker for survival after AECOPD requiring intensive care, a majority of patients with LTOT survived the AECOPD and 30% were alive after one year.


Introduction
Chronic obstructive pulmonary disease (COPD) is the third most common cause of death globally [1], and is associated with high morbidity and risk of hospitalization including intensive care [2].Patients with advanced COPD may develop chronic respiratory failure with need for long term oxygen therapy (LTOT) [3].Acute exacerbations of COPD (AECOPD) account for a significant part of mortality and lead to progressively impaired lung function [4,5].AECOPD are complex events, with severity ranging from mild to life threatening.
The decision to forego or admit to intensive care is difficult due to an uncertainty about the patient's prognosis in the setting of a life-threatening AECOPD.The Global Initiative for Chronic Obstructive Lung Disease (GOLD) provides international strategy documents, including indications for admission to respiratory or intensive care unit [6].In the individual case, considerations include evaluating expected benefits, reversibility of the precipitating event, patient's wishes and availability of intensive care.Previous research have concluded that clinicians may underestimate the survival prognosis for COPD patients admitted to intensive care [7], and GOLD highlights the risk that, because of such prognostic pessimism, patients who might otherwise survive their AECOPD may be denied admission to intensive care [6].In the absence of reliable prognostic models, observational data of current clinical practice is valuable for clinicians to support the decision-making process.
Data are scarce on survival after AECOPD requiring intensive care in relation to the presence/absence of LTOT [8,9].Thus, there is a lack of prognostic data to guide care of LTOT patients with an acute exacerbation of their disease.Information about short and long-term survival would be useful for clinicians and patients when discussing respiratory support and admission to intensive care.
The aim of this nationwide study was to evaluate short-term and long-term survival in current clinical practice after intensive care for AECOPD comparing patients without or with LTOT.

Design and population
This was a nationwide, registry-based, cohort study of patients admitted to Swedish intensive care units (ICUs) due to AECOPD in the Swedish Intensive Care registry (SIR).SIR is a prospectively collected nationwide database of individual intensive care patient records; it covered 44/69 (64%) medical and general ICUs during 2008 and 62/64 (97%) in 2015 [10].Inclusion criteria were patients with an ICU admission with COPD as the main diagnosis from January 1, 2008 to December 31, 2015.Exclusion criteria were patients aged <40 years, patients with recent surgery (during the same hospital admission before ICU care), and those who could not be identified in the Swedish Population Registry or were not resident in Sweden at the time of ICU admission.For each patient, only the first ICU admission for COPD captured in SIR during the study period was included.If the patient was transferred between ICUs or re-admitted within 24 h from discharge, these care episodes were merged in the analyses.

Assessments
Data on LTOT at the time of ICU admission were obtained from the Swedish Registry of Respiratory Failure (Swedevox), which covers about 85% of patients starting LTOT in Sweden since 1987 [11].According to national and international guidelines, the main indications for LTOT in COPD were severe chronic hypoxemia (PaO 2 <7.4 kPa at rest) or moderate chronic hypoxemia (PaO 2 7.4-8.0kPa at rest) with right-sided heart failure or polycythemia [12][13][14].Adherence to the indications was high and stable during the study period, with 97% of patients having hypoxemia (PaO 2 <8.0 kPa breathing air) when starting LTOT [11], and the Swedevox data has been validated against data in medical records, with a high concurrence and validity [15].
ICU data were obtained from SIR on date and time stamps, age, sex, ICU admission details including location before ICU (stratified as emergency department, ward or other).Severity of illness on ICU admission was assessed according to Simplified Acute Physiology Score 3 (SAPS3) model and corresponding probabilities of death [16,17].We used one modification of the original SAPS3: the Reaction Level Scale 85 (RLS85) [18] was our default for assessing cerebral responsiveness.The RLS scores were transformed to the Glasgow Coma Scale (GCS) using a previously published protocol [19].Unconsciousness was defined as RLS85 ≥ 4 or GCS ≤8.Respiratory support and support mode was available in SIR as recordings of nursing workload and therapeutic measures.This data, which was available for a majority of admissions, was categorized as "non-invasive ventilation (NIV) only", "invasive ventilation only", or "NIV followed by invasive ventilation".Diagnoses in SIR were registered at discharge from ICU according to a selected list from the Swedish version of the International Classification of Diseases, 10 th revision (ICD-10-SE).COPD was during the studied period coded with a single code (J44.9).Data were validated locally and transferred to the registry for central validation (data were confirmed to be within pre-specified limits, and inconsistencies and illogical entries were identified) and, if necessary, returned for correction and revalidation before being accepted and added to the master database.
Mortality was followed up longitudinally from the date of ICU admission using the mandatory National Population Registry until December 31, 2016.Follow-up was complete except for five patients censored at the date of emigration.

Statistical analyses
Patient characteristics were tabulated using descriptive statistics.Continuous variables were compared using Mann-Whitney's test, binary variables were compared using Fisher's exact test, and categorical variables with more than two categories were compared using chi-squared test.Survival after ICU admission was analyzed using medians (interquartile range; IQR) and Kaplan-Meier plots using the log rank test comparing patients with/without LTOT, and comparing men and women.Relative risk of mortality within 30 days from ICU admission was analyzed using logistic regression adjusted for sex, age, and SAPS3 score, and was reported as odds ratios (OR) with 95% confidence intervals (CI).Relative risk of mortality within one year from ICU admission was analyzed using Cox regression adjusted for sex and SAPS3 score, and was reported as hazard ratios (HR) with 95% CI.For patients with complete data on respiratory support in the ICU, these multivariable models were extended to include respiratory support mode as an independent variable in separate analyses.In the group of patients with LTOT, separate logistic regression and Cox regression models, adjusted for sex, age, and SAPS3, were used to analyze if time from start of LTOT was associated with mortality.In analyses using age as a separate variable, age groups were formed according to the SAPS3 scoring table and the points for age were subtracted from the total SAPS3 score.
In secondary analyses, mortality was estimated in subgroups of participants classified according to sex and age.Homogeneity between the subgroup estimates were evaluated using likelihood ratio tests.The proportional hazards assumption was assessed graphically by studying Kaplan-Meier curves and log minus log plots.Statistical significance was defined as two-sided p-value <0.05.Statistical analyses were conducted using the software packages R version 3.6.3[20].

Results
A total of 4,648 patients admitted to ICU due to AECOPD were included, of whom 450 (9.7%) had ongoing LTOT at the time of ICU admission (Figure 1); median time from start of LTOT was 366 (IQR 120-728) days.Median age was 72 (IQR 66-78) years, 60.5% were women, and 12.3% of the patients had co-morbid chronic heart failure.Age, sex, and distribution of SAPS3 score were largely similar between groups.Compared with patients without LTOT, patients with had a slightly higher median age and SAPS3 score at admission.Median follow-up time was 400 (range 0-3286) days.Detailed characteristics of participants are shown in Table 1.

Outcomes after ICU admission
The overall 30-day mortality was 26.1%.The median overall survival time from ICU admission was 415 (IQR 24-1502) days, 92 (IQR 6-460) days for LTOT patients vs. 520 (IQR 31-1811) days for patients without LTOT, as shown in Figure 2a and Table 1.Compared to those without LTOT, patients with LTOT had higher mortality in the ICU (17.1% vs. 9.2%; p < 0.001), within 30 days (38.4% vs. 24.7%;p < 0.001), and within one year from ICU admission (70.4% vs. 43.3%;p < 0.001).Long-term survival was better for women than for men in the group without LTOT, while no significant difference was seen between men and women in the LTOT group (Figure 2b).
Analyzed with multivariable models adjusted for sex, age, and SAPS3 score, patients with LTOT had an increased risk of both 30-day mortality (HR 1.83 [95% CI 1.46-2.28])and one-year mortality (HR 1.79 [95% CI 1.58-2.02])compared to patients without LTOT.Older age was strongly related to mortality in all analyses.One-year mortality was higher for men compared to women in the multivariable analyses (HR 1.14; 95% CI 1.02-1.28).Detailed data from the multivariable analyses are presented in Table 2.Among patients on LTOT, time from start of LTOT was not associated with mortality within 30 days or one year.
Respiratory support data was available for 2,631 (56.6%) of the participants (60.4% of patients with LTOT and 56.2% of patients without LTOT).Among these patients, 1,916 (72.8%) were treated with NIV only, 268 (10.2%) were treated with NIV first and then converted to invasive ventilation, and 447 (17.0%) were treated directly with invasive ventilation.Fewer patients with than without LTOT were treated with invasive ventilation (15.8% vs. 28.5%;p < 0.001), see Table 1.The effect of LTOT seen on mortality was similar to the main analyses when adjusted for respiratory support modality (Table 3).
Patients without available respiratory support data had lower SAPS3 score, spent less hours in the ICU, and had lower mortality both in the ICU (11% vs. 21% in the LTOT-group; 6% vs. 12% in the group without LTOT), at 30 days (26% vs. 47% in the LTOT-group; 20% vs. 29% in the group without LTOT), and at one year (65% vs. 74% in the LTOT-group; 41% vs. 50% in the group without LTOT) after ICU admission, compared with patients with available respiratory support data (Supplementary Table S1).

Secondary analyses
No statistical significant difference was observed regarding sex for 30-day mortality (test of homogeneity, p = 0.11) or one-year mortality (p = 0.16).Regarding age, statistical significant differences were observed for both 30-day mortality (p = 0.03) and one-year mortality (p = 0.03).The association between LTOT and mortality was more pronounced for patients <70 than ≥70 years of age, both in 30 days (

Discussion
This nationwide observational study provides information about survival after AECOPD requiring ICU care in Sweden.Patients on LTOT had a more severe prognosis, but 62% survived for at least 30 days and 30% were still alive after one year, which indicates that even patients with COPD requiring LTOT can benefit from ICU care and survive an acute exacerbation, although long term mortality was high.
These novel findings extend those of a study of COPD patients admitted to an Australian ICU, which reported that 17% of surviving patients were on LTOT prior to ICU admission, while LTOT prevalence in deceased patients was 39% [9].No survival data after discharge was presented, and LTOT was not included in the multivariable analysis.An American study focused on long-term outcome in LTOT patients treated with invasive ventilation found a one-year survival similar to the LTOT patients in our material (32%) [8].In our study, only a minority of the LTOT patients were treated with invasive ventilation, which may indicate a difference in exacerbation severity, but may also reflect differences in clinical practice.
The lower rate of invasive ventilation in the LTOT group compared to patients without LTOT in our cohort may reflect a difference in ceilings of care between the groups.It is likely that life-sustaining treatment limitations, including do not intubate orders, were more common in the LTOT group, although we lack such data.NIV is recommended as first-line treatment for acute hypercapnic respiratory failure in AECOPD, with a reported success rate of 80-85% in RCTs [6], and there may be more doubts about potential ‡ special therapy = systemic corticosteroids daily ≥0.3 mg/kg prednisolone within the last 6 months, radiotherapy for invasive malignancy in the last 6 months, or chemotherapy.Abbreviations: iCu, intensive care unit; iQr, inter-quartile range; ltOt, long-term oxygen therapy; niv, noninvasive ventilation; saPs3, simplified acute Physiology score 3. benefits of invasive ventilation for patients on LTOT than for patients without chronic hypoxemia.We found no evidence for associations between respiratory support modality and mortality at 30 days or one year.However, this must be interpreted with caution particularly due to the lack of data on life-sustaining treatment limitations, which are frequent in this setting.In addition, missing respiratory support data was associated with less illness severity and, therefore, missingness was more likely in patients without LTOT.Most baseline data was similar for patients with and without LTOT.Compared with patients without LTOT, those with LTOT had lower PaO 2 and PaO 2 /FiO 2 ratio, which could be a direct reflection of their chronic oxygen-dependent hypoxemia.The shorter length of stay in ICU for LTOT patients compared to patients without LTOT may be explained by the lower rate of invasive ventilation.In general, women in our study had a better long-term survival than men after ICU care.However, within the group of patients on LTOT, we found no difference in survival between men and women.
Strengths of the present study include the nationwide design with a large number of patients and a consequently good statistical precision, and a reliable material expected to be representative for a Swedish context.Second, despite the retrospective analysis, data were prospectively recorded and thus not subject to recall bias.As described in the methods section, data in SIR were validated in two steps to avoid incorrect registrations.COPD diagnosis in the Swedish Inpatient Registry has been validated with an accuracy of over 90%, and the consistency between SIR and the Swedish Inpatient Registry is high [21].Previous studies from Swedevox have shown that the adherence to therapeutic indications for LTOT in Sweden is high [11,15]; thus, the LTOT group is well defined and interpretable also in an international context.Third, the use of unique personal identification numbers in Sweden provide reliable data for long-term follow-up.
Potential limitations include a lack of information on advance directives, particularly do not intubate orders, making interpretation of differences in respiratory support difficult.Because of a gradual introduction of SAPS3 as the scoring system for disease severity, SAPS3 data were missing for a large proportion of the patients during the first years of the study and consequently, these patients were not included in the multivariable models.Data on LTOT prescribed before 2008 was not captured in this study, and previous data indicate that approximately 15% of LTOT-patients in Sweden are not captured in Swedevox [11].Thus, some patients with LTOT may be misclassified to the no-LTOT group in this study.Such misclassification could cause a regression dilution bias, i.e. an underestimation of the difference between the groups [22].

Implications
This study provides survival data based on current practice, of value for clinicians and patients in dialogues and decisions about potential restrictions in care.Deciding who would benefit from intensive care of AECOPD can be difficult in clinical practice.A study from the United Kingdom found that clinicians frequently underestimated the survival prognosis for COPD patients admitted to intensive care, and the difference was most pronounced in the group of patients with the poorest clinician-estimated prognosis (actual 180 day survival 36% vs. predicted 3%) [7].Availability of observational data can improve the situation by providing knowledge about outcomes, and reduce the risk of unwarranted prognostic pessimism.The data from this study suggest that COPD patients on LTOT may benefit from ICU care in terms of short and long-term survival.Further research is needed to identify patient characteristics and useful therapies to be able to differentiate between patients who profit by an ICU admission from those who suffer.

Conclusions
In current clinical practice in Sweden, 30-day survival was 62% for COPD patients with LTOT and 75% for patients without LTOT admitted to an ICU for AECOPD.Adjusted for age, sex and SAPS3 score, patients on LTOT had an approximately 80% increased risk of death both at 30 days and one year after admission.In summary, although need for LTOT was a negative prognostic marker for survival after AECOPD requiring ICU care, a majority of patients with LTOT survived the AECOPD and 30% were alive after one year.

Figure 1 .
Figure 1.flow diagram for inclusion and exclusion of patients.

Table 1 .
Data on baseline characteristics at iCu admission, treatment, and outcomes.
* Co-morbidities according to sir.missing data considered as no diagnosis.

Table 3 .
multivariable regression models including respiratory support on the subgroup of patients with complete data.