Association Between Systemic Corticosteroid Therapy and Mortality in Critically Ill Patients with Acute Exacerbation of Chronic Obstructive Pulmonary Disease: A Retrospective Cohort Study

Whether the patients admitted in intensive care unit (ICU) due to acute exacerbation of chronic obstructive pulmonary disease (AECOPD) would benet from systemic corticosteroid treatment remains controversial. In this study, we aim to investigate the relationship between the systemic corticosteroid treatment and mortality in patients with severe AECOPD, using the public Medical Information Mart for Intensive Care III database (MIMIC III). The baseline characteristics, laboratory examination and prognosis of AECOPD patients were retrieved from MIMIC III. Multivariate logistic regression was used to analyze the association between systemic corticosteroid treatment and prognosis. Propensity score matching (PSM) was employed to validate our ndings.


Abstract Background
Whether the patients admitted in intensive care unit (ICU) due to acute exacerbation of chronic obstructive pulmonary disease (AECOPD) would bene t from systemic corticosteroid treatment remains controversial. In this study, we aim to investigate the relationship between the systemic corticosteroid treatment and mortality in patients with severe AECOPD, using the public Medical Information Mart for Intensive Care III database (MIMIC III).

Methods
The baseline characteristics, laboratory examination and prognosis of AECOPD patients were retrieved from MIMIC III. Multivariate logistic regression was used to analyze the association between systemic corticosteroid treatment and prognosis. Propensity score matching (PSM) was employed to validate our ndings.

Conclusion
Our study observed that Systemic corticosteroid therapy did not improve the mortality of severe AECOPD patients in ICU and aggravated the mortality of patients with high eosinophils. The results need to be veri ed in randomized controlled trials.

Background
Chronic obstructive pulmonary disease (COPD) is one of the top three causes of death worldwide [1]. Exacerbations play an important role in COPD progression and accelerate the decline in lung function of the patients, resulting in reduced physical activity and an increased risk of death [2][3][4]. Treatments for acute exacerbation of chronic obstructive pulmonary disease (AECOPD) includes antibiotics, bronchodilators, corticosteroids and respiratory support (https://goldcopd.org/2021-gold-reports/).
The best available evidence on the e cacy of systemic corticosteroid therapy on the outcomes of AECOPD is summarized by Walters et al. in a cochrane meta-analysis of randomized controlled trials which contained sixteen studies [5]. It found that patients treated with systemic corticosteroids either injections or tablets compared with dummy treatment were less likely to experience treatment failure (122 fewer people per 1000 treated) with high quality evidence but corticosteroid treatment did not reduce the mortality up to 30 days [5]. However, most of the studies with regard to systemic corticosteroids for AECOPD excluded people who had signs of a very severe exacerbation or consistent with signi cant comorbidity. Only two of them [6,7] involved the patients mechanically ventilated or admitted to the intensive care unit (ICU). The results of these two studies were contradictory to some extent. The

Data Extraction and Participants
Data extraction was performed on BigQuery and reference codes could be accessed from the MIMIC Code Repository [9] (https://github.com/MIT-LCP/mimiccode). We obtained the related information on patients who were diagnosed as AECOPD on discharge using AECOPD ICD-9-CM code 491.21. The patients with missing data (basic information or blood eosinophil concentrations) were excluded. The de nition of AECOPD was episodes of worsening of respiratory symptoms that results in additional therapy [10]. Eosinophil initial was de ned es the rst record value of blood eosinophil concentrations of each hospital admission. Eosinophil max was de ned as the maximum value during the hospitalization and eosinophil mean the arithmetic mean of all values during hospitalization. Some patients had multiple admissions to hospital. For these patients, only the most recent hospital was used [11]. The detailed process of data extraction is shown in Figure 1. We collected the following variables for the rst day of ICU admission: age at the time of hospital admission, gender, admission type, sequential organ failure assessment (SOFA) score [12], simpli ed acute physiology score II (SAPS II) [13] and systemic in ammatory response syndrome (SIRS) score [14]. Severity of illness scores were from the materialized views generated by the code from the github website (https://github.com/MIT-LCP/mimic-code/tree/master/concepts) and were calculated based on the data obtained within the rst 24 h of each patient's ICU's stay. Comorbidities in categories proposed by Elixhauser et al. [15] used an algorithm by Quan et al. [16].
All the prescription records of systemic glucocorticoids (GC) are extracted and then merged into the patient information of AECOPD. In brief, the codes provided in Supplementary material 1 was used to extract the glucocorticoid prescription records from table PRESCRIPTION. Then, systemic glucocorticoid prescription records for patients diagnosed with AECOPD was extracted and merged into the basic information table. Cream, eye drops, suppository, inhalation and nasal spray types of glucocorticoids are excluded.

Outcomes
The primary outcome was 30-day mortality. The secondary outcomes included in-hospital mortality and hospital length of stay (LOS). Since a patient might have more than one admission, hospital mortality and length of hospital stay were determined only by the recent hospital admission. For patients who were discharged alive, the date of death was from the social security database. The hospital LOS was strati ed into two levels based on the median: hospital LOS level 1 (≤8 days) and level 2 (>8 days). The logistic regression models were selected to explore the relationship between systemic corticosteroid therapy and outcomes. Potential confounders were chosen based on univariate analysis and clinical experience. Two multivariate modes were used to adjust potential confounders: (1) model 1, variables were adjusted for age (years), Elixhauser comorbidity score, SAPS II, SIRS, invasive mechanical ventilation, and initial value of eosinophil concentration; (2) model 2, the same as model 1, but SAPS II was replaced with SOFA and the Elixhauser comorbidity score replaced with speci c comorbidities, including cardiac arrhythmias, liver disease, metastatic cancer, solid tumor, coagulopathy, obesity, drug abuse and psychoses.
Subgroup analysis was performed by restricting to patients with different EOS initial concentration. Two cut-off values were tried (EOS initial concentration < 0.35% or ≥ 0.35%, EOS initial concentration < 2% or ≥ 2%).
Relationship between the route of administration of systemic corticosteroids and outcomes was explored by categorizing GC group according to administration route (intravenously, orally, or combination of intravenous and oral administration).
Propensity score matching (PSM) was employed to validate our ndings. Propensity score (PS) measures the probability of a patient being assigned to systemic GC treatment. The propensity score for an individual was estimated based on the given covariates of SOFA score, SAPS II score, SIRS score, eosinophil initial concentration, congestive heart failure, peripheral vascular disease, hypertension, other neurologic disorder, hypothyroidism, obesity, blood loss anemia, depression using a logistic regression model. One-to-one nearest neighbor matching with a calliper width of 0.02 was applied in the present study. P-values between the and the GC cohort and non-GC cohort were calculated to assess the balance between groups after matching. Logistic regression was then performed to adjust for residual imbalance by including parameters with p < 0.05 and potential confounders judged by clinical expertise.
Data analyses were performed by STATA 15.1 and SPSS 22 software; a two-tailed p < 0.05 was considered statistically signi cant.

Baseline Characteristics and Outcomes
MIMIC-III contained 943 unique subjects with 1,198 hospital admissions due to AECOPD, and a total of 882 patients were included nally in our cohort. The baseline characteristics of included patients are summarized in Table 1  Of the study cohort, 293 patients had been exposed to systemic glucocorticoid treatment, and the remaining 589 patients did not have prescription records for systemic glucocorticoids. Table 2 shows the baseline characteristics for the GC and non-GC groups before and after propensity score matching. In general, patients in the GC group were more critically ill than the non-GC group, with signi cantly higher SOFA score (median 4 vs. 3; p = 0.002) and SAPS II score (median 40 vs. 37; p = 0.009). Patients in the GC group were more likely to be enrolled during 2000-2008. Admissions times due to AECOPD were balanced between the two groups. The results of multivariate logistic regression analyses for primary outcome and secondary outcomes are shown in Table 3 and Fig. 2

EOS initial initial value of eosinophil concentration
The logistic regression was adjusted for age (years), Elixhauser comorbidity score, SAPS II, SIRS, invasive mechanical ventilation.
In this study, intravenous administration did not signi cantly affect 30-day mortality compared to oral administration (p = 0.101). However, a signi cantly negative effect on 30-day and in-hospital mortality was observed when comparing intravenous administration to co-administration method (intravenous administration and oral administration) as shown in Table 5.

Propensity Score Analysis
The 258 patients who did not received systemic corticosteroid treatment were matched to 258 patients who received systemic corticosteroid treatment by PSM. After PSM, the imbalance in the covariates between the non-GC and GC groups was signi cantly minimized (Table 2). Since there were still residual imbalances between GC and non-GC groups, multivariate logistic regression was then used. As shown in Table 6, systemic corticosteroid treatment was not associated with improved 30-day mortality in overall population (OR = 1.144, 95% CI: 0.744-1.759; p = 0.541). However, the adjusted OR showed a close to statistically adverse effect on hospital LOS when comparing the GC group to non-GC group (OR = 1.395, 95% CI: 0.964-2.019; p = 0.077) (  The logistic regression was adjusted for SAPS II, SIRS, invasive mechanical ventilation, mean value of eosinophil concentration and psychoses.

Discussion
In this retrospective cohort study based on a large critical care database MIMIC-III (v1.4), we found that systemic corticosteroid treatment did not decrease 30-day mortality or shorten hospital length of stay in critically ill patients with AECOPD in ICU.
Biologically, glucocorticoids can relieve airways in ammation and in return reduce the airways resistance to improve the dyspnea and the work of breathing [17]. The use of systemic glucocorticoids has been Combining these two studies, the review by Walters et al. [5] drew the conclusion that there was no difference in length of ICU stay and the duration of assisted ventilation for corticosteroid treatment compared with the control for AECOPD patients requiring assisted ventilation in the ICU setting.
Short-term administration of systemic glucocorticoids may cause secondary infections, hyperglycemia, and a range of mood and behavioral changes [19]. Adverse effects of the long-term therapy include osteoporosis, hypertension, myopathy, and adrenal insu ciency [19][20][21][22]. Increased risk of hyperglycemia requiring treatment following systemic glucocorticoid usage was both reported by Alia et al.
[6] and Abroug et al. [7]. The risk-bene t balance of systemic steroids is negative from the results of the study by Abroug et al. [7]. In this study, we observed no statistically signi cant difference in 30-day mortality and in-hospital mortality between systemic corticosteroid group and the control group (Table 3 and Fig. 2). In addition, patients received systemic corticosteroid treatment had a greater chance of being hospitalized for more than eight days from the multivariate logistic regression (Table 3 and Fig. 2). The result was similar after PSM (Table 6). Taken the side effects into consideration, we recommend that the systematic administration of corticosteroids in severe COPD exacerbation in ICU needs to be cautious.
Some randomized studies showed that therapy with oral prednisolone is equally effective to intravenous administration of glucocorticoids [23,24]. A pharmacoepidemiological cohort study conducted at 414 US hospitals involving almost 80,000 AECOPD patients to a non-intensive care setting demonstrated that the risk of treatment failure (in-hospital mortality, initiation of mechanical ventilation, or readmission for AECOPD within 30 days of discharge) among patients treated with low doses of oral steroids was not worse than for those treated with high dose intravenous therapy [25]. However, the study regarding the route of steroid administration with AECOPDs in ICU has not been carried out. In this cohort study, oral administration presented similar e cacy with intravenous and the combined administration, with no difference in 30-day mortality and hospital LOS (> 8days). Besides, patients receiving intravenous steroids had higher risk of death within 30 days or in hospital, as well as a greater probability to stay in hospital for more than 8 days compared to the combined administration. Real-world information reminds us that studies with regard to steroids treatment should not be limited to the comparison of oral and intravenous administration, the co-administration method should also be included.
Eosinophil levels may be a helpful marker to predict outcomes in AECOPD and to direct corticosteroid therapy during exacerbations [26][27][28][29]. Singh et al. [30] assessed the prevalence of eosinophilic in ammation in COPD subjects and found that 37% of the COPD subjects had blood eosinophil counts persistently ≥ 2%. However, the prevalence of eosinophilic in ammation in COPD subjects requiring ICU admission has not been well de ned. In this cohort study, only 17.57% of the subjects with a COPD exacerbation requiring ICU had a peripheral blood initial eosinophil concentration greater than 2% and the median of initial blood eosinophil concentrations was 0.4% which was lower than other researches [31][32][33] aimed at AECOPD. A previous study drew the receiver operating characteristic (ROC) curve for the prediction of in-hospital mortality by eosinophil concentrations in hospital admissions diagnosed with AECOPD in MIMIC-III v1.4 database [34]. The area under the ROC curve for initial eosinophil concentration was 0.608 and the discriminatory eosinophil thresholds were 0.35% (sensitivity = 0.59, speci city = 0.61) for in-hospital mortality [34]. Thus, the cut-off value 0.35% was also used to distinguish subgroups.
Lower-eosinophilic patients were found to experienced poorer clinical outcomes in a prospective, multicenter, observational cohort study [31]. In addition, a retrospective observational cohort study showed that COPD exacerbations with acute respiratory failure requiring ICU admission had a shorter median length of ICU stay and lower mortality with a peripheral eosinophil level > 2% [24]. Similar results appeared in our research. As shown in Supplementary Table 1, increased blood eosinophil level was associated with decreased 30-day mortality in our study (OR = 0.688, 95% CI: 0.688-0.946, p = 0.008).
Targeting corticosteroid therapy in a subgroup of exacerbations dependent on the peripheral eosinophil count may be helpful to reduce inappropriate use of systemic corticosteroids [35]. Recent studies suggested that glucocorticoids may be more e cacious to treat acute COPD exacerbations in patients with higher levels of blood eosinophils ( ≥ 2%) [26][27][28][29]. However, things seem to be different in ICU. Little evidence was found that systemic corticosteroids provided bene t in patients with a blood eosinophil initial concentration < 2% in our study Table 4. What's more, in the subgroup with a blood eosinophil initial concentration ≥ 2%, patients treated with systemic corticosteroids had a signi cantly increased inhospital mortality compared with non-GC group (OR = 6.645, 95% CI: 1.537-28.723, p = 0.011). When the cut-off value set to 0.35%, the effect of eosinophil level on the e cacy of glucocorticoids on the mortality of severe AECOPD patients was more apparent. As shown in Table 4, there was no difference regarding 30-day mortality and in-hospital mortality in the steroid-treated and control groups in the subgroup of patients with initial eosinophil concentration < 0.35%. Critically, in the subgroup of patients with initial eosinophil concentration ≥ 0.35%, corticosteroid treatment resulted in worse outcomes compared with non-GC group. The adjusted odds ratio of 30-day mortality and in-hospital mortality for glucocorticoid therapy were 1.691 (95% CI: 1.002-2.855, p = 0.049) and 2.247 (95% CI: 1.218-4.147, p = 0.010) respectively.
Therefore, recommendation of systemic corticosteroid treatment is not supported in critically severe AECOPD with a high blood eosinophil initial concentration in ICU by these results until a more precise evidence emerges. More researches are needed to explore the cut-off value for eosinophils and the reasons for the con icting results between ICU patients and non-ICU patients.
Although this study was relatively considerable sample size, it has several limitations. First, our research is a database-based retrospective single-center study. The inherent bias could not be avoided. Outcomes in this study only included 30-day mortality, in-hospital mortality and hospital LOS (> 8days), which were not as exible as randomized controlled trials. Second, systemic steroids use in the 30 days prior to hospital admission was unknown. This may cause some bias in the results though it is closer to the real world. Third, only blood eosinophil concentrations were considered in this study because the numeric value of the blood eosinophil counts was hugely lacked in the MIMIC-III database. The subgroup analysis will be more complete if blood eosinophil counts were considered together.

Conclusion
In this cohort study, our results showed that systemic corticosteroid therapy was not associated with improved mortality in critically ill patients with acute exacerbations of chronic obstructive pulmonary disease. Patients who have received glucocorticoid therapy with AECOPD in the ICU have a higher probability of being hospitalized for more than eight days. In the subgroup of critically ill patients with high blood eosinophil concentration, corticosteroid treatment resulted in worse outcomes compared with non-GC group. Administration of systemic steroids in severe AECOPDs in ICU should be very cautious, especially for patients with high eosinophils. Further randomized controlled trials are needed to con rm these results. and Massachusetts Institute of Technology (protocol 0403000206) have approved the data collection and the use of MIMIC-III for research purposes and granted waiver of informed consent. This study is reported following the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) statement.

Consent for publication
Not applicable Availability of data and materials The data used for this study can be accessed from the MIMIC-III database (https://mimic.physionet.org/).

Competing interests
The authors declare that they have no competing interests.