In this cohort study we aimed to investigate the relative importance of comorbidities, age and sex for the odds of death within 60 days of ICU admission (60-day mortality) in COVID-19, sepsis and ARDS. 60-day mortality is an established mortality measure in COVID-1917. The study was approved by the Regional Ethics Committee of Uppsala (approval no. 2016/421) and the Swedish Ethical Review Authority (approval no. 2020-02144). Informed consent was waived by the same authority because of the nature of the study. We registered the study à priori at ClinicalTrials.gov (NCT04542538) and the research was conducted in accordance with the Declaration of Helsinki with subsequent revisions. Reporting follows the STROBE (strengthening the reporting of observational studies in epidemiology) guidelines18.
All general and most specialist ICUs report all admissions to the Swedish intensive care registry (SIR)19,20. The national patient registry (NPR), a research support tool, was established by the Swedish Board of Health and Welfare and reporting is governed by statutory and common law21. We collected data on ICU diagnoses, demographics, ICU care and mortality from the SIR and we received data on comorbidities reported in the five years preceding ICU admission from the in-patient sub-registry of the NPR for all patients.
We compared three groups of adult (age ≥ 18 years) ICU patients diagnosed with COVID-19, sepsis or ARDS. Sepsis was defined as severe sepsis or septic shock according to the Sepsis 2 criteria22, coded with International Statistical Classification of Diseases and Related Health Problems—tenth edition (ICD-10) A49.9, R65.1 or R57.2 in the SIR. ARDS was defined according to the American-European consensus conference on the ARDS definition23, 2011–2015, or the Berlin definition6, 2016 and coded with ICD-10 J80.9 × in the SIR. All ICU-admitted adult COVID-19 patients in Sweden from 6 February 2020 to 16 June 2021 were identified by ICD-10 code U07.1 in the SIR while virtually all Swedish ICU-admitted adult patients with severe sepsis, septic shock or ARDS were identified in the SIR from 2011 to 2016.
Any single Sepsis patient could be included in the ARDS cohort and vice versa. Accordingly, ARDS patients were non-COVID-19 ARDS patients and Sepsis patients were non-COVID-19 Sepsis patients. Patients were only included for their first admission for COVID-19, Sepsis, or ARDS. However, as the COVID-19 group stems from a separate time period a patient could be included in both the COVID-19 and Sepsis or ARDS groups. Exclusion criteria were lack of personal identification number and age < 18 years. ICU care episodes ending and starting in the same 24-h period were merged.
Data are reported as medians with interquartile range (IQR) or number with percent in brackets. The primary outcomes were the relative importance of age, sex and comorbidities (Table S1) for 60-day mortality in COVID-19, Sepsis or ARDS.
The relative importance of age, sex, Simplified Acute Physiology Score 3 (SAPS3)24 Box III, and comorbidities were assessed as an interaction with the disease group (COVID-19 or Sepsis and COVID-19 or ARDS) using logistic regression. COVID-19 was compared separately to Sepsis and ARDS. A significant interaction between disease group and a variable indicates a difference in effect between groups for that variable. Because we added age and comorbidities in the models and treatments preceding ICU admission might be related to diagnosis the SAPS3 Box III, representing the acute physiologic derangement at ICU admission, was used. SAPS324 is a risk score initially developed to perform risk adjusted comparisons of hospital mortality in ICU admitted patients between and within ICUSs, but is now widely used and validated also for 30- and 90-day mortality25,26.
We used restricted cubic splines in all continuous variables, age and SPAS3 Box III, as we could not rule out a non-linear relationship with the logit of outcome. To estimate individual risk factor p-values a linear representation of the variable was applied to the model adjusted for the splined variables. We found 14 marginally influential observations in the model on 60-day mortality in COVID-19 and ARDS using the rms-package. We found indications of multicollinearity in relation to age and SAPS3 Box III for all models. SAPS3 data was missing in 414 patients (1.3%), who were excluded from 60-day mortality modelling. Due to an imbalance between groups for the different hospital types, hospital type was added to the models.
Statistical significance was defined as p-value < 0.05 (two-sided). In analysis of crude differences between disease groups we used the Mann–Whitney U-test and Chi2-test as appropriate with Bonferroni-correction because of multiple comparisons. Odds ratios (ORs) were calculated between the 25th and 75th percentiles in variables for which restricted cubic splines were applied, i.e. age and the SAPS3 Box III. Data management and descriptive statistics were performed in SPSS for Windows version 27 (Microsoft Corp., IL, USA). For multiple imputations, regression models and graphics, we used the R Software version 4.0.3 (The R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org) with the mice, rms, Hmisc, and forest plot packages.
The specification and rationale for the performed sensitivity analyses are found in Supplementary Table S2 online.