Study Setting and Population
We conducted a retrospective, register-based study using the Finnish Intensive Care Consortium (FICC) database. The FICC database has previously been described in detail [
15]. Briefly, the FICC was established in 1994 as an ICU-benchmarking project, and all data were entered prospectively into the database. Today, all ICUs in mainland Finland, apart from one specialized unit, participate in the FICC. The database is maintained by TietoEVRY (Helsinki, Finland).
Neurosurgery and neurointensive care are provided only at five university hospitals in Finland. Four of these five units providing neurosurgical and neurointensive care (in the university hospitals of Kuopio, Oulu, Tampere, and Turku) participate in the FICC. These hospitals cover approximately two thirds of the population in Finland.
From the FICC database, we extracted data on patients admitted with a diagnosis indicating aSAH between 2003 and 2019 in these four units. aSAH was defined if the patient had an Acute Physiology and Chronic Health Evaluation III diagnosis indicating SAH and an International Classification of Diseases, 10th revision diagnosis of I60.0–I60.7.
We only included adult patients (age ≥ 18 years). We excluded foreigners and nonemergency admissions. Because of the low number of missing data, we excluded patients with missing data.
Definition of Covariates
We extracted all covariates from the FICC database. Age was measured on admission. The Glasgow Coma Scale (GCS) score was defined as the worst measured GCS score during the first ICU-day or as the last reliable GCS for intubated and/or sedated patients, according to the Simplified Acute Physiology Score (SAPS) II definition [
16]. Based on this GCS score, the patients were classified into patients with a good grade (World Federation of Neurological Surgeons [WFNS] grade I–III) and patients with a poor grade aSAH (WFNS grade IV–V) [
17].
Preadmission functional status was a modified version of the World Health Organization/Eastern Cooperative Oncology [
18] classification used in the FICC. Significant comorbidity was recorded if at least one of the SAPS II [
16] or Acute Physiology and Chronic Health Evaluation II [
19] comorbidities was present. Placement of an intracranial pressure probe or an EVD, as well as further information on ICU treatment, was obtained through the TISS-76 [
20] or TISS-28 [
21] recordings. For severity of illness adjustment, we created a modified SAPS II score without age, GCS, comorbidity, and admission type subscores.
Our primary outcome of interest was 12-month mortality. We also report crude hospital mortality rates.
Statistical Analyses
We used IBM SPSS Statistics for Macintosh (Version 26.0; IBM Corp, Armonk, NY) for the statistical analyses.
We report categorical data as numbers with percentages. We compared categorical data across groups by using a two-sided χ2 test. None of the continuous variables followed normal distribution according to the Kolmogorov–Smirnov test and visual inspection of histograms. Hence, we report medians with interquartile ranges and compared data across two groups by using the Mann–Whitney U-test and across several groups by using the Kruskal–Wallis test. p < 0.05 was considered significant in all analyses.
To test the association between admission period and 12-month mortality, we used univariate and multivariable logistic regression, reporting odds ratios (OR) with 95% confidence intervals (CIs). The multivariable model included age, sex, SAH grade (WFNS grades I–III and IV–V), preadmission dependency, significant comorbidity, and the modified SAPS II score. We included age and the modified SAPS II score as continuous variables.
We divided the admission period into three approximately equally long periods: 2003–2008, 2009–2014, and 2015–2019. This division ensured that every period had an adequate number of patients for the analyses. To test whether admission period was independently associated with 12-month mortality, we added the admission period (as a nominal variable) to the multivariable model described above. If this new model explained more of the variance in outcome (i.e., whether the difference between the log-likelihoods between the models was significant), this would suggest that admission period was independently associated with mortality. We report Nagelkerke’s R2 and Hosmer–Lemeshow (HL) test results, as well as the receiver operating characteristic area under curve (AUC) for both models.
We also calculated standardized mortality ratio (SMR), i.e., observed mortality divided by predicted mortality, for individual years of the follow-up period by using the multivariable logistic regression model without the admission period described above. For data on individual years, we used linear regression to test for a trend, reporting R2, and p values.
The association between admission period and mortality may depend on the treating hospital. Hence, as a sensitivity analysis, we also tested a generalized linear mixed model including the factors of the multivariable model as fixed effects and the treating hospital as a random effect.
We followed the Strengthening the Reporting of Observational studies in Epidemiology statement for reporting of results [
22].