Patient selection
This longitudinal, register-based, observational study included all patients with infratentorial or large supratentorial (≥30 mL) ICH, treated at three comprehensive stroke units, within the Sahlgrenska University Hospital in Gothenburg, Sweden, between 1 November 2014 and 30 June 2019. Sahlgrenska University Hospital has a catchment area of 800,000 inhabitants and is the only hospital in the region with a neurosurgical department. The neurosurgical department has a catchment area of 1.8 million inhabitants.
All patients were identified in the local Väststroke Quality Register, held by Sahlgrenska University Hospital. Additional data were collected retrospectively from medical records, the National Stroke Register in Sweden, the Longitudinal Integration Database for Health Insurance and Labor Market Studies, the National Patient Registry (NPR), and the Swedish Cause of Death Register. The data were merged by the National Board of Health and Welfare using Swedish personal identification numbers. Exclusion criteria comprised patients with restricted medical records and those whose care was initiated outside the region. Follow-up of all-cause mortality continued for 1 year after the incident ICH for all patients.
DNAR decisions
DNAR decisions regard initiation of cardiopulmonary resuscitation in case of cardiac arrest, and does not indicate withdrawal of other care or treatment limitations in Sweden [6]. A DNAR decision must be taken by a licensed medical doctor well acquainted with the patient’s status and medical history, in consultation with another licensed healthcare professional. If possible, it must be discussed with both the patient and his or her next-of-kin. The DNAR decision must be clearly documented in the medical records [14]. If no DNAR decision is mentioned in the medical records, the patient is to receive cardiopulmonary resuscitation in case of cardiac arrest.
Data collection
Data collection from electronic medical records was conducted using a standardized procedure. We first searched specific notes or modules for the relevant information. Second, a search was performed using the search function and pre-specified search terms. Third, if the relevant information was not clearly stated, an interpretation of the written notes was relied on, when possible.
Information concerning DNAR decisions, palliative care, living arrangements, friends or family present at the hospital, complications, and treatments were collected within 14 days after hospital arrival. Physiotherapist and occupational therapist evaluations were recorded, if performed within the first 14 days. Physiotherapist and occupational therapist evaluations should be performed on all stroke patients without referral within 48 h according to Swedish guidelines. Assessment of consciousness scores, typically documented using the reaction level scale, were collected from the medical records during the first 7 days. These scores were converted into GCS scores [15].
Being dependent was defined as receiving home care services, nursing, or living at a nursing home prior to stroke. Physical activity was defined as regular light physical activity for at least 4 h weekly [16]. Income was defined as the total income the year prior to stroke, and was divided into tertiles. Educational levels were trichotomized as follows: < 10 years (primary school), 10–12 years (secondary school), and > 12 years (postsecondary or university education). Comorbidity data were collected from the NPR and used to calculate Charlson comorbidity index (CCI) scores [17]. The CCI scores were categorized as follows: no comorbidity (0 points), mild comorbidity (1-2 points), and severe comorbidity (> 2 points).
Hematoma volume and location data were collected from initial computed tomography or magnetic resonance imaging scans on hospital admission. To calculate intraparenchymal hematoma volume, the ABC/2 formula was used. The ABC/2 formula multiplies the longest hemorrhage diameter (A) with the perpendicular diameter (B) and the number of slices multiplied by the slice thickness (C). Ventricular breakthrough was registered, but not included in the calculation of hematoma volume. Midline shift was determined as deviance > 3 mm.
Statistical analysis
Multiple imputation by chained equations (MICE) was used to handle missing observations in the dataset. The following variables were imputed: being dependent (5.1% missing), living situation (3.2% missing), presence of next-of-kin at hospital (0.6% missing), pre-stroke physical activity (14.6% missing), and education (1.0% missing). The variables were imputed separately for patients with and without DNAR decisions. Descriptive data are presented as median with interquartile range (IQR) or mean with standard deviation (SD) for continuous variables, and number and percentage for categorical variables. Binary logistic univariable and multivariable regression models were conducted to predict DNAR decisions. As next-of-kin were present at the hospital for almost all patients (96.4%), and only one patient with no DNAR received palliative care (0.6%), these variables were excluded from the analyses. Multicollinearity was assessed using Spearman’s rank correlation for ordinal variables, and Phi coefficient for nominal variables. Goodness of fit was determined using a Hosmer and Lemeshow test (p > 0.05 indicated good fit). The explained variance of the models was determined using Nagelkerke’s R2. Area under the curve was assessed using receiver operating characteristic curves. Kaplan-Meier curves were used to visualize cumulative survival rates. A multivariable Cox proportional hazards regression was used to model the risk of one-year mortality. Schoenfeld residuals were used to test independence between residuals and time. All statistical tests were two-tailed and interpreted at a significance level of 0.05.