TIPS was a clustered randomized controlled trial that involved 20 hospitals from three Australian states: Victoria, New South Wales, and Queensland. All hospitals that participated in the TIPS study had either a Stroke Care Unit or staffing equivalent to a stroke physician and a nurse, and an emergency department. Ethical approval for the TIPS study was obtained from relevant human research ethics committees in each state, from each participating hospital, and The University of Newcastle Human Research Ethics Committee. The study adheres to CONSORT guidelines (Supplement 1).
Hospitals were randomized, stratified by baseline intravenous thrombolysis rate, either to receive a multi-component multi-disciplinary collaborative intervention, which focused on the safe use of intravenous thrombolysis therapy for AIS patients; or to continue with standard care. Blinding was not possible because of the involvement of staff in the intervention activities. Pre-intervention data were collected for each hospital for 12-24 months before implementation of the intervention. Following the pre-intervention period, a 16-month active intervention period occurred during which intervention hospitals received the intervention while control hospitals continued with standard care. The intervention was then withdrawn, and outcomes monitored during a 12-month post-intervention period.
Data on all thrombolysed cases were entered into a TIPS study-specific database in a de-identified form by the hospital staff. The variables of interest were pre- and three-month post-thrombolysis mRS and rates of post-thrombolysis parenchymal haematoma as detected in routine clinical practice following guideline based post-thrombolysis imaging recommendations10. Hospital staff entering data were trained in the Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) characteristics of the European Cooperative Acute Stroke Study 2 (ECASS 2) classification system and were asked to review source imaging in the classification of haemorrhagic transformation11. The TIPS study-specific database included questions and an algorithm to ensure consistency in the classification of the mRS. At 90 days post-admission, patients were contacted by hospital staff to record the mRS either by phone or in a clinic. We used two dichotomous definitions of the 90-day clinical mRS outcome to reflect the most favourable and most catastrophic outcomes: excellent clinical outcome (mRS 0-1 vs 2-6), and poor outcomes (mRS 5-6 vs. 0-4). Patients who were thrombolysed also had their baseline, and where available follow-up, imaging recorded; all patients received a baseline non-contrast CT at a minimum. Hospital staff for the presence of a hemorrhage using the ECASS 2 scoring system assessed all 24-hour imaging11.
Haemorrhagic events were classified according to clinical and CT criteria. Haemorrhagic infarction 1 (HI1) was defined as small petechiae along the margins of the infarct; haemorrhagic infarction 2 (HI2) as confluent petechiae within the infarcted area but no space-occupying effect; parenchymal haemorrhage (PH1) defined as blood clots in 30% or less of the infarcted area with some slight space-occupying effect; and parenchymal haemorrhage (PH2) defined as blood clots in more than 30% of the infarcted area with substantial space-occupying effect11. For our study, we defined both PH1 and PH2 as post-treatment parenchymal haematoma (PH).
Other data of interest included age, gender, pre-stroke mRS, pre- and post-thrombolysis National Institutes of Health Stroke Scale (NIHSS) which measures the severity of stroke; and pre-thrombolysis systolic blood pressure (SBP) on admission which is a risk factor for poor outcome.
The intervention was a multi-component, collaborative intervention that was based on a knowledge translation approach and the behavioural change wheel12 (Figure 1). The intervention sites signed a written collaborative agreement and then participated in a site-specific situational analysis which was followed by a collaborative workshop, teleconferences, feedback, and monitoring as detailed in Figure 1. The control sites were not provided with any intervention and could be considered a ‘usual care’ condition.
The main study was powered to detect a difference in thrombolysis rates between the two groups6. Assuming the same parameters as the original power calculation (cluster coefficient of variation = 0.4; alpha=0.05; 10 clusters per arm), this secondary analysis of all thrombolysed cases (n=1559) had 80% power to detect absolute differences in these key secondary outcomes between intervention and control groups of between 22% and 33% for outcomes with prevalence’s ranging from 50% to 75%. The analysis population is all those that were treated with intravenous thrombolysis. Data was not available on patients that were considered for intravenous thrombolysis but not thrombolysed. Primary analyses compared outcomes between intervention and control group at the active intervention and post-intervention phases. Three mixed effects logistic regression models were used to assess the difference in the study outcomes, proportion of patients with excellent and poor clinical outcome and with PH, between the intervention and control arm during the both active and post intervention period separately. These three models included fixed effects for baseline thrombolysis rates (site-level), pre-morbid mRS, baseline NIHSS; treatment group. Another three mixed-effects logistic regression models were used to determine the effect of the educational intervention on changes in clinical outcomes from pre to active and pre to post-intervention period. These three models included fixed effects for baseline thrombolysis rates (site-level), pre-morbid mRS, baseline NIHSS; treatment group, period (pre vs post) and the interaction between treatment and period. All the above mentioned models included a random intercept for hospital site to account for correlations of individuals within the same site. Due to missing data on mRS outcome, multiple imputation analyses using the chained regression equations method were performed. The missing data was imputed based on hospital site, pre-morbid mRS, NIHSS, age and gender. The imputation process (n=200 imputations) was conducted assuming the data were missing at random and combined using Rubin’s method. Statistical significance was defined as a two-tailed p-value of <0.05. Statistical analyses were programmed using SAS v9.4 (SAS Institute, Cary, North Carolina, USA).