Stroke-like symptoms are a frequent cause of presentations to hospitals.[[1]] Prompt neurovascular imaging is essential to assist with diagnosis,[[2]] and to identify the cohort most likely to benefit from hyperacute therapies. While stroke treatment is associated with improved patient outcomes and reduced overall hospital costs,[[3]] diagnostic imaging is the fastest-growing expenditure in healthcare.[[4]]

Magnetic resonance imaging (MRI) provides greater sensitivity and specificity for the diagnosis of ischaemic stroke than computed tomography (CT),[[5]] however, it is resource intensive and can result in a bottleneck in the inpatient journey. Protocols of MRI sequences in the work-up of stroke vary between centres. Some of these sequences may contribute little to diagnostic clarification, and increase both resource utilisation and healthcare costs. There are some reports of stroke centres effectively using more selective MRI sequences in the evaluation of stroke.[[6,7]]  

In September 2019, a short stroke protocol (SSP) was introduced at Palmerston North Hospital Te Whatu Ora, which limited MRI sequences to axial T2 fluid attenuation inversion recovery, and susceptibility and diffusion-weighted imaging. While CT is the preferred imaging modality for stroke patients, MRI is used in cases of diagnostic uncertainty or where infarct localisation is considered necessary to aid management. The authors hypothesised that the SSP obtained sequences most helpful in diagnosis, and would result in more efficient healthcare delivery, without compromising patient care.

The primary aim of this study was to determine if SSP is associated with a difference in length-of-stay (LOS) or in hospital admission costs compared to standard MRI protocol. The secondary aim was to explore any differences in the safety outcomes for 1) readmissions with a neurological diagnosis, or 2) death, in the subsequent 6 months.

Methods

We performed a retrospective cohort study of patients admitted to Palmerston North Hospital between September 2019 and September 2021 with an initial clinical impression of stroke or transient ischaemic attack (TIA). Patients were identified using electronic databases and included if they underwent an MRI (standard protocol or SSP) during their index admission. Patients were excluded if the MRI was for an indication other than a working diagnosis of stroke/TIA (e.g., evaluation of an intracranial mass seen on CT).

Baseline demographic data including age, ethnicity and co-morbidities were obtained from electronic clinical records. The final diagnosis was obtained by review of electronic discharge summaries. This was divided primarily into stroke (including TIA), and non-stroke pathologies (e.g., migraine, functional neurology, delirium, peripheral vertigo, etc). Data on LOS, survival and readmissions with a neurological diagnosis at 6 months were also obtained using electronic records.

MRI scans were coded as SSP or non-SSP by reviewing PACS imaging and the radiology report. We also recorded any additional sequences obtained (e.g., angiography). The time taken to perform each scan was obtained by the difference in timing between the first and last images displayed on PACS. Admission costs were provided by the analytics and financial advisory department at Palmerston North Hospital.  

Continuous variables are presented as means (SD) for normally distributed data and median (interquartile range) for non-normally distributed data. Categorical data are presented as absolute numbers and frequencies. T-test and Wilcoxon Rank-Sum tests were used to test for associations for normally distributed and non-parametric continuous data respectively. Logistic and quantile regression (since outcome data were non-normally distributed) were used to test for associations. Only 1.9% of the data were missing (considered missing at random); we used complete case analysis in this project. All statistical analysis was performed using STATA BE/17. This quality improvement project was exempt from requiring ethical approval. This manuscript was written in accordance with SQUIRE guidelines.[[8]]

Results

One-hundred and two patients were eligible for inclusion. Eighty-five patients (84%) underwent routine MRI, and 17 SSP. One patient was planned for standard MRI protocol but switched to SSP due to claustrophobia, and was included in the SSP cohort.

Baseline characteristics were generally well balanced (Table 1), though there was a significantly higher proportion of patients with a history of ischaemic heart disease in the SSP cohort (35% vs 11% respectively, p=0.009).

View Tables 1–2 and Figure 1.

Rates of discharge diagnosis of stroke were similar between groups (58.8% vs 52.7%, p=0.33). Of note, no patients in the SSP cohort had a final diagnosis of TIA, compared to 17 (20%) in the standard cohort. There was no significant difference between LOS in days between SSP and standard MRI cohorts (4 vs 5, p=0.40).

Imaging duration was significantly less in the SSP cohort compared to the standard cohort, even after multivariable regression adjusted for potential confounders identified in the univariate analysis (12 minutes vs 24 minutes, p=0.001), Figure 1. Admission costs were equivalent in both groups. There was no significant difference in readmissions with a neurological diagnosis or mortality at 6 months between cohorts (Table 2).

Discussion

In this study, the use of the SSP (with high yield limited sequences) was associated with reduced imaging time and therefore MRI resource use when used as part of an inpatient workup for stroke, without an apparent sacrifice of diagnostic yield or adverse patient outcomes. An intangible strength of the short stroke protocol may be better acceptability to patients (particularly those with claustrophobia).

Strengths of this study include the face validity of the methodology. Other limited MRI sequence publications have focussed on utility in decision making for reperfusion treatment. Our study is original in that it evaluates the use of SSP versus standard care for stroke/TIA admissions.  

Limitations of this study include low patient numbers, preventing our ability to draw further inferences on admission costs and safety outcomes. Limitations of our statistical analysis included the potential for type I error inflation due to multiple statistical testing, meaning that significant findings may be spurious. Further, the modest sample size may mean non-significant findings may represent type II error. The retrospective cohort design with no control over why patients received limited versus more extensive imaging may bias results. Further, the multivariate adjustment should be treated with caution due to low numbers in the SSP group.

The authors anticipated a higher number of SSP scans completed, however, electronic records had inaccurate coding of the type of MRI scan performed. The SSP cohort in our study had a greater disease burden and fewer TIA patients, which may have contributed to the neutral primary outcome findings. Therefore, we suggest a larger study be performed before changes to MRI protocols are adopted at other centres for the work-up of stroke or TIA.

‍