Victorian patients residing in metropolitan area with symptoms suspected of stroke diagnosed by paramedics were simulated. The demographic characteristics of the simulated patients were defined according to a study examining the diagnostic performance of ACT-FAST in Victoria, Australia. Briefly, a total of 522 patients were screened between November 2017 and July 2019 by paramedics from Ambulance Victoria which is the sole service provider for medical emergency transfers in the state of Victoria. Study patients were transferred to 32 hospitals including 15 metropolitan and 17 rural with only the metropolitan hospitals having EVT and neurosurgery capacity (four CSCs in metropolitan Victoria have full-time EVT capabilities). LVO was defined as the intracranial internal carotid artery (ICA), first segment middle cerebral artery (M1-MCA) and basilar artery occlusions. Extended LVO includes complete or near-occlusion of extracranial or intracranial internal carotid, middle cerebral (M1/proximal-to-mid M2), basilar and proximal posterior (P1) cerebral arteries, cerebral artery dissection or symptomatic intracranial atherosclerosis. Diagnostic performance for diagnosing LVO was assessed in the simulation model. Ethical approval was provided by the Royal Melbourne Hospital Research Ethics Committee with a waiver of patient consent.

Data from the validation study may be requested from the corresponding author in the anonymised form by any qualified investigator.

ACT-FAST involves a two-step examination plus an eligibility and stroke mimic screen step. The former includes (1) unilateral arm drift to stretcher <10 s, (2) severe language deficit (if right arm is weak) or gaze deviation/hemineglect assessed by simple shoulder tap test (if left arm is weak), and (3) eligibility and stroke mimic screen (4). The eligibility and stroke mimic screen are intended to check whether patients meet the criteria for EVT procedure and to exclude patients with prior severe disability/brain tumor/seizures. The details of the algorithm are reported elsewhere (4). The sensitivity and specificity in identifying patients with LVO stroke were informed by the ACT-FAST validation study. It was assumed that LVO stroke constitutes 10% of all strokes. On average, the time to screen patients on the scene was <1 min.

A decision-tree combined with a Markov cohort model was developed to quantify the cost-effectiveness of a severity-based bypass strategy using ACT-FAST. The comparator was the current standard care pathway where all patients are transferred to the nearest thrombolysis center (i.e., primary stroke center, PSC) or CSC, and those requiring EVT who presented to PSC initially may need secondary transfer (i.e., drip and ship model). Patients who received EVT were also considered for intravenous thrombolysis using standard eligibility criteria (Figure 1). In the decision-tree component of the simulation model, the sensitivity and specificity of ACT-FAST screening tool were incorporated to comprehensively assess the variations in the long-term cost-effectiveness. Following either a positive or negative screening outcome at the scene, patients were then classified as true-positive, false-positive, true-negative and false-negative. In this way, the consequences associated with false-positives (i.e., patients without LVO but screened positive) and false-negatives (i.e., patients with LVO but screened negative) were considered and quantified. In addition, the proportion of patients being true-positive (patients with LVO and screened positive) but that did not receive EVT due to other considerations (e.g., established infarct) were also accounted for in the modeled simulation.

For the standard care pathway where ACT-FAST is not available, a proportion of patients in the model were transferred to the PSC initially while CSC was the closest option for the remaining metropolitan patients. Regardless of the capabilities of the hospital where the patient initially presents, an imaging examination will identify a proportion of patients who are eligible for EVT and, among these, a proportion will actually receive EVT. For patients not requiring EVT, thrombolysis or other evidence-based medical treatment will be administered. The proportion of patients who had EVT indicated but were firstly transported to the closest PSC, then require inter-hospital transfer to a CSC for EVT (i.e., a time delay in receiving EVT).

Functional outcomes in ACT-FAST and standard care pathways were determined by the 3-month modified Rankin Scale (mRS) grade with shifts estimated from known outcome-time relationship curves for both thrombolysis and EVT (1, 5). If patients survived the acute event, the age-dependent death rate was applied, adjusted by the hazard ratio (HR) of mortality due to stroke (6). The structure of the simulation model is shown in Figure 1. TreeAge pro healthcare was used for the model-based simulation (TreeAge Pro 2020, R2. TreeAge Software, Williamstown, MA).

A lifetime time horizon with a yearly cycle (i.e., modeled until all patients had died) was chosen to accumulate the costs and benefits associated with ACT-FAST triaging vs. current standard care pathway adjusted by the half-cycle correction. Quality-adjusted life year (QALY) calculated based on the life year (LYs) survived and quality of life of that survival. Costs and benefits were discounted by 3% annually. An incremental cost-effectiveness ratio (ICER) was estimated based on QALY and LY gains to determine the long-term cost-effectiveness of ACT-FAST triaging system only if it was positive. An often-quoted willingness-to-pay per QALY of A$50,000 was adopted as the decision criteria (13).

To examine the robustness of base case results, both one-way deterministic sensitivity analysis and probabilistic sensitivity analysis were undertaken. For the one-way deterministic sensitivity analysis, screening performance of ACT-FAST (i.e., sensitivity and specificity), proportion of LVO stroke, HR of mortality after stroke, utility weights by mRS score, costs of transfer and thrombectomy procedure, time horizon, and discount rate were tested (Table 1). In addition, the variation in time saved from stroke onset to EVT time (i.e., mean of 59 mins and 95%CI 40.0–61.5 mins) by ACT-FAST and minimal time saved (i.e., 30 and 10 mins) were tested in the sensitivity analysis (2). Moreover, the diagnostic performance of other triaging tools (RACE, LAMS, C-STAT) derived from paramedics validations studies was examined in the current simulation model (14, 15, 21).

For the probabilistic sensitivity analysis, distribution of key drivers of ICER identified from deterministic sensitivity analysis were incorporated (additional methods in Supplementary Material). A total of 5,000 parameters were drawn from each distribution to run through the simulation model and the average of these 5,000 iterations plus 95% confidence interval (CI) were reported. The probabilistic sensitivity analysis assumed that all the distributions were independent (i.e., the variation in one variable does not correlate with another).

Further, the diagnostic accuracy of ACT-FAST algorithm for an extended definition of LVO, reflecting those with evolving EVT eligibility, was examined in the sensitivity analysis. Diagnostic parameters used were a LVO prevalence of 15% and specificity (81.8 vs. 77.9%) was improved at the expense of lower sensitivity (75.8 vs. 82.6%).

Impact of the widespread application of the ACT-FAST screening algorithm across Australia was examined. The total number of strokes in Australia from 2019, together with the proportion of LVO strokes in metropolitan area were used to estimate the possible costs and benefits at the national level, excluding two regions (Tasmania and Northern Territory) that transfer interstate to access EVT. The possible point estimate of the weighted QALY gains from the simulation model was multiplied with the size of national population to gauge the potential costs and QALY gains across Australia. The Net Monetary Benefit from implementing ACT-FAST triaging was estimated.

All 517 patients from the ACT-FAST paramedic validation study were used to define a hypothetical cohort (N = 10,000) in this simulation analysis. The average age of patients was 72.3 year (standard deviation, SD 15.6) and 50.2% were males. A total of 168 (32.5%) patients were screened ACT-FAST positive. Baseline brain imaging identified 92/517 (17.8%) LVO strokes using the standard LVO definition. Of 74 patients who underwent EVT, 68 cases were in the metropolitan region and 42/74 (56.8%) patients received secondary EVT transfer. All the patients (including regional and metropolitan) were included for diagnostic accuracy measurement of ACT-FAST however the EVT time saving which was applied in the simulation modeling were solely based on metropolitan patients.

The mRS score across all the simulated patients at 3 months for ACT-FAST vs. the current standard care pathway showed that a similar proportion of patients in the ACT-FAST scenario (70.75 vs. 70.24%) achieved the functional independence (i.e. mRS ≤ 2).

Over the lifetime, ACT-FAST was associated with lower costs (–$45) and greater QALY gains (0.006) compared to the current standard care pathway, resulting in it being the dominant strategy (less costly but more health benefits). Due to a small proportion of extra patients arriving at EVT-capable centers within the EVT time window from ACT-FAST bypass strategy (who would otherwise miss the EVT treatment window), there would be additional EVT procedure performed. Our simulation estimated that implementing ACT-FAST triaging led to higher proportion of patients received EVT procedures (30 more additional EVT performed per 10,000 patients) (Table 3).

The one-way deterministic sensitivity analysis indicated that base-case ICER was sensitive to the sensitivity and specificity of ACT-FAST, hazard ratio for mortality post stroke defined by the mRS scores and time horizon (Figure 2). On the other hand, base case ICER were less sensitive to the utility weights associated with each mRS score post index stroke, and costs relating to thrombectomy and hospital transfer.

The probabilistic sensitivity analysis suggested that ACT-FAST bypass strategy had 99.1% probability of being dominant in triaging suspected stroke patients, indicating lower costs and higher benefits (Figure 3).

Moreover, comparing with other triaging tools (i.e., CPSSS, LAMS, and C-STAT), ACT-FAST bypass strategy remained to be optimal in terms of greater benefits and/or lower costs (Supplementary Table 2). Even if the time saved for EVT commencement from such bypass strategy was reduced to 40 mins, it was still better than the current usual care pathway. The results using average time saved (59 mins) for EVT are highly consistent with the base case results (Supplementary Table 3).

Results using the sensitivity and specificity for an extended definition of LVO were further improved than those from the base case where the LVO definition was employed. The results from this sensitivity analysis are shown in Supplementary Table 2.

Based on the number of hospitalisations in Australia due to stroke in 2016–2017 (n = 38,055) (22) then excluding patients with non-ischemic and non-LVO stroke, it was estimated that 4,567 patients may be eligible for EVT treatment at baseline in a single year. Assuming that a similar proportion of LVO would benefit from a bypass strategy (i.e., ~50% of LVO patients were from metropolitan areas) from these key states/territories (i.e., accounting for over 96.95% of all strokes nationwide), extrapolating the per patient cost saving and QALY gained would result in a maximum lifetime of A$99,623 cost saving and 13 QALYs gained. Based on a willingness-to-pay threshold of A$50,000 per QALY (23), the total Net Monetary Benefit from ACT-FAST care was estimated at A$0.76 million based on its implementation for a single year.