Current standard stroke care is compared with current standard stroke care supplemented with in-ambulance telemedicine using a cost-utility model measuring costs and quality adjusted life-years (QALYs) from a health care perspective. We combine a decision tree model (3 months) with a Markov model using a lifetime horizon (Figure 1). One possible implementation of the intervention (in-ambulance telemedicine) was previously described [4]. In-ambulance telemedicine allows the automated transmission of vital parameters (heart rate, blood oxygen saturation, and systolic and diastolic blood pressure), glycemia, electronic patient identification, functional assessments, and prehospital notification of the in-hospital team. Teleconsultants are not required to remain in the hospital to ensure 24/7 coverage because telemedicine support can be provided from any location with access to the Internet. A report of the teleconsultation containing all available information is immediately sent to the in-hospital team. Teleconsultants and ambulance personnel are trained to adequately use the telemedicine system. The in-hospital team is taught how to securely access and interpret the teleconsultation report.

All patients with suspicion of acute stroke are included in the model and are divided into 4 main categories: (1) stroke mimic, (2) transient ischemic attack (TIA), (3) ischemic stroke, and (4) hemorrhagic stroke. Ischemic strokes are divided into 3 treatment groups: (1) intravenous administration of IVT alone, (2) intravenous administration of IVT in combination with endovascular treatment (EVT), or (3) conservative care. Patients in the in-ambulance telemedicine model can either receive in-ambulance telemedicine on top of standard stroke care or standard stroke care alone. This additional arm accounts for the missed opportunities related to the accuracy of dispatchers to recognize a stroke and the proportion of patients not transported by an ambulance equipped with telemedicine technology.

To effectively model the time-sensitive nature of IVT with and without EVT, the treatment effect per time interval is modeled. For IVT alone, we assume that patients are treated up to 4.5 hours after stroke onset [13]. In combination with EVT, a positive effect is observed until 8 hours after stroke onset [12].

The comparator of standard stroke care is based on the performance indicators of Universitair Ziekenhuis Brussel (UZ Brussel), the university hospital of the Free University of Brussels (VUB), between 2011 and 2014. In this center, in-hospital prealerting of potential stroke patients and streamlined in-hospital workflows are part of standard medical practice [4].

For each end node in the decision tree, the modified Rankin Scale (mRS) at month 3 is used to classify patients into 3 categories: (1) favorable outcome (mRS 0-2), (2) unfavorable outcome (mRS 3-5), and (3) death (mRS 6). Data from the UZ Brussel Stroke Registry (2282 suspected stroke patients admitted to the UZ Brussel Stroke Unit between February 2009 and February 2014) are combined with literature data to populate the model for standard care (Table 1 and Multimedia Appendix 1). Each mRS group is associated with utilities. The cycle length of the Markov model is 1 year with a lifetime horizon. Patients enter the Markov model in the favorable or unfavorable state and transition to their current state, to a death state, or to a recurrent stroke state. Transition after a recurrent stroke is limited to the unfavorable or death state.

For the costs per treatment arm, we link hospital data and emergency claims data (including all payer costs) at the individual patient level (Multimedia Appendices 2 and 3). Claims for drugs, clinical biology, medical imaging, physicians’ honoraria, other claims charged to patients (copayments), and health insurances were included. To calculate the total hospital cost, a fixed day price was added according to the year of admission. This fixed day price covers the financing of nonmedical hospital activities. For this study, the weighted average per diem prices (across Belgian hospitals) was used [24]. All costs are expressed in and discounted to 2014, and Euro and US $ equivalents are calculated using the average 2014 exchange rate (€1=US $1.329). We do not model any productivity loss, as the average age of our patient cohort is 73 years. The cost-effectiveness threshold is set at US $47,747 (€35,927), the gross domestic product per capita of Belgium in 2014.

The impact of in-ambulance telemedicine on top of standard care is modeled by assuming an average time gain ranging from 5 to 60 minutes (Table 2 and Multimedia Appendix 4). This influences the treatment of stroke patients in 2 fundamental ways. First, probabilities of a positive outcome after treatment with IVT and/or EVT increase, since more patients are shifted into an earlier time window. Second, more patients can be treated with IVT and/or EVT as more patients shift into the applicable time windows, 4.5 hours and 8 hours, respectively, after symptom onset. Costs of the intervention are modeled by adding a fee per teleconsultation and a fixed fee per ambulance in which a telemedicine device is installed (Table 2 and Multimedia Appendix 5). Training of all stakeholders and mobile connectivity costs are included in the telemedicine installation cost.

Based on previous in-ambulance telemedicine pilot studies [4,5], we assume that 150 patients can be treated with 1 ambulance on a yearly basis, resulting in 3 ambulances to be equipped with the telemedicine technology for a patient cohort of 1000 suspected stroke patients (390 patients receiving in-ambulance telemedicine). All other parameters are assumed equal to standard medical stroke care.

Costs and QALYs are used to calculate the incremental cost-effectiveness ratio (ICER) after 3 months (decision tree only) and after a lifetime horizon (decision tree plus Markov model). Time gain after in-ambulance telemedicine is varied between 0 and 60 minutes, and cost for the implementation of in-ambulance telemedicine is varied between 50% and 400% of baseline cost in a 2-way sensitivity analysis, mapping the ICER for all combinations of both variables. Based on this analysis, we select the time gain for which in-ambulance telemedicine becomes cost effective. For this time gain, we perform a 1-way sensitivity analysis, varying all input parameters between 70% and 130% of their deterministic value and ranking the parameters according to the highest interval between calculated outcome parameters (both cost and QALYs are calculated). A probabilistic sensitivity analysis is applied using Monte Carlo simulations with 5000 bootstraps to account for the uncertainty around the input parameters and assess the robustness of the model. The cost-effectiveness acceptability curve is constructed for 5, 10, 15, and 30 minutes of time gain (1000 bootstraps). The health economic model was built and runs in Excel (Microsoft Office Professional Plus 2013, Microsoft Corp), and the UZ Brussel Stroke Registry was analyzed using Stata MP 13.0 for Windows (StataCorp LLC). The study was approved by the UZ Brussel ethical committee, and the model was validated by MF.

Under the base-case scenario (implementation cost of US $159,425) and taking a lifetime horizon into account, in-ambulance telemedicine is a cost-effective strategy compared to standard stroke care alone, starting from a time gain of 6 minutes (Figure 2).

After 12 minutes, in-ambulance telemedicine becomes a dominant strategy over standard best medical practice (Table 3). In a cohort of 1000 patients, 4.9 QALYs are gained (0.005 QALY/patient) and US $4040 (€3040) in long-term costs are avoided (–$4/patient). The savings of earlier stroke treatment outweigh the cost for implementation of in-ambulance telemedicine (cost equals US $159,425 [€119,959] for 390 patients receiving in-ambulance telemedicine) and higher utilization rates of specific stroke treatments (IVT and EVT). After 3 months, unfavorable outcome is avoided in 2.42 additional patients, resulting in long-term savings for society.

Not taking into account these long-term savings, in-ambulance telemedicine yields an ICER of US $201,557/QALY (€151,660/QALY) after 3 months. This incremental cost per saved QALY is explained by the cumulative cost of the intervention and the costs associated with more IVT and EVT after implementation of in-ambulance telemedicine.

One-way sensitivity analysis at 12 minutes time gain reveals that parameters involving outcome of ischemic stroke have the largest impact on calculated costs and QALYs (Multimedia Appendices 6-9). For likelihood of unfavorable outcome after conservative care, the incremental cost/patient varies between US –$252 (–€190) (130%) and US $244 (€184) (70%); for favorable outcome after conservative care, the incremental QALY/patient varies between 0.013 (70%) and –0.003 (130%). This is not surprising given the model’s rationale (shift from conservative care to IVT) and the time-sensitive nature of IVT. We note, however, that these parameters are taken from the analysis of pooled randomized controlled trials (RCT) [16,18], and the time-sensitive nature of IVT has been confirmed in larger clinical populations [11,25]. The proportion of patients receiving standard care (vs in-ambulance telemedicine) influences the outcome parameters of the model, indicating that regions with more ambulances equipped with telemedicine, with a higher ability of ambulance dispatchers to recognize a stroke, and with a higher proportion of patients being transported by ambulance will benefit more from in-ambulance telemedicine.

Under 4 times baseline costs (>US $500,000 implementation costs), in-ambulance telemedicine becomes cost effective after 19 minutes and is dominant after 39 minutes (Figure 2). For lower cost implementations (<US $70,000), in-ambulance telemedicine can be cost effective after 3 minutes and dominant after 7 minutes of achieved time gain.

The cost-effectiveness acceptability curves (Figure 3) reveal that under probabilistic analysis more than 90% of simulations are cost effective at the threshold of US $47,747, starting from 15 minutes time gain. This number drops below 80% under a scenario of 10 minutes time gain.

We report on the first ever comprehensive in-ambulance telemedicine cost-utility model combining a decision tree with a Markov model, using detailed cost information per treatment arm. In-ambulance telemedicine is dominant from a health care payer perspective starting from 12 minutes time gain.

Two previous publications describe health economic aspects of MSUs for improvement of prehospital stroke care [26,27]. Both models use the same source for calculation of improved outcome following faster IVT [16]. We applied the same methodology in our model by calculating the absolute risk difference based on the reported numbers needed to treat.

No impact of in-ambulance telemedicine on mortality was modeled, even though evidence exists that earlier stroke treatment reduces in-hospital mortality [11]. We chose not to model mortality because sufficiently specific information on mortality is not available in RCTs and because it may be unlikely that our population of 2200 patients would be sufficiently powered to model a possible impact on mortality.

The impact of in-ambulance telemedicine can vary greatly from 1 hospital or region to another depending on the standard quality and speed of care. Streamlining in-hospital workflows will always be a crucial part of a successful in-ambulance telemedicine implementation and can influence costs and potential benefit. For example, realized time gains could be more modest if prehospital notification through mobile phones or tablets is already part of current practice.

The major limitation of the presented model lies in the absence of information from RCTs evaluating the effects of in-ambulance telemedicine on costs and patient outcome. This drawback was addressed by implementing only solid criteria originating from RCTs for the outcome parameters. However, combining data from multiple trials is not without risk, and the results of this model should be interpreted carefully.

An advantage of in-ambulance telemedicine is the limited amount of additional resources needed from the hospital. If state-of-the art stroke care is available, no additional staff may be needed and existing ambulances can readily be equipped with the technology. However, depending on the catchment area and the number of stroke patients supported via in-ambulance telemedicine, additional teleconsultants on call may be needed. Although this could increase the organizational cost of in-ambulance telemedicine, more patients would be treated with in-ambulance telemedicine, which would further decrease the cost per patient and improve outcome. Training is required but was included in the cost of the telemedicine implementation.

Total implementation costs will decrease when in-ambulance telemedicine technology becomes more widely available. Lower cost alternatives such as tablet-based approaches are currently being investigated for remote stroke severity assessment in driving ambulances [6,7].

Widespread implementation of in-ambulance telemedicine will not only depend on its (cost-)effectiveness but also on the creation of the required legal framework. In Belgium, currently consultations are only officially recognized if face-to-face contact between the patient and treating physician occurs. Other issues include clear regulations for reimbursement and liability.

Our analysis only takes benefits of the expected time gain into account. Other benefits of in-ambulance telemedicine include a lower risk of stroke misdiagnosis and consequently missed opportunities for treatment with IVT or EVT and triage of patients to inadequate facilities [28]. Estimates of missed stroke diagnosis by emergency personnel range from 22% to 47%, indicating the potential for in-ambulance telemedicine to curtail the risk of misdiagnosis [29,30]. Reducing missed opportunities was not modeled here to avoid double counting when combined with faster treatment effects.

We excluded the implementation of rapid blood pressure lowering for hemorrhagic stroke patients in our model, even though indications exist on the benefit of this approach [19]. In-ambulance telemedicine could increase the proportion of patients receiving rapid blood pressure lowering, resulting in an underestimation of the potential cost effectiveness.

Further, expert prehospital care may help avoid secondary brain damage, as the teleconsultant can support the ambulance personnel in obtaining and maintaining homeostasis during ambulance transportation through optimal application of the standard operating procedures for airway protection, blood oxygen saturation, arterial blood pressure, heart rate, cardiac arrhythmia, decreased level of consciousness, dysglycemia, and other supportive measures (eg, antiemetics, analgesics).

The combination of in-ambulance telemedicine with MSUs is another interesting approach that could further decrease costs of the MSU as it would avoid sending highly trained physicians into the field for each individual patient. This approach is feasible, and preliminary analysis has shown that median time savings of 23 minutes between alarm-to-CT times can be attained when compared to standard care [31].

Recent clinical trials showing impressive benefits from EVT [32-34] herald a new era in acute ischemic stroke care. This highly effective treatment is resulting in a paradigm shift toward optimization of prehospital stroke diagnosis, identification of suitable candidates for IVT and EVT, and patient triage to appropriate centers [35]. A care model that avoids secondary transportations of stroke patients from primary stroke centers to comprehensive stroke centers for EVT is expected to result in better patient outcome for at least 0.2% of the patient cohort [10]. We did not take these effects of in-ambulance telestroke into account as their supportive evidence currently is insufficient to allow robust modeling. This probably results in an underestimation of the benefits yielded by in-ambulance telemedicine.

In-ambulance telemedicine has the potential to improve the organization of care for other medical emergencies, further strengthening the cost-effectiveness potential of this technology. The use of tablet computers as a support system for general emergency medical services and better patient triage have shown a decrease in transportation times by ambulances, showing the possibilities for further innovation in emergency care organization and delivery using telemedicine and mobile health solutions [36].

We believe that this positive health economic evaluation can inspire decision makers in hospitals and governments to actively pursue the implementation of and further research on in-ambulance telemedicine.

In-ambulance telestroke is highly cost effective from a health care perspective, resulting in more QALYs and less costs starting from a realized time gain of 12 minutes. The model is not directly based on results from RCTs on the effects of in-ambulance telemedicine, and trials to further crystalize these effects in various care models are needed. Support from governments and hospitals to facilitate implementation in clinical practice is indispensable and can be justified by the dominant cost effectiveness of in-ambulance telemedicine under several scenarios both in terms of implementation costs and time gain.