We have presented a model that combines geographic information systems (GIS) simulations of the lengths of emergency services times from an out-of-hospital cardiac arrest (OHCA) incident to defibrillation with data on survival rates from the Swedish Cardiac Arrest Register (SCAR). Simulations of ambulance alone as well as ambulance plus fire services were utilised as emergency resources. The results can be used to support rational planning processes regarding interventions that affect the alarm process for OHCAs, e.g. by using economic evaluations . When informed about the outcomes involved, the decision maker has the opportunity to make a more enlightened policy choice and select the least expensive way to achieve a specific objective.
Implementation of policies, e.g. changes in regions or dynamic changes such as location of emergency resources or which resources should carry a defibrillator, requires new GIS simulations of the type we showed for dual dispatch. Trimming of the static time intervals in the alarm process (alarm: 2 minutes; emergency call handling: 1.5 minutes; preparation: 1.5 minutes) is important and can be directly impacted by e.g. public education (to decrease time to first call) and optimizing call handling (to decrease time to emergency service first alert). Marginal effects of changes in these intervals on survival rates can be evaluated directly by our model.
The model has several limitations that could potentially bias the results. Among other things, it was assumed that all OHCAs occur in the patients’ homes and that the risk is identical in all homes, despite the fact that OHCAs have been shown to have definite time-geographic distribution patterns . E.g. commercial and business areas are more clustered during the day than at night and demographic factors such as age also matter. It has also been shown that there is a lower prevalence of and survival from initial VF from arrests at home versus public settings . This factor should be considered in further analysis, although the proportions of VF used in the model are general for all cases of OHCA. Inclusion of actual locations of OHCAs could be used to improve upon the specific geographic risk, but it is not certain that actual locations alone would improve the prediction of future OHCAs (compared to patients’ homes).
Also, although a flat rate reduction of the driving speed in localities was included in the model, we are uncertain whether it correctly captures variations in the driving speed of emergency vehicles. Traffic congestion, road works and the choices of route may be factors that complicate the simulation. The model does not take vertical distances, e.g. floors above street level, into account, which naturally leads to further delays in arrival at patients’ side. Actual call-to-on-scene times for real OHCA events could be used to validate the speed limit based numbers as well as the assumption that the response teams are always in their stations.
Moreover, there are uncertainties regarding the estimations of the static time intervals, e.g. the emergency call handling time. In our model the emergency call handling time is estimated to be 1.5 minutes on average, yet according to SOS Alarm AB, the company responsible for handling 112 emergency calls and coordinating rescue work, it might be as long as 4 minutes on average for ‘priority 1’ ambulance calls (Mikael Björkander, SOS Alarm AB, e-mail 23 October 2008). If so, there certainly exists potential to improve this time interval. Using SOS Alarm AB’s figure of 4 minutes, the simulation model yields the baseline numbers of survivors of 15 (ambulance) and 24 (ambulance plus fire services); the baseline level of survivors is as low as 2.2% and the additional lives saved through dual dispatch is 9 instead of 16 per year. One minute shorter emergency call handling time would yield a large benefit to the OHCA population, +3 survivors (ambulance) and +5 survivors (ambulance plus fire services).
It is clear that a sensitivity analysis of the results is necessary to provide a good basis for decisions. One further opportunity for improvement is to address variability in the model. We used point estimates for all components of the model. Variability around each of these estimates can have a compounding effect and ultimately affect the accuracy of the model. Assessment of the effects of errors in assumptions for values of variables used in the model should be addressed in some way, either internalized in the model or as a sensitivity analysis of the outcome e.g. .
With the limitations, the simulation results for dual dispatch comply well with the results from a ‘real life’ intervention. In addition, Pell et al. (2001) estimated that a reduced ambulance response time from arrival at the scene within 14 minutes in 90 per cent of all emergency calls to arrival within 8 minutes would increase the survival rate from 6 to 8 per cent; reducing the response time to 5 minutes would increase the survival rate to 10–11 per cent . Even if these results are not directly comparable with ours, they are in the same order of magnitude.
Although the analytic model was calibrated for the County of Stockholm, it can be generalised to other Swedish counties or regions, as well as other countries. The data that is needed to calculate the number of patients surviving as a result of a specific EMS design change in a region is very general and easily available (see Section 2). By adjusting the model specification for r (number of patients) and s (survival rate) we may calculate effects of other changes as well, e.g. the effect of earlier CPR or the effect of faster response to other diagnoses than OHCA. Before analysing the impact of the EMS design changes on other diagnoses, we need to establish a relation between survival rates as a function of time depending on the intervention considered.
The geographic location of defibrillators is possibly the most interesting factor to elaborate on and it also affects the marginal benefits of static response time changes, e.g. the increase in number of survivors was larger after introducing dual dispatch. When we consider geographic placement, the model is useful for analysing the effect of defibrillator placement in emergency vehicles (or at least professionals in moving vehicles, e.g. taxis, security personnel, home care staff). One natural step to further increase OHCA survival would be to introduce public access defibrillation (PAD), i.e. to place AEDs in the community for use by laypersons. The analytic model does not apply to this setting with stationary AEDs. However, although evaluations of PAD trials have shown a positive impact of survival, the cost-effectiveness of these programs remains uncertain. This is due to the large resources involved and the limited potential to save lives (less than 5 percent of OHCA occur in large public buildings) [21–23].
In summary, despite the complexity of modelling an intervention of this type, we believe that the results at hand are useful for deploying effective EMS system design strategies. The possibility of testing where defibrillators should be placed geographically is deemed particularly useful and focuses on first responder defibrillation, which has potential to have an effect in private homes where the vast majority of OHCAs happen . Extension of the tool to analyse changes in e.g. the proportion where CPR was started prior to arrival of emergency services is possible. More applications using GIS technology in time-sensitive emergency conditions would be equally interesting.