Stroke remains one of the leading causes of death and disability globally, underscoring the importance of timely and effective prehospital care to improve patient outcomes. This thesis aims to use artificial intelligence’s power to enhance prehospital stroke care. To accomplish this, we study challenges in prehospital stroke care by focusing on three interrelated research challenges: Mobile stroke unit (MSU) allocation, ambulance travel time estimation, and improving travel time calculations within emergency medical service (EMS) simulation. We develop and analyze different optimization and machine learning (ML) methods to achieve improved analysis and planning of prehospital stroke care. In particular, we propose methods to solve the MSU allocation problem, which aims to identify the optimal locations for a fixed number of MSUs at the existing ambulance station locations within a geographic region. Moreover, we develop a machine learning-based regression method for ambulance travel time estimation. Next, we apply our pre-trained ML-based regression method to improve ambulance travel time estimation within an EMS simulation framework.

For the MSU allocation problem, we first propose a mathematical model, which we apply to identify the optimal MSU locations in the Blekinge and Kronoberg counties of Sweden. The experimental findings show both the correctness of the suggested model and the benefits of placing MSUs in the considered regions. Second, we propose a Genetic algorithm (GA) method with an efficient encoding scheme for the input data, representing the number of MSUs and potential sites. Additionally, we develop custom selection, crossover, and mutation operators tailored to the specific characteristics of the MSU allocation problem. We present a case study on the Southern Healthcare Region in Sweden to demonstrate the generality and robustness of our proposed GA method. Particularly, we demonstrate our method’s flexibility and adaptability through a series of experiments across multiple settings. Third, we propose the enhanced genetic algorithm with clustering (EGAC). By leveraging clustering, the EGAC provides diverse and comprehensive coverage, avoiding the pitfalls of starting with closely located and potentially less optimal solutions, thereby effectively steering and accelerating its convergence towards the optimal MSU placements. Our experimental results show that the EGAC significantly outperforms the traditional genetic algorithm, which does not make use of cluster-based starting solutions, by achieving remarkably faster convergence towards the optimal solution for different numbers of MSUs to allocate. We illustrate the performance of the EGAC through qualitative and quantitative analyses.

For ambulance travel time estimation, we propose an ML-based regression method for estimating ambulance travel times. Ambulance travel time estimations play a pivotal role in ensuring timely and efficient emergency medical care by predicting the time needed by an ambulance to reach a specific location. Overlooking factors such as local traffic situations, day of the week, hour of the day, or the weather may create a risk of inaccurately estimating the ambulance travel times, which might lead to longer emergency response times, potentially impacting patient outcomes. We propose a machine learning approach to accurately estimate ambulance travel times, particularly using regression models and real-world spatiotemporal data from the Skåne region, Sweden. Our method includes data preprocessing and feature engineering, with a focus on variables significantly correlated with travel time. Through a comprehensive exploratory data analysis, we highlight the main characteristics, patterns, and underlying trends of the considered ambulance data set. We present an extensive empirical analysis comparing the performance of different machine learning models across different ambulance travel trips and feature sets, revealing insights into the importance of each feature in improving the estimation accuracy.

Another focus of this thesis is the use of our ML-based regression method to improve the ambulance travel time estimation within EMS simulation. To illustrate the effectiveness of the proposed regression modeling, we utilize a modeling construction framework to construct an EMS simulation model for stroke patients and applied it in a scenario study covering Skåne County, Sweden. The result of the simulation shows differences in am- bulance driving times when using the ML-based module compared to existing routing engines designed for passenger cars. The observed differences emphasize the impacts of integrating ML-based estimations into EMS simulations.