Spatiotemporal Epidemiological Similarity Based on Patient Trajectories

Healthcare-associated infections are a significant public health concern. This study presents the Spatiotemporal Epidemiological Similarity based on Patient Trajectories (StESPT) method to provide clinicians with support in investigating the epidemiological relationships between infected patients, making detecting outbreaks and identifying possible transmission routes. To this end, the StESPT retrieves the movements of infected patients through the hospital and transforms them into trajectories. Then, a Trajectory Distance Measure Algorithm (TDMA) was used to quantify the spatiotemporal similarity between the trajectories of each pair of patients. Finally, based on the results of each TDMA, the k-means clustering algorithm was used to group patients, which can help understand the spread of the infection. We compared the suitability of three commonly used TDMAs (Dynamic Time Warping (DTW), Spatiotemporal Linear Combine similarity (STLC), and Spatiotemporal Longest Common Subsequence (ST-LCSS)) that were modified to measure similarity instance of distance. For each TDMA, we also proposed a version that adapts better to the semantics of our problem. The StESPT method was tested with a synthetic simulation of Clostridium difficile infection in a hospital. The results of the modified version of the ST-LCSS-WTW best reflected the spatiotemporal relationships between patients and led to the k-means clustering revealing two potential outbreaks.