Future predictive modeling in emergency medicine will likely combine the use of a wide range of data points such as continuous documentation, monitoring using wearables, imaging, biomarkers, and real-time administrative data from all health care providers involved. Subsequent extensive data sets could feed advanced deep learning and neural network algorithms to accurately predict the risk of specific health conditions. Moreover, predictive analytics steers towards the development of clinical pathways that are adaptive and continuously updated, and in which healthcare decision-making is supported by sophisticated algorithms to provide the best course of action effectively and safely. The potential for predictive analytics to revolutionize many aspects of healthcare seems clear in the horizon. Information on the use in emergency medicine is scarce.

Aim of the study is to evaluate the performance of using routine-data to predict resource usage in emergency medicine using the commonly encountered symptom of acute neurologic deficit. As an outlook, this might serve as a prototype for other, similar projects using routine medical data for predictive analytics in emergency medicine.

Inclusion Criteria:

• Female and Male subjects
• Age ≥ 18 years

Exclusion Criteria:

- none

Retrospective analysis of routinely collected data. We aim to find data patterns associated with in-hospital resource utilization of patients hospitalized by emergency medical services for suspected acute neurologic deficit.

We will use only information from the routine electronic medical documentation system for this study. No linkage to other datasets will be performed. Data in the system includes patient demographics, initial symptoms, prehospital vital signs and scores, suspected diagnosis by emergency medical services, inhospital vital signs and scores, inhospital diagnostic (computed tomography CT, magnetic resonance imaging MRI) and therapeutic (catheter intervention) procedures and final diagnosis. See 'variables' for a full set of variables.

For the purpose of this study, only a fully pseudonomyzed dataset will be used.