Main goals: Hypoglycemia is among the most relevant acute complications of diabetes mellitus. It affects the neurocognitive and psychomotor function and has consistently been shown to be associated with an increased risk of driving mishaps and accidents. To prevent these, our project aims at designing, implementing and evaluating a vehicle hypoglycemia warning system “HEADWIND”. The overall goal is to detect hypoglycemia in an early stage with a sensing module and then trigger direct interventions through a support module. For this purpose, we draw upon current automotive sensor technology and innovative digital markers of vegetative function.
Novelty: Hypoglycemia has been widely studied in medical research, yet the focus has been limited to the ex-post understanding of how behavior and especially driving patterns are afflicted. This differs substantially from the current project, as our key innovation is to apply machine learning to neuro-cognitive and psychomotor function, affected automotive parameters, as well as physiological markers in order to reliably detect hypoglycemia and give early and effective warnings. First, we strive for advanced analytics that can generalize to unseen patients in order to serve as an early warning system. Second, we actively search for digital markers that entail the diagnostic capacity of hypoglycemia and have not been subject of previous studies. Examples include high-resolution time series with real-time driving data and physiological sensors, but also video streams of drivers’ facial expressions. Results will be compared to state-of-the-art continuous glucose sensors (CGM). Third, we ensure readily application of our findings by providing a functional, open source prototype of our system.
Broader impact: Diabetes mellitus is a chronic disease with a rapidly increasing prevalence. A considerable number of these patients are under treatments associated with an increased risk for hypoglycemia. Despite important developments in the field of diabetes technology (e.g. CGM), the problem of hypoglycemia during driving persists due to the limited availability of such technology and technical/behavioral limitations of available methods (e.g. non-continuous character of traditional self-measurement of blood glucose (SMBG), and delayed response to changes in glucose levels/non-adherence to warnings by CGM). Hence, alternative approaches to solve this issue are urgently needed. Our work promises to reduce driving accidents induced by hypoglycemia in the near future since the proposed approach natively integrates with current automotive technology. Furthermore, the availability of a reliable detection model in diabetic individuals will also shed light on similar problems in non-diabetic patients at risk for hypoglycemia, especially the rapidly growing population of post-bariatric patients (i.e. patients after surgery for obesity). Finally, the establishment of a reliable algorithm detecting medical conditions based on individual behavior may also be transferrable to other fields, especially in situations where humans control critical processes or infrastructure (e.g. when operating hazardous machines or handling demanding tasks).
Interdisciplinary research: This project combines medical research (clinical endocrinology) on diabetes mellitus and information technology (applied computer science, health information systems, ubiquitous computing). This is reflected in our work packages. Specifically, we first determine potential digital markers and evidence-based warnings in a clinical lab study with a driving simulator. We then cross-validate the findings and enhance the markers and warnings separately in a controlled field study on a test track and a driving school car with an instructor. Finally, markers and warnings are integrated and the clinical relevance of HEADWIND is evaluated in the field, i.e. by studying response rates of subjects to the warnings (such as safely stopping the vehicle).
Implementation: The resulting open source HEADWIND prototype can directly be integrated into state-of-the-art sensor technology and onboard systems of current cars (e.g. as used for detecting driver drowsiness). This is especially relevant as autonomous cars of level 4/5 are still forecasted to be several decades away.