Summarizing Complex Graphical Models of Multiple Chronic Conditions Using the Second Eigenvalue of Graph Laplacian: Algorithm Development and Validation

Understanding the Interactions Between Driving Behavior and Well-being in Daily Driving: Causal Analysis of a Field Study

BACKGROUND

Investigating ways to improve well-being in everyday situations as a means of fostering mental health has gained substantial interest in recent years. For many people, the daily commute by car is a particularly straining situation of the day, and thus researchers have already designed various in-vehicle well-being interventions for a better commuting experience. Current research has validated such interventions but is limited to isolating effects in controlled experiments that are generally not representative of real-world driving conditions.

OBJECTIVE

The aim of the study is to identify cause-effect relationships between driving behavior and well-being in a real-world setting. This knowledge should contribute to a better understanding of when to trigger interventions.

METHODS

We conducted a field study in which we provided a demographically diverse sample of 10 commuters with a car for daily driving over a period of 4 months. Before and after each trip, the drivers had to fill out a questionnaire about their state of well-being, which was operationalized as arousal and valence. We equipped the cars with sensors that recorded driving behavior, such as sudden braking. We also captured trip-dependent factors, such as the length of the drive, and predetermined factors, such as the weather. We conducted a causal analysis based on a causal directed acyclic graph (DAG) to examine cause-effect relationships from the observational data and to isolate the causal chains between the examined variables. We did so by applying the backdoor criterion to the data-based graphical model. The hereby compiled adjustment set was used in a multiple regression to estimate the causal effects between the variables.

RESULTS

The causal analysis showed that a higher level of arousal before driving influences driving behavior. Higher arousal reduced the frequency of sudden events (P=.04) as well as the average speed (P=.001), while fostering active steering (P<.001). In turn, more frequent braking (P<.001) increased arousal after the drive, while a longer trip (P<.001) with a higher average speed (P<.001) reduced arousal. The prevalence of sunshine (P<.001) increased arousal and of occupants (P<.001) increased valence (P<.001) before and after driving.

CONCLUSIONS

The examination of cause-effect relationships unveiled significant interactions between well-being and driving. A low level of predriving arousal impairs driving behavior, which manifests itself in more frequent sudden events and less anticipatory driving. Driving has a stronger effect on arousal than on valence. In particular, monotonous driving situations at high speeds with low cognitive demand increase the risk of the driver becoming tired (low arousal), thus impairing driving behavior. By combining the identified causal chains, states of vulnerability can be inferred that may form the basis for timely delivered interventions to improve well-being while driving.

A Functional Model for Structure Learning and Parameter Estimation in Continuous Time Bayesian Network: An Application in Identifying Patterns of Multiple Chronic Conditions

Bayesian networks are powerful statistical models to study the probabilistic relationships among sets of random variables with significant applications in disease modeling and prediction. Here, we propose a continuous time Bayesian network with conditional dependencies represented as regularized Poisson regressions to model the impact of exogenous variables on the conditional intensities of the network. We also propose an adaptive group regularization method with an intuitive early stopping feature based on Gaussian mixture model clustering for efficient learning of the structure and parameters of the proposed network. Using a dataset of patients with multiple chronic conditions extracted from electronic health records of the Department of Veterans Affairs, we compare the performance of the proposed network with some of the existing methods in the literature for both short-term (one-year ahead) and long-term (multi-year ahead) predictions. The proposed model provides a sparse intuitive representation of the complex functional relationships between multiple chronic conditions. It also provides the capability of analyzing multiple disease trajectories over time, given any combination of preexisting conditions.