Congratulations to Enrique Molina, who was awarded his PhD title for a thesis entitled “Efecto de los factores circadianos en la vigilancia durante la realización de una tarea de conducción” (The effect of circadian factors on vigilance in a driving task) this Friday 28th September 2018.
Enrique carried out his PhD at the University of Granada under the supervision of Dr Ángel Correa and Dr Daniel Sanabria.
English summary (taken from the thesis):
“Driver’s fatigue is one of the main causes of road accidents, as it can produce distractions and drowsiness, especially in long and monotonous roads. The ability to maintain a good performance while driving (or in any other task that requires attention) for a long period of time is called vigilance, and it can present fluctuations and a continuous decline with time on task, a phenomenon known as the vigilance decrement (Mackworth, 1948). These fluctuations are modulated by factors like motivation, sleep or circadian rhythmicity, which refers to the periodicity of several human functions, like sleep-wake cycle, hormone production or even cognitive performance (see for example (Goldstein, Hahn, Hasher, Wiprzycka, & Zelazo, 2007).
Therefore, a main goal in fatigue research is to identify these attentional fluctuations that can lead to poor performance states. One framework to attain this aim is monitoring subject’s psychophysiological variables, like the ocular movements, the body temperature or the electroencephalogram (EEG), and use them as indices of subject’s performance on the task. A system that, using these variables, could identify subject’s low attentional states and alert the driver of an imminent risk, would have important implications on safety, even more if such a system could identify these states in advance.
Within this framework, some neurophysiological models have been developed to predict a driver’s performance from records of the brain activity (Chin-Teng Lin et al., 2005). These models are, generally, subject- or task-specific and is not clear to what extent they can be transfer between subjects or tasks.
In this thesis we addressed the topic of predicting low performance states while driving. To do so, we first studied different factors that can modulate performance in a driving task; second, we studied two neurophysiologic indices of attentional state using a simple vigilance task; and third, we tested the predictive accuracy of a simple regression model when transferred between two different attentional tasks.
In Experiment 1, we studied the effect of time of day and chronotype on the vigilance decrement in a simulated driving task and in a simple reaction time task, the Psychomotor Vigilance Task (PVT; Dinges & Powell, 1985). We tested two extreme chronotype groups (morning-type and evening-type) at two different moments of day (8am vs 8pm). These two times of the day represented an optimum and sub-optimum cognitive state for each chronotype, being the morning-type in its optimum state at 8am, and the evening-type at 8pm. We also recorded the EEG throughout the driving task to analyze the power dynamics in the main frequency bands (i.e., theta, alpha and beta). Results from this study showed an interaction between the circadian factors (i.e., time of day and chronotype), but only for the evening-type group, while morning-type group showed a very stable performance at both moments of the day. This interaction appeared for both, the driving task and the PVT, but it was not reflected in the EEG, which only showed a generalized power increment with time on task for both chronotypes.
In Experiment 2 we analyzed two psychophysiological variables (EEG and skin temperature) as predictors of the attentional fluctuations while a long 45-min PVT, controlling for the effect of circadian factors by choosing only intermediate-type chronotypes and avoiding extreme hours (i.e., early in the morning or late in the evening) to run the experiment. The EEG and temperature were recorded continuously throughout the task. We examined the power dynamics of the theta, alpha and beta frequency bands from the EEG, and three different skin temperature measures, the distal temperature (recorded from the wrist, over the small superficial capillaries), the proximal temperature (recorded under the left clavicle as an estimator of the core body temperature), and the distal-proximal gradient (DPG), which is the difference between the distal and the proximal temperatures, being a more stable measure. Subsequently, these variables (the EEG frequency power and the three temperature recordings) were related to the performance in the PVT, measured as the reaction time.
Results showed evidence of a clear relationship between a slowing-down in the response time and an increment in power in theta and alpha frequency bands, in brain regions related to the attentional network. Likewise, a linear mixed-model showed that a reduction in the distal skin temperature and in the DPG resulted in an increment of the reaction time (i.e., a poorer performance). These results showed that both indices, the EEG frequency power and the skin temperature were potential predictors of subject’s performance in a vigilance task. However, while the EEG worked on a milliseconds scale, the temperature had a much slower time course, on the scale of minutes, which make the former a better choice as predictor for our next experiment.
In the Experiment 3, our main goal was to develop a model to predict performance in a driving task using some of the indices from Experiment 2, that is, the power dynamics from theta and alpha frequency bands. Also, we aimed to transfer this model between tasks, using the data from the PVT to build the model and then using it to predict performance in the driving task. Thus, in Experiment 3, subjects performed a 20-min PVT and a 60-min simulated driving task, while the EEG was recorded throughout the tasks.
The EEG analyses were similar to those used in Experiment 2, and results from the power dynamics replicated those found in Experiment 2. More precisely, the EEG analyses showed an increment of alpha and theta frequency power related to an increment in the reaction time for both tasks, the PVT and the simulated driving. Also, these frequency dynamics were located in the frontal-parietal attentional network and in the default mode network.
The data from the PVT were used to fit an individualized linear regression model, using the EEG frequency power as predictor of the reaction time. Then, this model was feed with the EEG data from the driving task to obtain estimates of the reaction time, which was finally correlated with the actual reaction time from the driving task. This correlation was positive and significant for barely a third of the subjects. It is possible that the complexity of the model was a key factor when transferring it between tasks, and a simple model as ours only works when the attentional fluctuations are not influenced by other factors unrelated to fatigue.
In sum, in this thesis we have proposed an individualized predictive model of performance in a simulated driving task built from a simpler task like the PVT. In Experiment 2 we have shown that EEG and temperature analyses can grasp the attentional state and can predict (even 2 seconds in advance in the case of the EEG) the reaction time in the PVT. Then, in Experiment 3, results showed that it is possible to transfer a simple regression model and make accurate predictions between tasks, at least for a few subjects. We have likewise shown for the first time that circadian factors play an important role on driving performance, and that this could also be related to subjects’ personality. Thus, if personality could affect the performance stability between sessions, then, although circadian factors were controlled in our last experiment, other personality or mood states could have influence the difference in outcomes between tasks. Therefore, a more complex regression model that accounts for variables related to the subject could be the key for the model to work in most of the subjects.
The approach proposed in this thesis is based on the generalization of a predicting model using a simple task. This model would allow anyone to adjust an individualized and instantaneous system for fatigue prevention, not only for driving tasks, but for many other daily tasks that require of a continuous level of alertness. These tasks are common in many jobs and in many cases, an attentional lapse can result in serious accidents, and therefore, research in such a model will have a big impact in safety.“
