Using EEG to characterise drowsiness during short duration exposure to elevated indoor Carbon Dioxide concentrations

Drowsiness which can affect work performance, is often elicited through self- reporting. This paper demonstrates the potential to use EEG to objectively quantify changes to drowsiness due to poor indoor air quality. Continuous EEG data was recorded from 23 treatment group participants subject to artificially raised indoor CO2 concentrations (average 2,700 ± 300 ppm) for approximately 10 minutes and 13 control group participants subject to the same protocol without additional CO2 (average 830 ± 70 ppm). EEG data were analysed for markers of drowsiness according neurophysiological methods at three stages of the experiment, Baseline, High CO2 and Post-Ventilation. Treatment group participants’ EEG data yielded a closer approximation to drowsiness than that of control group participants during the High CO2 condition, despite no significant group differences in self-reported sleepiness. Future work is required to determine the persistence of these changes to EEG over longer exposures and to better isolate the specific effect of CO2 on drowsiness compared to other environmental or physiological factors. Practical implications This study introduces EEG as a potential objective indicator of the effect of indoor environmental conditions upon drowsiness Participants exposed to 2,700 ppm for 10 minutes showed greater evidence of a progression towards drowsiness (as measured by EEG) than that of participants who received the same protocol without additional CO2 (mean 830 ± 70 ppm), despite similar ratings of subjective sleepiness. Subjective and objectively measured indications of drowsiness were reduced following ventilation of the room. Future work could explore the potential of regular ventilation episodes in knowledge work spaces to retain alertness.


128
Participants 129 A total of 47 subjects were recruited and participated in the study between October 2016 and 130 February 2017. Usable EEG data was available from 36 of the 47 participants, reflective of the 131 sensitivity of EEG to movement artefacts and the researchers' wish for data reliability. The 132 study protocol and conditions of participation were approved by the University of Southampton 133 Ethical Research Governance Office (ERGO# 30443). Sampling was achieved by advertising 134 the study on billboards throughout the University, a local supermarket and a departmental 135 mailing list. Convenience sampling was used for contacts of the research team who were 136 unaware of the study protocol. The final sample was comprised mostly of students and staff 137 from the University. Written consent was gathered from each participant prior to their 138 participation in the study. Exclusion criteria for the study were adapted from those used by 139 Garner et al. [36], a study where participants were subjected to 7.5% CO 2 (75,000 ppm) level 140 of CO 2 . Exclusion criteria included current or historic drug/alcohol abuse or panic attacks, 141 current treatment for migraine headaches, pregnant, current neurological conditions (e.g. 142 epilepsy), and recent severe illness. Participants were compensated £10 in vouchers for an 143 online retailer for their participation.
144 Participants were split into two groups. Of the participants with usable EEG data, this involved: 145 23 participants in the "treatment group" (TG) who received artificially raised CO 2 146 concentrations and 13 participants in the "control group" (CG) for whom CO 2 concentrations 147 were not artificially raised (Table 1). The variance in the size of the groups is due to which of 148 the participants had sufficiently clean EEG for inclusion and the difficulty in recruiting a larger 149 sample.   181 Carbon dioxide was introduced using a cylinder of ultrapure CO 2 (greater than 99.99% purity) 182 located in the corner of the room with the outlet attached to pedestal fan to achieve mixing.
183 The fan was pointed away from the participant and in operation only for the duration of 184 Condition 3 (see Table 2), when CO 2 was being released, in order to minimise any influence 185 of air movement on perception or produce possible thermal comfort effects during subsequent 186 conditions. The target CO 2 concentration once mixed was 2,700 ppm (mean: 2,700 ± 300 ppm 187 for the duration of Condition 5). Participants were instructed to sit at the table in the middle of 188 the room while the researcher operated the computer and the gas cylinder behind the 189 participant. In this way participants were aware the air quality was going to be changed 190 somehow during the experiment, but were not aware how.

192
Experimental Procedure 193 The experimental protocol took place in the one study room (Figure 1). The study protocol is 194 summarised below for TG participants (   Figure 1. The loggers were positioned so as to avoid influence from direct respiration. The 217 heights of the loggers from the floor were 720 mm (logger 1), 1,545 mm (logger 2) and 1,995 218 mm (logger 3). The distance from logger 2 to logger 3 was 2,100 mm and logger 1 was 219 approximately 1,300 mm perpendicular to the participant's heads ( Figure 1). Instrument 220 accuracies for the CL11 are ± 0.3 ° C (temperature), < 2.5% RH (humidity) and ± 30 ppm ±  Figure 2).
231 In order to minimise movement artefacts in the EEG, participants were asked to sit quietly and 232 remain still throughout the experiments except during the short break for the questionnaire 233 following Condition 5 (refer Table 2, Figure 2). Indoor conditions by analysis segment 294 Table 3 below summarises the measured indoor environment parameters at each of the two-295 minute analysis segments: Baseline, High-CO 2 and Post-Ventilation (Figure 2), for TG and 296 CG participants:  Table 3, TG participants were 304 exposed on average to an additional 1,898 ppm of pure CO 2 to that generated by human 305 respiration alone.
306 To control for possible temperature effects, all participants were able to adjust clothing as they 307 wished prior to the experiment to ensure comfort. A 3 (analysis segment) by 2 (group) mixed 308 model ANOVA was run to assess temperature fluctuations. Results show that CG participants 309 were tested at a significantly higher temperature than TG participants (see Table 3 and Section 310 0; F (1, 34) = 6.30, p = .02, η p 2 = .16). This was due to the majority of CG participants being 311 tested following the activation of the building's heating systems. Results also showed that 312 temperature varied significantly between each of the analysis segments irrespective of group 355 Overall results, irrespective of group, show no changes in the temporal electrode region for any 356 frequency. The strongest effects from Baseline to High-CO 2 are an increase of frontal high-357 delta, theta and beta, central high-delta, and occipital high-delta and theta, as well as global 358 increases in high-delta, theta, and alpha. Despite a lack of significant group effects in the overall 359 model, the data presented in Table 4 show a clear difference in the pattern of frequency power 360 changes across the brain in the two groups. According to the definition of drowsiness employed 361 (Section 0), the results show the EEG of the TG shows a closer approximation to drowsiness 362 compared to that of the CG, considering: (a) the increase in delta and theta is more global than 363 the CG and (b) CG also has a significant overall increase in alpha and beta, while TG increase 364 is theta and high-delta only.