shows the demographic information for the participants. There were no statistically significant differences between OSA and comparison drivers in terms of gender or age. The mean AHI for OSA drivers was 21.2 +/− 19.9 (range 5.07 – 96.57), mean minimum oxygen saturation was 81.0 +/− 8.58% (range 58.0 – 94.0) and mean sleep latency was 10.6 +/− 4.0 (range 2.2 – 16.2). SOREMP’s were seen in only one driver, who had 2 associated with an MSL of 4.8. Note that MSLT data were available for only 15 OSA subjects, as the remaining 10 underwent a “split night” protocol on the preceding PSG.
OSA drivers were significantly sleepier than controls as evidenced by higher ESS scores (see ). SSS did not differ significantly between groups immediately before the drive, but was significantly greater after in OSA drivers compared to controls. In each group, sleepiness increased, as SSS values were higher after the drive than before (p<0.01). However, OSA drivers had significantly higher SSS scores at the end of the drive than did comparison drivers (p = 0.03).
Subjective sleepiness measures by OSA status. ESS = Epworth Sleepiness Scale; SSS = Stanford Sleepiness Scale.
Vigilance during driving
describes the outcomes on the vigilance task in OSA and comparison drivers. There was no difference in the number of targets presented to each group. Between group analyses showed that HR on the vigilance task was lower in drivers with OSA than in the comparison group (p = 0.03). The number of false positive responses was very small; a total of 25 subjects had false positives, but only 5 (7.58%) had more than one and there was no difference in the number of false positives between the two groups (p = 0.79). HR correlated with D-prime (p = 0.03), indicating that subject responses were not significantly influenced by false positives; consequently, HR was used as the measure of response accuracy in subsequent analyses. There was no significant difference in RT between the two groups (p = 0.25).
Results of the vigilance task by OSA status. Targets (#), HR = hit rate (%), FP = false positive responses (#), RT = reaction time (seconds);
We also examined driver performance on the vigilance task as a function of target appearance in central versus peripheral locations of the simulator display. Preliminary results showed that the three central locations (center, 12.5 degrees left, and 12.5 degrees right) had similar outcomes for HR and RT (p = 0.07 and p = 0.71, respectively), as did the two far left locations (p = 0.08 and p = 0.27) and the two far right locations (p = 0.15 and p = 0.80). Consequently, we compared the 3 central locations with the 4 peripheral locations (see ). The mean HR’s for central and peripheral targets were 0.955 and 0.814 in the comparison group and 0.923 and 0.723 in the OSA group. Both groups performed similarly on the central targets (p = 0.21), but the OSA group performed more poorly than the controls on peripheral targets (p = 0.02). A mixed model aimed at assessing whether the two groups had similar differences in performance between central and peripheral targets, showed a trend towards different HR based upon target location (p = 0.09 for interaction). For RT, both groups performed in a similar fashion on central vs. peripheral targets (p = 0.84 for interaction). Adjusting for group status, subjects responded to central targets 0.14 seconds faster than peripheral targets (p < 0.01).
Results of the vigilance task as a function of target location. HR = hit rate (%), RT = reaction time (seconds).
In terms of the relationship between changes in vigilance performance and drive duration, HR did not significantly differ between OSA and control drivers over the course of the drive (p = 0.83 for interaction), and did not significantly change over time for either group (p = 0.19 for main effect of time). Similarly, the trend in RT over the drive was not significantly different between OSA and control drivers (p = 0.22 for interaction). There was marginal evidence for a decrease in RT from the first to the last drive segment for both groups (p = 0.08). However, this amounted to a mean decline in RT of only 0.046 seconds.
There were no significant differences between OSA and comparison drivers in either the number of lane deviations (49.16 +/− 39.576 vs. 45.976 +/− 38.225, p = 0.72) or speed errors (2.96 +/− 4.996 vs. 5.098 +/− 8.139, p = 0.39), while there was a trend towards higher SDLP in the OSA drivers (0.3697 +/− 0.0920 vs. 0.3288 +/− 0.0827, p = 0.07).
Relationship between vigilance and sleepiness
There were no relationships between self-reported sleepiness and either vigilance measure for either OSA drivers or controls, except for HR, which was inversely correlated with ESS for the entire group (Spearman r = −0.28154, p = 0.02) (See ). However, there was a relationship between changes in HR and sleepiness over time during the drive. In the OSA drivers, we found a negative correlation between change in SSS (“after” minus “before”) and change in HR (from the first to the last drive segment) (Spearman r = −0.49, p = 0.01). In other words, the OSA subjects who became sleepier over the drive had lower HR’s in the final segment. In contrast, there was no significant relationship between changes in SSS and HR for controls (Spearman r = 0.09, p = 0.59). No significant relationships were found for RT (Spearman r = .13, p = .45 for OSA drivers, Spearman r = −.11, p = .48 for controls).
Table 5 Relationship between vigilance and subjective sleepiness measures. Spearman Correlation Coefficients with p-values in parentheses. HR = hit rate (%); RT = reaction time (seconds); ESS = Epworth Sleepiness Scale; SSS = Stanford Sleepiness Scale; OSA = (more ...)
Relationship between vigilance and objective measures of disease severity (OSA drivers only)
AHI was divided into 3 categories; 5–15, >15–30, >30 events per hour. There were no significant relationships between AHI and either of the vigilance measures; HR (p = 0.11) and RT (p = 0.46). Minimum oxygen saturation did not correlate with RT (r = −0.03, p = 0.8986), but was significantly correlated with HR (r = 0.54, p < 0.01). There were no correlations between MSL and either vigilance measure (HR; r = 0.34, p = 0.22, RT; r = 0.05, p = 0.87). Too few SOREMP’s were recorded to make any comment on their relationship to performance on the vigilance tasks.
Relationship between driving errors, vigilance and sleepiness
The main predictors of interest for driving performance were HR, RT, and a measure of sleepiness (SSS before the drive, SSS after the drive, and ESS). SSS after the drive was used as the measure of sleepiness as the other 2 did not have a significant relationship with the driving outcomes. The variables age, OSA status, and gender were looked at as possible confounders. SSS scores after the drive were grouped into 3 levels. Subjects with scores of 1–2 were termed “Awake”, 3–4 “Marginally Sleepy”, and 5–7 “Sleepy”.
There were no relationships between lane deviations and age (p = 0.73), gender (p = 0.43) and OSA status (OSA vs. controls) (p = 0.75), so these variables were omitted from the final Poisson models. As there were no differences between OSA and comparison drivers, results below are for all 66 participants.
HR was a significant predictor of lane deviations (p = 0.01). A 10 % absolute decrease in HR yielded a 15.8% increase in the estimated mean number of lane deviations. RT was also a significant predictor of lane deviations (p = 0.03). A 0.1 second increase in reaction time yielded a 5.0% increase in the estimated mean number of lane deviations. However, the effect of RT was largely due to 2 outliers (1 OSA and 1 control). With these 2 subjects removed, RT was not a significant predictor (p = 0.21). SSS after the drive was a significant predictor of lane deviations (p < 0.01). The “Sleepy” subjects had an estimated 2.61 times the mean number of lane deviations as the “Awake” subjects and an estimated 1.62 times the mean number of lane deviations as the “Marginally sleepy” subjects. To determine the best predictors of lane deviations between HR, RT, and SSS, a model was fit that included all three predictors. With all three predictors in the model, SSS remained significant (p = 0.04), while HR and RT did not, indicating that SSS after the drive was the strongest predictor of lane deviations.
Few drivers had any low speed errors; the majority of speed errors were high-speed errors. Consequently, both low and high-speed errors were combined for subsequent analyses. Age was significantly related to speed errors (p = 0.04), so all results were adjusted for age. OSA status (p = 0.23) and gender (p = 0.20) were not significantly related to speed errors. As in the analysis of lane deviations, the results given below are for all 66 drivers.
HR was not a significant predictor of speed errors (p = 0.44) but RT was (p < 0.01). A 0.1 second increase in reaction time yielded a 19.4% decrease in the estimated mean number of speed errors. In other words, slower responses on the vigilance task predicted fewer high-speed errors. SSS after the drive was also a significant predictor of speed errors (p < 0.01). In order to determine the best predictor of speed errors, SSS and RT were modeled simultaneously. RT remained significant (p < 0.01) while SSS did not (p = 0.28). The results were very similar whether or not outliers were included in the analysis.
Standard Deviation of Lane Position
Gender was significantly related to SDLP (p = 0.03) and OSA status was marginally significantly related (p = 0.07). There was not a significant relationship between SDLP and age (p = 0.93). The following results are for all 66 drivers, adjusted for age and OSA status.
HR was a significant predictor of SDLP (p = 0.02). A 10% absolute decrease in HR yielded an estimated .022 increase in SDLP. RT was also a significant predictor of SDLP (p = 0.04). A 0.1 second increase in RT yielded a .006 increase in SDLP. SSS after the drive was a marginally significant predictor of SDLP (p = 0.08). The "Sleepy" subjects had an estimated SDLP that was .074 higher than the "Awake" subjects and .052 higher than the "Marginally Sleepy" subjects. To determine the best predictors of SDLP between HR, RT, and SSS, a model was fit that included all three predictors. With all three predictors in the model, HR (p = 0.16), RT (p = 0.12), and SSS (p = 0.28) did not remain significant.