In the current study, we investigated whether or not the driving simulator was a sensitive instrument for assessing drug-induced effects on car driving by comparing effects of a drug on similar measures taken in the simulator and on the road. The tasks used to compare driving environments were the ‘road tracking task’ as devised by O’Hanlon et al. (1982) and the ‘car following task’ adapted from Brookhuis et al. (1994).
We used the two-threated approach to validity testing proposed by Blaauw (1982). This means that averages in the simulator and on the road were compared. If averages did not differ significantly, then the simulator was considered to be valid on an ‘absolute’ level of correspondence. If averages were influenced by an independent variable in the same way, then the simulator was considered to be valid on a ‘relative’ level of correspondence. In the case of the current research, the simulator was considered as valid on a relative level when the drug-induced effects that were found on the road were also seen in the simulator.
The results of the road tracking task showed that SDLP was comparable between driving conditions on an absolute level since averages did not significantly differ and the effect size of the main effect equalled the preset level of 0.01 (effect size considered too small to be meaningful). Also, the average treatment effects were not significantly different. Even so, there was a difference between on-the-road and simulator driving regarding the strength of the treatment effects. On the road, both treatment conditions significantly decreased performance on SDLP, but in the simulator, only the 20-mg dronabinol treatment effect had a statistically significant effect.
This lack of effect in the simulator for the lowest dronabinol dosage can be explained by a higher variance in the simulator than on the road as was illustrated by the difference in size of the 95 % confidence intervals around the treatment averages. It is relevant to find out if the higher variances in the simulator are due to random error or due to variance that can be attributed specifically to simulator driving since the upper bound of the confidence interval is used to test for equivalence of drug-induced impairments to alcohol-induced impairments so as to be able to assess clinical relevance of the results. In this case, for example, the higher upper bound of the 95 % confidence interval for the 20-mg treatment condition in the simulator would lead to more extreme conclusions about the possible impairing effects of the treatment on driving performance than equivalence tests for on-the-road driving data would.
Another possible explanation for not finding an effect at a lower dose of dronabinol on SDLP in the simulator compared to results found on the road is that not the driving environments but the drug concentrations differed. Even though previous research had shown that orally administered THC concentrations in blood peak after 1.5 h postdrug intake and then remain stable for about 4–5 h (Haney et al. 1999), average THC levels in serum were already decreased at 4.25 h postdrug intake. However, absolute differences in mean THC concentrations during on-the-road driving and simulated driving where minimal, i.e. 1 ng/mL after the 10-mg dose and about 4 ng/mL after the 20-mg dose. Therefore, differences in THC concentration may have caused some additional variance but are not likely to fully account for differential effects of dronabinol 10 mg during actual and simulated driving tests.
Car following results were less clear in terms of comparability. Coherence was comparable between driving conditions but phase and delay were not. Coherence was high and comparable in relative terms since both simulator and on-the-road driving condition showed no treatment effects. But also in absolute terms, coherence results were comparable since average treatment effects were non-significantly different—at least for the highest dose (20 mg dronabinol); the treatment effect of 10 mg dronabinol showed an effect size above the preset level of 0.01. Gain was not comparable in absolute terms since it was higher on the road than in the simulator, but was in relative terms since for both on-the-road driving and simulator driving, no treatment effect was found. For reaction to speed changes of the lead car, however, neither absolute nor relative correspondence was found. An increase in reaction time with dose was found when driving in the simulator, but not on the road.
These results indicate that participants were well able to follow the lead car in both the road and simulator driving conditions irrespective of dronabinol treatment. However, they tended to overreact to speed changes of the lead car more on the road as compared to in the simulator and had longer reaction times to speed changes of the lead car. These differences in car following performance in absolute terms are not problematic since the treatment effects of the car following task are generally not compared on an absolute level but in relative terms. However, since the reaction time to speed changes of the lead car was not comparable in relative terms either, one could conclude that car following results from simulator and on the road driving did not correspond.
It could be that differences in the way the car following task was presented on the road versus in the simulator influenced the outcome measures. For example, in the on-the-road car following task, participants were subjected to six to ten cycles of sinusoidal speed changes of the lead car in 25 min, whereas in the simulator car, the same amount of cycles were introduced in about 10 min of driving. The cycles in the simulator were also a bit shorter (i.e. 20–40 s) compared to on-the-road driving (50 s). Also, the posted speed in the simulator was lower (80 km/h) than on the road (100 km/h). Moreover, the simulator car following task is more controlled than the on-the-road driving task. Measurement disturbances such as other cars driving too slow in front of the lead car or cars merging between the lead and follow cars are not present in the simulator.
Also, the ability to estimate distance was different in the simulator versus on the road since the stereoscopic input differs between the two driving conditions. The simulator has a 2D vision on the car in front whereas in real life this is 3D. This might make it more difficult to perform this task in the simulator than on the road.
Another possible explanation is again that not the task but drug effects differed. However, this does not explain why a dose-related treatment effect of dronabinol on reaction time to speed changes of the lead car was found in the simulator and not or to a lesser extent on the road. The latter finding would actually argue for a better sensitivity of the simulator car following task in picking up drug-induced effects as compared to the on-road driving car following task.
For practical reasons, the on-the-road driving task always preceded the simulator driving task. This lack of counterbalancing may have influenced the outcomes. Furthermore, as explained, small differences in drug-induced effects may have occurred due to taking the simulator test somewhat later than the on-the-road driving test, which in turn may have affected the outcomes. Also, the investigated group was quite hetrogeneous since both occasional and regular cannabis users were included. This hetrogeneity of groups could have influenced the influence of the drug on performance somehow despite the within-subjects design.
Furthermore, factors that influence driving in both devices differentially could have influenced outcomes. Although the aim of the study was to investigate if outcomes of the two driving tasks (OR and SIM) could generate similar results irrespective of these factors, they are worth mentioning as potential limitations.
First, the devices can influence behaviour on a motivational level differentially. In the on-the-road driving task, for example, an instructor is sitting next to the participant, whereas in the simulator, there is no instructor present. This may prompt some participants to try and perform better for the instructor in the on-the-road driving task. On the other hand, the fact that one cannot have an accident in the simulator could influence motivation in the simulator differentially than while driving on the road. Although the chances of becoming involved in an accident are also nihil on the road since an instructor is there to intervene-, the percieved risk may be different i.e. driving in a real car may prime reactions that are associated with real driving while the simulator may not.
Second, the aforementioned difference in sensorial input between the simulator and on-the-road driving could also influence performance differentially since on-the-road driving occurs in a 3D and simulator driving in a 2D setting. Also, there is no tactical input in the simulator (which was fixed-based) while on the road there is. Also, for simulator driving participants can differ in their the susceptibility to simulator sickness, experience with driving in a simulator or gaming experience which can create extra variance that is not present in the on the road driving task. Another difference between the two devices is the controllability of traffic circumstances. In the simulator, everything can be controlled for, the weather, lighting, traffic flow etc., that cannot be controlled for on the road, leaving a gap between driving conditions on the road versus in the driving simulator.
For future research, it may be interesting to find out if the difference in variance between simulator and on-the-road driving that seemed to be present in this data is due to random error or due to some of the above-described factors in simulator driving.
The driving simulator was proven to be sensitive for demonstrating dronabinol-induced effects particularly at higher doses. Treatment effects of dronabinol on SDLP were comparable with driving on the road but inter-individual variability seemed higher in the simulator than on the road which may have potential effects on the clinical inferences made from simulator driving. Car following on the road and in the simulator were, however, not comparable.