There has been a fair amount of hype surrounding the potential for video games to enhance cognitive performance, although a lot of previous work has focused on how these games affect simpler forms of cognition like visual abilities and considerable debate still exists about whether playing video games can actually produce meaningful benefits.
This past fall, researchers from the Gazzaley Lab at UCSF reported in Nature that playing a simple, multi-task video game (Neuroracer) for just one hour a month could help elderly adults to selectively enhance their multi-tasking abilities. If replicated, these results could be extremely useful, as they would suggest a readily available strategy for combating cognitive decline in the elderly. Likely for this reason, the September cover of Nature suggests, tongue-in-cheek, that this finding is a “game changer”, and when reported in the New York Times, one MIT neuroscientist apparently stated that playing the game was powerful enough to make older individuals “cognitively younger”. In contrast, one author of the study, Gazzaley, provides a cautionary note: “Video games shouldn’t now be seen as a guaranteed panacea”.
What exactly is this wondergame?
Neuroracer is a relatively simple 3D, first-person driver game, which requires participants to stay on the road while they respond to signs flashed up on a computer screen. Participants might be asked to complete either a ‘sign discrimination task’ without stepping behind the driver’s wheel, a second ‘drive only’ task, or to participate in a combined ‘multi-task’ condition. While Neuroracer lacks the graphics and acceleration of popular games like Forza or Gran Turismo, driving along the track at high speed levels still looks fairly challenging (see here for a high speed demo, and this WSJ interview displays clips of lead author Jose Anguera playing the game from 1:16-1:59).
Using this setup, Anguera et al. were able to have participants complete each of the individual and combined task conditions which they use to calculate a measure of performance decline or ‘multi-tasking cost’ associated with completing both individual tasks at once. In other words, how much did participants’ ability to correctly identify signs flashed up on the screen decline if they had to drive the Neuroracer car at the same time?
Multi-task performance declines with age
As a first proof-of-concept study, the researchers recruited 174 healthy participants (aged 20-79) for a single day of trials. There were about 30 participants per decade of life and (elderly) participants were screened for cognitive, psychiatric and motor deficits. Participants completed both individual versions of the game during a 30 min training session, and the difficulty of each task was varied to match individuals’ performance to approximately 80% accuracy.
In the critical multi-task trial, the researchers found that the multi-task cost increased linearly with increasing age (from, on average, ~26% cost for individuals in their 20s to >60% cost for individuals in their 70s), despite the fact that the older cohorts had large handicaps on either the ‘drive’ or ‘sign’ tasks.
Overall, this first trial provides good evidence that Neuroracer can be used to detect differences in multi-tasking performance associated with age.
As part of their main study, the researchers also sought to investigate whether long-term training on Neuroracer could improve multi-task performance.
They recruited an additional cohort of 482 older adults (aged 60-85), who were screened on a large battery of tests. After screening, 60 total participants were randomized to each of three training conditions (multi-task, both single tasks, no training) and 46 participants performed well enough on the Neuroracer training tasks to be retained for the entire study. Participants were trained on the task at home, three hours per week for one month (12 hours total). Difficulty levels were adjusted to individual performance throughout training.
When performance was assessed at one month after the start of training, the researchers found that the handicap for each group had been eliminated and the multi-tasking cost had declined for the single-task training group (average ~40% cost), and had virtually disappeared in the multi-task training group (average ~10% cost). This level of performance was nearly as good, or better, than that for an untrained group of younger adults (ages 20-29, average cost ~24%). Moreover, at a six month follow-up test, this improved performance was maintained only in the multi-task training group.
This result presents pretty clear evidence that training on Neuroracer leads to improved performance on Neuroracer, which is pretty much expected. The only surprising finding here is that only the multi-task group sustained improvements, even though both training groups practiced the same tasks. This argues that there was a selective multi-task benefit as a result of this specific kind of training.
Neuroracer multi-task training alters brain activity
The researchers also used ERPs [http://en.wikipedia.org/wiki/Event-related_potential] to assess multi-task performance. The ERP measures were time-locked to moments when a sign was presented while participants were driving, and were used to assess theta power (brain wave magnitude) over the medial prefrontal (mPF) cortex as well as frontal-posterior (FP) theta coherence (correlated brain activity). Each of these activity measures has previously been related to cognitive control performance, with theta power believed to reflect reduced brain activation (i.e. possibly a marker of increased efficiency).
Before any training had occurred, elderly participants had lower levels of mPF theta power and FP theta coherence during Neuroracer multi-tasking than an untrained cohort of younger adults. In contrast, after one month of practice, both measures had improved; although this improvement was only significant for participants who had completed the multi-task training. Further, this improvement in MF theta power was correlated with a high Neuroracer performance at six months in the multi-task training group only (r = .76). MP coherence doesn’t appear to have correlated with 6 month performance.
Overall, this finding suggests that one-month improvements in Neuroracer multi-tasking performance were correlated with increased efficiency of processing while performing this task.
Does training generalize across tasks?
Of course, the bigger question here is not whether training can improve performance on the same task (a ubiquitous phenomenon), but whether training on one task can lead to performance improvements on another (a “transfer” of benefits). To test this question, the authors had participants complete a number of “cognitive control” tasks, before and after training. Six key tasks assessed working memory, attention, “dual-tasking” and interference from distraction, while two additional tasks assessed “speed of processing”, a more generic performance measure that was not expected to be specifically affected by multi-task training.
In support of task-general cognitive improvements, the authors found that the multi-task training group improved significantly more on 2 out of 6 “cognitive control tasks” (“test of variables of attention” or TOVA and a “working memory task”, or WMT), than either the single training or no training control groups. Other cognitive tasks assessing susceptibility to distractors, “dual-tasking” and attention did not show significant differences between groups. Additionally, performance on the TOVA was correlated with the change in MF theta power while playing Neuroracer in the multi-task training group on (r = .56). However, on the other hand, there was no relationship between TOVA performance and theta coherence, nor any relationship between WMT performance and any theta measure.
Overall, the authors find evidence that some tasks are enhanced following multi-task training only, but it isn’t entirely clear whether these “selective” enhancements reflect generalized cognitive improvement or simply improvement on cognitive control tasks that are most similar to the skill practiced while playing Neuroracer.
But does it actually work?
Overall, this is a pretty solid study. It looks as though it was well-designed and the researchers provide some pretty convincing evidence that Neuroracer can be used to document age-related differences in multi-tasking ability. The long-term study also provided reasonably strong evidence to suggest that practicing the game on an adaptive mode with progressively increasing difficulty levels leads to better performance on the game and changes in brain activity while playing it.
The bigger question here is whether these performance improvements affect older adults’ abilities on a wide variety of tasks—in other words, general cognitive enhancement. The data to support this interpretation are certainly suggestive, but a little mixed.
For one, it’s unclear that any of the brain measures of “efficiency of processing” during multi-tasking are good indicators of performance on the cognitive control tasks. There were two brain measures and 6 tasks, so that’s 2×6 = 12 possible correlations. The researchers found one correlation, so there’s not much evidence to suggest that the neural measures indicate general cognitive improvement. Instead, these measures are probably reflective of improvements in Neuroracer performance.
A second point is that while the authors clearly show an improvement in performance on two of the cognitive control tasks, they don’t report all of their data in the analysis, which focuses mostly on reaction time. It would have been nice to see accuracy data as well; even if this data was analyzed separately, an ideal practice would be to regress accuracy against reaction time to fully control for any speed/accuracy trade-offs. Moreover, the authors are a little sneaky in the main paper, where they analyze multiple “levels” of the TOVA and WMT in their main statistical model (most of which are significant), but ignore multiple levels of several other cognitive control tasks (none of which are significant). This is unlikely to have affected their significant interaction effect, but it could have affected their follow-up analyses and it certainly serves to “beef up” the appearance of their results.
Finally, the most important question is whether the results of this small, preliminary study will generalize to a larger, independent replication cohort. It is common for studies of this sort to report results that are robust, and then to later demonstrate that the evidence was weaker than believed.
There are a few reasons for this. For one, small samples tend to be highly variable, and so don’t give us a great picture of how things work in the wider population. Also, without assigning specific endpoints, there are literally dozens of “positive” outcomes to choose from (in this case, 6 tasks, with up to 15 difficulty levels, 2 outcome measures, etc). This study was actually pretty good in their analysis, but still far from perfect.
Moreover, while statistical significant tells us if what we were looking for is present (did Neuroracer enhance performance?), it doesn’t tell us how much. The authors of this paper report that the game has a big effect, mostly because they find “medium to large effect sizes (all cohen’s d’s: 0.50–1.0)”. However, it is well known that effect size estimates will be inflated when power is low (as tends to be true in studies with small sample sizes and in most neuroscience research).
Thus, some caution should be taken before interpreting these results to suggest that Neuroracer can provide a big boost in brain power. This study presents evidence that tasks such as this are “promising”, but it doesn’t provide evidence to truly answer this bigger question. Additional, well-powered, hypothesis-driven studies are necessary to tell us whether Neuroracer should serve as an ideal holiday present for your aging relatives.