The relationship between vigilance capacity and physical exercise: a mixed-effects multistudy analysis

Introduction

The current trend toward a sedentary lifestyle in modern societies clashes with the human natural tendency to be physically active (Andersen, Mota & Pietro, 2016; Blair, 2009). This pervasive lack of regular physical activity has been related to numerous chronic physical and mental diseases and, relevant to this article, to suboptimal cognitive functioning (Booth, Roberts & Laye, 2012). Indeed, there is a substantial body of work depicting a positive association between regular practice of physical activity and cognition (Gomez-Pinilla & Hillman, 2013; McMorris, 2016). The key factors driving that link, however, are still a matter of scientific debate (Stillman et al., 2016). The aim of this brief report was to further contribute to that topic by testing the role of two critical variables, cardiovascular fitness and sport type participation, on the association between the regular practice of exercise and the level of vigilance (i.e., the ability to stay focused, and to detect and respond efficiently to target stimuli in order to attain the goals of the task). The interest in the study of vigilance (or sustained attention) was motivated by its crucial role in general cognitive capacities and to achieve optimal performance in many daylife activities (e.g., driving, attending to an academic lesson, etc.) (Larue, Rakotonirainy & Pettitt, 2011). To accomplish the objective of the present study, we performed a mixed-effects analysis including data from seven studies conducted by our research group (total N = 361) that used the same reaction time (RT) measure from the Psychomotor Vigilance Task (PVT) (Luque-Casado et al., 2013).

Exercise is typically defined as physical activity performed in a structured, planned and repetitive manner (Caspersen, Powell & Christenson, 1985). Enhanced physical or cardiovascular fitness is one of the main consequences of the regular practice of physical exercise that has been related to cognitive performance (Gomez-Pinilla & Hillman, 2013; Åberg et al., 2009; Tarumi & Zhang, 2015; Voss et al., 2016). In the particular case of vigilance, our studies showed that higher-fit individuals outperformed lower-fit individuals in the PVT (Luque-Casado et al., 2013; Ciria et al., 2017; Luque-Casado et al., 2016b). The evidence comes from both behavioral (RT) and electrophysiological measures (event related brain and cardiac potentials). The outcome of this research appears to support the cardiovascular fitness hypothesis, by virtue of which, physiological adaptations (e.g., increased VO₂, increased brain-derived neurotrophic factor (BDNF), etc.) induced by regular exercise are assumed to be responsible for observed cognitive improvements (North, McCullagh & Tran, 1990; Voss, 2016).

Given the variety of exercise-related contexts available, however, exercise practice is far more than just a way for enhancing cardiovascular fitness (Hartman et al., 2017). In fact, given its inherent and varying perceptual and cognitive demands, it is expected that cognitive enhancement would follow sustained practice (Alesi et al., 2015). For instance, optimal performance in football or basketball requires, together with a sufficiently good level of fitness, rapid adaptation and response to the constantly varying exercise environment (i.e., they are instances of externally-paced activities). In contrast, endurance cycling or long distance running involve self-regulation of the effort in a relatively consistent and predictable environment (i.e., they are instances of self-paced exercise). It is therefore reasonable to expect that any cognitive improvement related to the regular practice of exercise would depend on the particular activity and the associated cognitive demands. Our research on vigilance (Ballester et al., 2017) has taken that possibility into consideration, reporting that individuals (children) who practice externally-paced exercise regularly outperform those who practice self-paced exercise (i.e., they show shorter RTs in absence of differences in cardiovascular fitness between the two groups of athletes). These findings are consistent with the “cognitive skill” hypothesis (Mann et al., 2007; Voss et al., 2010) whereby the learning of basic cognitive abilities through practice of one particular activity can be transferred to other domains. This result may in turn jeopardize the notion that cardiovascular fitness is critical for differences in vigilance performance to occur as a function of exercise practice.

At this point, our research has not hitherto provided a clear answer to the issue of whether cardiovascular fitness, sport type participation, or both, are among the key factors determining the relationship between regular practice of exercise and vigilance performance. For this reason, we have decided to perform a mixed-effects multistudy analysis on data from seven studies (six published and one non-published) that our research group has conducted so far on this topic (Table 1) involving a total of 361 participants.

The use of the same task (the PVT) in all studies enabled us to incorporate the raw RT data, which we believe represents an advantage over the use of standardized effect sizes that are included in typical meta-analytical reports. To maintain homogeneity between studies, we only analyse RT from the first 5 min of the task, that corresponds to the shortest version of the PVT we have used so far. Note that the 5 min version of the task has been reported to be a reliable tool to assess vigilance performance (Loh et al., 2004). Importantly, apart from including cardiovascular fitness and sport type participation as main variables of interest in the analysis, we controlled for age, sex and the method of cardiovascular fitness estimation (that differed between studies).

In brief, this report aims to advance our current understanding of the research linking physical exercise and cognition, with a focus on vigilance. The details of the analysis are reported below and the raw data and the R scripts can be downloaded here (https://osf.io/wcbev/).

Method

Design and statistical analyses

Reaction times from the seven studies conducted in our laboratory were collapsed into a single dataset to estimate the effects of sport type (i.e., self-paced, externally-paced and sedentary conditions) on them. In face of the diversity of samples’ characteristics and study features, we fitted RTs using multilevel linear mixed-effects modelling, as implemented in the lme4 R package (Bates et al., 2015).

In order to take into account the dependencies potentially generated by any procedural differences between studies, we treated RTs as obeying to a multilevel data structure (Fig. 1), with participant (level 3), nested into study (level 2), and study nested into estimation method (level 1). Thus, the random part of the model consisted of intercepts for participant (nested in), study (nested in), VO₂ estimation method. This random part was common to all models. The fixed part in a first, baseline model (H₀) consisted of VO₂, age and trial number (to facilitate model convergence, gender was not included as a fixed factor, as it is mostly controlled for by participant and VO₂, and not included in any further interactions). Additionally, the effect of age consisted of a linear and a quadratic component, to allow for non-linearity in the age-RT association. This baseline model was pitched against a second, H₁ model with the same random and fixed parts as H₀, plus sport type (self-paced, externally-paced, sedentary) as an added fixed-effects factor. A third, saturated model further included the sport type × age, the VO₂ × age, and the sport type × trial interactions (please, note that interactions involving age actually refer to two different interaction effects, one involving the linear component and the other involving the quadratic component of the age effect, that were always jointly included/excluded for model comparisons). Interactions were later removed one-by-one (Interaction 1, Interaction 2 and Interaction 3 models; see Table 2). If any of the three interactions contributed to the model fit, it was kept in the final, best-fitting model. This last, best-fitting model thus consisted of the same random and fixed parts as the H₁ model plus the interactions identified as substantially contributing to model fit. This model was used to estimate effects.

Table 2:

Models, fitting indices, and likelihood ratio comparisons.

Model	Fixed part	df	AIC	L. ratio	p
Baseline	Age + VO2 + trial	9	34,881.04
H1	baseline + sport type	11	34,866.47	18.569	<0.001	>baseline
Interaction 1	saturated − (sport type × age)	15	34,865.81	8.089	0.088	<saturated*
Interaction 2	saturated − (VO2 × age)	17	34,863.70	1.977	0.372	= saturated
Interaction 3	saturated − (sport type × trial)	17	34,863.59	1.868	0.393	= saturated
Saturated	H1 + (sport type × age) + (VO2 × age) + (sport type × trial)	19	34,865.72
Best-fitting	H0 + sport type + (sport type × age)	15	34,861.58	31.465	<0.001	>baseline

DOI: 10.7717/peerj.7118/table-2

Notes:

All models have been adjusted with the Maximum Likelihood (ML) method. Age effects include a linear and a quadratic component, jointly included/excluded for model comparisons. The random part is common to all models (see text).

“>” Indicates better fit.

“<” worse fit.

“=” not substantially worse fit.

* For factor inclusion/exclusion comparison, a relatively lenient p < 0.010 significance level has been used.

Prior to fitting, RTs larger than 1,000 ms (i.e., omission errors; 0.5% trials) were removed from further analyses, and the remaining ones (16,239) were log-transformed. To enable model convergence and facilitate interpretation of regression coefficients, trial number and log-transformed RTs, were scaled and zero-centered. Models were compared using the Akaike Information Criterion (AIC), and a Likelihood ratio test. For model comparisons performed to identify the best-fitting model, a relatively lenient 0.010 p-value criterion was adopted. Inferences were subsequently made based on the results of the best-fitting model.

Results

Table 2 shows the fitting indices for all the models included in the relevant comparisons described above (baseline, H1, saturated, interaction 1–3, and best-fitting models). Table 3 (columns 1–5) displays unstandardized regression coefficients (B), their standard errors and significance levels (according to t-tests), for each fixed component in the best-fitting model. The best-fitting model included the effects of sport type, and the sport type × age interaction.

In the best-fitting model, the contrast corresponding to the comparison between externally paced sports and the other two conditions (sedentary and self-paced sports, pooled together) yielded a significant t-test (contrast 1); whereas self-paced sports did not differ from the sedentary condition (contrast 2). Neither C1 nor C2 interacted with either the linear or the quadratic effect of age.

The absence of a significant effect of VO₂ in the best-fitting model (p = 0.247) was also confirmed by an approximated Bayes Factor (Wagenmakers, 2007). In order to compute it, we built a third model (H1_b), equivalent to the best-fitting model, but with VO₂ removed from the fixed part. The AIC value for this model was 34,861.02. The Bayes Factor approximation for this comparison between the best-fitting model and H1_b was BF₁₀ = 0.016, which supports H0_b relative to the best-fitting model, and suggests the absence of any substantial influence of VO₂ on RTs if sport type is controlled for.

The upper panel of Fig. 2 displays predicted (standardized) RTs across age for the three sport types. Although the age × sport type interactions were not significant, the effect of sport type seems to vanish for intermediate ages. Additionally, given there were virtually no practitioners of externally paced sport among 40–50 years-old participants, predictions for that age group mostly resulted from fitting the effect of age on RT for younger participants. In order to ensure the effect of sport type was not inflated in the previous analysis, we run the best-fitting model with a dataset restricted to the studies in which there were participants in the three sport types (study 6 and 7). As shown in columns 6–9 in Table 3, the effect of sport type remained largely significant, but there also was an interaction between the sport type C1 contrast and the linear component of age. The lower panel of Fig. 2 shows the shape of this interaction. Importantly, qualitative predictions for these studies were virtually identical to the ones made from the full dataset for the 10–35 age range.

Discussion

Sedentarism has been related to numerous health issues and, according to a wealth of literature, to poorer cognitive function (with respect to active individuals) (Gomez-Pinilla & Hillman, 2013). The aim of this paper was to contribute to that body of work by pooling the data from seven studies (N = 361) to examine whether VO₂ (index of cardiovascular fitness), type of sport participation (externally-paced and self-paced), or both (controlling for sex, age and the method of VO₂ estimation), are crucial factors to explain the association between the regular practice of exercise and vigilance capacity (measured by means of the PVT).

The results were straightforward. Sport type was significantly related to RT, although only externally-paced sport differed from the sedentary condition (and also from the self-paced sport condition). Sport type and age interacted, showing that the (same) pattern of differences between sport types was more evident in children and adolescents than in older participants. Both the multilevel linear mixed-effects modelling and Bayesian analysis confirmed the absence of effect of VO₂ and self-paced sport practice (with respect to the sedentary condition) on RT.

Maximum oxygen consumption (VO_2max) has been the primary index of cardiovascular fitness to associate with cognitive and brain functioning (and anatomy) over the last years. The positive findings to date establish that the greater the VO_2max, the higher the cardiovascular fitness and the better the cognitive and brain functioning (Gomez-Pinilla & Hillman, 2013; Åberg et al., 2009; Tarumi & Zhang, 2015; Voss et al., 2016). Together with the outcome of randomized controlled trials (RCT; where increased VO_2max after exercise intervention was accompanied by enhanced cognition) (Erickson et al., 2011) this evidence has fueled the cardiovascular (selective) hypothesis, that is, the regular practice of cardiovascular exercise (not other forms of physical activity such as stretching; cf. Kramer et al., 1999) positively affect cognition (with a selective influence on executive function) by means of its physiological effects at neural level. The null effect of VO₂ and the non-significant RT difference between self-paced exercise and the sedentary condition (well differentiated in terms of cardiovascular fitness) in our study appear to challenge that hypothesis (cf. Etnier et al., 2006).

A simple explanation of our null result is that VO_2max is a measure of cardiovascular fitness that is not sensitive enough to variations in cognitive performance (RT in the PVT here), even though, as we said above, it is the main index used to date. Proponents of the cardiovascular selective hypothesis could also argue that our null finding was due to the low executive demands of the PVT. The evidence from neuroimaging data speaks against that account. There is indeed work reporting that PVT performance involves selective activation of brain areas related to sustained attention and cognitive control (Drummond et al., 2005; Zhu et al., 2018), presumably driven by the (large) temporal uncertainty of the target appearance, the repetitive and monotonous nature of the task and the need of generating temporal expectations. Of course, in defense of the cardiovascular selective hypothesis, it could still be claimed that the executive demands of the PVT are much lower than those of the conflict or working memory tasks used in previous research on this topic. This is a matter open to opinion and debate as no study to date has addressed this issue directly.

The fact that self-paced exercise was not related to improved RT with respect to the sedentary condition is not surprising if one considers that only half of the studies in which we compared a group of self-paced sport athletes with a group of sedentary individuals revealed statistically significant RT differences. Assuming that the large N in the present analysis ensures sufficient statistical power, one could argue that those reported significant group RT differences were false positives (at least considering the first 5′ of the task). However, those RT group differences from the single studies might still be meaningful but explained by the influence of an unknown (uncontrolled) variable. Moreover, in the studies showing positive results, group differences were not only seen in terms of RT but also in accuracy performance in an oddball task (with much lower RT demands than the PVT; see Ciria et al., 2017), and, more importantly, in task-related cardiac and electroencephalographic measures (Luque-Casado et al., 2013, 2016a, 2016b). If those group differences were to be true, the result of the present analysis suggests that they were not due to mere differences in cardiovascular fitness.

Even if it involves executive functioning, the PVT is clearly a task demanding visuomotor coordination to react as rapidly as possible to the target appearance. This could explain that only externally-paced sport practice associates to (enhanced) RT performance in our study, supporting the cognitive skills hypothesis whereby sport practice is just another medium for cognitive training (Voss et al., 2010), over and above its effect due to cardiovascular physiological adaptations. Obviously, this result cannot be taken as evidence of the effect of externally-paced sport practice on the vigilance capacity of our participants for other variables not related to the sport practice itself could well account for the reported positive relationship. For instance, pre-existing individual differences that biased the choice of the particular sport practice (Belsky et al., 2015 for a related argument) might explain the sport-type effect in our multi-study analysis, and also the interaction between sport type and age. In fact, most of the younger participants in the externally-paced sport type groups were football players from (two) Spanish 1st Division League junior teams (with strict selection criteria and talent identification programs) while older participants were amateur/recreational athletes. However, the above and any other alternative explanation are speculative, for only well-designed RCTs would establish cause-effect relationships between exercise practice and cognitive performance. Also, future research should clarify whether indexes of response accuracy (beyond psychomotor response speed typical of the PVT) from other cognitive tasks measuring sustained attention (e.g., oddball task) discriminate between the type of sport practice and cardiovascular fitness in relation to the ability to maintain attention over time.

In the absence of that RCT, one could still argue that the extant evidence on the effect of chronic exercise on cognition supports the hypothesis that practicing exercise and sport regularly would have a positive effect on vigilance capacity. However, that evidence is not as conclusive as it may appear, with some systematic reviews and meta-analysis reporting positive results (Colcombe & Kramer, 2003; Guiney & Machado, 2013) and other showing null effects (Angevaren et al., 2008; Singh et al., 2018; Verburgh et al., 2014; Young et al., 2015).

Conclusions

The results of the multistudy analysis reported here point to the type of sport practice as a major factor to explain differences in vigilance performance as a function of regular exercise over and above the level of aerobic fitness, which in turn challenges the cardiovascular hypothesis. In any case, this topic warrants further well-designed research (RCTs) to unveil whether chronic exercise (and sport practice) has a true effect on cognition in general, and vigilance in particular. Interested researchers are facing a challenging task, as they would have to take into account the cognitive demands of the sport (exercise) activity as we highlighted here, the many other factors related to exercising (such as the intensity and duration of each exercise session or the periodization of the exercise program) and all potential mediators (BDNF, sleep, motivation, etc.) (Stillman et al., 2016). Only after solid evidence is acquired, researchers would be ready to pave the road for prescription of exercise as a potential tool to enhance cognition and prevent cognitive decline.