Introduction
Aerobic exercise appears to benefit cognitive function in children (Erickson et al., Reference Erickson, Milham, Colcombe, Kramer, Banich, Webb and Cohen2004; Fedewa & Ahn, Reference Fedewa and Ahn2011; Lees & Hopkins, Reference Lees and Hopkins2013; Sibley & Etnier, Reference Sibley and Etnier2003; Tomporowski, Davis, Miller, & Naglieri, Reference Tomporowski, Davis, Miller and Naglieri2008) and healthy older adults (Bherer, Erickson, & Liu-Ambrose, Reference Bherer, Erickson and Liu-Ambrose2013; Kramer, Erickson, & Colcombe, Reference Kramer, Erickson and Colcombe2006). Epidemiological evidence from children regarding exercise and cognitive function is unavailable, but case-control (Geda et al., Reference Geda, Roberts, Knopman, Christianson, Pankratz, Ivnik and Rocca2010) and prospective cohort (Flicker et al., Reference Flicker, Almeida, Acres, Le, Tuohy, Jamrozik and Norman2005; Weuve et al., Reference Weuve, Kang, Manson, Breteler, Ware and Grodstein2004) studies of older adults suggest that aerobic physical activity delays age-related cognitive decline (Haskell, Reference Haskell2008; Sofi et al., Reference Sofi, Valecchi, Bacci, Abbate, Gensini, Casini and Macchi2011), loss of gray matter volume in the prefrontal cortex (Erickson et al., Reference Erickson, Raji, Lopez, Becker, Rosano, Newman and Kuller2010; Rovio et al., Reference Rovio, Spulber, Nieminen, Niskanen, Winblad, Tuomilehto and Kivipelto2010) and neurodegenerative disease (Hamer & Chida, Reference Hamer and Chida2009), while also reducing mortality risk (Samitz, Egger, & Zwahlen, Reference Samitz, Egger and Zwahlen2011). Physical inactivity in the teen years has been related to cognitive problems in late life, and this relationship was attenuated by mid-life physical activity (Middleton, Barnes, Lui, & Yaffe, Reference Middleton, Barnes, Lui and Yaffe2010). However, most of the epidemiological studies have examined physical activity and cognitive performance in adults aged 60 and older (Prakash, Voss, Erickson, & Kramer, Reference Prakash, Voss, Erickson and Kramer2015). Reviewers have pointed out there is inadequate information about the extent to which age moderates the relationship between physical activity and cognition, and that more research is needed about exercise and cognitive function from large samples with a broad age range that better represents the general population (Smith et al., Reference Smith, Blumenthal, Hoffman, Cooper, Strauman, Welsh-Bohmer and Sherwood2010; Snowden et al., Reference Snowden, Steinman, Mochan, Grodstein, Prohaska, Thurman and Anderson2011).
Whether resistance exercise is effective in delaying declines in cognitive function associated with aging or promotes better cognitive function at younger ages is unclear based on the epidemiological evidence. In randomized trials conducted with older adults, resistance training added to short-term aerobic exercise training led to larger improvements in cognitive performance when compared to aerobic training alone (Colcombe & Kramer, Reference Colcombe and Kramer2003). At least seven randomized trials show that short-term resistance training alone improved cognitive performance (Busse et al., Reference Busse, Filho, Magaldi, Coelho, Melo, Betoni and Santarem2008; Cassilhas et al., Reference Cassilhas, Viana, Grassmann, Santos, Santos, Tufik and Mello2007; Lachman, Neupert, Bertrand, & Jette, Reference Lachman, Neupert, Bertrand and Jette2006; Liu-Ambrose & Donaldson, Reference Liu-Ambrose and Donaldson2009; Moul, Goldman, & Warren, Reference Moul, Goldman and Warren1995; Peig-Chiello, Perrig, Ehrsam, Staehelin, & Krings, Reference Peig-Chiello, Perrig, Ehrsam, Staehelin and Krings1998; Tsutsumi, Don, Zaichkowsky, & Delizonna, Reference Tsutsumi, Don, Zaichkowsky and Delizonna1997). In the most recent experiment of this type, the participants were presented with four possible stimuli (60 total in 3 blocks of 20), either a single digit (1 or 3) or three digits (111 or 333). Participants were given two instructions that switched during the task: indicate how many digits appeared (1 or 3) or identify the number present on the screen (1 or 3). In 40% of the trials, these instructions changed which required a change in cognitive strategy. This short-term randomized trial with older adults showed small, statistically insignificant improvements in task switch time and accuracy after resistance training, in part because similar improvements occurred in the health education controls (Kimura et al., Reference Kimura, Obuchi, Arai, Nagasawa, Shiba, Watanabe and Kojima2010).
Much of the knowledge about relationships between cognitive function and physical activity or aerobic fitness is based on small, cross-sectional, laboratory studies that in general targeted homogenous samples of older adults, college students or children (Bauermeister & Bunce Reference Bauermeister and Bunce2014; Brown et al., Reference Brown, McMorris, Longman, Leigh, Hill, Friedenreich and Poulin2010; Chodzko-Zajko, Reference Chodzko-Zajko1991; Colcombe et al., Reference Colcombe, Erickson, Raz, Webb, Cohen, McAuley and Kramer2003; Davis & Cooper, Reference Davis and Cooper2011; Elsayed, Ismail, & Young, Reference Elsayed, Ismail and Young1980; Erickson et al., Reference Erickson, Prakash, Voss, Chaddock, Hu, Morris and Kramer2009; Etnier, Nowell, Landers, & Sibley, Reference Etnier, Nowell, Landers and Sibley2006; Hillman, Castelli, & Buck, Reference Hillman, Castelli and Buck2005; Hillman, Kramer, Belopolsky, & Smith, Reference Hillman, Kramer, Belopolsky and Smith2006; Van Boxtel et al., Reference Van Boxtel, Paas, Houx, Adam, Teeken and Jolles1997; Voss et al., Reference Voss, Chaddock, Kim, VanPatter, Pontifex, Raine and Kramer2011). Results from these investigations, and the available evidence from randomized controlled trials, suggest that physical activity and aerobic fitness are associated with modest improvements in memory, information processing speed and executive function (Angevaren, Aufdemkampe, Verhaar, Aleman, & Vanhees, Reference Angevaren, Aufdemkampe, Verhaar, Aleman and Vanhees2008; Colcombe & Kramer, Reference Colcombe and Kramer2003; Smith et al., Reference Smith, Blumenthal, Hoffman, Cooper, Strauman, Welsh-Bohmer and Sherwood2010). These observations are buttressed by rodent experiments and human neuroimaging studies with older adults and children showing that exercise training induced changes in brain anatomy and physiology that could plausibly support better cognitive control (e.g., altered activity in prefrontal and anterior cingulate cortices) (Chaddock-Heyman et al., Reference Chaddock-Heyman, Erickson, Voss, Knecht, Pontifex, Castelli and Kramer2013; Krafft et al., Reference Krafft, Schwarz, Chi, Weinberger, Schaeffer, Pierce and McDowell2014) and memory (e.g., increased hippocampal and medial temporal lobe volumes and enhanced hippocampal dentate gyrus neurogenesis) (Erickson et al., Reference Erickson, Voss, Prakash, Basak, Szabo, Chaddock and Kramer2011; Pereira et al., Reference Pereira, Huddleston, Brickman, Sosunov, Hen, McKhann and Small2007; ten Brinke et al., Reference ten Brinke, Bolandzadeh, Nagamatsu, Hsu, Davis, Miran-Khan and Liu-Ambrose2014).
There are few population-based studies that have examined the association of physical activity and cognition across a broad range of ages. One study analyzed data from the mobile app BrainBaseline (Lee et al., Reference Lee, Baniqued, Cosman, Mullen, McAuley, Severson and Kramer2012). The participants (n=15,346; 34%≥40 years of age) first completed a short practice block on several cognitive tests during which performance feedback was provided. Next, a main test block involving the same cognitive tests was completed, and these results were used for the analysis. High- and low-exercise groups were created from self-reported prior week frequencies of strenuous, moderate and mild intensity exercise bouts lasting 15 min or more. Using statistical models that controlled for gender and level of education as covariates, the high-exercise group showed faster information processing speed compared to the low-exercise group. There was no difference between the two exercise groups on measures of memory or attention (Lee et al., Reference Lee, Baniqued, Cosman, Mullen, McAuley, Severson and Kramer2012).
Here, we extend prior research by examining associations between cognitive function and both resistance and aerobic exercise in a large (n=8752), heterogeneous group of online game players ages 13 to 89 years. The sample was older (57.6%>40 years of age) compared to the Lee and colleagues study (Lee et al., Reference Lee, Baniqued, Cosman, Mullen, McAuley, Severson and Kramer2012). In addition to gender and education level, we controlled for several other putative confounding variables including smoking (Anstey, von Sanden, Salim, & O’Kearney, Reference Anstey, von Sanden, Salim and O’Kearney2007), caffeinated coffee and tea use (Beydoun et al., Reference Beydoun, Gamaldo, Beydoun, Tanaka, Tucker, Talegawkar and Zonderman2014; Ritchie et al., Reference Ritchie, Artero, Portet, Brickman, Muraskin, Beanino and Carrière2010), and sleep (Fortier-Brochu, Beaulieu-Bonneau, Ivers, & Morin, Reference Fortier-Brochu, Beaulieu-Bonneau, Ivers and Morin2012). The flanker task was used because physical activity, exercise training, aerobic fitness, and mobility in older adults have been associated with better flanker performance in several studies of small homogeneous samples (Bauermeister & Bunce, Reference Bauermeister and Bunce2014; Chaddock-Heyman et al., Reference Chaddock-Heyman, Erickson, Voss, Knecht, Pontifex, Castelli and Kramer2013; Chaddock et al., Reference Chaddock, Erickson, Prakash, Vanpatter, Voss, Pontifex and Kramer2010; Chaddock et al., Reference Chaddock, Erickson, Prakash, Voss, VanPatter, Pontifex and Kramer2012; Colcombe & Kramer, Reference Colcombe and Kramer2003; Colcombe et al., Reference Colcombe, Kramer, Erickson, Scalf, McAuley, Cohen and Elavsky2004; Gothe et al., Reference Gothe, Fanning, Awick, Chung, Wójcicki, Olson and McAuley2014; Hillman, Belopolsky, Snook, Kramer, & McAuley, Reference Hillman, Belopolsky, Snook, Kramer and McAuley2004; Hillman, Buck, Themanson, Pontifex, & Castelli, Reference Hillman, Buck, Themanson, Pontifex and Castelli2009; Hillman, Motl, et al., Reference Hillman, Motl, Pontifex, Posthuma, Stubbe, Boomsma and de Geus2006; Krafft et al., Reference Krafft, Schwarz, Chi, Weinberger, Schaeffer, Pierce and McDowell2014; McAuley, Szabo, et al., Reference McAuley, Szabo, Mailey, Erickson, Voss, White and Kramer2011; Niemann, Godde, Staudinger, & Voelcker-Rehage, Reference Niemann, Godde, Staudinger and Voelcker-Rehage2014; Pontifex et al., Reference Pontifex, Kamijo, Scudder, Raine, Khan, Hemrick and Hillman2014; Voss et al., Reference Voss, Chaddock, Kim, VanPatter, Pontifex, Raine and Kramer2011). The flanker task is a measure of executive function (Denckla, Reference Denckla1996), or cognitive control (Botvinick, Braver, Barch, Carter, & Cohen, Reference Botvinick, Braver, Barch, Carter and Cohen2001), that assesses the ability to selectively attend to relevant information, to ignore irrelevant information and to inhibit incorrect responses (Eriksen & Eriksen, Reference Eriksen and Eriksen1974).
A novel feature of the present investigation is the examination of change across five blocks of a 45-s flanker task. Participants in studies of cognitive function often initially complete several practice trials in a single, short practice block to provide familiarization and reduce improvements in performance that result from practicing a task (i.e., practice effects). Subsequent cognitive tests are often presented in blocks (e.g., of 20 to 160 trials) with analyses performed on a criterion measure which averages performance across all the blocks. The practice trials often are not fully analyzed because they are thought to represent error variance. Practice effects, however, have been shown to be potentially important indices of cognitive health. Practice effects decrease with age (Rönnlund, Lövdén, & Nilsson, Reference Rönnlund, Lövdén and Nilsson2007) and are smaller among people with lower IQ (Rapport, Brines, Theisen, & Axelrod, Reference Rapport, Brines, Theisen and Axelrod1997) or mild cognitive impairment (Howieson et al., Reference Howieson, Carlson, Moore, Wasserman, Abendroth, Payne-Murphy and Kaye2008; Wilson, Leurgans, Boyle, & Bennett, Reference Wilson, Leurgans, Boyle and Bennett2011). Practice effects also appear to have diagnostic utility (Duff et al., Reference Duff, Lyketsos, Beglinger, Chelune, Moser, Arndt and McCaffrey2011; Shuttleworth-Edwards, Radloff, Whitefield-Alexander, Smith, & Horsman, Reference Shuttleworth-Edwards, Radloff, Whitefield-Alexander, Smith and Horsman2014) and can significantly predict prospective changes in cognitive performance in groups at increased risk for cognitive decline (Duff et al., Reference Duff, Beglinger, Moser, Paulsen, Schultz and Arndt2010, Reference Duff, Beglinger, Schultz, Moser, McCaffrey, Haase and Paulsen2007; Granholm, Link, Fish, Kraemer, & Jeste, Reference Granholm, Link, Fish, Kraemer and Jeste2010; Machulda et al., Reference Machulda, Pankratz, Christianson, Ivnik, Mielke, Roberts and Petersen2013; Newman et al., Reference Newman, Kirchner, Phillips-Bute, Gaver, Grocott, Jones and Blumenthal2001; Suchy, Kraybill, & Franchow, Reference Suchy, Kraybill and Franchow2011). Similarly, analyses for the first block of trials or changes across the primary test blocks typically have not been reported, although the potential usefulness of considering such changes has been recognized for years (Spirduso, Reference Spirduso1980). The relationship between regular (i.e., frequent) physical activity and change in performance during the initial practice trials or across repeated blocks of cognitive tasks has rarely been investigated (Herting & Nagel, Reference Herting and Nagel2012), and whether age moderates the effect is unknown.
The primary aim of this study was to examine whether people differed in their change in performance across the first five blocks of a flanker task in a large heterogeneous group and whether those trajectories of change were associated with self-reported aerobic or resistance exercise frequency according to age.
Method
Study Population
A data set of 10,000 game players who completed the 45-s Lost in Migration game 5 times between January 1, 2012, and September 24, 2013, was received from Lumos Labs. The data set was trimmed to exclude those who indicated on a lifestyle questionnaire that they had a medical condition that prevented getting regular activity and those reporting an age less than 13 or older than 89.99. The final sample included in the analysis consisted of 8752 Lumosity users. The research was completed in accordance with the Helsinki Declaration.
Outcome Measures
Online Flanker Task
Participants freely chose to log onto the Lumosity Web site to play or train on one or more cognitive games (Sternberg et al., Reference Sternberg, Ballard, Hardy, Katz, Doraiswamy and Scanlon2013). The present analysis focused on the game Lost in Migration (http://blog.lumosity.com/the-science-behind-lost-in-migration) because it is a modified Ericksen flanker task (Eriksen & Eriksen, Reference Eriksen and Eriksen1974). Before the task, instructions were provided. A first screen was presented with the instructions to “spot the direction of the center bird, do not let the others distract you—some are lost in migration!”. A second screen was presented that included instructions to “press keyboard arrows to input the direction of the central bird” and showed icons of the four arrow keys of a standard computer keyboard. The arrow keys were the only potential valid response options, other key presses were inaccurate responses. Pressing the start button on the third screen started the 45-s task. During the task, each screen shows a flock of five birds. The goal was to use the arrow keys on a keyboard to indicate the direction that the central bird was facing as accurately and quickly as possible. Once the game started, a new flock of birds appeared immediately after each response, regardless of accuracy, until 45 s had elapsed at which time that game was over. Each 45-s game was a block. The head of the central bird faced either up, down, left or right, and the direction of the central bird could change with each flock presented. In all presentations, there were four flanking birds, two on each side of the central bird. On half of the presentations in each block, the birds all faced in the same direction (congruent trials), and on the other presentations the central bird faced a direction opposite of that of the other four flanking birds (incongruent trials). Here, we focused on performance accuracy, which often differs between young and old (Spreng, Wojtowicz, & Grady, Reference Spreng, Wojtowicz and Grady2010), during the congruent and incongruent presentations. Reaction time data were not included because the precise timing of visual stimuli presentation and the accuracy of reaction time data can be influenced by differences in software and hardware configurations and internet connections speeds (Birnbaum, Reference Birnbaum2004). The day on which the game was played was recorded and used in the analysis. Some players completed five blocks on the same day while others took days or months to complete five blocks. Some players completed some or all of their five blocks as part of a “daily workout” in which the types of games played are determined by a proprietary algorithm. Other players made the decision to play the Lost in Migration game for some or all of their blocks after navigating to the “free play” part of the website. This variable (i.e., “daily workout” vs. “free play”) was used in the analysis.
Physical activity and lifestyle questions
Participants were asked two questions about their physical activity: “In a typical week, how many times do you engage in aerobic exercise (e.g., running, cycling, aerobics, swimming, brisk walking, hiking, etc.)?” and “In a typical week, how many times do you engage in resistance exercise (e.g., lifting weights, push-ups, sit-ups, etc.)?” Responses used were never, rarely and 1–7 days per week.
Participants were asked “Do you smoke?” (possible responses were - yes or no)”, “How many hours of sleep do you typically get each night?” (less than 4 hr in hour increments up to more than 10 hr), “What is your education level?” (1=some high school, 2=high school, 3=some college, 4=associate, 5=bachelors, 6=masters, 7=professional, 8=Ph.D.), and “On average, how many cups of caffeinated coffee or tea do you drink each day?” (0 to ≥5 cups).
Statistical Analysis
Latent transition and class analysis
Participants in the cohort were classified as performing “daily workouts” or “free play” or as completing sessions within a day or between days across the blocks using latent transition analysis (LTA) by Mplus 7.3 (Muthén & Muthén, 1998–Reference Muthén and Muthén2012). LTA provided Bayesian probability estimates of people being classified into those groups based on their observed status at each block. Model fit was tested using a robust maximum likelihood ratio test. The number of classes was tested by a significant chi-square change (χ2 Δ) estimated by a bootstrapped likelihood ratio test.
Latent growth modeling
Trajectories of change in percent accuracy were estimated using latent growth modeling (Bollen & Curran, Reference Bollen and Curran2006) in Mplus 7.3 All models were adjusted for between- and within-participant variation in time between blocks using a multilevel random slopes model that uses individually varying time periods of observations within each person to model change. Aerobic activity and age were each centered on their sample means. Other multivariate adjustments were made for between-participant differences by including the following covariates in the growth model: age at block 1, gender, education level, smoking, sleep, Lumosity training status (i.e., “daily workout” or “free play”), completing sessions within a day or between days, and caffeinated coffee or tea consumption. Near complete data were obtained for accuracy of the flanker task (0.002% missing on one block, 0% on the other blocks) and the lifestyle questions, which ranged from 0% missing for aerobic exercise frequency to 0.8% missing for resistance exercise frequency. The MLR estimator uses full-information to impute missing values and is robust to non-normality and up to 25% missing data (Enders & Bandalos, Reference Enders and Bandalos2001). Robust maximum likelihood parameters and their standard errors were estimated for initial status (i.e., mean at baseline), change (i.e., trajectory of differences across the five blocks), and the variances (i.e., inter-individual and intra-individual differences) of initial status and change. Model fit was evaluated using the chi-square (χ2) statistic, comparative fit index (CFI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR) (Hu & Bentler, Reference Hu and Bentler1999). Values of CFI ≥0.90 were judged to be acceptable, while values >0.95 indicated good fit. Values of the RMSEA ≤0.06 and ≤0.08 indicated close and acceptable fit. Concurrent values ≥0.95 for CFI and ≤0.08 for SRMR provide optimal protection against type I and type II error rates. A significant interaction was decomposed using standard procedures (Aiken & West, Reference Aiken and West1991).
Results
Demographics and Descriptive Findings
Demographics and raw cognitive performance data for the sample are presented by age group, aerobic exercise frequency, and resistance exercise frequency in Tables 1, 2, and 3, respectively. The mean (SD) number of total flanker trials completed within each of the five blocks were 42.8 (15.7), 45.6 (15.2), 46.8 (16.2), 47.4 (15.9), and 48.1 (16.0) for blocks 1–5, respectively.
Table 1 Demographics of participants and unadjusted cognitive performance data by age group
Note. Means are reported unless otherwise specified. Standard deviations are in parentheses except for the “number of participants” variable.
Table 2 Demographics of participants and unadjusted cognitive performance data by typical weekly frequency of aerobic exercise
Note. Means are reported unless otherwise specified. Standard deviations are in parentheses except for the “number of participants” variable.
Table 3 Demographics of participants and unadjusted cognitive performance data by typical weekly frequency of resistance exercise
Note. Means are reported unless otherwise specified. Standard deviations are in parentheses except for the “number of participants” variable.
Latent Transition and Class Analysis
The best fitting models indicated two classes of cases for sessions [3347 participants (38%) completed blocks within the same day and 5405 participants (62%) completed blocks on different days; χ2 Δ (6)=8968.5; p<.001] and for Lumosity training status [6111 participants (70%) played the game as part of a “daily workout” and 2641 participants (30%) played as part of free play; χ2 Δ (6)=55729; p<.001]. Classification probabilities were 95% for sessions and 99% for training status. The mean (±SD) time between sign-up to play and block 1 of the Lost in Migration game was 8.2 months (±3.7 months). Thirty-nine percent to 43% of the five blocks occurred within 24 hr. Between blocks 2 to 5, the mean (±SD) inter-day time between blocks ranged from 14 to 16 days (±27 to 30) days. The mean (±SD) intra-day time between blocks ranged from 3.8 to 4.0 hr (±7.6 to 7.9 hr). Sixty-seven percent to 71% of the blocks occurred during Lumosity training.
Participants who subsequently were more likely to complete blocks on different days were 2% more accurate at block 1 [95% confidence interval (CI), 1.8% to 2.30%; z=15.5; p<.001], but they had similar linear (p=.649) and quadratic (p=.856) change compared to participants who were more likely to complete blocks during the same day. Similarly, participants who were more likely to have performed the blocks as part of a “daily workout” were 1.8% more accurate at block 1 (95% CI, 1.54% to 2.14%; z=12.05; p<.001), but they had similar linear (p=.660) and quadratic (p=.972) change compared to those who were not enrolled.
Growth Models
Incongruent presentations across five blocks
Figure 1 shows unadjusted accuracy during incongruent presentations during blocks 1 through 5. Accuracy increased from 94.4% at block 1 to 95.4% at block 2 and was stable thereafter (ICC-2, 5=.70, 95% CI, .68–.72). Growth modeling indicated linear increase [B=0.600%, 95% CI, 0.42 to 0.78; z=6.53; p<.001] from the estimated initial score of 94.6% (95% CI, 94.4 to 94.8) followed by a quadratic deceleration [B=−0.109%; 95% CI, −0.015 to −0.002; SE=0.021; z=5.32; p<.001], resulting in a net 0.491% increase in accuracy on blocks 2–5. Variance was 14%, (SE=4%, p=.001) for the linear slope and 0.5% (SE=0.19%; p=.005) for the quadratic slope. Model fit was good (χ2 (6)=15.5; p=.02; CFI=0.992; RMSEA=.013 (.005–.022), SRMR=.044). After adjustment for variation in time between blocks, aerobic activity was unrelated to accuracy on block 1 (p=.807), but it was positively related to linear increase in accuracy (B=0.713%; 95% CI, 0.256 to 1.170) and inversely related to the quadratic deceleration of accuracy gains (B=−0.767%; 95% CI, −1.251 to −0.283).
Fig. 1 Observed and predicted accuracy during the incongruent flanker trials presented within five 45-s blocks. Mean ±95% confidence interval. The 95% CIs are too small to be observed in the figure but are ~0.20% for the observed and ~0.10% for the predicted data.
Figure 1 shows predicted accuracy during incongruent presentations during blocks 1 through 5, after adjustment for time between blocks, covariates, sessions, and training status. Aerobic physical activity remained unrelated to accuracy during block 1 (p=.692) (Table 4). Accuracy on block 1 was higher in older players, those with more education, in women, those who drank more coffee or tea, and those who were more likely to have played the game as part of a “daily workout” or complete blocks on different days. Aerobic activity was positively related to linear increase in accuracy (B=0.577%; 95% CI, 0.112 to 1.25) and inversely related to the quadratic deceleration of accuracy gains (B=−0.619%; 95% CI, −1.117 to −0.121). For each day of aerobic physical activity above the mean of 2.8 days, there was approximately six-tenths of a percent linear increase in accuracy and six-tenths of a percent less deceleration in accuracy. Thus, people who reported above-average aerobic physical activity attained and maintained a higher accuracy during incongruent blocks 2–5 compared to people who reported below-average activity. In addition, the interaction of aerobic physical activity with age was inversely related to the linear increase in accuracy (B=−0.031%; 95% CI, −0.05 to −0.01; p=.005) and positively related to the quadratic deceleration of increased accuracy (B=0.03%; 95% CI, 0.01 to 0.05; p=.01). Figures 2 and 3 show that physically active participants younger than age 45 had a larger linear increase and a smaller quadratic deceleration compared to other participants.
Fig. 2 Aerobic activity-by-age linear effect on predicted accuracy during incongruent trials.
Fig. 3 Aerobic activity-by-age quadratic effect on predicted accuracy during incongruent trials.
Table 4 Fully adjusted growth model for accuracy during incongruent trials within the first block of the flanker task in relation to aerobic exercise
Resistance activity was unrelated to accuracy at block 1 (B=0.042; SE=0.033; p=.205) or to linear (B=0.239; SE=0.261; p=.361) and quadratic (B=−0.143; SE=0.305; p=.638) change.
Congruent presentations across five blocks
After adjustment for variation in time between blocks, aerobic activity was unrelated to accuracy on block 1 (B=0.010%; SE=0.023; p=.653), but it was positively related to linear increase in accuracy (B=0.427%; 95% CI, 0.084 to 0.770) and inversely related to the quadratic deceleration of accuracy gains (B=−0.449% CI, −0.817 to −0.081). Thus, people who reported above-average aerobic physical activity attained and maintained a higher accuracy during congruent blocks 2–5 compared to people who reported below-average activity.
After adjustment for varying time between blocks, covariates, sessions, and training status Table 4), aerobic physical activity was unrelated to congruent accuracy during block 1 (B=0.017; SE=0.022; p=.445), linear change (B=0.289; SE=0.183; p=.113), and quadratic change (B=−0.315; SE=0.187; p=.092). Accuracy on block 1 was higher in players with more education, in women, those who drank more coffee or tea, and those who were more likely to have been those to have played the game as part of a “daily workout” or complete blocks on different days.
Resistance activity was unrelated to accuracy during congruent presentations during block 1 (B=−0.022; SE=0.034; p=.523), linear change (B=0.252; SE=0.452; p=.578), and quadratic change (B=−0.142; SE=0.688; p=.837).
Discussion
Results for Change across Blocks 1 to 5
The primary finding from the present investigation was that for the more cognitively demanding, incongruent flanker task items, the frequency of aerobic exercise was positively related to increased accuracy that occurred between block 1 and 2, and inversely related to deceleration of accuracy gains across blocks 2 to 5. Thus, people who reported above-average aerobic physical activity attained and maintained a higher accuracy during incongruent blocks 2–5 compared to people who reported below-average activity. More particularly, those under age 45 who reported higher than average frequency of aerobic activity showed the largest increase in incongruent accuracy from block 1 to 2 and also less deceleration in accuracy from blocks 2 to 5 compared to those over age 45 who reported below average aerobic activity.
We have been unable to find prior studies documenting changes in flanker performance accuracy analyzed across multiple blocks of ~46 trials, either within a session on a single day or across days. Some studies have not specified the number of trials or blocks (McAuley, Mullen, et al., Reference McAuley, Mullen, Szabo, White, Wójcicki, Mailey and Kramer2011), others have analyzed a single block of 40 to 160 flanker trials (Gothe et al., Reference Gothe, Fanning, Awick, Chung, Wójcicki, Olson and McAuley2014) and some have averaged multiple blocks of fewer trials per block (e.g., 17–20 trials) (Colcombe et al., Reference Colcombe, Kramer, Erickson, Scalf, McAuley, Cohen and Elavsky2004). Thus, it is not possible to directly compare the present findings to similar published studies. Performance on the flanker test can improve with practice, especially in incongruent presentations for which there is less potential for practice effects to be attenuated by a ceiling effect. Test–retest changes in mean-level performance have been quantified using nearly 1600 effects from published neuropsychological test results, and executive function and attention tasks were found to be reduced by ~.01 SD across a 1-year time frame (Calamia, Markon, & Tranel, Reference Calamia, Markon and Tranel2012). In this meta-analysis, separate effects were not provided for flanker tasks and no quantitative review of shorter-term changes in flanker performance, such as within a single day or across a week or a month, has been published. In the one study examining flanker accuracy, practice effects were found to be largest during the first six trials compared to subsequent trials (Cohen-Kdoshay & Meiran, Reference Cohen-Kdoshay and Meiran2009). The potential moderating effect of age was not considered.
There appears to be inadequate prior experimental evidence to draw a strong conclusion as to whether the magnitude of flanker performance is moderated by age. The larger increase in accuracy between block 1 and 2 and the attenuated deceleration of accuracy gains between blocks 2 and 5 for aerobically active participants younger than age 45 found here is generally consistent with prior research on age and practice effects. In a sample of 1616 adults ranging from 18 to over 80 years of age, the magnitude of practice effects were negatively correlated with age across a range of cognitive tasks, including memory (Salthouse, Reference Salthouse2010). Although tasks focused on cognitive control, such as the Stroop or flanker, were not examined, the group-level correlation between mean age and the estimated practice effect across 2.5 years for memory was −.86 (Salthouse, Reference Salthouse2010). The findings in the literature are not unanimous, for example, a larger practice effect, defined by the time it took to complete the Stroop task, was reported in 638 participants ≥80 years of age compared to 1174 participants ≤71 years of age in the PROSPER trial (Houx et al., Reference Houx, Shepherd, Blauw, Murphy, Ford, Bollen and Westendorp2002).
Aerobic fitness has not been linked previously to learning across blocks of a flanker task but has been considered in relation to the initial trials of at least two cognitive tasks. Aerobic fitness in 30 adolescents was significantly and positively associated with learning during the first six trials of a virtual Morris Water Maze task which measures aspects of visuospatial memory (Herting & Nagel, Reference Herting and Nagel2012). In a separate study of eight adults under age 46, both aerobic fitness at the end of 3-months of exercise training and changes in aerobic fitness were significantly and positively associated with improvements in immediate recall of the first 20 words presented in a modified Rey Auditory Verbal Learning Test (Pereira et al., Reference Pereira, Huddleston, Brickman, Sosunov, Hen, McKhann and Small2007).
Although the flanker task has been described as emphasizing cognitive control, performance on the version of the task used here also required adequate motivation, working memory and sustained attention (Botvinick & Braver, Reference Botvinick and Braver2015). Physical activity may have been associated with any or all of these processes thought to underlie flanker task performance. Working memory historically is thought to be a trait (Baddeley, Reference Baddeley2003) with a strong genetic component (Friedman et al., Reference Friedman, Miyake, Young, DeFries, Corley and Hewitt2008). Contemporary findings suggest that working memory can be improved to a degree and suggest that the underlying neural circuity has some plasticity in response to interventions such as intense cognitive training or exercise training (Jaeggi, Buschkuehl, Jonides, & Shah, Reference Jaeggi, Buschkuehl, Jonides and Shah2011; Mahncke et al., Reference Mahncke, Connor, Appelman, Ahsanuddin, Hardy, Wood and Merzenich2006; McNab et al., Reference McNab, Varrone, Farde, Jucaite, Bystritsky, Forssberg and Klingberg2009; Nagamatsu et al., Reference Nagamatsu, Chan, Davis, Beattie, Graf, Voss and Liu-Ambrose2013; Padilla, Pérez, & Andrés, Reference Padilla, Pérez and Andrés2014; Sprenger et al., Reference Sprenger, Atkins, Bolger, Harbison, Novick, Chrabaszcz and Dougherty2013; Volkers & Scherder, Reference Volkers and Scherder2014) or their combination (Smith et al., Reference Smith, Spiegler, Sauce, Wass, Sturzoiu and Matzel2013).
The mechanisms by which physical activity or inactivity could alter working memory are unclear, but are being investigated. Research with older adults, for instance, found that higher aerobic fitness levels attenuated age-associated reductions in N-acetylaspartate, a metabolite found in neuronal cell bodies that is decreased in Alzheimer’s disease, and that higher N-acetylaspartate concentrations mediated the association between aerobic fitness and working memory (Erickson et al., 2012). Whether this observation generalizes to practice effects is unknown.
Some evidence, although not all (Pontifex, Hillman, & Polich, Reference Pontifex, Hillman and Polich2009), suggests that exercise training can influence brain regions known to be involved in attention (Colcombe & Kramer, Reference Colcombe and Kramer2003; Pérez, Padilla, Parmentier, & Andrés, Reference Pérez, Padilla, Parmentier and Andrés2014). The present findings might be explained by improvements in attention processes needed to select or inhibit a correct response, including a specific motor response. The basal ganglia are thought to be involved in the selection of a needed motor program and the active inhibition of an unneeded one (Boraud, Bezard, Bioulac, & Gross, Reference Boraud, Bezard, Bioulac and Gross2002; Mink, Reference Mink1996). Also, studies of rodents and patients with Parkinson’s disease show that neuroplasticity in basal ganglia results from exercise-induced alterations in dopamine and glutamate (Petzinger et al., Reference Petzinger, Fisher, Van Leeuwen, Vukovic, Akopian, Meshul and Jakowec2010). Compared to less aerobically fit older adults, more highly fit older adults showed different activation of the anterior cingulate cortex and better flanker performance (Colcombe et al., Reference Colcombe, Kramer, Erickson, Scalf, McAuley, Cohen and Elavsky2004). In contrast, anterior cingulate cortex activation was not altered after resistance training in older adult despite improvement in flanker performance (Liu-Ambrose, Nagamatsu, Voss, Khan, & Handy, Reference Liu-Ambrose, Nagamatsu, Voss, Khan and Handy2012), consistent with the observation here that resistance training was unrelated to flanker performance.
Practice of attentional tasks alters brain activation responses. For example, decreases in anterior cingulate cortex activity and increases in dorsolateral prefrontal cortex activity during the Stroop were reduced in later practice trials compared to early practice trials (Erickson et al., Reference Erickson, Milham, Colcombe, Kramer, Banich, Webb and Cohen2004). The neural basis for physical activity or fitness influences on practice effects in tasks involving cognitive control rarely has been investigated. Compared to less fit children and during a flanker task, higher fit children showed greater activation of the anterior cingulate cortex but only in the early task block. Early was defined as the first 195 s (Chaddock et al., Reference Chaddock, Erickson, Prakash, Voss, VanPatter, Pontifex and Kramer2012). Whether those findings extend to adults is unknown.
The literature shows that the potential influence of aerobic fitness and physical activity on practice effects has frequently been ignored. The present findings underscore the potential usefulness of systematically including an analysis of practice effects in investigations of physical activity and cognitive function. Such data may be useful in predicting important cognitive health outcomes (Duff et al., Reference Duff, Beglinger, Moser, Paulsen, Schultz and Arndt2010) and also may offer an alternative hypothesis when exercise appears to have effects on cognitive performance at one age but not another.
Block 1 Results
A second finding was that accuracy at block 1, for both the congruent and incongruent flankers task, was unrelated to the frequency of either aerobic or resistance exercise. Only a few studies have examined the relationship between a single block of flanker data and physical activity or fitness. One study that analyzed a single block of 64 flanker trials found that lower aerobic fitness was associated with greater variability in flanker reaction time, and this effect increased with age but no information was reported about flanker accuracy (Bauermeister & Bunce, Reference Bauermeister and Bunce2014). A study of 179 older adults reported a weak correlation (r=.112) between incongruent flanker performance and estimated maximal oxygen consumption and the flankers appeared to have been presented in a single block (Verstynen et al., Reference Verstynen, Lynch, Miller, Voss, Prakash, Chaddock and Erickson2012).
Although it is unknown if any caffeine was consumed before the flanker tasks were performed, it was perhaps not surprising that block 1 performance was better among participants who reported consuming more caffeinated coffee or tea. Caffeine acts on dopamine rich brain areas involved in the executive control of visual attention (Lumme, Aalto, Ilonen, Någren, & Hietala, Reference Lumme, Aalto, Ilonen, Någren and Hietala2007) and can have dose-response effects on flanker performance (Brunyé, Mahoney, Lieberman, & Taylor, Reference Brunyé, Mahoney, Lieberman and Taylor2010). It also was not surprising that incongruent accuracy at block 1 was better among participants with a higher level of education because attention and motivation is crucially important for many types of learning and self-regulation that contributes to educational success (Posner & Rothbart, Reference Posner and Rothbart2005). Less certain is why older adults, women, those likely to complete blocks on different days and those likely to be performing the Lost in Migration game as part of a “daily workout” were better performers at block 1. Those electing to play cognitive games as part of a “daily workout” may have done so because of worry about cognitive decline or they had stronger perceptions of wanting to maintain cognitive health and consequently been more strongly motivated from the outset to perform well. Relatively little is known about why some older adults are motivated to engage in cognitive training using online games (Herman, de Kort, & Ijsselsteijn, Reference Herman, de Kort and Ijsselsteijn2009).
Demographic and Descriptive Results
The proportion of the sample reporting never or rarely engaging in physical activity (22.1%) was lower than U.S. population estimates for adults (26.1%) (Carroll et al., Reference Carroll, Courtney-Long, Stevens, Sloan, Lullo, Visser and Dorn2014) and higher than high school students (15.2%) (Kann et al., Reference Kann, Kinchen, Shanklin, Flint, Hakins and Harris2014). A higher percentage of the sample reported engaging in two or more weekly bouts of strengthening physical activity compared to estimates for the U.S. population (48% vs. 29.3%) (Centers for Disease Control, 2013). The number of hours slept and number of daily drinks of caffeinated coffee or tea were generally consistent with U.S. population estimates (Frary, Johnson, & Wang, Reference Frary, Johnson and Wang2005; Kripke, Garfinkel, Wingard, Klauber, & Marler, Reference Kripke, Garfinkel, Wingard, Klauber and Marler2002). The proportion of the sample that smoked was lower than the 2005–2012 estimated prevalence for adults in the U.S. (10.5% vs. 18.1%) (Agaku, King, & Dube, Reference Agaku, King and Dube2014). The lower accuracy in the more cognitively challenging incongruent flankers compared to the congruent flankers also was consistent with expectations based on prior research (Sanders & Lamers, Reference Sanders and Lamers2002).
Limitations and Conclusions
The present investigation had several limitations. Participants were not randomly selected from a well-defined population but did have access to the internet and chose to play the Lost-In-Migration game at least five times. Thus, the generalizability of the findings to the population at large, which includes people without internet access and those with no interest in online game playing, is uncertain.
The self-reported demographic and health data obtained were not verified and are potentially inaccurate. A systematic and quantitative review of studies comparing self-reported to objectively measured body weight, for example, showed that people have a tendency to report their weight as lower than their actual body weight (Gorber, Tremblay, Moher, & Gorber, Reference Gorber, Tremblay, Moher and Gorber2007). Whether this is true for the demographic and health data obtained here is uncertain. Notwithstanding, the measurement limitations of self-reported variables, there is little reason to think that such measurement errors would systematically bias the relations seen here with flanker performance.
Although we controlled for several potential confounding variables, no adjustments were made for others such as socioeconomic status or alcohol use. These variables are associated with both physical activity and cognitive performance and may have contributed to some of the findings observed here. For example, the Whitehall II study found that alcohol intake was associated with less physical activity and a reduced risk of poor cognitive function (Britton, Singh-Manoux, & Marmot, Reference Britton, Singh-Manoux and Marmot2004; Steinmo, Hagger-Johnson, & Shahab, Reference Steinmo, Hagger-Johnson and Shahab2014).
The single-item self-reported physical activity frequency questions that were used lacked information about duration and intensity. Thus, the present results are limited to self-reported exercise frequency and provide no insight into relationships that exercise duration, intensity, or total volume of exposure may have on trajectories of change in flanker performance. Single-item scales reduce participant burden and allow for data collection from large numbers of individuals who otherwise may be unwilling to provide physical activity information. Single item physical activity frequency questions similar to the questions used in this study were found to have strong repeatability and moderately strong associations with aerobic fitness or markers of cardiometabolic health (Churilla, Magyari, Ford, Fitzhugh, & Johnson, Reference Churilla, Magyari, Ford, Fitzhugh and Johnson2012; Kohl, Blair, Paffenbarger, Macera, & Kronenfeld, Reference Kohl, Blair, Paffenbarger, Macera and Kronenfeld1988; Loprinzi, Loenneke, & Abe, Reference Loprinzi, Loenneke and Abe2015; Milton, Bull, & Bauman, Reference Milton, Bull and Bauman2010).
The conditions under which the online game was played were unsupervised and uncontrolled (e.g., the size of the screen, lighting conditions, distractions in the environment), factors which may have altered or weakened effects that might have been observed if the data were obtained under laboratory conditions. No comparison was made between the incongruent trials of the flanker task that followed congruent trials and those that followed incongruent trials or visa-versa. Consequently no insight was obtained about potential aerobic or resistance exercise effects on phenomena such as feature repetition bias (Mayr, Awh, & Laurey, Reference Mayr, Awh and Laurey2003) or congruency switch costs (Schmidt & De Houwer, Reference Schmidt and De Houwer2011). Also, the analysis did not account for potential confounders such as prior experiences with online game playing or potential age-related variations in regular video game participation, engagement and interest (Festl, Scharkow, & Quandt, Reference Festl, Scharkow and Quandt2013; Kim, Park, & Baek, Reference Kim, Park and Baek2009).
Bearing in mind the limitations mentioned, this investigation suggests that (i) accuracy during the first block of an online flanker task, for both the congruent and incongruent flankers, was unrelated to the frequency of either self-reported aerobic or resistance exercise; and (ii) age moderates the association between, self-reported aerobic, but not resistance, exercise and changes in cognitive control that occur during the initial five blocks of an online flanker task. The present findings underscore the potential usefulness of systematically including an analysis of practice effects in investigations of physical activity and cognitive function.
Acknowledgments
All authors report no disclosures. The work reported here was completed without financial support. The data were provided by Lumos Labs, Inc. with the help of Faraz Farzin and Daniel Sternberg.
