Can Intraindividual Variability in Cognitive Speed Be Reduced by Physical Exercise? Results From the LIFE Study

Abstract

Objectives

Findings are mixed regarding the potential to improve older adults’ cognitive ability via training and activity interventions. One novel sensitive outcome may be intraindividual variability (IIV) in cognitive speed, or moment-to-moment changes in a person’s performance. The present article evaluated if participants who participated in a moderate physical activity intervention showed a reduction in IIV, compared with a successful aging education control group.

Method

For approximately 2.6 years, sedentary adults aged 70–90 years participated in the Lifestyle Interventions and Independence for Elders (LIFE) Study (n = 1,635), a multisite Phase 3 randomized controlled trial to reduce major mobility disability. They completed 4 reaction time tests at baseline and at approximately 24 months post-test.

Results

Analyses were conducted following both the intent-to-treat principle and complier average casual effect modeling. Results indicated that participants in the physical activity group did not show a reduction in their IIV.

Discussion

The lack of a significant reduction in IIV may be due to the mild nature of the physical activity program and the cognitively healthy sample. It is also possible that other types of lifestyle activity interventions (e.g., social and cognitive engagement) can elicit reductions in IIV for older adults.

A large number of studies have attempted to demonstrate the potential to improve cognitive ability in older adulthood via cognitive training and activity interventions. The majority of cognitive interventions involve extensive practice of tasks, or training of particular strategies in a cognitive domain to improve performance (Ball et al., 2002; Schmiedek, Lövdén, & Lindenberger, 2010), but their success is not entirely clear (Gross et al., 2012; Papp, Walsh, & Snyder, 2009). An alternative perspective is to introduce activities whereby stimulation is achieved via an intellectually and socially complex environment (Park, Gutchess, Meade, & Stine-Morrow, 2007; Stine-Morrow et al., 2014). Some activity-focused interventions, such as learning to act (Noice, Noice, & Staines, 2004) have been successful in demonstrating cognitive improvement in older adults, and there is a rich literature showing positive results from physical activity interventions (Smith et al., 2010). However, not all activity-based interventions have found cognitive improvement (Sink et al., 2015; Young, Angevaren, Rusted, & Tabet, 2015) or only observed limited cognitive change (Snowden et al., 2011). Further, the National Institutes of Health State-of-the-Science Conference concluded that there was “limited but inconsistent evidence” for cognitive activity engagement, and only preliminary evidence for physical activity, regarding their association with slower cognitive decline (Daviglus et al., 2010).

However, it is possible that potential benefits of an intervention may be missed due to a lack of sensitivity in the cognitive outcomes. For example, it has been proposed that neuroimaging measures may be more sensitive to intervention effects than traditional laboratory cognitive tests (Lustig, Shah, Seidler, & Reuter-Lorenz, 2009). Such assessments are costly and even in large funded trials, only feasible for a subpopulation of participants. Therefore, it is critical to explore other measures that may be sensitive to cognitive plasticity in lifestyle interventions.

One promising approach is intraindividual variability (IIV) in cognitive speed, or moment-to-moment changes in a person’s performance on a reaction time (RT) task. IIV reflects fluctuations in attention or executive control (Bunce, MacDonald, & Hultsch, 2004; West, Murphy, Armilio, Craik, & Stuss, 2002), and is a stable individual characteristic that provides unique information beyond mean RT (Hultsch, MacDonald, Hunter, Levy-Bencheton, & Strauss, 2000). Although IIV gradually increases throughout adulthood (Bielak, Cherbuin, Bunce, & Anstey, 2014), it is associated with maladaptive traits in older age, such as poorer cognition and everyday functioning (Burton, Strauss, Hultsch, & Hunter, 2009; Rabbitt, Osman, Moore, & Stollery, 2001), and poorer health (Bunce, Tzur, Ramchurn, Gain, & Bond, 2008) and engagement (Bielak, Hughes, Small, & Dixon, 2007). Higher levels of IIV are also found for individuals with neurological conditions, such as dementia (Hultsch et al., 2000) and mild cognitive impairment (MCI; Dixon et al., 2007). IIV is hypothesized to be a sensitive behavioral indicator of neurological integrity (Li & Lindenberger, 1999) and greater IIV in adulthood has been linked to maladaptive structural, functional, and neuromodulatory characteristics (MacDonald, Li, & Bäckman, 2009). There is evidence that greater IIV is correlated with white matter integrity in several brain regions, including the prefrontal cortex (Tamnes, Fjell, Westlye, Østby, & Walhovd, 2012), poorer brain functioning (Weissman, Roberts, Visscher, & Woldorff, 2006), and poorer dopamine modulation (MacDonald, Karlsson, Rieckmann, & Nyberg, 2012). Although the precise mechanisms of IIV are unclear, these studies support several fundamental neurological underpinnings of IIV.

IIV is also sensitive to variations in cognitive performance as it negatively covaries with cognitive ability in longitudinal studies (i.e., on occasions when IIV is high, cognitive scores are correspondingly low; Bielak, Hultsch, Strauss, MacDonald, & Hunter, 2010a; Lövdén, Li, Shing, & Lindenberger, 2007). Moreover, IIV has been shown to be a strong predictor of maladaptive outcomes, including attrition, MCI status, multiple types of dementia (Bielak, Hultsch, Strauss, MacDonald, & Hunter, 2010b; Kochan et al., 2016), and even impending death (MacDonald, Hultsch, & Dixon, 2008).

Given the maladaptive associations with IIV, the majority of the literature has thus far focused on using IIV as a predictor of cognitive decline. However, two studies have evaluated if it is possible to reduce IIV: one via experimental means, and the other via a mindfulness training intervention. Garrett, MacDonald, and Craik (2012) provided half of their older participants with their own median score after each block of 40 RT trials, and instructed them to focus their attention for the next block to improve their speed and consistency. Participants who received feedback showed reductions in IIV over the four blocks, demonstrating that IIV is indeed modifiable. However, Garrett and colleagues acknowledged that long-term reductions in IIV are unlikely to arise from extensive practice on RT tasks, and that training older adults to be more consistent likely had no effect on neurological functioning. Smart, Segalowitz, Mulligan, Koudys, and Gawryluk (2016) compared the results of an 8-week mindfulness training intervention for adults with subjective cognitive decline with an education control group. Although the sample was small, they found that the mindfulness treatment group showed decreased IIV, demonstrating that change in IIV was possible via a lifestyle change. These findings demonstrate that IIV can be improved, and suggest the possibility of a theoretical shift from using IIV only as a predictor of cognitive decline to an indicator of cognitive and neurological plasticity.

In sum, IIV is a facet of cognitive functioning that is different from traditional cognitive outcomes, and a decrease in IIV as a result of a lifestyle intervention could indicate an improvement in brain integrity. The present article evaluated if change in IIV occurred as a result of participating in the Lifestyle Interventions and Independence for Elders (LIFE) Study, a large, randomized controlled trial (RCT) that asked sedentary older adults to engage in group and individual physical activity for a minimum of 24 months. Previous investigation of this data set found no improvements in cognitive function as a result of the physical activity intervention (Sink et al., 2015). Consequently, the LIFE study is a prime data set to evaluate the hypothesis that IIV may be sensitive to minute cognitive changes from a lifestyle intervention for older adults. We hypothesized that participants who received the exercise condition would show a reduction in IIV compared with participants in the education control condition. Further, we expected to see a variation of this effect by the degree of compliance. Although the intention-to-treat (ITT) protocol is commonly used to evaluate RCTs, this approach includes all participants who completed pre- and post-treatment testing, regardless of actual protocol adherence. Therefore, in addition to using ITT, we analyzed complier average causal effect (CACE) models, which take into account the degree of compliance and can therefore provide information regarding the dose of exposure (Jo, 2002). CACE models compare those who complied with treatment to those in the control group who would have complied if given the opportunity (Stuart, Perry, Le, & Ialongo, 2008). Specifically, we expected that compliers of the protocol would show a reduction in IIV. Finally, because past findings have shown that higher IIV is associated with older age (Bielak et al., 2014) and poorer mental health (Bunce et al., 2008), we controlled for these factors, along with sex, race, education, overall health, center site, and post-test timing.

Method

The LIFE Study was a Phase 3, multicenter single-blinded RCT designed to evaluate the effects of a physical activity program compared with a health education program (Fielding et al., 2011). The primary aim was to reduce major mobility disability, and cognitive functioning was a secondary outcome. Given the secondary nature of the present analyses, our institution’s research ethics committee declared these analyses exempt.

Participants

Participants were recruited via mass mailings to the community, through print, radio, and television advertisements, and also presentations at local senior centers, fairs, and community organizations. Inclusion criteria consisted of (a) being 70–89 years old; (b) living a sedentary lifestyle (<20 min/week of regular physical activity and <125 min/week of moderate physical activity); (c) being at high risk of mobility disability based on lower extremity function (score ≤ 9 on the Short Physical Performance Battery; Guralnik, Ferrucci, Simonsick, Salive, & Wallace, 1995); (d) having the capacity to walk 400 m in less than 15 min; (e) having no major cognitive impairment (score no less than 1.5 SDs below education and race-specific norms on the Modified Mini-Mental State Examinations (3MSE) (Teng & Chui, 1987)); and (f) being able to safely participate in the intervention based on medical history and physical examination. Additionally, those with 9 or more years of education were excluded if their 3MSE score was less than 80 (<76 for black individuals), and those with less than 9 years of education were excluded if their 3MSE score was less than 76 (<70 for black individuals and native Spanish speakers).

Fielding and collegues (2011) described that recruitment of 1,600 participants would provide 80% power to detect the primary study outcome of major mobility disability. The original sample tested at baseline comprised of 1,635 adults, of whom 818 received the activity intervention, and 817 received the education intervention. Of these, 1,394 returned for cognitive testing at Year 2 (695 in the activity group and 699 in the education control group). After removing participants with missing covariate data, the available sample for analysis was 1,329. The average age of the participants was 78.7 years (SD = 5.2), and the majority of participants were women (66.8%). Approximately 96.3% of participants were non-Hispanic, and 75.8% were Caucasian. Table 1 describes further baseline characteristics about the participants (see Sink et al., 2015, for flow of participants through the study). The intervention period varied by individual as those who enrolled earlier in the recruitment period received the intervention for longer. Although the average intervention time across the entire LIFE sample was 2.6 years, participants were only required to participate for a minimum of 24 months.

Table 1.

Demographic Characteristics of the Two Groups

	Physical activity group (n = 662)	Education group (n = 667)
Variable	M (SD)	M (SD)
Age in years	78.6 (5.2)	79.8 (5.3)
Sex (% female)	66.2%	67.5%
Race (% Caucasian)	73.9%	77.7%
Education (% attended college)	67.1%	67.2%
Self-rated overall health (% good or better)	85.3%	83.8%
Depressive symptoms	8.27 (7.64)	8.72 (7.87)
Months between testing periods	21.5 (5.5)	21.1 (5.3)

Interventions

Participants were randomized to either the physical activity intervention or the health education program using a web-based data management system, stratified by field center and sex. The physical activity intervention involved moderate-intensity walking, flexibility and balance exercises, and stretching. Participants were expected to attend two center-based sessions per week, with home-based activity 3–4 times per week over the course of the intervention. The sessions were individualized and progressed toward a goal of 30 min of daily walking at moderate intensity, 10 min of lower extremity strength training using ankle weights, and 10 min of balance training and flexibility exercises. Participants were encouraged to reach a walking goal of 150 min/week, with this goal approached progressively over the first 3 months.

The health education control condition focused on learning about successful aging, and involved 60–90 min workshops on topics relevant to older adults (e.g., navigating the health care system). Participants attended the workshops weekly for 26 weeks, and then monthly for the remainder of the intervention. Topics related to physical activity were avoided. The program also included a 5–10 min instructor-led gentle stretching and flexibility exercises.

Measures

Intraindividual variability

IIV was calculated from the trial latencies on four different RT tasks administered on a computer, and resulted in five different IIV outcomes for analysis. All RT measures were administered at baseline and at either 18 or 30 months, depending on when the participant enrolled in the study. As the Year 2 visit was found to be too lengthy, participants completed the RT tasks for a second time at either 18 months (n = 961) or 30 months (n = 368). Participants were tested at the research clinic during their assessment visits, or at their home if they could not come to the clinic. For all tasks, participants were asked to respond as quickly and accurately as possible.

Flanker

Participants were asked to indicate which direction a central arrow was pointing by pushing one of two keys corresponding to the left or the right. The central arrow was surrounded by arrows pointing the same direction (congruent), or the opposite direction (incongruent). Eight practice trials were followed by 40 congruent trials and 40 incongruent trials administrated in a random order. The congruent and incongruent trials were analyzed as separate outcomes. This task involved selective attention and inhibition of the arrows alongside the central arrow.

N-back

Two versions of this task were given. For both, individual letters were presented one at a time for 2 s. In 1-back, participants were asked to indicate via button press whether the presented letter was the same or different from the letter that appeared immediately before it. Participants were required to get 6 of 10 practice trials correct before completing the test trials. The first trial did not require a response, and the remaining 45 trials were assessed. In 2-back, participants were asked to indicate whether the presented letter was the same or different from the letter that was presented two letters earlier. Participants were required to correctly answer 5 of 10 practice trials before completing the test trials. The first two trials did not require responses, and the remaining 45 trials were recorded. Both n-back tasks involved working memory to recall the previously presented letters.

Task-switching

This task involved alternating between making a judgment about the letter or the digit in the presented letter/digit pair. Participants were shown a four-box grid, and a letter/digit pair was presented inside one of the boxes. When the stimulus was in either of the top two boxes, participants were to indicate whether the number was odd or even by pressing the corresponding key. When the stimulus was in either of the two lower boxes, participants were to indicate whether the letter was a vowel or consonant by pressing the corresponding key. Three practice blocks were administered: (a) 16 trials of only responding to the number judgment in the top half of the grid; (b) 16 trials of only responding to the letter judgment in the bottom half of the grid; and (c) 32 trials of switching between completing the number and letter judgments as determined by the target pair placement in the grid. One hundred nineteen test trials were then presented. This task involved attention and cognitive flexibility in switching between completing the two judgment types.

Covariates

We controlled for the effects of age group, sex, race, education, overall health, field center, post-test evaluation time point, and depressive symptoms. Participants were divided into young–old (70–79.99 years) and old–old (80 years and older) age groupings, consistent with previous LIFE analyses (Sink et al., 2015). Education was assessed with a categorical variable, and divided into participants who had up to 12 years of education, and those who had at least some college education. Overall health was determined from a self-rated health estimate from 1 (excellent) to 5 (poor). Field center accounted for variations between the eight different centers throughout the country where the RCT took place. The post-test evaluation for the RT tasks was completed either at 18 months or 30 months due to a protocol change, and thus included a covariate. Depressive symptoms were evaluated by the CES-D (Radloff, 1977) at the intake testing session, and included as a continuous covariate.

Data Preparation

Calculation of IIV

The calculation procedures were completed separately for each task at each time point, including the congruent and incongruent trials of the flanker task, and the two follow-up assessment time points of 18 and 30 months. Incorrect responses were removed to address the confound of variability due to accuracy, and a lower limit of 150 ms was applied. Upper bounds were set using two methods: At the group level, latencies that exceeded the mean +5 SDs were deleted (except for the n-back tasks where the upper limit was 2,000 ms for a response). At the individual level, means and standard deviations were next calculated for each individual across all trials, and any latencies that exceeded +3 SDs for that individual were deleted. Participants missing data on more than 50% of the trials for the task were removed from the IIV calculation for that wave and task, as imputation accuracy has been shown to be significantly reduced when there is greater than 50% of item-level missingness (Burns et al., 2011). The missing values were imputed using a regression substitution method that forms individualized equations of the RTs across all trials, and uses these equations to predict the missing values (Hultsch et al., 2000). The approach used for data cleaning and missing data imputation was conservative by decreasing within-subject variation and is consistent with other research (Bielak et al., 2014; Bunce et al., 2008; MacDonald et al., 2008).

Following the methodology developed by Hultsch and colleagues (2000), the confounding mean trends of age and trial were removed by regressing the RT data onto categorical age group, categorical trial, and their interactions. The resulting residuals therefore were independent of any systematic within (i.e., trial) and between-subject (i.e., age group differences in mean RT) sources of variance that could influence the SD (e.g., higher mean RT is associated with higher SD; see Hultsch, Strauss, Hunter, & MacDonald, 2008, for detailed justification of this method). Past comparisons have shown that the additional residualization of within-person linear trial effects produces identical IIV values (Bielak et al., 2014). The residuals were transformed to standardized T-scores (M = 50, SD = 10) to enable comparisons across tasks. Each individual’s standard deviation (ISD) across all trials was calculated and used as the indicator of IIV. ISD values were computed for all RT tasks at each assessment.

Calculation of compliance for the physical activity group

We calculated compliance in two different ways to account for the possibility that participants may have had different compliance rates with each portion of the physical activity intervention: center-based attendance, and individual at-home activity.

Participants were required to attend the centers for an exercise session twice per week for a minimum of 24 months, totaling 208 sessions. Attendance was recorded by research staff. A high rate of activity adherence in the pilot of the LIFE Study was defined as attending 70% or more of center-based exercise sessions (Fielding et al., 2007), and 62% of participants in the pilot achieved this adherence. However, only 40.6% of participants attended at least 70% of the center-based exercise sessions in the current data set (i.e., 146 sessions or 1.4 sessions/week). Consequently, we chose to define compliance as participating in 60% or more of required center-based exercise sessions (i.e., 125 sessions or 1.2 sessions/week), which over half the sample achieved (64.3%). We also ran the analyses with 70% compliance to provide a comparison.

Participants were expected to participate in home-based activity multiple times per week over the course of the intervention. Participants were given paper physical activity logs that were to be completed at home and brought to one of their center-based visits each week. The log sheet requested information on daily walking time. Exercises and walking completed while at the center-based sessions were not to be included. The frequency of the expected at-home activity sessions increased over time: Once/week for Weeks 1–4, twice/week for Weeks 5–8, and 3–4/week for the remainder of the intervention period. Participants were encouraged to gradually reach a goal of walking 150 min/week. Using 150 min/week as the required activity frequency, and adjusting for the gradual transition to this goal, we calculated that full compliance with the home-based activity sessions would total 14,700 min of walking over the 2-year period. Only 44.3% of participants managed to reach 70% of this total walking goal (i.e., 10,290 min or approximately 90 min/week). We consequently chose to define compliance in the home-based exercise sessions as 60% or more of the required walking time (i.e., at least 8,820 min or approximately 85 min/week), which slightly over half (52.0%) of the sample achieved. For comparison, we also ran the analyses with 70% compliance.

Statistical Analyses

The data were analyzed with the ITT principle using repeated measures analysis of covariance (RM-ANCOVAs) in SPSS version 25, and with CACE models and sensitivity analyses in Mplus 8 (Muthén & Muthén, 1988–2017). All RM-ANCOVAs and CACE models included sex, race, age group, education, health, depression, and time between pre- and post-test as covariates. The CACE models additionally included baseline performance as a covariate. One RM-ANCOVA was run for each of the cognitive tasks. As such, a Bonferroni-corrected alpha level of p = .01 was used. For each of the five cognitive tasks, two CACE models were run with different compliance variables: one using center-based activity and another using individual activity. Given that a total of 10 CACE models were run for each compliance criteria, a Bonferroni-corrected alpha level of p = .005 was used. As the aim of the study was to examine between-groups effects of physical exercise after potential covariates were accounted for, the following two-step statistical procedure was used when conducting sensitivity analyses. First, a CACE model including both treatment assignment and covariates was run with the exclusion restriction in place, or where the effect of the intervention on the noncompliers is constrained to zero. The exclusion restriction assumes that the treatment effect is zero for noncompliers. This restriction was then relaxed to determine the tenability of this assumption.

Results

The ISD values at baseline and post-test are presented by group in Table 2. Table 2 also shows that the number of participants with valid IIV data varied considerably by RT task. Therefore, the following analyses include different ns for each task, ranging from n = 1,283 for congruent flanker to n = 782 for 2-back.

Table 2.

ISDs for the Five RT Measures at Baseline and Post-Test

	Physical activity group			Health education group
		Baseline	Post-test		Baseline	Post-test
Task	n	M (SD)	M (SD)	n	M (SD)	M (SD)
Congruent flanker	630	5.69 (4.43)	4.61 (4.49)	653	5.98 (4.73)	4.39 (3.76)
Incongruent flanker	622	5.18 (4.44)	4.50 (4.51)	647	5.69 (4.82)	4.32 (3.78)
Task-switching	575	13.45 (6.30)	9.65 (5.07)	566	12.72 (5.96)	9.24 (4.78)
1-back	566	10.16 (2.60)	7.71 (1.94)	577	10.09 (2.66)	7.81 (2.05)
2-back	387	9.40 (1.68)	8.70 (1.61)	395	9.45 (1.53)	8.70 (1.49)

Note. ISD = individual’s standard deviation; RT = reaction task.

ITT Analyses

Of the five RM-ANCOVAs (one per task), only IIV on the incongruent flanker reported a significant Time × Group interaction [F(1, 1,259) = 8.41, p = .004, ${\eta }_p^2$ = .007]. Post hoc paired-samples t-tests showed decreases in IIV in both groups for incongruent flanker [exercise group, t(621) = 3.56, p < .001, Cohen’s d = 0.14] but a larger decline for the education group [t(646) = 8.20, p < .001, Cohen’s d = 0.33]. No main effects of group were observed in any of the RM-ANCOVAs (F = 0.00–4.38, p = .037–.997, ${\eta }_p^2$ = .000–.004). Regardless of group, IIV decreased over time only on the task-switching [F(1, 1,131) = 7.43, p = .007, ${\eta }_p^2$ = .007] and 1-back tasks [F(1, 1,133) = 35.61, p < .001, ${\eta }_p^2$ = .030].

CACE Models

The CACE estimates of the 10 tested models and sensitivity analyses are presented in Tables 3 and 4. The results across the two exercise types (i.e., center-based and home-based) and compliance cutoffs (60% and 70%) were very similar. When the exclusion restriction was applied, the 60% compliance CACE estimates for the compliers showed significant but minimal increases in ISD (i.e., worsened performance) for the congruent and incongruent flanker and task-switching tasks, and no change on the 1-back or 2-back tasks. The sensitivity analyses—conducted by relaxing the exclusion restriction to test its tenability (Jo, 2002)—resulted in similar CACE estimates for the compliers on congruent and incongruent flanker, and task-switching. Compliers also showed significant small increases in ISD for 1-back. CACE estimates remained nonsignificant for noncompliers for both of the flanker ISDs, but significant decreases in ISD were observed in the noncompliers on the task-switching task (for both forms of exercise) and the 1-back task (home-based exercise only). The significant values for noncompliers indicated that these results are sensitive to the exclusion restriction (Stuart et al., 2008). The 70% compliance CACE estimates generally showed similar results: compliers did not improve on any task, and in some cases performed significantly worse post-intervention.

Table 3.

CACE Model and Sensitivity Analysis Results for Center-Based and Home-Based Activity by Each Reaction Time Task at 60% Compliance

		With exclusion restriction	Without exclusion restriction
		Compliers	Compliers	Noncompliers
Exercise type	Task	CACE (SE)	CACE (SE)	CACE (SE)
Center-based	Congruous	0.46 (0.11)**	0.47 (0.11)**	−0.75 (0.71)
	Incongruous	0.51 (0.09)**	0.51 (0.09)**	−0.94 (0.67)
	Switching	0.98 (0.21)**	1.19 (0.19)**	−1.95 (0.51)**
	1-back	−0.31 (0.19)	−0.61 (0.22)	0.62 (0.27)
	2-back	0.04 (0.17)	−0.11 (0.49)	0.28 (0.80)
Home-based	Congruous	0.49 (0.11)**	0.50 (0.11)**	−0.63 (0.57)
	Incongruous	0.51 (0.14)**	0.52 (0.13)**	−0.90 (0.49)
	Switching	1.09 (0.22)**	1.32 (0.25)**	−1.41 (0.46)*
	1-back	−0.34 (0.28)	0.68 (0.17)**	−1.00 (0.22)**
	2-back	−0.04 (0.21)	−0.29 (0.30)	0.35 (0.29)

Note. Values are unstandardized CACE estimates. Negative values indicate reductions in ISD (i.e., improvements in performance). CACE = complier average causal effect; ISD = individual’s standard deviation.

*p < .005. **p < .001.

Table 4.

CACE Model and Sensitivity Analysis Results for Center-Based and Home-Based Activity by Each Reaction Time Task at 70% Compliance

		With exclusion restriction	Without exclusion restriction
		Compliers	Compliers	Noncompliers
Exercise type	Task	CACE (SE)	CACE (SE)	CACE (SE)
Center-based	Congruous	0.46 (0.11)**	0.47 (0.11)**	−0.40 (0.46)
	Incongruous	0.52 (0.11)**	0.52 (0.09)**	−0.64 (0.46)
	Switching	1.17 (0.29)**	1.33 (0.25)**	−0.80 (0.33)
	1-back	−0.60 (0.33)	−1.00 (0.54)	0.35 (0.26)
	2-back	0.06 (0.26)	−0.03 (0.53)	0.08 (0.36)
Home-based	Congruous	0.52 (0.11)**	0.53 (0.11)**	−0.53 (0.50)
	Incongruous	0.47 (0.09)**	−0.70 (0.57)	0.66 (0.14)**
	Switching	−0.85 (0.55)	−1.33 (0.48)	1.02 (0.20)**
	1-back	−0.45 (0.26)	0.82 (0.19)**	−0.87 (0.19)**
	2-back	−0.04 (0.25)	−0.31 (0.38)	0.30 (0.29)

Note. Values are unstandardized CACE estimates. Negative values indicate reductions in ISD (i.e., improvements in performance). CACE = complier average causal effect; ISD = individual’s standard deviation.

*p < .005. **p < .001.

Discussion

The current study aimed to examine the possibility that IIV in older age could be reduced via a 2-year physical exercise intervention. We expected that participants assigned to the physical exercise group would show improvements in cognitive ability (i.e., a reduction in IIV) compared with the education group. Further, we expected the magnitude of IIV reduction to vary by the degree of compliance with the treatment protocol. Neither hypothesis was supported.

Analyses that followed the ITT principle showed only one small interaction effect, with the education group showing a slightly larger decline in IIV over time. The CACE models that considered compliance found either no change or significant but minimal increases in IIV in participants who complied with the physical activity intervention. Moreover, noncompliers (individuals who did less than 60% or 70% of the expected protocol for center- and home-based exercise) showed significant reductions in IIV for two of the RT tasks. Overall, the results were inconsistent with our expectations and prior theory. The fact that noncompliers showed a significant reduction in IIV but the compliers did not demonstrates that the intervention did not have the intended effect on cognition. The LIFE Study was designed to reduce mobility disability, with cognitive improvement as a secondary possible outcome. Unfortunately, our lack of change in IIV as a result of engaging in the LIFE study is consistent with the null results found for the more traditional cognitive outcomes in the study (Sink et al., 2015).

There are a number of possible considerations regarding the unexpected null findings. First, the physical activity intervention, consisting of moderate-intensity walking, flexibility and balance exercises, and stretching, may not have been strenuous enough to cause significant cognitive change. In the program, participants began with light intensity exercise and built up to a goal of 30 min of daily walking. Colcombe and Kramer (2003) found that short sessions of exercise (i.e., less than 30 min) did not have a significant impact on cognitive function, and the analysis by Sink and colleagues using this data did not find cognitive change. Next, it is possible that a greater degree of compliance with the exercise protocol was required to observe a reduction in IIV. However, the results were similar when we repeated the CACE analyses with a 70% compliance cutoff. Additionally strict compliance values were not attempted as few participants adhered to higher levels of the protocol.

Third, it is possible that physical activity may not result in improvements in IIV specifically. Kramer and colleagues (1999) found that tasks measuring executive function showed significant improvements after a 6-month walking exercise program, but nonexecutive tasks, such as those measuring processing speed, were only minimally affected by the intervention. Northey, Cherbuin, Pumpa, Smee, and Rattray (2018) also found that the effects of physical activity on cognition were moderated by the measure of cognition, but notably, significant effects were also evident for attention-based outcomes (but see Young et al., 2015 who failed to find any cognitive benefits of aerobic exercise). However, there is substantial evidence that physical exercise can modify the brain: Batouli and Saba (2017) reviewed 53 studies and found that up to 80% of gray matter is modifiable by physical activity. Given that IIV is believed to be a behavioral indicator of neurological integrity (Hultsch et al., 2000) and has noted neurological correlates (Weissman et al., 2006), it was surprising that change in IIV was not also apparent. Of note, lower IIV was associated with higher scores on traditional cognitive measures (i.e., WAIS, rs −.30 to −.46; verbal fluency, rs −.16 to −.34), faster walking speed (rs −.08 to −.18) and less time to walk 400 m (rs .07 to .12), demonstrating that a reduction in IIV would have been indicative of a positive change in the current sample.

Finally, it is possible that reduction in IIV via a lifestyle intervention among cognitively healthy older adults is limited. Although Smart and colleagues (2016) found that IIV decreased after eight weeks of mindfulness training in older adults, these participants reported subjective cognitive decline. The current LIFE sample was cognitively healthy, but relatively sedentary and with physical limitations. Therefore, a reduction in IIV following a lifestyle intervention may be apparent only for those with cognitive complaints or objective cognitive deficits. Indeed, previous research has found IIV is greater amongst older adults diagnosed with dementia and MCI compared with cognitively healthy older adults (Dixon et al., 2007). Consequently, greater cognitive benefits and thus improvements might be more readily observed in those with compromised cognitive ability.

The present study had many strengths, including using data from a multicenter Phase 3 RCT, a large sample, a substantial intervention period and follow-up, and IIV based on four different RT tasks. The analysis also incorporated CACE modeling, which considered the degree of compliance with the treatment protocol for both center- and home-based exercise. However, the present analyses were limited in that the compliance metrics for the home-based exercise were based on self-report. Additionally, the LIFE study did not calibrate exercise intensity to each participant’s fitness level and change through objective measures (e.g., heart rate monitors) as the intervention progressed, limiting the potential impact of the intervention. Other fitness interventions have found individualized changes in fitness correlate with training-induced neurological change (Kleemeyer et al., 2016, 2017), but we were unable to evaluate this fitness change in relation to change in IIV. There may have also been social factors that prevented participants from fully engaging with the intended activity protocol. For example, some adults may have preferred the center environment and completed relatively few at-home exercise sessions, while other adults may have disliked the large center exercise sessions and chose to complete the majority of their exercise at home. These social factors could have contributed to unknown differences between compliers and noncompliers for the center-based activity, and/or between the center-based and home-based activity compliance rates. It is also worth noting that the noncompliers in the present study showed reductions in IIV on a small number of tasks. These reductions were likely due to practice effects rather than being indicative of true change, but it is unclear why the compliers did not also exhibit similar change due to practice.

The current investigation is one of the few research studies to focus on using IIV as a metric of cognitive plasticity rather than cognitive decline, and the first to evaluate if IIV can be reduced via a lifestyle intervention for cognitively healthy older adults. Although we did not find that participants who complied with a 2-year exercise program showed any reductions in IIV, a plethora of research has demonstrated that IIV is a sensitive measure of neurological and cognitive decline (Kochan et al., 2016; MacDonald et al., 2009). Therefore, it remains possible that reductions in IIV may be evident from other types of interventions, either with more vigorous physical exercise, or those that primarily involve social or cognitive engagement. Rather than implementing entirely new interventions with modified protocol and techniques to hopefully induce cognitive change using traditional cognitive outcomes, the use of novel cognitive indices such as IIV may be a valuable approach.

Funding

This work was supported by the National Institutes of Health (NIH; R03AG055748 and R03AG055748-01S1 to A. A. M. Bielak). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. The Lifestyle Interventions and Independence for Elders Study was funded by the NIH (UO1AG22376 and 3U01AG022376-05A2S), and sponsored in part by the Intramural Research Program, NIH.

Conflict of Interest

None reported.