Secondary Logo

Journal Logo

Aging

Late-life Cognitive Activity and Dementia

A Systematic Review and Bias Analysis

Sajeev, Gautam; Weuve, Jennifer; Jackson, John W.; VanderWeele, Tyler J.; Bennett, David A.; Grodstein, Francine; Blacker, Deborah

Author Information
doi: 10.1097/EDE.0000000000000513

Abstract

An estimated 47 million people worldwide have dementia in 2015, and this number is projected to triple by 2050.1 Thus, the search for modifiable preventive factors is becoming increasingly urgent. One such potentially modifiable factor is cognitive activity, which is thought to contribute to building and maintaining brain structure and function,2,3 sometimes viewed as analogous to the contribution of physical exercise to cardiopulmonary and neuromuscular structure and function.

A variety of websites and publications encourage older adults to be mentally active (“use it or lose it!”) to protect brain health. Although epidemiologic studies often show reduced risk of dementia with late-life cognitive activity, concerns about residual confounding and reverse causation cast doubt on these findings. Several characteristics and behaviors linked with greater late-life cognitive activity, such as greater formal education, higher socioeconomic status (SES), and better general health, also appear to be associated with reduced dementia risk, suggesting that residual or unmeasured confounding may influence study findings.4 Furthermore, falloff in cognitive activity in late life may be a consequence of cognitive losses before dementia onset, raising the possibility that reverse causation also contributes to the observed inverse associations.

To investigate these issues, we conducted a systematic review of epidemiologic studies of cognitive activity and incidence of Alzheimer’s disease dementia (AD) and other dementias, characterized and quantified the potential bias from confounding and reverse causation, and identified additional methodologic issues relevant to the interpretation of these studies.

METHODS

Systematic Search Strategy

We conducted a systematic search of PubMed and EMBASE through June 2014 to identify relevant studies. To build our search strategy, we first obtained lists of index terms relevant to cognitive activity, AD, and dementia, and nested case–control and cohort studies using the PubMed Medical Subject Headings (MeSH) database and the EMBASE EMTREE thesaurus. We also identified common synonyms for cognitive activity from primary and review articles and incorporated these terms into our search strategy (eTables 1 and 2; https://links.lww.com/EDE/B61).

Selection of Studies

To be eligible for inclusion, studies had to be published in a peer-reviewed journal, be cohort or nested case–control studies in well-defined cohorts, have a clearly described operationalization of cognitive activity, evaluate cognitive activity prospectively in relation to dementia, evaluate all participants (or a systematically drawn sample) for AD or dementia using established diagnostic criteria, provide effect estimates with confidence intervals (CIs) or standard errors for the association between cognitive activity and AD or dementia, and adjust for at least age and sex. This review was conducted in part for the database of AD epidemiology findings AlzRisk (www.alzrisk.org), which uses these prespecified criteria and complies with current standards for systematic review and meta-analyses of observational studies.5,6

Data Extraction

For each eligible study, we extracted the following: publication year; study cohort; study design; number of participants; baseline age distribution; follow-up time; method of ascertainment and operationalization of cognitive activity; number of AD and dementia cases; effect estimates, with CIs or standard errors; and model covariates. Due to the heterogeneity of the exposure definitions used, we did not compute meta-analysis summary estimates.

Bias Analysis Method

Unmeasured confounding is likely present in all observational studies; more important is the degree to which such confounding biases estimated effects. We conducted a bias analysis7 to evaluate how reported hazard ratios (HR) for the effect on AD of “high” versus “low” levels of cognitive activity—obtained from one representative study8—would change under different assumptions about the magnitude of unmeasured confounding. As incipient dementia may be a shared cause of reduced cognitive activity and clinical dementia, we used this same approach to quantify how the observed association between cognitive activity and AD would be influenced by reverse causation.

We used a formal quantitative approach applicable to the case of a binary exposure, a hypothetical unmeasured binary confounder U, and a rare time-to-event outcome.7 It can be shown that the degree of confounding by U is a function of (1) the HR reflecting the association of U with AD, HRU-AD, conditional on exposure and measured covariates; (2) the prevalence of U in the high-activity group, p1, conditional on measured covariates; and (3) the prevalence of U in the low-activity group, p0, conditional on measured covariates. To quantify how confounding by U would change an observed estimate reflecting the effect of cognitive activity on AD (HRCA-AD), we calculated a bias-corrected estimate (HRCA-AD|U), the estimated effect of cognitive activity on AD that would have been obtained with adjustment for the hypothetical confounder U (in addition to the measured covariates). This bias-corrected estimate is a function of the observed estimate of the effect of cognitive activity on AD, and the three parameters defining the degree of confounding and is given by HRCA-AD|U = HRCA-AD/{[1 + (HRU-AD – 1)p1]/[1 + (HRU-AD – 1)p0]}. This approach entails an additional simplifying assumption of no interaction between the effects of U and cognitive activity on the HR scale. The impact of unmeasured confounding by U can be assessed in two ways. The first approach compares the observed estimate with the bias-corrected estimate under different specifications of U, and asks “given HRCA-AD, and specifying HRU-AD, p0 and p1, what would HRCA-AD|U be?” The second approach calculates how much confounding would be required for the observed association to be eliminated entirely, and asks “given HRCA-AD, what values of HRU-AD, p0 and p1 would make HRCA-AD|U become 1?”

To use this approach, we needed an effect estimate from a study that reported cognitive activity as a categorical exposure in relation to the time-to-event outcome of AD, and in which AD occurred relatively rarely (in general, this approach provides correct HR estimates when the disease occurs in up to 10% of both exposure groups in the study population). Among the cohort studies in our review, these conditions were most adequately met in Akbaraly et al.,8 in which approximately 2% of the study population developed AD over a 4-year follow-up period. (AD incidence was in the range of 10%–15% over the follow-up periods of the other studies, which would lead to only minor violations of the rare disease assumption.) Notably, Akbarly et al.8 was similar to the other studies with respect to the types of cognitive activities assessed (eTable 3; https://links.lww.com/EDE/B61), the operationalization of cognitive activity as an exposure and the estimated magnitude of the association between cognitive activity and dementia (Table 1).

TABLE 1
TABLE 1:
Summary of Results: Cognitive Activity and Incidence of AD and All-cause Dementia

Akbaraly et al.8 compared AD risks corresponding to what the authors defined as “high” versus “low” (reference) engagement in both “stimulating” (HRCA-AD = 0.39 [95% CI: 0.21, 0.71]) and “passive” (HRCA-AD = 0.68 [0.41, 1.13]) activity. We estimated how much confounding would be needed to increase these point estimates to 1. To be more conservative in our bias assessment, we also performed similar calculations for the upper confidence limit of the HR for stimulating activity. We did not perform this calculation for the upper confidence limit for the passive activity HR as it already exceeded 1.

Application to Unmeasured Confounding

For our bias analysis, we posited the presence of a binary confounder U to represent an unmeasured risk factor with either a small (HRU-AD = 1.5), moderate (HRU-AD = 2), or large harmful effect on AD (HRU-AD = 3). These three values encompass a range of effect estimates (risk, odds, or hazard ratio) reported in studies of individual or multiple cardiovascular risk factors in relation to AD incidence.15,16 Depression,15 low education,15 and lifecourse socioeconomic position17,18 are other potential confounders of the cognitive activity–AD association with effect estimates in this range. The studies we reviewed adjusted for confounding by these variables to varying degrees, so the specific unmeasured confounders represented by U, and the degree of residual confounding will differ across individual studies. We used these values of HRU-AD, and allowed p0 and p1 to vary between 0 and 1 in the formula above, to calculate how the HRCA-AD estimates from Akbaraly et al.8 would change under different amounts of confounding.

Application to Reverse Causation

The magnitude of bias due to reverse causation will depend on the degree to which individuals with mild cognitive impairment (or potential incipient dementia more broadly, including subjective cognitive concerns,19) engage in less cognitive activity (p1p0), and on the magnitude of their increased risk of developing clinical dementia (HRU-AD). Such individuals are not excluded at baseline in most studies of the effect of cognitive activity on AD, and may account for a substantial portion of the study population; in three cohorts reviewed here, individuals with mild cognitive impairment made up 19%, 26%, and 39% of all nondemented individuals.8,20,21 Indications of incipient dementia in cognitively normal individuals are strongly associated with eventual AD or dementia diagnosis, with hazard or odds ratio (OR) estimates ranging from 2.8 to 6.7.19,20,22–24 On the basis of these reports, we used an HR of 4.5 for the effect of incipient dementia on AD (HRU-AD = 4.5), and again calculated how reverse causation would change the HRCA-AD estimates from Akbaraly et al.8

RESULTS

Literature Search Results

Using our search strategy, we identified 926 citations. After removing 238 duplicate citations, and reviewing article titles and abstracts against our inclusion criteria, we identified 56 citations for full-text review. Of these, we excluded articles not in English. Twelve of the remaining English-language article met criteria for inclusion in our review (Fig.).

FIGURE
FIGURE:
Flow chart showing how studies were selected for inclusion.

Study Design Details

The 12 studies we reviewed included a total of 13,939 participants from 11 cohorts8–14,21,25–28 (Table 1; eTable 4; https://links.lww.com/EDE/B61). Two studies analyzed the same participants from the Kungsholmen cohort; we included both, as they examined different aspects of cognitive activity.12,26 Of these 12 studies, 10 were prospective cohort studies including a total of 13,431 participants8–12,21,25–28 and two were nested case–control studies.13,14 The prospective studies assessed cognitive activity at a baseline visit late in life. Almost all studies inquired only about current or habitual participation in leisure activities; one additionally inquired about activity at ages 6, 12, 18, and 40.21 The two nested case–control studies assessed cognitive activity earlier in life; one inquired about current mid-life activity (at mean age 45),13 while the other asked at a mean age of 57 about activity before age 40.14

In the 10 prospective cohort studies, follow-up for dementia began immediately after the baseline visit. Six studies reported mean follow-up time (range, 2.5–6.1 years). In the two nested case–control studies with mid-life measures of cognitive activity, dementia follow-up began approximately 20 years after assessment of cognitive activity. All but one of our reviewed studies adjusted for education. Four studies additionally adjusted for SES, or a marker of SES such as occupation or income.

Dementia Assessment

Most studies used multistage assessments for dementia, consisting of initial cognitive screening followed by more detailed clinical examinations under a standard protocol. Two studies conducted full annual examinations on all participants.21,28 Dementia and AD were diagnosed by standard clinical and research criteria.29–32 Diagnoses were available for AD alone in three studies,21,27,28 for both AD and all-cause dementia in three studies,8,11,14 and for all-cause dementia alone in six studies.9,10,12,13,25,26

Definition, Dimensions, and Operationalization of Cognitive Activity

In all studies, popular leisure activities considered to require information seeking and processing were characterized as “cognitive.” Most studies assessed participation in a relatively narrow range of leisure activities, such as reading books, newspapers, or magazines; doing crosswords or playing cards; and watching television and listening to the radio. However, a few studies also inquired about activities such as participation in group discussions and attending cultural and social events (eTable 3; https://links.lww.com/EDE/B61).

The ways that studies operationalized cognitive activity varied widely. One study categorized activity based on frequency of participation across all cognitive activities, and did not distinguish different frequencies of participation in the specific activities assessed.12 Most commonly, investigators started with reported frequency of participation in each activity, assigned a score corresponding to each frequency, and then combined scores across all activities into a composite score.8,9,11,21,27,28 Three studies used the total number of reported cognitive activities10,13,25; one study used time devoted to cognitive activities25; and one used a cognitive factor index score derived from an exploratory factor analysis of a mixture of leisure items.14 A few studies also assessed the intensity of cognitive activity and further subclassified activities as stimulating or passive (e.g., reading vs. watching television), but these terms were not used consistently across studies.8,13,26

Summary of Review Findings

The 12 reviewed studies included a total of 1,663 dementia cases of which 565 were specifically AD. Most studies found inverse associations of late-life cognitive activity with AD and/or all-cause dementia (Table 1). In the six studies that operationalized cognitive activity using composite measures of participation frequency, greater participation was generally associated with lower incidence of AD and all-cause dementia.8,9,11,21,27,28 Similarly, in three studies that examined the number of cognitive activities, higher activity count corresponded to lower dementia incidence.10,13,25 In one study, time spent on cognitively engaging hobbies was also related to a lower dementia rate.25 Among investigations of the intensity of cognitive activity, two found lower dementia incidence with engagement in stimulating but not passive activity,8,26 whereas another found the opposite.13

Bias Analysis Results

Bias Due to Confounding

Table 2 shows how an observed HRCA-AD of 0.39 would change under our bias model if adjustment were also made for a strong unmeasured confounder, U (HRU-AD = 3), calculated under different scenarios for the prevalence of U among the active and inactive groups. If U were more prevalent among the cognitively active, the observed HRCA-AD of 0.39 would actually underestimate the true protective effect (HRCA-AD|U < 0.39). Under the more likely scenario that a harmful U is more prevalent among the cognitively inactive, the observed HRCA-AD of 0.39 would overestimate the protective effect (HRCA-AD|U > 0.39). In these instances, the estimated protective effect would diminish with adjustment for U, but it would be eliminated entirely or reversed only if there were extreme imbalances in the prevalence of U between activity groups. One such scenario would require the prevalence of U to be 81% in the low-activity group when its prevalence was 1% in the high-activity group. Prevalences of U in the low-activity group greater than 81% would make HRCA-AD exceed 1 in U-adjusted analyses, reversing the observed inverse association. Increasing the observed upper 95% confidence limit for this estimate from 0.71 to 1 would require smaller but still quite substantial prevalence differences; for U prevalences in the high-activity group of 1%, 10%, 25%, and 50%, the corresponding prevalences required in the low-activity group would be 22%, 35%, 56%, and 91%, respectively (eTable 5; https://links.lww.com/EDE/B61). If U were only moderately (HRU-AD = 2) or weakly (HRU-AD = 1.5) associated with AD, an observed HRCA-AD of 0.39 would be weakened under U adjustment but would not be not nullified or reversed under any combination of U prevalences in the high- and low-activity groups.

TABLE 2
TABLE 2:
Bias-corrected AD HR (Comparing High vs. Low Participation in Cognitive Activity), Adjusted for U, When HRU-AD = 3 or HRU-AD = 4.5, Given that Hazard Ratio Unadjusted for U = 0.39

Similarly, eTable 6 (https://links.lww.com/EDE/B61) shows how an observed HRCA-AD of 0.68 would change under this bias model if adjustment were made for a strong confounder U (HRU-AD = 3). When the prevalence of U in the high-activity group is 1%, a prevalence of U in the low-activity group of 25% would be required to eliminate an observed association of this magnitude. For U prevalences in the high-activity group of 10%, 25%, and 50%, the corresponding prevalences required in the low-activity group to eliminate the observed association would be 39%, 61%, and 98%, respectively. If U were more weakly associated with AD (e.g., HRU-AD = 2 or HRU-AD = 1.5), larger prevalence differences than these would be required to eliminate an observed HRCA-AD of 0.68 entirely.

Bias Due to Reverse Causation

Table 2 and eTable 6 (https://links.lww.com/EDE/B61) also show how estimated HRCA-AD of 0.39 and 0.68 could reflect the presence of incipient dementia as specified (HRU-AD = 4.5), under a range of incipient dementia prevalences in the active and inactive groups specified in our bias model. If the incipient dementia prevalence was only 1% in the active group, then a prevalence of 48% in the inactive group would be required for an observed HRCA-AD of 0.39 to increase to 1 upon adjustment for incipient dementia. An incipient dementia prevalence of 10% in the high-activity group would require an incipient dementia prevalence of 71% in the low-activity group for an HRCA-AD of 0.39 to increase to 1 (Table 2).

Under our bias model, smaller prevalence differences are required to increase an HRCA-AD of 0.68 to 1 (eTable 6; https://links.lww.com/EDE/B61). Calculating similarly, for incipient dementia prevalences in the high-activity group of 1%, 10%, 25%, or 50%, the corresponding prevalences required in the low-activity group to eliminate the observed association, upon adjustment, would be 15%, 28%, 50%, and 87% respectively. Similar prevalence differences would be required to nullify a HR of 0.71, the upper 95% confidence bound of the HRCA-AD point estimate of 0.39 (eTable 5; https://links.lww.com/EDE/B61).

DISCUSSION

These epidemiologic studies generally reported associations of greater participation in late-life cognitive activities with lower risk of both AD and all-cause dementia. We consider the implications of our bias analysis on the extent to which these findings might result from confounding or reverse causation, and issues relevant to the definition, ascertainment and operationalization of cognitive activity that might also have led to biased results.

Bias Due to Confounding and Reverse Causation

Confounding by Shared Causes of Cognitive Activity and Dementia

Characteristics associated with higher cognitive activity such as formal education and higher SES have also been consistently associated with reduced dementia risk.33,34 Almost all studies adjusted their estimates for formal education, and a few additionally adjusted for occupational history,8 or income and indicators of early-life SES,21 with generally unchanged results. However, some concern about residual confounding remains warranted, given the strong associations between different dimensions of SES and dementia risk,33,34 and because education and SES are often measured and modeled inadequately.4 Given the correlation between higher SES and lower cardiovascular risk, adjustment for cardiovascular risk factors may also help reduce residual confounding by SES. Another potential source of confounding is longstanding intellectual ability, which is likely correlated with cognitive activity,35 education, and a variety of dementia risk factors; in one study, adjustment for intelligence quotient at age 11 mostly eliminated the positive associations between late-life cognitive activity and late-life cognitive performance.35 In addition, as some of the cognitive activities frequently assessed in these studies also involve social engagement or the ability to travel (e.g., attending group activities, visiting the cinema/theatre), failure to adjust for general health status (or a proxy such as physical or social activity), which is also related to dementia,36 may lead to an overestimated influence of cognitive activity on dementia risk.

However, under the bias model used in our analyses, the bias from residual confounding by these variables is unlikely to be large enough to completely explain inverse associations of the magnitude observed in the reviewed studies. Our calculations indicate that to eliminate protective HRs of 0.68 and 0.39, a strong harmful confounder, U, would need to be at minimum 24 and 80 percentage points more prevalent, respectively, among inactive individuals. Such large prevalence differences are indicative of very strong associations between U and cognitive activity; a 24 percentage point greater prevalence of U among inactive individuals would correspond to, at minimum, a U cognitive activity OR of 2.66 (when p1 = 0.38 and p0 = 0.62). Imbalances between activity groups of this size are also larger than those reported for important confounders in other studies. In Akbaraly et al.,8 the largest imbalance across activity groups reported for any measured confounder was 17 percentage points for low education, and imbalances of just 2-, 5-, and 6-percentage points were reported for diabetes, hypertension, and vascular disease, respectively.8 In a different study of cognitive function in which the size of similar confounder-cognitive activity relationships was reported, the prevalence imbalances were similarly modest.37 In addition, our bias analysis suggests that even the bias-corrected upper confidence limit for this estimate would remain under 1. Arguably, correcting the upper CI from a single study may be overly conservative; a hypothetical pooled estimate based on several studies would have a smaller upper confidence limit than the one used in this calculation.

Given the interrelationships between SES markers such as income and occupational attainment, as well as the association of these variables with cardiovascular risk factors and health status, confounding by collective effects of multiple unadjusted or incompletely adjusted SES-related variables may be of concern. While our simple analyses cannot quantify such complex biases we would expect that adjustment for education, which was done in almost all studies, would capture much of the confounding attributable to SES or related variables. In addition, correlations among these imperfectly adjusted variables might further decrease the expected bias.

Reverse Causation

Individuals in the prolonged preclinical stages of dementia might be more likely to eschew participation in cognitive activities than their healthy counterparts,21,38 resulting in an inverse association between activity and dementia risk. In most reviewed studies, analyses excluding individuals judged more likely to be in the prodromal stages of dementia (e.g., those with a mild cognitive impairment diagnosis, or poorer screening test performance, or dementia diagnosis early in the follow-up period) returned largely unchanged results. However, such exclusions may be insufficient, considering that average follow-up time in the late-life cognitive activity studies was between 2 and 7 years, whereas detectable cognitive decline begins as early as 5–8 years before dementia diagnosis,39–41 and even more subtle impairments, perhaps noticeable only to affected individuals themselves, occur still earlier.42,43 If some of those with incident dementia in these studies were impaired enough at baseline to affect their cognitive activity, it is likely that the observed relationship at least partially reflects reverse causation.

We also analyzed how robust the observed associations might be to such reverse causation observing that associations of incipient dementia with both late-life activity and AD diagnosis would need to be large to completely explain findings of the magnitude observed. The magnitude of the effect of incipient dementia will differ between studies based on their average duration. We used an HR (4.5) from the middle of the range of reported estimated effects of indications of incipient dementia on risk for clinical dementia.19,20,22–24 Smaller HRs for the effect of incipient dementia yield results similar to those described for confounding, and would be unlikely to entirely explain the observed associations. However, under our bias model, with larger HRs for the effect of incipient dementia, smaller prevalence differences would be sufficient to fully explain an HRCA-AD of 0.68, attenuate an HRCA-AD of 0.39 and increase its upper confidence bound from 0.71 to 1. Overall, contingent on the validity of our model, it is plausible that reverse causation could bias these findings; the degree of bias could be substantial if the disparity in incipient dementia prevalence between high- and low-activity groups is large.

However, quantifying the degree to which incipient dementia is more prevalent among low- rather than high-activity individuals is difficult. In Akbaraly et al.,8 cognitive impairment at baseline, defined as scoring less than 24 on the mini-mental state examination, was more common among those with low (7.8%) rather than high (1.9%) levels of stimulating activity. Data on how late-life cognitive activity changes with declining cognition are sparse. In Wilson et al.,21 baseline mild cognitive impairment was associated with lower cognitive activity at baseline, although not with decline in activity over a 3-year follow-up. The same investigators also showed that whereas cognitive activity predicted global cognitive function over the following year, global cognitive function did not predict subsequent degree of cognitive activity. However, performance in two specific cognitive domains, working memory and perceptual speed, was associated with decline in cognitive activity over the following year.44 To better understand the influence of reverse causation, longitudinal data on the extent and timing of cognitive activity changes will have to be evaluated in parallel with changes in cognition in longer prospective follow-up studies.

While the calculations reported here provide some measure of the potential degree of bias for a typical effect estimate, analyses using study-specific parameters will be better able to assess the degree of bias in individual studies. Nonetheless, our bias analyses can be informative in interpreting other studies in which the conditions required for use of this approach are met. These conditions are unlikely to be severely violated in the cohort studies reviewed here, where AD incidence over the relatively short study follow-up periods remains quite low. In addition, the formula we used to calculate a bias-corrected HR applies analogously to ORs when the assumptions specified are met.7,45 Overall, under the specifications of our bias analysis and conditional on the validity of its assumptions, we found that the observed inverse associations are likely robust to unmeasured confounding, and likely only partially explained by reverse causation.

Methodologic Considerations in Measurement of Cognitive Activity

Type and Context

Although cognitive activity scales are being developed,46 further investigation into valid measurement of the construct is warranted. In an effort to isolate the effect of purely cognitive activities,46 most studies focused on leisure activities that are primarily or exclusively cognitive. Some of the more frequently assessed leisure activities (e.g., reading, attending theater) are more likely to be favored by individuals of higher SES or educational attainment. Failure to assess other leisure or nonleisure activities requiring cognitive processing (e.g., home repair, gambling) would result in underestimating overall late-life cognitive activity for many individuals, and could bias estimates of its effect on dementia risk depending on the assessed activities’ associations with higher SES/education.

Self-reporting

Valid measurement of cognitive activity is particularly challenging in populations that may have cognitive deficits. All reviewed studies used self-reported data, but validation against other sources (e.g., diary records, electronic devices) was rare even though individuals with memory loss would be expected to self-report less accurately.47,48 If individuals with mild cognitive impairment or subjective cognitive concerns systematically under-report cognitive activity, lower participation would be spuriously linked with greater dementia risk. Alternatively, if participation frequency is overstated because of failure to accurately recognize or recall declines from lifetime activity levels, this would attenuate any protective association.

Timing and “Dose”

A critical question from a prevention perspective is whether the inverse association between late-life cognitive activity and dementia can be attributed specifically to greater late-life cognitive activity, which would suggest that even cognitively “sedentary” individuals might benefit from increasing their cognitive activity late in life. Two studies in the same cohort concluded that the effect of cognitive activity depends on late-life activity because even after adjusting for activity earlier in life, late-life activity was strongly associated with lower AD incidence.21,49 Similarly, more detailed investigation of the “dose” of cognitive activity associated with lower AD risk will be required to inform prevention guidelines. Questions about timing and dose of activity are germane to possible benefits of intervening throughout the lifecourse, such as recommending engagement in specific types of activities in midlife and possibly even policies and practices that shape childhood intellectual development.

Cognitive Training Versus Cognitive Activity

These epidemiologic findings, coupled with the hypothesized neuroprotective effects of cognitive activity, have inspired interest in cognitive training as a means to enhance late-life cognitive function. The largest trial to date, the Advanced Cognitive Training for Independent and Vital Elderly (ACTIVE) study, compared three group-based behavioral interventions in 2,786 cognitively normal, independently living adults (mean age 74). The interventions individually targeted memory, reasoning, and processing speed, and were compared with a no-contact regime.50 Post training improvements were maintained for at least 5 years in all three arms.51–53 In addition, self-reported difficulty performing instrumental activities of daily living was lower in the training arms than among controls from year 2 onwards. However, the training interventions had no effects on performance-based measures of neurocognitive function at any of the follow-up points,52 and a secondary analysis including 189 incident dementia cases found no difference between trained and untrained groups in the rate of dementia occurrence over 5 years of follow-up.54 Expanding training to simultaneously target multiple cognitive domains, lengthening intervention time, and targeting an older-old population, have been suggested as modifications that may yield treatment effects on dementia incidence, but such strategies remain untested.

Although evidence from randomized trials so far does not support the efficacy of specific cognitive training interventions in preventing dementia, this does not rule out potential benefits of cognitively active lifestyles. The short-term influence of prescribed training interventions may differ from the cumulative stimulation attained via habitual performance of personally selected cognitive leisure activities. Interventions in line with a way of life that more fully incorporates cognitive activity as practiced in observational studies are difficult to define and operationalize in randomized trials. Evidence for any salutary effects of this broader notion of “engagement” (social, cognitive, and otherwise) on late-life cognition may be better assessed by trials of more holistic interventions.55

CONCLUSION

Our systematic review and bias analyses suggest that late-life cognitive activity may offer some reduction in risk of AD and all-cause dementia. Our bias analyses appraised these epidemiologic findings with respect to two critical potential sources of bias—confounding and reverse causation—by quantifying the potential degree of these biases for a typical effect estimate. Under the parameters specified in our bias analysis, the observed inverse associations are unlikely to be explained entirely by unmeasured confounding, but reverse causation remains a more plausible but still little understood source of bias. Although a number of limitations in the measurement of cognitive activity might have led to bias, they do not appear sufficient to account for the observed findings. It remains possible, however, that confounding, reverse causation, and measurement limitations could combine to substantially bias estimates. As with any bias analysis, our conclusions are conditional on the validity of our bias model.

Over time, better characterization of the type, duration, intensity, and timing of activity56,57 associated with late-life cognitive benefit will be required to develop more specific recommendations applicable over the lifecourse. Observational studies like those reviewed here can shed light on benefits associated with the performance of habitual, self-selected cognitive activities. Interventions under development extend this work by targeting cognitive activity and other modifiable dementia risk factors simultaneously.55,58 This type of multimodal approach was used in the Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability, in which the intervention group received diet, exercise, and vascular risk monitoring regimens in addition to cognitive training.58 Another interventional program, Experience Corps, aims to increase older adults’ cognitive, physical, and social activity by giving them meaningful volunteer roles in public elementary schools.55 Evidence from these different types of approaches will be critical to help us better understand the relationship of cognitive activity with late-life cognition.

REFERENCES

1. World Alzheimer Report 2015: The Global Impact of Dementia. 2015.Alzheimer’s Disease International.
2. Park DC, Bischof GN. The aging mind: neuroplasticity in response to cognitive training. Dialogues Clin Neurosci. 2013;15:109–119.
3. Stern Y. Cognitive reserve in ageing and Alzheimer’s disease. Lancet Neurol. 2012;11:1006–1012.
4. Kaufman JS, Cooper RS, McGee DL. Socioeconomic status and health in blacks and whites: the problem of residual confounding and the resiliency of race. Epidemiology. 1997;8:621–628.
5. Stroup DF, Berlin JA, Morton SC, et al. Meta-analysis of observational studies in epidemiology: a proposal for reporting. Meta-analysis of observational studies in epidemiology (MOOSE) group. JAMA. 2000;283:2008–2012.
6. Systematic reviews: CRD’s guidance for undertaking reviews in health care. 2009.Centre for Reviews and Dissemination.
7. VanderWeele TJ. Unmeasured confounding and hazard scales: sensitivity analysis for total, direct, and indirect effects. Eur J Epidemiol. 2013;28:113–117.
8. Akbaraly TN, Portet F, Fustinoni S, et al. Leisure activities and the risk of dementia in the elderly: results from the three-city study. Neurology. 2009;73:854–861.
9. Sörman DE, Sundström A, Rönnlund M, Adolfsson R, Nilsson LG. Leisure activity in old age and risk of dementia: a 15-year prospective study. J Gerontol B Psychol Sci Soc Sci. 2014;69:493–501.
10. Scarmeas N, Levy G, Tang MX, Manly J, Stern Y. Influence of leisure activity on the incidence of Alzheimer’s disease. Neurology. 2001;57:2236–2242.
11. Verghese J, Lipton RB, Katz MJ, et al. Leisure activities and the risk of dementia in the elderly. N Engl J Med. 2003;348:2508–2516.
12. Wang HX, Karp A, Winblad B, Fratiglioni L. Late-life engagement in social and leisure activities is associated with a decreased risk of dementia: a longitudinal study from the Kungsholmen project. Am J Epidemiol. 2002;155:1081–1087.
13. Carlson MC, Helms MJ, Steffens DC, Burke JR, Potter GG, Plassman BL. Midlife activity predicts risk of dementia in older male twin pairs. Alzheimers Dement. 2008;4:324–331.
14. Crowe M, Andel R, Pedersen NL, Johansson B, Gatz M. Does participation in leisure activities lead to reduced risk of Alzheimer’s disease? A prospective study of Swedish twins. J Gerontol B Psychol Sci Soc Sci. 2003;58:P249–P255.
15. Barnes DE, Yaffe K. The projected effect of risk factor reduction on Alzheimer’s disease prevalence. Lancet Neurol. 2011;10:819–828.
16. Luchsinger JA, Reitz C, Honig LS, Tang MX, Shea S, Mayeux R. Aggregation of vascular risk factors and risk of incident Alzheimer disease. Neurology. 2005;65:545–551.
17. Karp A, Kåreholt I, Qiu C, Bellander T, Winblad B, Fratiglioni L. Relation of education and occupation-based socioeconomic status to incident Alzheimer’s disease. Am J Epidemiol. 2004;159:175–183.
18. Zeki Al Hazzouri A, Haan MN, Kalbfleisch JD, Galea S, Lisabeth LD, Aiello AE. Life-course socioeconomic position and incidence of dementia and cognitive impairment without dementia in older Mexican Americans: results from the Sacramento area Latino study on aging. Am J Epidemiol. 2011;173:1148–1158.
19. Reisberg B, Shulman MB, Torossian C, Leng L, Zhu W. Outcome over seven years of healthy adults with and without subjective cognitive impairment. Alzheimers Dement. 2010;6:11–24.
20. Bennett DA, Wilson RS, Schneider JA, et al. Natural history of mild cognitive impairment in older persons. Neurology. 2002;59:198–205.
21. Wilson RS, Scherr PA, Schneider JA, Tang Y, Bennett DA. Relation of cognitive activity to risk of developing Alzheimer disease. Neurology. 2007;69:1911–1920.
22. Boyle PA, Wilson RS, Aggarwal NT, Tang Y, Bennett DA. Mild cognitive impairment: risk of Alzheimer disease and rate of cognitive decline. Neurology. 2006;67:441–445.
23. Jessen F, Wiese B, Bachmann C, et al.; German Study on Aging, Cognition and Dementia in Primary Care Patients Study Group. Prediction of dementia by subjective memory impairment: effects of severity and temporal association with cognitive impairment. Arch Gen Psychiatry. 2010;67:414–422.
24. Manly JJ, Tang MX, Schupf N, Stern Y, Vonsattel JP, Mayeux R. Frequency and course of mild cognitive impairment in a multiethnic community. Ann Neurol. 2008;63:494–506.
25. Hughes TF, Chang CC, Vander Bilt J, Ganguli M. Engagement in reading and hobbies and risk of incident dementia: the MoVIES project. Am J Alzheimers Dis Other Demen. 2010;25:432–438.
26. Karp A, Paillard-Borg S, Wang HX, Silverstein M, Winblad B, Fratiglioni L. Mental, physical and social components in leisure activities equally contribute to decrease dementia risk. Dement Geriatr Cogn Disord. 2006;21:65–73.
27. Wilson RS, Bennett DA, Bienias JL, et al. Cognitive activity and incident AD in a population-based sample of older persons. Neurology. 2002;59:1910–1914.
28. Wilson RS, Mendes De Leon CF, Barnes LL, et al. Participation in cognitively stimulating activities and risk of incident Alzheimer disease. JAMA. 2002;287:742–748.
29. Hughes CP, Berg L, Danziger WL, Coben LA, Martin RL. A new clinical scale for the staging of dementia. Br J Psychiatry. 1982;140:566–572.
30. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM. Clinical diagnosis of Alzheimer’s disease: report of the NINCDS-ADRDA work group under the auspices of department of health and human services task force on Alzheimer’s disease. Neurology. 1984;34:939–944.
31. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, 3rd ed. rev.: DSM-III-R. 1987.Washington, DC: American Psychiatric Association.
32. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders: DSM-IV. 1994.Washington, DC: American Psychiatric Association.
33. Meng X, D’Arcy C. Education and dementia in the context of the cognitive reserve hypothesis: a systematic review with meta-analyses and qualitative analyses. PLoS One. 2012;7:e38268.
34. Stern Y, Gurland B, Tatemichi TK, Tang MX, Wilder D, Mayeux R. Influence of education and occupation on the incidence of Alzheimer’s disease. JAMA. 1994;271:1004–1010.
35. Gow AJ, Corley J, Starr JM, Deary IJ. Reverse causation in activity-cognitive ability associations: the Lothian Birth Cohort 1936. Psychol Aging. 2012;27:250–255.
36. Montlahuc C, Soumaré A, Dufouil C, et al. Self-rated health and risk of incident dementia: a community-based elderly cohort, the 3C study. Neurology. 2011;77:1457–1464.
37. Saczynski JS, Jonsdottir MK, Sigurdsson S, et al. White matter lesions and cognitive performance: the role of cognitively complex leisure activity. J Gerontol A Biol Sci Med Sci. 2008;63:848–854.
38. Hultsch DF, Hertzog C, Small BJ, Dixon RA. Use it or lose it: engaged lifestyle as a buffer of cognitive decline in aging? Psychol Aging. 1999;14:245–263.
39. Hall CB, Lipton RB, Sliwinski M, Stewart WF. A change point model for estimating the onset of cognitive decline in preclinical Alzheimer’s disease. Stat Med. 2000;19:1555–1566.
40. Wilson RS, Segawa E, Boyle PA, Anagnos SE, Hizel LP, Bennett DA. The natural history of cognitive decline in Alzheimer’s disease. Psychol Aging. 2012;27:1008–1017.
41. Wilson RS, Leurgans SE, Boyle PA, Bennett DA. Cognitive decline in prodromal Alzheimer disease and mild cognitive impairment. Arch Neurol. 2011;68:351–356.
42. Reisberg B, Prichep L, Mosconi L, et al. The pre-mild cognitive impairment, subjective cognitive impairment stage of Alzheimer’s disease. Alzheimers Dement. 2008;4(1 suppl 1):S98–S108.
43. Rajan KB, Wilson RS, Weuve J, Barnes LL, Evans DA. Cognitive impairment 18 years before clinical diagnosis of Alzheimer disease dementia. Neurology. 2015;85:898–904.
44. Wilson RS, Segawa E, Boyle PA, Bennett DA. Influence of late-life cognitive activity on cognitive health. Neurology. 2012;78:1123–1129.
45. Vanderweele TJ, Arah OA. Bias formulas for sensitivity analysis of unmeasured confounding for general outcomes, treatments, and confounders. Epidemiology. 2011;22:42–52.
46. Schinka JA, McBride A, Vanderploeg RD, Tennyson K, Borenstein AR, Mortimer JA. Florida Cognitive Activities Scale: initial development and validation. J Int Neuropsychol Soc. 2005;11:108–116.
47. Tabert MH, Albert SM, Borukhova-Milov L, et al. Functional deficits in patients with mild cognitive impairment: prediction of AD. Neurology. 2002;58:758–764.
48. Reitz C, Schupf N, Luchsinger JA, et al. Validity of self-reported stroke in elderly African Americans, Caribbean Hispanics, and Whites. Arch Neurol. 2009;66:834–840.
49. Wilson RS, Boyle PA, Yu L, Barnes LL, Schneider JA, Bennett DA. Life-span cognitive activity, neuropathologic burden, and cognitive aging. Neurology. 2013;81:314–321.
50. Jobe JB, Smith DM, Ball K, et al. ACTIVE: a cognitive intervention trial to promote independence in older adults. Control Clin Trials. 2001;22:453–479.
51. Ball K, Berch DB, Helmers KF, et al.; Advanced Cognitive Training for Independent and Vital Elderly Study Group. Effects of cognitive training interventions with older adults: a randomized controlled trial. JAMA. 2002;288:2271–2281.
52. Willis SL, Tennstedt SL, Marsiske M, et al.; ACTIVE Study Group. Long-term effects of cognitive training on everyday functional outcomes in older adults. JAMA. 2006;296:2805–2814.
53. Rebok GW, Ball K, Guey LT, et al.; ACTIVE Study Group. Ten-year effects of the advanced cognitive training for independent and vital elderly cognitive training trial on cognition and everyday functioning in older adults. J Am Geriatr Soc. 2014;62:16–24.
54. Unverzagt FW, Guey LT, Jones RN, et al. ACTIVE cognitive training and rates of incident dementia. J Int Neuropsychol Soc. 2012;18:669–677.
55. Fried LP, Carlson MC, McGill S, et al. Experience Corps: a dual trial to promote the health of older adults and children’s academic success. Contemp Clin Trials. 2013;36:1–13.
56. La Rue A. Healthy brain aging: role of cognitive reserve, cognitive stimulation, and cognitive exercises. Clin Geriatr Med. 2010;26:99–111.
57. Bielak AA. How can we not ‘lose it’ if we still don’t understand how to ‘use it’? Unanswered questions about the influence of activity participation on cognitive performance in older age–a mini-review. Gerontology. 2010;56:507–519.
58. Ngandu T, Lehtisalo J, Solomon A, et al. A 2 year multidomain intervention of diet, exercise, cognitive training, and vascular risk monitoring versus control to prevent cognitive decline in at-risk elderly people (FINGER): a randomised controlled trial. Lancet. 2015;385:2255–2263.

Supplemental Digital Content

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.