Relationships among young adult general cognitive ability, midlife physical activity, and early late‐life cognitive functioning: A four‐decade longitudinal cohort study in men

Abstract

INTRODUCTION

Research on whether physical activity (PA) is associated with cognition is abundant but very few studies have examined the extent to which prior cognitive ability may account for PA participation in midlife.

METHODS

Over 800 men self‐reported PA at average ages of 40 and 56. General cognitive ability (GCA) was assessed at an average age of 20. Specific cognitive abilities and GCA were assessed at average ages of 56 and 68. Relationships among age 20 GCA, midlife PA, and cognitive functioning in mid‐ and late‐life were examined with generalized estimating equations.

RESULTS

Age 20 GCA was significantly associated with age 56 leisure metabolic equivalent of energy expenditure (MET)‐hours of PA (b = 0.14, p = 0.027). Age 56 leisure MET‐hours were positively (b = 0.04, p = 0.021) and age 40 vigorous leisure PA was inversely (b = −0.10, p = 0.012) associated with age 68 GCA (b = 0.04, p = 0.021).

DISCUSSION

There are reciprocal associations between PA and cognitive functioning.

Highlights

• Young adult general cognitive ability (GCA) predicts midlife physical activity (PA).

• Midlife PA and cognition were not associated after adjusting for young adult GCA.

• Midlife PA is associated with later‐life cognition, adjusted for young adult GCA.

• Work‐related PA was inversely associated with later‐life cognitive functioning.

• The relationship between PA and cognitive function is bidirectional.

1. BACKGROUND

Leisure‐time physical activity (PA) is associated with a decreased incidence of dementia ¹ and cognitive decline, ² while, for work‐related PA, a possible association is opposite. ¹ The cognitive domain most affected by PA has been suggested to be executive function. ³ The association between PA and better late‐life cognition or decreased dementia incidence in observational studies has, however, not been uniformly replicated in randomized controlled trials studying the effect of aerobic exercise on cognition. ⁴ , ⁵

One reason for the different results between observational studies and randomized controlled trials of the relationship between PA and cognition or dementia may be that the majority of PA‐dementia studies control for the influence of education but not for prior level of cognitive function. Young age general cognitive ability (GCA), that is, cognitive reserve, ⁶ is a strong predictor of late‐life cognition. ⁷ , ⁸ Using education as an approximation for baseline level of cognitive ability may not be sufficient to account for an individuals’ starting level of cognitive abilities. Most cognitive abilities stay relatively stable or change rather slowly through much of adulthood ⁷ , ⁹ and PA interventions lasting months, or even a year or 2, may be too short to result in significant changes in cognitive function.

Very few previous observational studies of PA and cognition have been able to consider young age GCA. In the British 1946 Birth Cohort, PA was associated with a slower rate of decline in episodic memory from age 43 to age 53 even after taking into account young age GCA. ¹⁰ In another study from the same cohort, the diversity of PA (number of different physical activities) in midlife was not associated with rate of decline in episodic memory from age 43 to age 60–64 years but was associated with a slower rate of decline in a letter cancellation test after taking into account young age GCA. ¹¹ In a more recent study from the British 1946 Birth Cohort, PA at all ages from 36 to 69 years was positively associated with age 69 global cognition, episodic memory, and processing speed over and above young age GCA. ¹² In contrast, a study from the Lothian Birth Cohort made an opposite finding, suggesting a sensitivity period for PA rather than a cumulative model. ¹³

Given the importance of young adult GCA for late‐life cognition, our primary hypothesis was that PA contributes little to late‐life cognitive functioning over and above young adult GCA. Importantly, we examined whether this association is present in the reverse direction: from young adult GCA to midlife PA. Therefore, we examined whether age 20 GCA is associated with midlife PA as well as whether midlife PA (at an average ages of 40 and 56) is associated with midlife cognitive function (average age 56) and early late‐life cognitive function (average age 68) after taking into account average age 20 GCA. We also examined if the following aspects of PA associate differently with GCA because contrasting associations have been reported for these aspects ² , ¹⁴ , ¹⁵ : PA in different contexts (work‐ versus leisure‐time PA), the amount, and whether or not an individual engages in any vigorous PA.

2. METHODS

2.1. Study population

The Vietnam Era Twin Study of Aging (VETSA) is an ongoing longitudinal twin study focused on early identification of risk for Alzheimer's disease (AD) and understanding factors that affect cognitive ageing. ¹⁶ , ¹⁷ , ¹⁸ VETSA comprises members of the Vietnam Era Twin Registry (VETR), a registry of male‐male twin pairs who served in the military at some time during Vietnam era (1965–1975). ¹⁹ VETSA participants were randomly sampled from a large study of all VETR members. ²⁰ To examine relationships among young adult GCA (age 20), midlife PA (ages 40 and 56), and late‐life (age 68) cognition, we used data from enlistment (mean age 20), from mean age 40 (range 33–46) when VETSA participants had participated in a mailed heath survey ²¹ and from 2 waves of VETSA (at mean ages of 56 [range 51–60], and 68 [range 61–70]) (Figure 1 and Figure S1). The lifestyle questions at a mean of age 40 and VETSA study waves were different, and we used information on PA from both time points in midlife to check for consistency of the results. The age 20 and 40 data were utilized because they were available from prior data. Young adult GCA provides cognitive ability data prior to any aging effects. The age 40 health survey (1990) provided earlier exercise data. VETSA began when participants were in their 50s and were followed every 5–6 years; hence, the data collection at ages 56, 62, and 68. Recent evidence suggests that there are major biological shifts in the human body around ages 40 and 60. ²² This evidence for the timing of biological shifts suggests the value of some of the time periods assessed.

FIGURE 1.

Flowchart of the study. MET, metabolic equivalent of energy expenditure; PA, physical activity; VETSA, the Vietnam Era Twin Study of Aging; WHO, World Health Organization.

Participants gave written informed consent to participate at the beginning of each study wave's face‐to‐face examinations. The study was approved by the Institutional Review Boards at the University of California, San Diego and Boston University.

2.2. Physical activity

Due to differences in PA questions, the variables differ in the age 40 and age 56 assessments (see Table 1).

TABLE 1.

Description of PA variables used in this study.

Age 40 and age 56 physical activity measures
	Construct	Measure name	Context	Item	Variable construction
Age 40	Vigorous physical activity (VPA)	Age 40‐VPA	Only leisure‐time PA	For at least 3 months, which of the following activities have you performed regularly? Jog or run at least 10 miles per week Play strenuous racquet sports at least 5 h per week Play other strenuous sports (basketball, soccer, etc.) Ride a bicycle at least 50 miles per week Swim at least 2 miles per week	0/1 = If participant engaged in any of these.
	Work‐related physical activity	Age 40 work‐PA	Work‐related PA	On a typical day, how much time on the job do you spend walking? Practically all the time More than ½ the time ½ the time Less than ½ the time Almost never How often do you have to lift heavy weights or carry heavy things (greater than 50 lb)? Frequently Sometimes Very infrequently or never	0 = walking almost never on the job and lifting or carrying heavy weights never of infrequently, 1 = walking less than ½ of the time on job and lifting or carrying heavy weights never of infrequently, 2 = walking at least ½ of the time but not practically all the time on the job or lifting or carrying heavy weights sometimes, 3 = walking practically all the time on the job or lifting or carrying heavy weights frequently.
Age 56	Vigorous physical activity	Age 56‐VPA	Mainly leisure‐time PA	Please list any sports, fitness or recreational activities in which you participated last week. We are interested only in times you were physically active (up to six responses). Sports, fitness or recreation Times/week Average minutes per episode	Activity (a) recoded as light, moderate, vigorous intensity. 23 VPA = 1 if participant has any vigorous physical activity minutes.
	Metabolic equivalent of energy expenditure (MET) per day	Age 56 MET‐hours	Mainly leisure‐time PA	(same question as above)	Activity (a) assigned MET: light = 2.3; moderate = 4.5; vigorous = 8. Total MET‐hours of PA in MET‐hours per day calculated (MET) * frequency per day * duration).
	Physical activity meets WHO recommendations	Age 56 WHO‐PA	Mainly leisure‐time PA	(same question as above)	0/1 = 150 min of moderate exercise or 75 min of vigorous exercise or equivalent combination of moderate and vigorous exercise per week.
	Physical activity frequency	Age 56 PA‐Freq	Mainly leisure‐time PA	How frequently did you do [sports participation, physical fitness, walking and hiking, outdoor activities, dancing] in the past 30 days? (Never, Once a month, Once a week, Several times per week, Daily).	Sum of the five activity items

Note: For continuous variables means and ranges are presented (mean, [range]). For categorical variables numbers and percentages are presented (n, [%]).

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed the literature using traditional (e.g., PubMed) sources. While many studies associate physical activity (PA) with cognitive functioning, there are few data on how young age general cognitive ability (cognitive reserve) is associated with midlife PA and how young age cognitive reserve affects associations between PA and cognitive functioning in later life.

2. Interpretation: Our findings show that the association between PA and cognitive functioning is reciprocal from young age general cognitive ability to midlife PA and from midlife PA to early late‐life, but not midlife, cognitive functioning. The associations were different depending on the way to measure PA.

3. Future directions: Future studies should endeavor to understand (a) mechanisms by which young age general cognitive ability may influence later PA participation and (b) why the associations differ for midlife and early late‐life cognitive functioning and for different contexts of PA.

2.2.1. Age 40

In 1990, VETR members received a mailed and telephone survey. Survey questions allowed us to create 2 PA measures at age 40: vigorous PA (VPA) (age 40 VPA) and work‐related PA (age 40 work‐PA) (see Table 1 and Materials and Methods in Supporting Information for details).

2.2.2. Age 56

At age 56, we created 4 PA measures: total amount of mainly leisure‐time PA in metabolic equivalent of energy expenditure (MET)‐hours (age 56 MET‐hours), age 56‐VPA (mainly leisure‐time PA), meeting of World Health Organization PA recommendations (age 56 WHO‐PA, mainly leisure‐time PA), and frequency of leisure‐time PA (age 56 PA‐Freq, mainly leisure‐time PA) (see Table 1 and Materials and Methods in Supporting Information for details). ²³

2.3. Cognitive measures

GCA was based on the Armed Forces Qualification Test (AFQT) ²⁴ administered at an average age of 20 years (range 17–26), 56, and 68 (see Materials and Methods in Supporting Information for more information). ²⁵ , ²⁶

Function in specific cognitive domains was based on previously derived factor scores for executive function, ²⁷ processing speed, ²⁸ episodic memory, ²⁹ verbal fluency, ³⁰ and visuo‐spatial ability. Briefly, these factor scores were created based on factor loadings from 3 to 6 neurocognitive tests per domain. The neurocognitive tests are presented in Materials and Methods in Supporting Information and are described in detail elsewhere. ³¹ Factor scores were standardized in relation to VETSA wave 1 and adjusted for practice effects due to repeated testing at subsequent assessments. ³²

2.4. Covariates

2.4.1. Time‐invariant covariates

Race/ethnicity was used as a dichotomous variable (non‐Hispanic White vs. other). Socioeconomic status (SES) was measured with the Hollingshead‐Redlich index. ³³ This index was created by applying the formula 5*average occupation + 3*average education ³³ with education on a 1 to 7 scale. Apolipoprotein E (APOE) genotype was used as a dichotomous variable (no ε4 allele vs. at least 1 ε4 allele). APOE genotyping in this cohort has been described previously. ³⁴ , ³⁵

2.4.2. Age 40 covariates

We created a multimorbidity index by summing 15 chronic medical conditions from the Charlson comorbidity index ³⁶ (i.e., diabetes, emphysema, asthma, cancer, osteoarthritis, rheumatoid arthritis, stroke, heart attack, heart failure, heart surgery, angina pectoris, hypertension, peripheral vascular disease, cirrhosis, acquired immune deficiency syndrome [AIDS]). ⁷ Body mass index (BMI) was calculated from self‐reported weight and height. Alcohol consumption was treated as a continuous variable (number of drinks during the past 2 weeks). Smoking was categorized as current versus never or former smoker. Depression was used as a dichotomous variable at age 40: the individual was categorized with depression if they reported a doctor ever having told them they had depression.

2.4.3. Age 56 and 68

Multimorbidity index, alcohol consumption, and smoking were categorized similarly as at age 40. BMI was calculated using weight measurements with a digital scale and height measurements with a stadiometer. Depressive symptoms were measured with the Center for Epidemiologic Studies Depression Scale (CES‐D) ³⁷ with higher scores indicating more depressive symptoms.

2.5. Statistical analyses

We used generalized estimating equations (GEE) with an identity link function to analyze the relationship between PA measures and cognition. ³⁸ There was heteroscedasticity in several responses and the semiparametric GEE was adopted to ensure valid inference. ³⁸ The GEE model also addresses the clustered data structure we have (twin pairs). We ran all analyses primarily with two different working correlation structures: independent and exchangeable. GEE provides valid inference regardless of the choice of working correlation structures. For all models, the working independent correlation structure yielded more power and thus was chosen. The results are presented with regression coefficients and p‐values. Age 56 MET‐hours and CES‐D depression scale were both skewed. Although GEE provides valid inference regardless of the distribution of the data, we used square root transformations of these predictors to improve power.

First, we examined whether age 20 GCA is associated with the different midlife PA measures at ages 40 and 56. These analyses were adjusted for age and race/ethnicity (Model 1). Then, we examined whether age 40 and age 56 PA measures are associated with ages 56 and 68 GCA and whether age 40 and age 56 PA measures are associated with function in specific cognitive domains at ages 56 and 68. We tested three different models. Model 1 was adjusted for age at follow‐up and race/ethnicity. Model 2 additionally adjusted for age 20 GCA. Model 3 was further adjusted for SES. Fully adjusted Model 4 was further adjusted for the multimorbidity index, BMI, smoking status, alcohol consumption, APOE genotype, and depression. Although PA may exert some of its effect indirectly through BMI or depression, we were primarily interested in the direct effect of PA on cognition.

Participants with over 40 MET‐hours per day were excluded from the analyses as extreme values (this corresponds to cycling at 30 km/h for 5 h every day of the week). Variance inflation factors were checked with no indication of serious multicollinearity. We used interaction testing to test for moderation by young adult GCA and APOE genotype. GCA and specific cognitive domain scores were standardized before analyses. All analyses were carried out in Stata 18.0 (StataCorp LLC). The threshold for significance was set at two‐sided α = 0.05 and the analyses for the association of PA measures and cognitive functioning in specific domains were corrected for multiple testing (see Materials and Methods in Supporting Information). We followed Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cohort studies.

3. RESULTS

3.1. Descriptive statistics

Age 40 and age 56 characteristics of the participants are presented in Table 2. The follow‐up length from age 40 to age 68 was on average 27.4 years (SD = 0.7, range 25.5–29.7), and from age 56 to age 68, it was on average 11.4 years (SD = 1.4, range 5.6–14.4). GCA declined from age 56 to age 68 on average 0.27 SD (SD = 0.43).

TABLE 2.

Characteristics of the study sample.

Characteristics	Age 40 (n = 805) a	Age 56 (n = 785) b
Age	40.0 (33–46)	55.9 (51–61)
Education (years)	14.0 (2.1)	14.0 (2.1)
SES	43.1 (11.1)	43.0 (11.2)
Occupation (on the Hollingshead‐Redlich index)	5.6 (1.8)	5.6 (1.9)
Ethnicity
White	758 (94.2)	728 (92.7)
Non‐White	47 (5.8)	57 (7.3)
BMI	26.0 (3.6)	29.2 (4.7)
Multimorbidity index	0.2 (0.5)	1.0 (1.1)
Alcohol consumption (drinks/2 weeks)	3.4 (1.5)	11.8 (20.9)
APOE genotype
No ε4 allele	557 (69.7)	556 (71.4)
At least one ε4 allele	242 (30.3)	223 (28.6)
Smoking status
Never or former	585 (72.7)	641 (81.7)
Current	220 (27.3)	144 (18.3)
Depression
Yes	64 (8.0)
No	734 (92.0)
CES‐D		14.2 (5.0)
Age 40‐VPA
Yes	222 (27.6)
No	583 (72.4)
Age 40 Work‐PA
0	72 (9.4)
1	162 (21.1)
2	267 (34.7)
3	268 (34.9)
Age 56‐VPA
Yes		102 (13.0)
No		683 (87.0)
Age 56 MET‐hours		2.5 (0–32.1)
Age 56 WHO‐PA
Yes		435 (55.4)
No		350 (44.6)
Age 56 PA‐Freq		10.7 (5–24)

Notes: For age means and ranges are presented. For other continuous variables means and standard deviations are presented (mean, [SD]). For categorical variables numbers and percentages are presented (n, [%]).

Abbreviations: APOE, APOE genotype; BMI, body mass index; CES‐D, Center for Epidemiologic Studies Depression Scale; MET, metabolic equivalent of energy expenditure; PA, physical activity; SES, socioeconomic status; VPA, vigorous physical activity; WHO‐PA, World Health Organization – physical activity.

^a Some covariates for age 40 PA had missing information: APOE genotype was known for 799 persons, and multimorbidity index was known for 756 persons.

^b Some covariates for age 56 PA had missing information: APOE genotype was known for 818 persons, and depression was known for 790 persons.

Age 40: The proportion of subjects engaging in age 40 VPA was 28% (see Table 1 for measure descriptions). The work participants did at age 40 was on average quite physical: 70% walked at least half of the time and/or lifted weights heavier than 22.7 kg at least sometimes at work.

Age 56: The number of participants meeting the age 56 WHO‐PA recommendation for aerobic exercise was 55%. Age 56 MET‐hours were on average 2.5 (SD = 3.7), corresponding to, for example, running at 10 km/h for 1 h and 45 min per week or walking briskly at 5.7 km/h for 4 h and 23 min per week. The percentage of participants engaging in age 56 VPA was 13%.

3.2. Age 20 GCA and PA participation at age 40 and age 56

In Model 1, age 20 GCA was positively associated with age 56 MET‐hours (b = 0.14, p = 0.027) (Figure 2A), age 56 WHO‐PA (b = 0.09, p = 0.002), age 56‐VPA (b = 0.04, p = 0.036) and age 56 PA‐Freq (b = 0.48, p = 0.006), but inversely associated with age 40 work‐PA (b = −0.16, p = 0.006) (Table 3). Participants with higher age 20 GCA worked at less physically strenuous jobs at age 40.

FIGURE 2.

The associations between age 56 MET‐hours and GCA at ages 20, 56, and 68. (A) The association between age 20 GCA (x‐axis) and age 56 MET‐hours squared (y‐axis) (age‐ and ethnicity‐adjusted). (B) The association between age 56 MET‐hours squared (x‐axis) and age 56 GCA (y‐axis) (age‐ and ethnicity adjusted). (C) The association between age 56 MET‐hours squared (x‐axis) and age 68 GCA (y‐axis) (age‐ and ethnicity‐adjusted). GCA, general cognitive ability; MET, metabolic equivalent of energy expenditure.

TABLE 3.

Association of age 20 general cognitive ability and midlife physical activity: Regression coefficients and their significance from generalized estimating equations models.

Parameter	Age 40‐VPA	Age 40 Work‐PA	Age 56 MET‐hours	Age 56‐VPA	Age 56 WHO‐PA	Age 56 PA‐Freq
Age 20 general cognitive ability	0.01 (0.719)	−0.16 (0.006)	0.14 (0.027)	0.04 (0.036)	0.09 (0.002)	0.48 (0.006)

Notes: The results are presented in regression coefficients with p‐values (the threshold for significance has been set at < 0.05). The analyses are adjusted for age and ethnicity.

Abbreviations: MET, metabolic equivalent of energy expenditure; PA, physical activity; VPA, vigorous physical activity; WHO‐PA, meets World Health Organization recommendations.

3.3. Age 40 and age 56 PA associations with age 56 GCA

In Model 1, age 40 work‐PA, age 56 WHO‐PA, age 56‐VPA, and age 56 PA‐Freq were associated with age 56 GCA (Table 4, Figure 2B, and Tables S1–S4). After adjusting for age 20 GCA in Model 2, only the association between age 40 work‐PA and age 56 GCA remained statistically significant (Table 4). In Model 3, after adjusting additionally for SES, there were no longer any statistically significant associations between any PA measure and age 56 GCA. The interaction tests showed no modification by APOE genotype or age 20 GCA. In the fully adjusted Model 4, no PA measures at age 40 or age 56 were associated with performance in any specific cognitive domain at age 56 (Table S5).

TABLE 4.

Association of physical activity and general cognitive ability: Regression coefficients and their significance from generalized estimating equations models.

	PA regressed on age 56 GCA:				PA regressed on age 68 GCA
Parameter	Model 1 a	Model 2 b	Model 3 c	Model 4 d	Model 1 a	Model 2 b	Model 3 c	Model 4 d
Age 40‐VPA	−0.03 (0.300)	−0.03 (0.300)	−0.05 (0.146)	−0.04 (0.209)	−0.06 (0.214)	−0.05 (0.136)	−0.08 (0.033)	−0.10 (0.012)
Age 40 Work‐PA	−0.03 (0.049)	−0.03 (0.049)	−0.02 (0.247)	−0.02 (0.229)	−0.10 (0.00004)	−0.05 (0.003)	−0.04 (0.070)	−0.04 (0.079)
Age 56 MET‐hours	0.04 (0.074)	0.010 (0.800)	−0.00 (0.900)	−0.00 (0.898)	0.08 (0.0001)	0.05 (0.003)	0.04 (0.016)	0.04 (0.021)
Age 56‐VPA	0.15 (0.012)	0.04 (0.300)	0.02 (0.613)	0.04 (0.411)	0.20 (0.001)	0.11 (0.019)	0.07 (0.148)	0.06 (0.228)
Age 56 WHO‐PA	0.12 (0.011)	0.02 (0.441)	0.01 (0.780)	0.01 (0.726)	0.16 (0.006)	0.07 (0.042)	0.05 (0.202)	0.05 (0.198)
Age 56 PA‐Freq	0.02 (0.008)	0.01 (0.221)	0.00 (0.454)	0.00 (0.360)	0.01 (0.025)	0.00 (0.437)	0.00 (0.908)	−0.00 (0.987)

Notes: The results are presented in regression coefficients with p‐values. Bolded results are statistically significant (the threshold for significance is set at < 0.05).

Abbreviations: APOE, Apolipoprotein E; BMI, body mass index; GCA, general cognitive ability; MET, metabolic equivalent of energy expenditure; PA, physical activity; SES, socioeconomic status; VPA, vigorous physical activity; WHO‐PA, World Health Organization – physical activity.

^a Model 1 is adjusted for age and ethnicity.

^b Model 2 is additionally adjusted for young adult cognitive ability.

^c Model 3 is additionally adjusted for SES.

^d Model 4 is additionally adjusted for BMI, multimorbidity index, smoking, alcohol consumption, APOE genotype, and depression.

3.4. Age 40 and age 56 PA associations with age 68 (early late‐life) GCA

In Model 1, age 40 work‐PA and all age 56 PA measures were associated with age 68 GCA (Table 4, Figure 2C, and Tables S6–S9). The associations of age 40 work‐PA, age 56 MET‐hours, age 56 WHO‐PA, and age 56‐VPA with age 68 GCA remained significant after additional adjustments in Model 2 (Table 4). In Models 3 and 4, age 40‐VPA and age 56 MET‐hours were significantly associated with age 68 GCA. Age 40 VPA, however, was negatively associated with age 68 GCA, and age 56 MET‐hours was positively associated with age 68 GCA. Since Models 3 and 4 are adjusted for age 20 GCA, the results can be interpreted as indicating that higher vigorous activity at 40 was associated with worse than expected cognitive ability at age 68, while participants who engaged in more MET‐hours of leisure activity at 56 performed better than expected. In the fully adjusted Model 4, age 40 work‐PA was significantly negatively associated with executive function and episodic memory (Table S10). None of the other PA measures were significantly associated with performance in any specific cognitive domains at age 68.

We also ran interaction tests to examine if age 20 GCA or APOE genotype moderate the association between aspects of PA at midlife and age 68 GCA. Age 20 GCA or APOE genotype did not moderate the association between PA and age 68 GCA.

4. DISCUSSION

Young adult GCA was associated with more leisure‐time PA and less work‐related PA in midlife. This finding is compelling: the direction of the association is opposite to what is often presumed, in that cognitive ability predicted the extent of PA that a person was engaged in. Early cognitive reserve may contribute in complex ways to later PA, though, whether the mechanisms are related to occupation and resources, better life skills, or to better awareness of the role of fitness is unknown. Consequently, the most fruitful window of opportunity for prevention seems to arise from early life.

Age 40 work‐PA and age 56 leisure PA (i.e., MET‐hours) were associated with age 56 GCA but only age 40 work‐PA survived adjustment for young adult GCA. However, after taking into account young adult GCA, SES, and health, there were no significant associations between midlife PA and midlife GCA or functioning in specific cognitive domains. The only earlier study that has examined the association between midlife PA and midlife cognition did not find a significant association between midlife PA and midlife episodic memory after taking into account young age GCA and other spare‐time activities. ¹⁰ Taken together, the current research evidence does not support PA as a booster of midlife cognitive performance.

While there were no significant PA‐cognition associations in midlife after taking into account young adult GCA and other covariates, age 56 MET‐hours was positively associated with age 68 GCA after taking into account young adult GCA and other covariates. Age 56 MET‐hours was the most precise PA measure (it is our only PA variable taking into account frequency, duration, and intensity of PA). These results suggest that PA likely does not enhance cognitive ability but rather delays the accumulation of neurodegenerative changes or improves the ability to cope with these changes. Regardless, early adult cognitive ability functions as a cognitive reserve and remains the strongest predictor of age 68 cognitive ability. The small regression coefficient for the association between age 56 MET‐hours and age 68 GCA, and the fact that our cruder PA measures (age 56 WHO‐PA, age 56 PA‐Freq, age 56‐VPA) were not significantly associated with age 68 GCA suggests that the association between midlife PA and early late‐life GCA is not strong. Cognitive trajectories may differentiate more between cognitive decliners and maintainers later and PA‐cognition associations may become more evident.

It is unclear why higher age 40‐VPA (only leisure‐time PA) was associated with worse than expected age 68 GCA. This is a surprising finding given that leisure‐time PA is usually positively associated with cognitive function. ² High MET‐hours do not necessarily imply high VPA: someone could accumulate high MET‐hours through long durations of light or moderate activity without any VPA. Alternatively, short durations of VPA might result in lower total MET‐hours compared to longer durations of moderate activity. The people who exercise very vigorously may be different than moderate exercisers in other aspects of life than PA and these other differences may explain the seemingly paradoxical associations. For example, in early midlife, participants engaging in vigorous sports may have been at more risk for accidents or injury, or may be engaged in other activities that were not protective. High levels of vigorous PA, which age 40‐VPA reflects, have been associated with an increased risk for adverse cardiovascular outcomes and several possible cardiac maladaptations. ³⁹ Very high levels of vigorous PA may not be beneficial for cognitive health either.

Age 40 work‐related PA was negatively associated with age 56 GCA and age 68 GCA, although the associations did not survive all the adjustments. This type of inverse association between work‐related PA and worse health outcome has also been reported for cardiovascular mortality. ⁴⁰ Midlife work‐related PA was also inversely associated with executive function and episodic memory. Reasons for the opposite effects of work‐related PA and leisure‐time PA could be low worker control over PA, insufficient recovery times, and harmful effects on resting heart rate and blood pressure ⁴¹ combined with low resources for the health burdens of physically strenuous jobs. Work‐related PA logically stops at retirement and post‐retirement PA may be required for PA to benefit late‐life. ⁴² It has also been suggested that the intensity of work‐related PA may often be too low to induce similar cardiovascular benefits as leisure‐time PA. ⁴² According to many studies, more intense PA is more beneficial for cognitive ageing, ⁴³ , ⁴⁴ , ⁴⁵ although conflicting evidence has also been published. ⁴⁶ In cognitive ageing specifically, the lack of regular cognitive stimulation and intellectual work has also been suggested to explain this “health paradox” of work‐related PA and leisure‐time PA. ⁴² Moreover, it may simply be that people with lower cognitive ability tend to be in more physically demanding jobs.

The only prior study addressing the amount (not diversity) of leisure‐time midlife PA and early late‐life cognition while adjusting for young age GCA showed a significant association for overall cognition, episodic memory and processing speed. ¹² Two‐thirds of the association was explained by young age GCA, SES, and education. ¹² These findings are very similar to ours: half of the association of age 56 MET‐hours and age 68 GCA was explained by young adult GCA and SES. Further adjustment for other covariates such as multimorbidity, cardiovascular health, and APOE genotype did not significantly change our results and do not, thus, explain the association seen between midlife PA and early late‐life GCA. The other studies examining the direction of the association between PA and cognition have examined the direction of the association in older adults. ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ Therefore, earlier studies suggesting prominent benefits from PA to cognitive ageing may be exaggerated, as few earlier studies have been able to take into account young age GCA.

Our study has limitations. VETSA consists of veteran male twin pairs who served in the military for an average of 35 years prior to study participation. However, the study cohort is reasonably well representative of American men in their age cohort with respect to health, education, and lifestyle characteristics, and approximately 80% report no combat experience ²⁵ . The retention rates of VETSA have been high ⁷ and twins are also proven to have similar mortality as the general population ⁵¹ and the same prevalence of most diseases as singletons. ⁵² GEE handles the correlated data by modeling the average response within twin pairs, a well‐established method that does not affect the magnitude of correlations but uses more stringent significance levels.

The PA measures were all self‐report. The PA question that was the basis for age 56 MET‐hours, age 56 WHO‐PA, and age 56‐VPA was open‐ended and limited to the preceding week. Short walks during everyday life may not be well reported and these PA variables may be limited in their ability to grasp lower‐end PA. The number of persons reporting zero PA was rather high (∼33.5%). On the other hand, the frequency of leisure‐time PA measure taps the frequency of walking, sports, fitness, dancing, and outdoor activities per month, but it does not measure the intensity or duration of PA. The open‐ended activity question in VETSA could not distinguish between aerobic exercise and muscle strengthening exercise intensity. Therefore, the two‐category variable of meeting the age 56 WHO PA recommendation is only an estimate of meeting the recommendation in this study. Hence, there is imprecision in measuring PA of which direction and magnitude is difficult to decipher. Questions on PA at ages 40 and 56 were different and did not allow the creation of a long‐term PA variable. In addition, the study population was limited to mainly white, non‐Hispanic twin men who enlisted during the Vietnam Era. This limits the generalizability of the results to women and other ethnicities. Gender differences have been formally tested in very few studies ² and based on earlier research, PA‐cognition relationship is similar across different ethnicities despite unequal distribution in social drivers of dementia risk between ethnic groups. ⁵³

Our study also has several strengths. We had very long follow‐ups (of 11 and 27 years), and we had multiple measures of PA (including MET‐hours, work‐related PA), GCA, and specific cognitive domain scores based on multiple tests. We were able to adjust our results for young adult cognitive reserve (young adult GCA) and examine if PA is associated with later‐life cognition independent of young adult cognitive reserve. We also accounted for many possible confounding factors. The study highlights the importance of a lifecourse perspective on the association of PA and cognitive performance.

This cohort study shows a modest but significant reciprocal relationship from young adult GCA to midlife PA and from midlife PA to mainly later life cognition. The direction of the association was opposite for work‐related and other PA, suggesting the importance of careful characterization of PA. These findings are important from a dementia prevention perspective: the seeds for healthy late‐life cognition, thus, seem, in part, to be planted at early ages. Both clinicians and society should encourage leisure‐time PA, but strategies to influence young adult GCA may work to increase midlife PA levels and postpone cognitive decline more effectively. Future research should examine the potential for dementia prevention early in development, not just in midlife and later life. ⁵⁴

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

All study participants provided informed consent.

Supporting information

Supporting Information

Supporting Information

Iso‐Markku P, Buchholz EJ, Tu XM, et al. Relationships among young adult general cognitive ability, midlife physical activity, and early late‐life cognitive functioning: A four‐decade longitudinal cohort study in men. Alzheimer's Dement. 2025;17:e70169. 10.1002/dad2.70169