Table 2 Overview of the seven studies included

Andel et al. (2015) [21]

The US Health and Retirement Study (HRS). general population aged 55+ at baseline (n = 3779)

Self-reported retirement. Comparing the trajectories of cognitive change of participants with high and low job strain before and after retirement. Control for age, gender, education, marital status, race, income, length of occupation, depressive symptoms, cardiovascular disease, manual/non-manual work.

Episodic memory (immediate and delayed recall test). (F)^b

10 data collection points at 2 years intervals. The participants averaged 7.4 interviews; 3.8 before retirement, 4.4 after retirement.

Growth curve model parameter estimates:

Pre-retirement change was significant (−5.72, p > 0.01) Post-retirement change was not significant (0.92, p = 0.75)

Greater job strain associated with worse episodic memory (−0.49, p < 0.001); not associated with pre-retirement change in episodic memory (0.39, p = 0.181); associated with greater decline in episodic memory post retirement (−0.65, p < 0.05).

Bonsang et al. (2012) [22]

The US Health and Retirement Study (HRS). General population aged 51–75 (n = 14,710)

Self-reported retirement (year and month when last employment ended). Compare retired with average score of sample. In the model they include eligibility for social security to provide evidence against reverse causality and they control for time in-variant heterogeneity.

Episodic memory (10 word recall test –immediate and delayed) (F)

Working memory (subtract 7 from 100 up to five times). (F)

6 data collection points at 2 years intervals.

Length of retirement included.

Retirement for one year or more has a negative effect on episodic memory score (Coefficient estimate: −0.942: 95%-confidence interval: −1.61 to −0.28). Magnitude of effect: −0.278, SE 0.100.

Working memory: (coefficient estimate

−0.279, SE 0.126). Magnitude of effect −0.230, SE 0.104 (p < 0.05).

Finkel et al. (2009) [29]

Swedish Adoption/twin study of Aging (SATSA). Twins aged 55+ (n = 462)

Self-reported year of retirement). Comparing slope of cognitive change before and after retirement, and comparing the trajectories of high and low complexity occupations. Control for dementia and practice effect.

Spatial ability (Figure Logic, Block design, Card Rotation) (F)

Verbal ability (Information, Synonymous, Analogies) (C)^b

Memory (Digit Span, Picture Memory, Names & Faces) (F)

Processing speed (Symbol Digit and Figure Identification) (F)

5 measure points at 3 years intervals (one 7 years)

Length of retirement measured (15 years before retirement to 20 years after retirement modelled).

Growth curve model parameter estimates:

Verbal ability: Mean at retirement: (L)^a: 53.7; (H)^a: 55.3 (ns); pre-retirement change: (L): −0.13; (H): +0.07 (p < 0.05). Post retirement change: (L):-0.1: (H): −0.2 (ns).

Spatial ability: Mean at retirement: (L): 51.36; (H): 54.36 (p < 0.05); pre-retirement change (L): −0.35, (H): −0.35 (ns); post retirement change (L): −0.29; (H): −0.51 (p < 0.05).

Memory: So significant findings.

Processing speed: Mean at retirement: (L): 52.93; (H): 55.47 (p < 0.05); pre-retirement change: (L): −0.27; (H): −0.52 (ns); post retirement change: (L): −0.60; (H): −0.69 (ns).

Fisher et al. (2014) [24]

US Health and Retirement Study (HRS). General population aged 51–61 at entry (n = 4182)

Self-report of year and month of retirement. Comparing rate of cognitive change before and after retirement, and comparing the trajectories of high and low mental demand occupations. Control for practice effect and socioeconomic, demographic, and health variables.

Episodic memory (immediate and delayed word recall test) (F)

Mental status (telephone Interview of Cognitive Status)

Data from 1992 to 2010 collected at 2 years intervals. Participants included if completed min. 2 waves.

Length of retirement included. (Mental status only post retirement measures)

Episodic memory: Difference in pre- and postretirement decline (Estimate = 0.06; p < 0.05). Higher mental demands associated with better memory (Estimate = 0.06; p < 0.05) and less steep decline (Estimate = 0.01; p < 0.05).

Mental status: Higher mental demands associated with better cognitive status at point of retirement (Estimate = 0.11; p < 0.05). There was a general decline of 0.27 SD per 10 years post retirement. Higher mental work demands was associated with slower rate of decline (Estimate = 0.004; p < 0.05).

Roberts et al. (2010) [25]

The UK Whitehall II study. London-based civil servants aged 38–60 at entry (n = 2031)

Employment status. (still working vs fully retired at follow up, all working at baseline).

Difference between baseline and follow up cognitive score comparing retirees and those still working. Control for adult IQ, age, mental and physical health, self-rated health, social class, education, psychosocial job characteristics, and leisure activities.

Short term verbal memory (free recall test) (F)

Inductive reasoning (AH4 – part 1) (F)

Verbal fluency (“s” words and animal names) (F)

5 years

(159 had been retired between <1–115 weeks at follow up; 151 between 115 and 218 weeks; 160 between 219 and 309 weeks)

General trend of improved cognitive functioning at follow up.

Inductive reasoning: retirees less improvement than those still working (Regression coefficient − 0.7; 95% CI −1.2 to −0.09).

Short term verbal memory: no significant findings

Verbal fluency: no significant findings

Wickrama et al. (2013) [26]

US HRS (Health and retirement study). General population aged 62+ at entry (n = 8524)

Self-reported work status: working full-time, working part-time, fully retired. They use structural equation models to investigate the reciprocal association between change in work status and cognitive change. Control for age, education, gender, race/ethnicity, depressive symptoms and physical disability.

Immediate memory (Recall from 10-word lists) (F)

6 data collection points at 2 years intervals.

No information on length of retirement period.

Over three time intervals the level of working at one point in time predicted subsequent changes in immediate memory (β = .04, & .06, both p < 0.01, and β = .07, p < 0.001).

Ryan (2008) [27]

The Seattle Longitudinal Study (SLS). White middle and upper class individuals aged 60+ at entry (n = 271)

Work status (retired vs working). Comparing the cognitive scores of those working all three waves with those retiring during study. Control for gender, education, perceptual speed, subjective and objective health.

Verbal memory (PMA, immediate recall, delayed recall) (F)

Inductive reasoning (PMA, ADEPT, word series, number series) (F)

Verbal ability (PMA, ETS) (C)

3 data collection points at 7 years intervals.

Length of retirement not measured.

Inductive reasoning: Participants employed in all three waves averaged 0.14 t-score units higher than those who worked only during one or two waves (p < 0.05).

Verbal ability: Working more than one wave was associated with a 2.4 t-score unit gain (p < 0.01).

Verbal ability: No significant findings