Self-Reported Change in Quality of Life with Retirement and Later Cognitive Decline: Prospective Data from the Nurses’ Health Study

The Nurses’ Health Study cohort and the cognitive subcohort

The Nurses’ Health Study (NHS) began in 1976, when 121,700 US registered nurses, aged 30–55 years, completed a mailed questionnaire about their health and lifestyle. Follow-up questionnaires were mailed biannually. From 1996 to 2000, participants who were ≥70 years old and free of stroke were invited to participate in a telephone-based study of cognitive function (Fig. 1). For the first interview, 93% of all eligible women participated. We conducted three follow-up interviews at 1.5-year intervals (median cognitive follow-up time: 6 years) (Fig. 1). Participation in the follow-up interviews was >85% in living women. The study was approved by the Institutional Review Board of Brigham and Women’s Hospital (Boston, MA).

Cognitive assessment

The telephone interview consisted of a battery of six cognitive tests, which have been validated and previously described in detail [28–30]. Overall cognitive status was assessed with the Telephone Interview of Cognitive Status (TICS, range 0 to 41 points) [31], which is a telephone adaptation of the Mini-Mental State Examination. Verbal memory, a strong predictor of Alzheimer’s disease [32], was assessed with the TICS 10-word list (immediate and delayed recalls) and the East Boston Memory Test (immediate and delayed recalls) [33]. A test of category fluency [34], which partially assesses executive function, and the digit span backwards [35], a measure of attention, were also administered.

Our primary outcome was the global composite score, computed as the mean of the z-scores from all cognitive tests. As secondary outcomes, we considered the TICS score, the verbal memory composite score (mean of the z-scores from the immediate and delayed recalls of both TICS-10 word list and the East Boston Memory Test), the category fluency score and the digit span backwards score.

Composite scores (i.e., global score and verbal memory score) were constructed only for women who completed all contributing tests. Participation rates were identical across all cognitive assessments and remained stable over time.

Assessment of change in QOL at retirement

To be eligible for the present study, participants needed to have answered the 1996 biennial NHS questionnaire and retired before it (Fig. 1). In this questionnaire, along with retirement status and age at retirement, participants were asked to rate their satisfaction regarding retired life: “Overall, how would you say the quality of retired life compares with life when you were working?”. Possible responses were “much worse”, “somewhat worse”, “about the same”, “somewhat better”, and “much better”. Due to the skewed distribution of the responses, the categories of “much worse” and “somewhat worse” were combined into one category and the “somewhat better” and “much better” categories were combined into another.

Population for analysis

Of the 19,415 participants who completed the initial cognitive interview, we excluded women who did not complete any follow-up cognitive interviews (n = 2,274). We then excluded those whose baseline cognitive function evaluation occurred before the 1996 questionnaire or more than 5 years after the 1996 questionnaire response (n = 2,609). We further excluded those who were not fully retired or those whose age at retirement was unknown (n = 5,268 women) and those whose age at retirement was below 60 or above 70 years (n = 2,631) as it is possible that early or late retirement compared to the most common retirement ages (60–69 years) may be associated with cognitive outcomes [36]. We then excluded those who did not complete the retirement QOL assessment (n = 200) and those with missing data on important work-related factors (from the 1982 NHS questionnaire) such as occupation type before retirement and history of shift work (n = 1,507). After these exclusions, 4,926 women were included in the analysis. At the time of the initial cognitive interview, these women were similar in age and in cognition when compared with the larger population of 19,415 women (mean age (standard deviation) = 74.4 (2.2) years versus 74.3 (2.3) years in the larger population, mean TICS score (standard deviation) = 33.9 (2.5) versus 33.7 (2.8) points in the larger population).

In our analysis sample, the time elapsed between the 1996 questionnaire and the first cognitive interview ranged from 1 month to 4.7 years, with a median time of 3.6 years.

Correlation between post-hoc assessment of change in QOL and prospective pre-and post-retirement health-related QOL

Because the 1996 question on change in QOL with retirement was retrospectively assessed at a single timepoint, we aimed to test its validity in a subset of participants in whom prospective data was available on a QOL proxy both before retirement and after retirement. The QOL proxy we used was the health-related QOL as measured by the Short Form (36 items) Health Survey (SF-36) [37, 38]. General QOL is broader in concept than just health-related QOL because it encompasses non-health related life features, such as environment or social relationships, whereas health-related QOL is directly connected to an individual’s health or disease status. However, they both aim to capture a respondent’s subjective perception of well-being, so a correlation is likely. Therefore, in the subset of 777 women of our analysis sample in whom the SF-36 information was available in 1992 and in 1996, and who retired between 1992 and 1996, we considered whether the changes in physical and mental health component summary measures [39] from 1992 (pre-retirement) and 1996 (post-retirement) were consistent with participants’ responses in 1996 of the question on how the quality of retired life compared with working life, using the Jonckheere-Terpstra test, a nonparametric test for ordered differences among classes [40].

Covariates

We also obtained information on multiple potential confounders plausibly linked with both cognition and QOL at retirement, including demographic, socioeconomic, occupational, lifestyle, and health-related factors. As adjustment variables, basic models included age (continuous, in years), education (registered nurse, Bachelor’s degree, Master’s or Doctorate), age at retirement (60–64, 65–69 years), as well as the delay between retirement and post-hoc assessment of change in QOL at retirement (<5, 5-<10, ≥10 years). In multivariable-adjusted models, we further adjusted for 1) factors assessed using the most updated available data as of the first cognitive assessment: a) lifestyle and health factors (smoking status, alcohol intake, body mass index, physical activity, diet quality indicator based on the Alternate Healthy Eating Index [41, 42], age at menopause, postmenopausal hormone, vitamin E, ibuprofen, aspirin, diabetes, hypertension, hyperlipidemia, myocardial infarction, SF-36 Physical Functioning score), b) psychosocial factors (SF-36 Vitality score, SF-36 Mental Health score, antidepressant use, Berkman-Syme social network index [43], stress from caregiving) and c) socioeconomic factors (husband’s education level, median income of US census tract, whether participants ever had to forego medical treatment for financial reasons); 2) pre-retirement occupational factors (1982 questionnaire responses to current type of work and any history of shift work ≥20 years as assessed in 1988), 3) pre-retirement stress (1982 questionnaire response to stress at home or at work in daily life) and 4) early life socioeconomic factors (birth below 37° N latitude, father’s occupation at 16 years of age).

Statistical analysis

To evaluate the association between QOL at retirement and cognitive decline in later life, we used generalized mixed models for repeated measures. For each outcome, we evaluated basic and multivariable-adjusted models with covariates previously described. We estimated mean differences in rates of cognitive decline across categories of QOL at retirement: worse, same, and improved QOL; in all analyses, the “same” category was the reference group. Such models assumed that a participant’s change in cognitive function followed that of the population mean except for random effects for initial cognitive levels (i.e., random intercepts) and rates of change (i.e., random slopes). They also allowed for missing values in cognitive scores during follow-up, optimizing use of available cognitive data. We calculated 95% confidence intervals (95% CIs) for all models and performed linear tests of trend.

Given the possibility that various factors could modify the association between change in the QOL with retirement and subsequent cognitive function, we evaluated, in separate multivariable-adjusted models, the potential effect modification by seven factors: 1) age at retirement (60–64 versus 65–69 years), 2) delay between retirement and post-hoc assessment of change in QOL at retirement (<5 years versus ≥5 years), 3) work status as assessed in 1982 (working versus homemaker), 4) vitality (SF-36 Vitality score ≥50 versus <50), 5) mental health (SF-36 Mental Health score ≥53 versus <53), 6) physical functioning (SF-36 Physical Functioning score >30 versus ≤30) and 7) subjective memory complaint at first cognitive interview (which was appraised as the number of positive response(s) to seven items: change in memory, difficulties in remembering a short list of items, difficulties in remembering things from one second to the next, difficulties in remembering recent events, difficulties in understanding instructions, difficulties in following a conversation, difficulties in finding the way around familiar streets) that has been shown to be a risk factor for long-term cognitive decline [44, 45].

Finally, to reduce any potential bias due to reverse causation [16] where women may have first had impaired cognition that led to early retirement and/or a poorer subjective assessment of QOL after retirement, we conducted several restricted analyses excluding women who had the worst cognitive function at the initial assessment (defined as those in the worst 10% of the distribution or alternatively as those whose TICS were below 34), women who completed less than all four of the cognitive telephone interviews, women with >1 subjective memory complaints or women whose retirement age was below 65. To evaluate if QOL change with retirement and cognition may be mediated by depression, we also restricted the analysis to those without evidence of depression or severe depression symptoms as of baseline cognitive assessment (no antidepressant use and SF-36 Vitality score ≥50 and SF-36 Mental Health score ≥53).

All models were fitted by maximum likelihood method using the SAS software (SAS release 9.3, SAS Institute Inc., Cary, NC).

Correlation between post-hoc assessment of change in QOL and prospective pre-and post-retirement health-related QOL in a subsample

Post-hoc self-reported change in QOL at retirement was associated with prospectively-observed change in SF-36 score before and after retirement (Table 1). Whereas differences in change in SF-36 between the “worse” and the “same” groups were clear for both physical and mental health components (with less favorable SF-36 change in the worse group as expected), the difference in change in SF-36 between the “same” and “improved” groups were only in the mental health component. Across the QOL change categories, the relation with the physical component was not significant (p-value for the Jonckheere-Terpstra test = 0.22) while the relation with the mental component was significant (p < 0.001).

Factors associated with change in QOL at retirement

Compared to women reporting no change in QOL at retirement, those who reported worsened QOL showed clear differences in socioeconomic and health profile, with less favorable circumstances overall (Table 2): they showed less healthy behaviors (higher rates of smoking and obesity, lower levels of physical activity and diet quality), were more likely to have cardiovascular conditions or risk factors (myocardial infarction, hyperlipidemia, etc.), as well as mental health risk factors (caregiving burden, low social network, stress in daily life). On average, their SF-36 scores for vitality, mental health, and physical functioning were lower. In contrast, the same and improved groups were very similar as far as socioeconomic, behavioral, and health-related factors were concerned. Some slight differences occurred at the occupational/retirement level; in particular, women who reported an improved QOL at retirement were more likely to have been working as a nurse in 1982 (versus other work/homemaker) and had on average retired at an earlier age than the “same” group.

Age at retirement and cognitive decline

In our study sample (restricted to women whose retirement age was 60–69 years), no significant association between retirement age and cognitive decline was observed (p = 0.36), but in the larger sample of all women who retired before 1996 (n = 7,557), those who retired at age 65–69 experienced slower cognitive decline overall as compared to those who had retired earlier (at ages <50–64 years) (data not shown). Thus, in all analyses for this study, we adjusted for age at retirement (65–69 years versus 60–64 years).

Change in QOL at retirement and cognitive decline in later life

Among 4,926 women followed for a median of 6 years, the overall cognitive change during follow-up in the global composite score was –0.123 on average.

Basic and multivariable-adjusted models of differences in rate of cognitive decline showed very similar results. Based on model fit statistics, the multivariable-adjusted models significantly improved the fit to the data (Supplementary Table 1). In multivariable-adjusted models (Table 3), when compared with women who reported no change in QOL at retirement, women who reported improvement showed a significantly slower rate of decline in the global composite score (+0.011 standard unit/year; 95% confidence interval [0.004, 0.019]), the TICS (+0.041 point [0.007, 0.075]) and the verbal memory score (+0.012 standard unit [0.003, 0.021]). Fixed effects for intercept, time from baseline, QOL change at retirement and time by QOL change at retirement from the multivariable-adjusted model of global composite score are provided in Supplementary Table 2. Because mean differences in rate of cognitive decline can be difficult to interpret, particularly when z-scores were used, we compared these effect estimates to that for age and cognitive decline, thus using the effect of age on cognitive decline as a “benchmark” for interpreting the mean differences. The mean difference in cognitive rates was equivalent to that observed between women who were  2 years apart in age.

In contrast, no significant differences in cognitive decline rates were observed for the women who reported worsened QOL (Table 3) when compared to those who reported no change.

While the “worse” group was small, we observed significant trends of slower cognitive decline across the three categories of change in QOL, going from worse to same to better QOL with retirement for the global composite score and the verbal memory score.

Effect modification and restricted analyses

We observed no significant interactions with age at retirement, delay between retirement and post-hoc assessment of change in QOL at retirement, work status as assessed in 1982, vitality, mental health, physical functioning and subjective memory complaint at first cognitive interview.

Models limited to women with preserved cognitive function at baseline (in the top 90^th percentile of global score or alternatively with TICS ≥34), to women without evidence of severe depression symptoms, as well as to women with full cognitive follow-up yielded similar results as the primary analyses (Table 4). Among those who had no or little (≤1) subjective memory complaints and among those who retired aged ≥65 years, results were no longer significant, but the magnitude and direction of the estimates were virtually unchanged.

Conclusion

The present study is the first to investigate in a cohort sample cognitive decline in relation to change in QOL at retirement. Our results suggest that QOL improvement at retirement may contribute to maintaining cognitive function into old age. Thus, our study highlights the period of retirement, a very common life transition, as a critical window of opportunity to promote healthy ageing. Because all our participants were female and most worked as nurses, these findings should be replicated in other populations. If confirmed, they would support greater behavioral counseling (e.g., to improve diet or increase physical activity [57]) by health care providers or the implementation of programs (e.g., general wellness or exercise programs) for helping to maintain or enhance QOL throughout the retirement transition.