Introduction

In the last decades, a growing body of literature has reported a protective association between several proxies of cognitive reserve and the development of dementia among elderly people. In particular, educational attainment[1], active lifestyle (i.e., social, mental, physical engagement)[2–5], bilingualism[6] and complex occupation[7,8] have been repeatedly associated with decreased risk of developing Alzheimer’s disease (AD) and with delayed onset of AD.

The professional setting can be perceived as an “enriched” environment given that, for many individuals, it provides an important source of social support and cognitively stimulating activities, both of which are considered to contribute to cognitive reserve. Leaving the work setting is therefore a major life course transition that causes substantial changes in one’s life and may affect cognitive functioning. This outcome was reported in several studies showing that retirement has an impact on cognition of older adults[9–11]. However, to our knowledge only two recent studies addressed this issue in the context of pathological aging and reported that retirement age is associated with both the risk of developing dementia[12] and the onset of AD[13]. The first ever study in this area was the study conducted by Lupton and colleagues, which revealed that each additional year of employment delayed the onset of AD symptoms by 0.13 years. However, although interesting, this study has major limitations. First, the female population (N = 938) was excluded from the analyses, which considerably reduced the sample size (N = 382) and impeded generalization of the results. A second limitation is that it is still unclear whether the age at onset of AD was based upon a subjective parameter (i.e., reported age at onset of symptoms) or upon a more objective one (i.e., age at diagnosis). Moreover, their study only considered individuals who had already been diagnosed without considering the normal aging population. In addition, they attempted to address the reverse causality issue (i.e., to limit the possibility that people may have retired due to prior cognitive impairment) by excluding individuals, who were diagnosed before retirement. However, due to the particularly long prodromal phase of AD, which lasts more than a decade[14], this strategy is not sufficient to discard the possibility that patients may have been in a prodromal phase of AD that inclined them to retire long before the diagnosis of dementia was made. In other words, the causality issue is not completely resolved. Finally, the selection strategy used by Lupton and colleagues (i.e., considering only AD patients and excluding those who retired after the disease onset) generates a selection bias that leads to an overestimation of the effect of retirement on the onset of AD.

The selection bias observed in Lupton et al.’s study[13] is close to the “healthy worker effect”[15]. The healthy worker effect describes a continuing selection process such that individuals who remain employed tend to be healthier in comparison to those who are retired. In the context of Lupton et al.’s study, the selection bias stems from the fact that as all subjects had already been diagnosed with the disease, and that those who had been diagnosed before retirement were excluded from the analyses, subjects who took later retirement were necessarily those who were diagnosed at a later age. In more concrete terms, an individual who retired at 60 years of age could only be diagnosed after this age, whereas an individual who retired at 80 years of age could only be diagnosed after the age of 80.

In this context, the aim of the present paper was to study the association between retirement age and age at onset of AD (i.e., onset of first symptoms and diagnosis of AD) by taking into consideration the two biases (i.e., the selection bias and the bias of the reverse causality) that may explain the results in favor of the cognitive reserve hypothesis. To that end, we first examined whether deferred retirement is associated with delayed age at onset of AD in the whole sample (i.e., without the inclusion of patients who developed dementia before retirement). Then, we tested whether this association persists when the two biases are taken into account. In order to control for selection bias, we restricted our analyses to subjects who retired at or before 65 years of age. To address the issue of reverse causality, we re-ran our analyses on a subsample of patients who retired at or before 65 years of age and were diagnosed after the age of 75.

Materials and Methods

Results

Characteristics of the sample

Of the 1,380 AD cases, the selection procedure has led to a study sample of 815 patients (see Fig. 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart of participant selection for the analysis.

https://doi.org/10.1371/journal.pone.0115056.g001

The characteristics of the sample (e.g., sociodemographic variables, health status variables) and baseline differences between AD patients discarded from the analyses (n = 565) and those who remained in the study sample (n = 815) are shown in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Study sample characteristics.

https://doi.org/10.1371/journal.pone.0115056.t001

The number of years of education, the frequency of diabetes, hypertension, stroke, household income and depression were not significantly different in the two groups. However, excluded subjects were younger, retired at a younger age (25% of the excluded subjects who were professionally active retired before the age of 50), had a slightly lower MMSE score at inclusion and had a higher frequency of hypercholesterolemia but a lower rate of ischemic heart disease. They also had a lower age at diagnosis and a lower age at onset of first symptoms, which is expected since those excluded refer to young cases of AD who were diagnosed before retirement (14.6% of those excluded developed dementia before retirement and 9.8% before turning 60) (see Table 1).

Of the 815 patients, 383 were male (47%). Mean retirement age was 61.3 years (sd = 5.1). Age at diagnosis and age at onset of symptoms followed a Gaussian distribution with a mean age of, respectively, 77.1 (sd = 6.2) and 74.9 (sd = 6.2). Mean MMSE score at the first clinic appointment was 20.6 (sd = 3.8).

Association between retirement age and age of onset of AD

Step 1. Fig. 2A displays the age at diagnosis as a function of the age at retirement in the sample of 815 subjects. A positive association was observed between the age at retirement and the age at diagnosis (β = 0.33, p < 0.0001) and a similar association was found for the age at onset of first symptoms (β = 0.30, p < 0.0001). Controlling for potential confounding factors did not change the intensity of the associations: deferred retirement was associated with delayed age at diagnosis (β = 0.31, p < 0.0001) and delayed age at onset of symptoms (β = 0.30, p < 0.0001). In other words, each additional year of employment delayed the age at onset of symptoms by 0.30 years and the age at diagnosis by 0.31 years. Fig. 2B also shows that due to the selection of subjects who developed dementia after retirement, no data are observed in the lower right-hand side of the chart, suggesting that the subjects who retired later in life influence part of the observed association. As mentioned earlier, a patient who retired at 60 years of age could only develop AD after this age whereas a patient who retired at 80 years of age could not develop AD before turning 80. Therefore, the patient who retired later necessarily has a greater probability of being diagnosed at a later age in comparison to the patient who retired sooner.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Distribution of the age at diagnosis of Alzheimer’s disease according to age at retirement (N = 815).

https://doi.org/10.1371/journal.pone.0115056.g002

Step 2. In order to limit this selection bias, we selected a subsample of 637 subjects who retired at or before the age of 65 and who did not develop AD before the age of 65, thus restricting the analysis to the heart of the scatterplot (see shaded area in Fig. 2C). This way, we removed the influence of subjects who retired later (these individuals are also those who developed dementia at a later age). In the multivariate analyses, the age at retirement was still associated with the age at diagnosis, even though the association was weaker (β = 0.15, p = 0.004). A similar result was found with the age at first symptoms (β = 0.15, p = 0.007).

Step 3. A last analysis was performed in order to determine whether the association between retirement age and age at onset of AD could be explained by a reverse causality phenomenon. Among the 447 subjects who retired before the age of 65 and who developed AD at least 10 years after retirement (see shaded area in Fig. 2D), the association between retirement age and age at diagnosis was weaker (β = 0.06) and no longer significant (p = 0.18).

Discussion

The association between retirement age and age at onset of AD (i.e., onset of symptoms and diagnosis) was examined using data collected from baseline information in a sample of 815 patients with probable AD from the longitudinal observational European ICTUS study. Our first aim was to carry out the analyses using a similar strategy to that of Lupton and colleagues[13] (i.e., the first study that addressed this issue in pathological aging). Consequently, our analyses were conducted on AD patients who were diagnosed after retirement. The empirical results highlight that each additional year of employment delayed the age at onset of symptoms by 0.30 years and the age at diagnosis by 0.31 years. Such results were obtained after controlling for gender, education, occupation, country, clinical center, income household, and prior history of diabetes, hypercholesterolemia, hypertension, ischemic heart disease, depression and stroke. These results are consistent with those of Lupton and colleagues[13], as they show that each additional year of employment delayed the onset of dementia symptoms by 0.13 years. In Lupton et al.’s study, the exclusion of patients who were diagnosed before retirement led the authors to assume that it is retirement that has an impact on the onset of AD and not the other way around. Nonetheless, this modality of exclusion of patients is not sufficient to exclude the possibility that patients may have had cognitive impairment that caused them to retire (i.e., reverse causality).

Furthermore, this strategy of patient selection (i.e., AD patients who were diagnosed after retirement) creates a selection bias that leads to an overestimation of the effect of retirement on the onset of AD. In Lupton et al.’s study[13], this selection bias stems from the fact that due to their selection of patients, those who retired later were necessarily those who developed AD at a later age.

Within this context, the second aim of this paper was to replicate similar analyses, taking into account these two biases (i.e., the selection bias and the bias of the reverse causality).

To this end, we first took into consideration the selection bias by restricting our analyses to subjects who retired before the age of 65 and did not develop AD before 65 years old. By doing this, we were still able to find a significant association between retirement age and age at onset of AD, but to a lesser degree. More specifically, we found that each additional year of employment delayed the age at onset of diagnosis and the age at first symptoms by 0.15 years. Therefore, taking into account the selection bias reinforces the hypothesis of an association between deferred retirement and further age at onset of AD, as observed in Lupton et al.’s study[13]. However, even though this association persists and remains significant, it is still plausible that patients left the workforce due to prior cognitive impairment, with several studies showing that the prodromal phase of AD lasts more than a decade[14,17,18]. Hence, the same analysis was conducted after exclusion of individuals who were diagnosed within 10 years after retirement. This latter analysis shows that the association between retirement age and age at onset of AD is reduced by half and is no longer significant. Although a lack of power could be argued due to the considerable reduction of our sample size, it still comprises 447 subjects. This finding highlights that the conclusion of Lupton and colleagues[13] is probably overstated. Indeed, the selection of AD patients who developed the disease before retirement (in order to ensure that people did not leave the workforce due to prior cognitive impairment) only leads to an automatic association between a later age at retirement and delayed onset of AD. As a result, it was critical to use similar design in order to show that this selection strategy induces a selection bias, which leads to an overestimation of the effect of retirement. Thus, our study clearly points out that caution is needed when studying the effect of retirement on AD patients’ health outcomes. Based on these observational data, it is still unclear whether it is retirement that has an impact on the onset of AD or whether it is cognitive impairment that inclines individuals to retire.

Nevertheless, several studies in “healthy” aging suggested that it is retirement as such that has an impact on the cognitive functioning of elders[9,10,19]. First, a country level analysis highlighted that cognitive score of elders was better in countries in which eligibility age for retirement benefits tended to be higher (e.g., 65 years of age in Sweden) compared to those where it tended to be lower (e.g., 60 years of age in France). Given that countries do not set the eligibility age for retirement based on the cognitive vitality of individuals, the authors argue that this finding is a strong argument in favor of the role of retirement on cognition. Second, longitudinal data from the Health and Retirement Study (HRS) shows that the curves of the rate of retirement and memory decline can be overlaid, with a lag of about one year. For instance, while the rate of retirement in the United States is higher at 62, 65, and 66 years of age, a significant cognitive decline with a lag of about one year at the ages of 63, 66, and 67 is observed. These data suggest that retirement is not a consequence of prior cognitive decline, but rather that departure from the workforce causes a decrease in cognitive performance with a delay of about one year post-retirement.

Although these studies highlight that retirement plays a role in cognitive functioning of older adults, it is indisputable that in the context of pathological aging, there is a lack of empirical evidence. Further prospective studies (including both subjects with dementia and “healthy” older adults) are thus mandatory in order to better address the effect of retirement on the onset of AD. Before concluding that deferred retirement has a protective effect on the aging population’s health, some issues have to be addressed. For instance, future studies should seek to determine whether the association between retirement and cognition is direct and/or whether some intermediate variables such as social networks or depression influence this association. Indeed, work is known to increase social interaction and sense of self-efficacy[20], both variables associated with cognitive functioning of elders[21,22] and decreased risk of dementia[5,23]. Above all, further studies should test whether the impact of retirement on cognitive functioning depends on different types of retirement (e.g., voluntary versus involuntary departure), on participation in cognitively stimulating activities, and/or on occupational activity undertaken while employed (e.g., physical versus intellectual work). For instance, some studies have reported that cognitively stimulating occupation during adulthood and active lifestyle are associated with lower age-related cognitive decline[24,25] and also with lower risk of developing dementia[4,7,26]. Nonetheless, a longitudinal study found differential cognitive trajectories prior and following retirement in individuals with high complexity of work. Specifically, high complexity of work with people was associated with better verbal performances during employment followed by a faster rate of cognitive decline after retirement on spatial skills[27]. Therefore, further studies should seek to better understand the relationships between complexity level of occupation during adulthood and retirement and its influence on subsequent trajectories of cognitive decline.

In conclusion, our results support that, in the context of pathological aging, the association between retirement age and age at onset of AD requires further studies, with prospective design, to better assess the impact of retirement on AD patients’ health.

Acknowledgments

The authors would like to thank the physicians who recruited the patients and the multidisciplinary team (e.g., physicians, neuropsychologists, neurologists) who assessed them. Special thank are also provided to the ICTUS/DSA group who provided the database.

ICTUS Study Group: Vellas B., Reynish E., Ousset PJ., Andrieu S. (Toulouse), Burns A. (Manchester), Pasquier F. (Lille), Frisoni G.(Brescia), Salmon E. (Liège), Michel J.P., Zekry D.S. (Geneva), Boada M. (Barcelona), Dartigues J.F. (Bordeaux), Olde-Rikkert M.G.M. (Nijmejen), Rigaud A.S. (Paris), Winblad B. (Huddinge), Malick A., Sinclair A. (Warwick), Frölich L. (Mannheim), Scheltens P. (Amsterdam), Ribera C. (Madrid), Touchon J. (Montpellier), Robert P. (Nice), Salva A. (Barcelona), Waldmar G. (Copenhagen), Bullock R.(Swindon), Costa-Tsolaki M. (Thesaloniki), Rodriguez G. (Genoa), Spiru L. (Bucharest), Jones R.W. (Bath), Stiens G., Stoppe G. (Goettingen), Eriksdotter Jönhagen M. (Stockholm), Cherubini A. (Perugia), Lage P.M., Gomez-Isla T. (Pamplona), Camus V. (Tours), Aguüera-Morales E., Lopez F. (Cordoba). DSA Group: Andrieu S., Savy S., Cantet C., Coley N.

Eric Salmon and Stéphane Adam are members of the European Alzheimer’s Disease Consortium, Cognitive Task Force