Late-life deficits in cognitive, physical and emotional functions, childhood intelligence and occupational profile: a life-course examination of the Aberdeen 1936 Birth Cohort (ABC1936)

Abstract

Objectives: the ‘triad of impairment’ phenomenon describes the co-occurrence of age-related cognitive, emotional and physical functioning deficits. We investigated how occupational profile and childhood intelligence contribute to the triad of impairment in late life.

Methods: we analysed data of a subsample of the Aberdeen Birth Cohort of 1936 (n = 346). Data were collected on participants' childhood intelligence, late-life cognitive ability, physical functioning, depressive symptoms and main lifetime occupation. We summarised the various occupational and impairment measures into two latent variables, ‘occupational profile’ and the ‘triad of impairment’. We used a series of data reduction approaches and structural equation models (SEMs) of increasing complexity to test both the validity of the models and to understand causal relationships between the life-course risks for the triad of impairment.

Results: occupational profile had a significant effect on the triad of impairment independent of childhood intelligence. Childhood intelligence was the predominant influence on the triad of impairment and exerted its effect directly and indirectly via its influence on occupation. The direct effect of childhood intelligence exceeded the independent influence of the occupational profile on impairment by a factor of 1.7–1.8 and was greater by a factor of ∼4 from the indirect pathway (via occupation).

Conclusions: childhood intelligence was the predominant influence on the triad of impairment in late life, independently of the occupational profile. Efforts to reduce impairment in older adults should be informed by a life-course approach with special attention to the early-life environment.

Introduction

Studying separate health and well-being factors in late life is difficult because of covariance between measures. Non-resilient ageing is characterised by ‘frailty’—a state leading to the accumulation of simultaneous disorders because of multisystem dysregulation [1]. Although frailty is a central concept in clinical assessment of older people, there is no consensus definition of frailty [2]. The concept is multifactorial, although physical components dominate. Although physical deficits are often accompanied by cognitive and emotional decline in older people, their co-occurrence is rarely investigated [3, 4] despite ongoing debate over the clinical need to understand cumulative deficits [5]. Cognitive decline [6] and depressive symptoms [7] independently contribute to frailty but are not yet incorporated into its definition [8]. Here, we explore the ‘triad of impairment’ (triad), a concept introduced in previous studies, as a useful alternative to ‘frailty’. Triad recognises the co-occurrence of cognitive, emotional and physical deficits in late life [3, 4].

Identification of pathways to prolong healthy living in old age will have benefits for individuals and society [9]. Employment is often undertaken until ∼70 years of age. Occupation, a major feature of socioeconomic circumstances, is determined partly by work experience, cognitive skills, education, socioeconomic opportunities and sociocultural constraints. Childhood intelligence (‘childhood-IQ’) is a major influence on educational achievements, work experience and opportunities for work-related training. Both factors are potentially modifiable in early life and, due to their accumulating impact over the life-course, may determine a delayed onset of functional decline.

Childhood-IQ and occupational characteristics are strong predictors of late-life health and therefore might constitute important factors in preventing the concurrent declines in cognitive, emotional and physical functions. First, different occupational characteristics independently influence cognitive [10], emotional [11] and physical [12] function, but few studies examine how the occupational profile influences concurrent impairments [13]. Second, the interplay between childhood-IQ and life-course socioeconomic factors in determining health outcomes in old age is demonstrated [14] but not in the context of triad. Of interest is the evidence for a positive link between childhood-IQ and late-life health and survival [15]. This relationship was not attenuated after adjusting for early socioeconomic circumstances [16, 17] and modestly attenuated after adjusting for adult socioeconomic position [17]. This suggests that life-course socioeconomic circumstances only partially account for the association between early-life intelligence and mortality. The question then arises as to whether a relationship exists between childhood-IQ, occupation and triad.

In a sample of adults approaching retirement age, we examined how the occupational profile, described by employment type, occupational complexity and occupational stress contribute to cognitive, emotional and physical functions in late life, over and above childhood mental ability, using a life-course approach. We used various levels of reduction within a structural equation model (SEM) that hypothesised that childhood-IQ is a direct influence on triad and an indirect influence via an individual's occupational profile (Figure 1).

Figure 1.

Hypothetical model (general terminology).

Methods

Aberdeen 1936 Birth Cohort (ABC1936)

ABC1936 was established to study brain ageing and health among participants who sat the Moray House Test in 1947 [15]. Of 496 people with occupational data, 352 had complete scores on the variables of interest collected at their first assessment visit. We excluded those with Hospital Anxiety and Depression Scale (HADS) [18] scores (HADS ≥ 11) and those whose occupation was coded as ‘housewife’ (n = 6). The analysis included 346 participants. A full description of ABC1936 including sources of attrition from the study is available [19]. The local Research Ethics Committee approved the ABC1936 programme.

Occupation variables

Primary lifetime occupation and years of full-time education were provided at the first visit. Lifetime ‘best-ever’ occupations were coded on a 1−9 scale, 1 representing highest occupational socioeconomic position [20]. Occupational complexity in terms of working with ‘data’ (Complexity Data), ‘people’ (Complexity People) and ‘things’ (Complexity Things) was derived from the Dictionary of Occupational Titles [21]. Occupational stress was classified using external scales [22].

Cognitive function tests

The childhood intelligence data from the Scottish Mental Survey (1947) were accessed from the Scottish Council for Research in Education [23]. These scores were standardised: z-scores were multiplied by 15 and added to 100 to create an IQ-like score. In late life, the Mini-Mental State Examination (M-MSE) was administrated and tests assessing four domains of cognitive function: (i) non-verbal reasoning [Raven's Standardised Progressive Matrices (RPM) [24]], (ii) spatial ability, [Block Design Test (BLK)[25]], (iii) mental speed [Digit Symbol Test (DS) [25]] and (iv) verbal memory [Rey's Auditory Verbal Learning Test (AVLT) [26]].

Physical and mental health variables

Time to perform a 6 m walk at usual pace was recorded in seconds, normalised for height and multiplied by 100 (WT). Participants completed the SF-36 Health Survey, which measures general health in nine domains including limitation in physical activities due to health problems [27]. Total raw scores for each domain were transformed as outlined in the SF-36 Health Survey Manual [28]. Depressive symptoms were scored using a self-rating scale (HADS).

Statistical analysis

If triad is true, then it should be possible to reduce the cognitive, physical and emotional measures to a single score reflecting the extent to which a person is impaired on all dimensions. If the occupational profile exists, then we should be able to describe all occupational variables with a single number. Therefore, for the highly correlated health and occupational variables, principal component analysis (PCA) was performed for data reduction to calculate a single representative value. We extracted the first un-rotated principal components, triad and occupational profile, which represented the shared variance among the health and occupational data. PCA allowed us to determine whether data reduction was appropriate, i.e. whether the measured variables were driven by the same underlying latent constructs.

To obtain the single summary score for triad, cognitive, physical and emotional functions were reduced in two steps. First, PCA was performed on the cognitive (RPM, BLK, DS, AVLT) and physical (WT, SF36) variables to extract the first un-rotated principal components ‘g’ and ‘pf’ for general intelligence and physical functioning, respectively, in SPSS (IBM SPSS, 21). Next, HADS, ‘g’ and ‘pf’ were subjected to PCA to extract the component triad. Similarly, facets of occupation were subjected to PCA to extract the component occupational profile. To test our main hypotheses, we used the SEM in AMOS (IBM AMOS 21). We assumed that occupational and health variables could be adequately summarised by two latent variables: occupational profile and triad. Each SEM included measurement errors of the indicators (e) and residual errors in the prediction of the unobserved factors (res). We first examined the hypothesised relationships using the most parsimonious model (Model 1), i.e. a model with as few parameters as possible for the most general explanation. To quantify triad in Model 1, HADS, ‘g’ and ‘pf’ were reduced to a summary score via the regression imputation technique in AMOS [29]. The occupational profile factor was calculated using the same approach. Childhood-IQ remained an observed variable. We then formulated a series of models with progressively added complexity (Supplementary data, Appendix A, available in Age and Ageing online).

Our goal was to show consistency in results across all models with different degrees of parsimony by demonstrating in each model a consistency in the strength and direction of each association and by showing that each model variation produces an acceptable fit to the data. In the least parsimonious model (Model 5), triad was calculated using raw scores (i.e. no data reduction) describing cognitive (RPM, BLK, DS, AVLT), emotional (HADS) and physical (WT, SF36) functioning. The occupational profile was calculated using raw scores including (1) complexity data, (2) complexity people, (3) complexity things, (4) profession and (5) employment stress.

The goodness of fit was assessed according to Hu and Bentler guidelines [30], which suggest combined values of CFI ≥ 0.95 with an RMSEA ≤ 0.06 indicating an acceptable fit of the model to the data. The models were tested first, and then modification indices [29] were examined and applied. To confirm the validity of the models between genders, we assessed critical ratios for sex differences between estimated parameters. All relationships were tested for significance at the α = 0.05 level (two-tailed). For each model, we obtained bootstrap estimates for the standard errors of the standardised direct, indirect and total effects of childhood-IQ on triad. The bootstrap approximation was obtained by constructing two-sided percentile-based confidence intervals (200 bootstraps samples, confidence level of 90%).

Results

Participant characteristics

The characteristics of the 346 included participants are given in Table 1. Female participants demonstrated significantly higher mean MMSE scores and lower occupational status. Compared with cases with incomplete data, complete cases had significantly higher childhood-IQ and higher occupational status, and performed better on RPM and DS tests at the α = 0.05 level, and based on median values, had higher Complexity Things and higher employment stress (Supplementary data, Appendix B, available in Age and Ageing online).

Table 1.

Descriptive summary of the sample (Complete cases)

Characteristics	Mean (SD) Median; IQR	Occupation	Mean (SD) Median; IQR	Cognition	Mean (SD) Median; IQR	Physical and emotional	Mean (SD) Median; IQR
Age	64.79 (0.83)	Occupational	4.56 (2.20)	RPM	36.66 (7.87)	WT (seconds)	3.05 (0.71)
	64.92; 1.17	Code [1−9]	4.00; 3.00		37.00; 11.00		2.96; 0.86
Education (years)	11.22 (2.04)	Data	2.95 (1.78)	BLK	25.22 (8.33)	SF36 [0−100]	77.11 (23.30)
	10.00; 1.00	Complexity [0−6]	3.00; 3.00		24.00; 11.00		85.00; 25.00
MMSE [0−30]	28.97 (1.38)	People	5.96 (1.81)	DS	44.65 (11.27)	HADS [0−21]	2.83 (2.16)
	29.00; 1.00	Complexity [0−8]	6.00; 1.00		44.00; 16.00		2.00; 3.00
MHT	43.95 (12.46)	Things	4.43 (2.59)	AVLT	59.40 (12.70)	TOI (Factor Score)	−0.64 (0.37)
	46.00; 15.00	Complexity [0−7]	4.00; 5.00		60.00; 17.00
		Occupational	2.82 (1.03)
		Stress [0−4]	3.00; 2.00
		OP (Factor Score)	1.57 (1.25)

Possible range of scores for a feature in square brackets. All indicators of the occupational profile (OP)/triad of impairment (TOI) reduced to a single factor score via the regression imputation technique in AMOS. IQR refers to the inter-quartile range.

MHT, Moray House Test; MMSE, Mini-Mental State Examination; RPM, Raven's Standardised Progressive Matrices; BLK, Block Design Test; DS, Digit Symbol Test; AVLT, Rey's Auditory Verbal Learning Test; WT, walk time normalised for height; SF36, SF-36 Health Survey; HADS, Hospital Anxiety and Depression Scale.

Data reduction

The component ‘g’ accounted for 55% of the total variance among cognitive domains, and each test had a comparable component loading: RPM = 0.833, BLK = 0.755, DS = 0.787 and AVLT = 0.566. The component ‘pf’ explained 71% of the total variance. The component triad accounted for 50% of the total variance, and each score had a comparable component loading: HADS = 0.697, ‘g’ = −0.640, ‘pf’ = 0.781. PCA extraction of the component occupational profile revealed that Complexity Things was almost entirely loaded on the second component. Thus, Complexity Things was excluded from the occupational profile component and considered a separate profile factor. After exclusion of Complexity Things, the component occupational profile accounted for 61% of the total variance with component loading: Complexity Data = 0.833, Complexity People = 0.711, profession = 0.795, employment stress = −0.789.

Structural equation modelling

Results for modified Models 1–3 are given in Table 2, for modified Model 4 in Table 3, and for modified Models 5–6 in Supplementary data, Appendix C, available in Age and Ageing online. Here, we present results of the model with best fit indices (Model 2). Descriptive reports for other models are in Supplementary data, Appendix D, available in Age and Ageing online.

Table 2.

SEM regression weights/covariance estimates and direct/indirect/total effects of childhood-IQ on latent TOI (Models 1–3)

Regression weights and covariancesa							Standardised direct, indirect and total effects of childhood-IQ on TOIb
Model no.			Estimate	S.E.	P	Std. estimate	Effect	Std. effect	Std. S.E	Lower bound CI	Upper bound CI	P
Model 1
Childhood-IQ	→	OP	−0.050	0.005	<0.001	−0.494	Direct	−0.623	0.027	−0.671	−0.578	0.010
Childhood-IQ	→	TOI	−0.019	0.001	<0.001	−0.623	Indirect	−0.170	0.019	−0.201	−0.139	0.010
OP	→	TOI	0.102	0.010	<0.001	0.344	Total	−0.793	0.020	−0.829	−0.761	0.010
Model 2
Childhood-IQ	→	OP	−0.041	0.004	<0.001	−0.496	Direct	−0.545	0.070	−0.669	−0.426	0.010
Childhood-IQ	→	TOI	−0.015	0.005	0.004	−0.545	Indirect	−0.157	0.032	−0.217	−0.105	0.010
OP	→	TOI	0.104	0.039	0.007	0.317	Total	−0.702	0.080	−0.851	−0.576	0.010
TOI	→	HADS	1.000			0.190
TOI	→	‘pf’	0.612	0.205	0.003	0.252
TOI	→	‘g’	−2.146	0.724	0.003	−0.881
e5	↔	e6	0.537	0.112	<0.001
e6	↔	res1	0.207	0.058	<0.001
Model 3
Childhood-IQ	→	OP	−0.041	0.006	<0.001	−0.446	Direct	−0.647	0.029	−0.695	−0.597	0.010
OP	→	Profession	1.011	0.099	<0.001	0.634	Indirect	−0.146	0.023	−0.184	−0.109	0.010
OP	→	occupational stress	−0.518	0.045	<0.001	−0.694	Total	−0.793	0.020	−0.829	−0.761	0.010
OP	→	People	0.766	0.078	<0.001	0.582
OP	→	Data	1.000			0.772
Childhood-IQ	→	TOI	−0.016	0.001	<0.001	−0.647
OP	→	TOI	0.088	0.011	<0.001	0.328
e4	↔	childhood-IQ	−7.795	1.652	<0.001

Model 1: all occupational measures and all indicators of the triad of impairment (HADS, ‘g’, ‘pf’) reduced to a single number (composite score) via the regression imputation technique in AMOS; Model 2: occupational profile represented by a composite score occupational profile and the triad of impairment by a latent variable triad; Model 3: occupational profile represented by a latent variable occupational profile and the triad of impairment by a composite score triad.

Covariances indicated as ‘↔’. Acronyms [the meaning of each variable's higher score]: OP, occupational profile [lower overall complexity]; TOI, triad of impairment [higher impairment]; RPM, Raven's Standardised Progressive Matrices [higher functioning]; BLK, Block Design Test [higher functioning]; DS, Digit Symbol Test [higher functioning]; AVLT, Rey's Auditory Verbal Learning Test [higher functioning]; WT, walk time [lower functioning]; SF36, SF-36 Health Survey [better functioning]; HADS, Hospital Anxiety and Depression Scale [more depressive symptoms]; ‘pf’, first un-rotated principal component for physical functioning [higher impairment]; ‘g’, first un-rotated principal component for general intelligence [higher ability]; e, ‘error’, i.e. measurement error of the indicator; res, residual (‘disturbance’).

^aThe unstandardised effect is significantly different from zero at the 0.05 level (two-tailed).

^bThe standardised effect is significantly different from zero at the 0.05 level (two-tailed). This is a bootstrap approximation obtained by constructing two-sided percentile-based confidence intervals.

Table 3.

SEM regression weights/covariance estimates and direct/indirect/total effects of childhood-IQ on latent TOI (Model 4)

Regression weights and covariancesa							Standardised direct, indirect and total effects of childhood-IQ on TOIb
Model no.			Estimate	S.E.	P	Std estimate	Effect	Std effect:	Std S.E	Lower bound CI	Upper bound CI	P
Model 4
Childhood-IQ	→	OP	−0.041	0.006	<0.001	−0.445	Direct	−0.583	0.061	−0.689	−0.484	0.010
OP	→	TOI	0.087	0.034	0.011	0.294	Indirect	−0.131	0.039	−0.203	−0.074	0.010
Childhood-IQ	→	TOI	−0.016	0.005	0.003	−0.583	Total	−0.713	0.071	−0.843	−0.596	0.010
OP	→	Profession	0.993	0.100	<0.001	0.629
OP	→	occ_stress	−0.509	0.045	<0.001	−0.690
OP	→	People	0.757	0.078	<0.001	0.581
OP	→	Data	1.000			0.780
TOI	→	HADS	1.000			0.191
TOI	→	‘pf’	0.758	0.237	0.001	0.312
TOI	→	‘g’	−2.085	0.693	0.003	−0.860
e5	↔	e6	0.559	0.115	<0.001
e4	↔	childhood-IQ	−7.890	1.663	<0.001

Model 4: both the occupation profile and the triad of impairment represented by two latent variables occupational profile and triad, respectively.

Covariances indicated as ‘↔’. Acronyms [the meaning of each variable's higher score]: OP, occupational profile [lower overall complexity]; TOI, triad of impairment [higher impairment]; RPM, Raven's Standardised Progressive Matrices [higher functioning]; BLK, Block Design Test [higher functioning]; DS, Digit Symbol Test [higher functioning]; AVLT, Rey's Auditory Verbal Learning Test [higher functioning]; WT, walk time [lower functioning]; SF36, SF-36 Health Survey [better functioning]; HADS, Hospital Anxiety and Depression Scale [more depressive symptoms]; ‘pf’, first un-rotated principal component for physical functioning [higher impairment]; ‘g’, first un-rotated principal component for general intelligence [higher ability]; e, ‘error’, i.e. measurement error of the indicator; res, residual (‘disturbance’).

^aThe unstandardised effect is significantly different from zero at the 0.05 level (two-tailed).

^bThe standardised effect is significantly different from zero at the 0.05 level (two-tailed). This is a bootstrap approximation obtained by constructing two-sided percentile-based confidence intervals.

Model 2: reduced occupational profile and latent triad. After an initial analysis, the modification indices indicated that a significantly better fit would be obtained if correlations between the error terms e5 and e6 and subsequently between res1 and e6 were introduced. Correlating e5 and e6 is plausible and suggests that individual's mental and physical health may be influenced by an unmeasured common factor. This correlation occurs after adjusting for the common variable, which means that the two indicators are strongly associated in addition to triad, explaining their individual variances. The covariance between res1 and e6 suggests that the variance in the occupational profile not explained by childhood-IQ and the variance in physical functioning not explained by the latent variable triad are correlated. An example of an unmeasured variable explaining this covariance might be a health condition that limits one's occupational opportunities and contributes to frailty in late life. After addition of these correlations, the final model met the pre-specified criteria, suggesting an excellent model fit. None of the critical ratios for differences between parameters differed significantly between genders. All causal relationships hypothesised by this model were significant (P < 0.05). A reduced occupational profile had a direct effect on latent triad with a standardised regression weight (β) of 0.317. Childhood-IQ had a total β of 0.702; of which, 0.545 was direct and 0.157 indirect (Table 2).

Overall, all causal relationships hypothesised by Models 1–5 were significant. No relationships between the individual indicators over and above any effects of latent variables were found. Therefore, there was no evidence of measurement-specific relationships not explained by the latent summary scores. Childhood-IQ was the predominant influence on triad both directly and indirectly via the occupational profile. Its direct effect exceeded the influence of the occupational profile on triad by a factor of 1.7–1.8 and was greater by a factor of ∼4 from the total indirect pathway (smallest 3.47, highest 4.45). A more complex occupational profile had a smaller but significant effect on triad, independent of childhood-IQ. Complexity Things (Model 6) did not significantly predict triad in late life (P = 0.544) (Supplementary data, Appendix C, available in Age and Ageing online).

Discussion

We found that childhood-IQ was the predominant influence on triad in late life, described by contemporaneous cognitive, emotional and physical deficits. Primary lifetime occupations with greater complexity associated with working with data or people, higher social status and employment stress in mid-life, though less influential than childhood-IQ, also had a significant effect on triad independently of childhood-IQ. All models with differing degrees of parsimony confirmed this relationship, suggesting that the data reduction to summarise the occupational profile and triad is a reasonable approach in this context and that the latter may provide a useful alternative to a frailty measure. A major implication is that clinical studies should consider a life-course perspective (with a focus on early life) to accurately assess and to improve the health of older adults.

Plausible mechanisms underlying the influence of the occupational profile on triad in late life merit further study. Occupational complexity associated with working with things had no effect on triad and did not explain any variance. Complexity in terms of data, which reflects the levels of cognitive involvement and information processing, had the highest load on the occupational profile. Overall, the most significant occupational factors that affect triad are associated with cognitive endeavour. Studies suggest that maintaining brain health might constitute crucial pathways in preventing age-associated cognitive [31], emotional [32] or physical [33] deficits, whereby triad's components depend on brain lesion location [34]. Also, occupational characteristics mitigate the negative influences of neuropathology in non-demented [35] and demented populations [36]. Our findings suggest that more mentally demanding jobs are beneficial to cognitive, physical and emotional health into late life. Therefore, job re-design aimed at challenging the brain might offer an opportunity to address functional deficits associated with ageing. Importantly, childhood-IQ had a predominant effect on triad in late life, emphasising the importance of successful cognitive development. Brain structure in early- [37] and late [38] life is influenced by childhood socioeconomic status pointing to the importance of early-life environment such as better nutrition, access to health care and education. Further research should establish whether the occupational profile acts as a moderator in the relationship between brain burden and triad, and how the developing brain influences this relationship.

Major strengths of our study include the availability of childhood-IQ in a narrow-age cohort of older people, prospectively collected life-course data and the careful statistical testing using a series of data reduction and SEM. Despite the relatively small sample size, childhood-IQ has the predominant influence on triad in late life independently of the occupational profile, and this emphasises the need for including prospectively collected measure of pre-morbid intelligence in life-course models. The occupational profile latent variable captured a wide range of economic and psycho-social aspects of the occupational environment. However, we did not consider other characteristics of early-life adversity, and we expect that people with access to more resources in childhood might have higher childhood-IQ. It is unclear how best to capture reliable descriptors of the occupational profile and triad. We performed data reduction on physical functioning variables consisting of self-reported measures, but even when the data fit the model, data interpretation may be problematic [39]. Lastly, we could not dissociate the potential negative impact of retirement on late-life cognition [40] from the effects of the occupational profile. To our knowledge, this is the first study to examine the influence of the occupational profile and childhood-IQ on triad and requires validation across other cohorts.

Key points.

• A ‘triad of impairment’ recognises the co-occurrence of cognitive, emotional and physical functioning deficits in late life and may provide an alternative to a frailty measure.

• We have demonstrated that childhood intelligence is the predominant influence on the triad of impairment in late life.

• Occupations with greater complexity, though less influential than childhood intelligence, also had a significant independent effect on the triad of impairment.

• The results suggest that efforts to improve health in older adults should be designed with a life-course perspective and special attention to the early-life environment.

Supplementary data

Supplementary data mentioned in the text are available to subscribers in Age and Ageing online.

Conflicts of interest

None declared.

Funding

This work was supported by the PhD studentship Pathways to a Healthy Life, University of Aberdeen (2013–2016), and Pump Priming Grant, University of Aberdeen (2013). The Aberdeen 1936 Birth Cohort studies were supported by Biotechnology and Biological Sciences Research Council (1999–2002); Scottish Health Department (2000–2001); Wellcome Trust (2001–2006); Medical Research Council (2002–2003); Alzheimer Research UK (2003–2006); Alzheimer Society (2006–2008) and University of Aberdeen Development Trust (2006–2009) contributions. C.B. was supported by The Farr Institute @ Scotland. The Farr Institute @ Scotland is supported by a 10-funder consortium: Arthritis Research UK, the British Heart Foundation, Cancer Research UK, the Economic and Social Research Council, the Engineering and Physical Sciences Research Council, the Medical Research Council, the National Institute of Health Research, the National Institute for Social Care and Health Research (Welsh Assembly Government) and the Chief Scientist Office (Scottish Government Health Directorates), the Wellcome Trust, (MRC Grant No: MR/K007017/1).