Study design

This study used data from wave 6 of ELSA (2012–2013) to derive an IC score, and Mini-mental State Examination (MMSE) scores [12], provided by the ELSA-HCAP sub study (2018) to investigate whether global cognitive function can be predicted by IC measured between 5 and 6 years previously using linear regression models.

DatasetsELSA study

ELSA is a population-based longitudinal study of individuals aged 50 and over living in the UK. The original dataset contained 11,578 households and 18,813 individuals, and the ELSA dataset was refreshed at waves 3, 4, 6, 7 and 9 using respondents from the HSE [13].

ELSA-HCAP sub-study

The ELSA Harmonised Cognitive Assessment Protocol (ELSA-HCAP) is a sub-study of ELSA, implemented between waves 8 and 9 [14]. This sub-study aimed to measure cognitive impairment and dementia prevalence of the older ELSA participants.

Participants

In the current study, participants were selected from ELSA wave 6 (2012–2013). Participants were excluded if they had not taken part in the ELSA-HCAP and had been diagnosed with dementia or Alzheimer’s disease by wave 6. Participants without nurse data available were also excluded. Only participants aged 60–89 years were included to avoid missing data for gait speed, which was not available for ≥ 90 years (n = 6). Note also that according to ELSA codification, all participants aged ≥ 90 years are collapsed into one nominal category, limiting the possibility to draw accurate conclusions when age is considered a continuous variable. Participants aged < 60 (n = 5) were excluded according to the definition of older adults by the WHO (aged 60 or over). Lastly, participants who did not have data for all the IC variables available were excluded from the study. The final study involved 731 participants (see Fig. 1 for further details).

Fig. 1

Sample selection criteria

Ethical statement

All participants gave written consent prior to the ELSA study. ELSA wave 6 received ethical approval from the NRES Committee South Central—Berkshire on 28th November 2012 (11/SC/0374). For the ELSA-HCAP sub-study, ethical approval was given from the South Central-Berkshire National Health Services (NHS) Research Ethics Committee and was conducted in accordance with the ethical standards of the Declaration of Helsinki. Informed verbal consent was obtained from all participants or their guardians [14].

Intrinsic capacity indicators and domains

To create a comprehensive IC score, 12 of the variables measured in ELSA were selected to represent the five domains: cognition, locomotion, psychological wellbeing, sensory and vitality. Domains, indicators and cut-off points used in the item response theory (IRT) method are summarised in Table 1.

Table 1 Intrinsic capacity domains, indicators and summary scoresCognition domain

Temporal orientation and delayed recall variables were used. Temporal orientation involved asking the participant to orientate themselves regarding year, month, day and time [8]. A score of less than 4 (maximum score) was considered impaired [8, 15]. Delayed recall was assessed using lists of nouns given aurally to the participant who must recall as many as possible from memory after a short delay [6]. The bottom tertile of scores was considered impaired [8].

Locomotion domain

Gait speed, the Balance test (part of the Short Physical Performance Battery) and the Chair-Stand test were used. Gait speed was measured as the time taken in seconds to walk a certain distance (2.4 m) at the participant’s regular pace [6]. A score of less than 0.8 m/s was recorded as impaired [8, 16]. Static balance was measured using balance tests from the Short Physical Performance Battery. These included a side-by-side stand, a semi-tandem stand and a full-tandem stand [6]. A cut-off score of less than 4 was classed as impaired [8]. The Chair-Stand test was measured as the time taken to repetitively rise from and sit down in a chair 5 times without the participant using their hands. The time taken to reach full standing position was recorded in seconds [6]. Five rises in more than 16.7 s were used as a cut-off score and considered impaired [8].

Psychological domain

Sleep, depression and satisfaction with Life were included in the psychological domain. The Jenkins Sleep Problem Scale [17] was used to measure sleep and involved questions such as “how many times during the last month had the participant experienced trouble falling asleep”. This was reported in categorized answers [17], and a score of over 6 was considered impaired. Depression was assessed using a self-reported depression scale with 8 items [18] where a score of over 3 was considered impaired [8, 19]. Satisfaction with Life was measured using the Satisfaction with Life Scale [20]. This involved a self-reported questionnaire asking questions such as “in most ways my life is close to my ideal”. Answers are categorized into 7 categories from strongly disagree to strongly agree, and overall scores range from extremely satisfied to extremely dissatisfied [20]. Following a previous approach, the scale was reversed for higher scores to represent lower satisfaction levels, setting a cut-off score of over 20 points for impairments [8].

Sensory domain

Sensory was assessed through Eyesight and Hearing. Eyesight was measured via self-reported questions such as “how good is your eyesight for seeing things at a distance?”, with categorized responses ranging from 1 (excellent) to 5 (poor) [6]. A score of 4 or 5 was considered impaired [8]. Hearing was also assessed via self-reported questions, where participants were asked to rate their hearing in categorized responses [6]. As before, a score of 4 or 5 was considered impaired [8].

Vitality domain

Vitality was assessed through Forced Expiratory volume (FEV) and Hand Grip Strength. Forced expiratory volume was the maximum volume of air expired by the participant in 1 s using a spirometer [6]. The bottom quartile of results was considered impaired. Hand grip strength was measured using a handheld dynamometer [6]. Each hand (dominant and non-dominant) was trialled 3 times and averaged. The value from the dominant hand was used [6]. A score of less than 30 kg in men or less than 20 kg in women was considered as impaired [8, 16].

Intrinsic capacity scores derivationZ-score algorithm

The continuous nature of the variables available for each of the domains facilitated the derivation of an IC using z-scores, where a higher score indicated higher intrinsic capacity. Measurements for eyesight, hearing, chair stand, gait speed, sleep, depression and satisfaction with life were reverse coded. z-scores of each raw variable were calculated and z-scores belonging to each domain were then summed and averaged. This value was taken to be the overall score of that domain, and all domain scores were finally summed to create an overall IC z-score, giving equal weighting to all domains.

Item response theory (IRT) algorithm

A second algorithm to calculate the IC was implemented using cut-off scores for each of the 12 variables to create dichotomous indicators. Impairment for each variable was coded as 0, whilst capacity was coded as 1 (see details in Table 1). Using these dichotomous variables, an Item Response Theory (IRT) model was applied to calculate a value for the underlying trait for each participant, labelled as intrinsic capacity [21]. Specifically, a 2-parameter logistic model was applied to capture both the difficulty and discriminations of each item.

Global cognition (GC)—Mini-mental State Examination (MMSE)

The MMSE is a widely used screening test to assess global cognitive function [22]. It consists of 30 questions, with total scores comprising integer values between 0 and 30, where higher values indicate better cognition [12]. From the MMSE total score, it is also possible to categorize the global cognitive status of individuals ranging from normal to severe impairment [23], with high sensitivity for moderate to more severe cognitive decline [24]. Given the lack of normality and ceiling effects of MMSE scores, associations with IC were assessed against these previously defined global cognitive status categories: Normal = MMSE scores between 30 and 24, Mild impairment = MMSE scores between 23 and 19, Moderate impairment = MMSE scores between 18 and 10, Severe impairment = MMSE scores between 9 and 0 [23]. Moreover, given MMSE is generally used as a cognitive screening to differentiate normal vs impaired global cognition, an approach using established categories can provide more precise information regarding relevant cognitive changes that the continuous scale cannot detect.

Covariates

Sociodemographic and lifestyle covariates included age (measured at the time of the ELSA-HCAP assessment), sex, education, smoking and frequency of engagement in physical activity. Diagnosed conditions of lung disease, cancer, Parkinson’s disease, psychiatric conditions, heart attack (including myocardial infarction and coronary thrombosis), brain stroke, hypertension, diabetes and high cholesterol levels were dichotomously scored as present/diagnosed = 1, or absent = 0, to obtain a summed score of comorbidities frequently associated with age-related decline. Given the wealth of evidence supporting a strong correlation between socioeconomic status (SES) and educational level in the English population that has been maintained since the beginning of the last century to date [25], these covariates were not included to avoid multicollinearity issues. Moreover, this trend is confirmed by the 2024 Organisation for Economic Co-operation and Development (OECD) report, showing higher educational levels positively correlated with SES and income [26].

Statistical analysis

The two versions of IC scores were derived through the R Statistical Software (v4.3.2) [27]. For the 2-parameter IRT model IC scores, the ltm package was used [28]. Statistical analyses were conducted using IBM SPSS Statistics v.27.0 [29].

To select the best IC scoring algorithm, each version of IC scores was regressed separately on age, sex and education, given previous evidence reporting strong associations to these covariates [6, 30]. For this analysis, age at the time of IC indicators assessments was considered. Given that the MMSE test contains similar questions to those included within the IC cognition domain, the MMSE score was also included as an additional covariate to evaluate potential collinearities with the IC algorithms. Model assumptions were examined following recommended practice [31, 32].

Overall model fit and model selection were determined by the highest adjusted R2, lowest Bayesian Information Criterion difference (ΔBIC) > −10 [33], the highest log-likelihood value and greater effect size (Cohen’s f2) [34].

After algorithm selection, the association between IC and GC categories was assessed through a Multinomial Logistic Regression (MLR) model, with the Normal category set as reference. Since only two participants were categorized with severe impairment, this category was collapsed into moderate to severe impairment. Age at time of MMSE assessment, sex, education, smoking status, frequencies of mild, moderate and vigorous physical activities and the number of comorbidities were included as covariates.

To assess the individual effects of IC and all the covariates included in the fully adjusted model, bivariate logistic regressions were conducted between GC categories. To discard potential lack of statistical power leading to false negative results in effect of IC on the less frequent GC category (moderate to severe), an additional fully adjusted MLR analysis was performed collapsing mild and moderate to severe GC categories into impaired GC.

Multiple comparison analysis of IC between cognition categories was evaluated by the Kruskal–Wallis test for independent groups, followed by the Bonferroni corrected Dunn’s test for multiple comparisons (α = 0.05).