Risk Factors of Progression to Cognitive Frailty: Singapore Longitudinal Ageing Study 2

Abstract

Introduction

Cognitive frailty is associated with higher risk of dementia and adverse health outcomes. However, multidimensional factors that influence cognitive frailty transitions are not known. We aimed to investigate risk factors of incident cognitive frailty.

Methods

Prospective cohort study participants were community-dwelling adults without dementia and other degenerative disorders and baseline and follow-up, including N = 1,054 participants aged ≥55 free of cognitive frailty at baseline, with complete baseline (March 6, 2009, to June 11, 2013) and follow-up data at 3–5 years later (January 16, 2013, to August 24, 2018). Incident cognitive frailty was defined by one or more criteria of the physical frailty phenotype and <26 of Mini-Mental State Examination (MMSE) score. Potential risk factors assessed at baseline included demographic, socioeconomic, medical, psychological and social factors, and biochemical markers. Data were analyzed using least absolute shrinkage selection operator (LASSO) multivariable logistic regression models.

Results

A total of 51 (4.8%) participants, including 21 (3.5%) of the cognitively normal and physically robust participants, 20 (4.7%) of the prefrail/frail only, and 10 (45.4%) of cognitively impaired only, transited to cognitive frailty at follow-up. Risk factors for transition to cognitive frailty were having eye problem (OR = 2.6, 95% CI 1.24–5.43) and low HDL cholesterol (OR = 4.1, 95% CI 2.03–8.40), while protective factors for cognitive frailty transition were higher levels of education (OR = 0.3, 95% CI 0.10–0.74) and participation in cognitive stimulating activities (OR = 0.4, 95% CI 0.17–0.82).

Conclusion

Multi-domain modifiable factors especially related to leisure activities predict cognitive frailty transition and may be targeted for prevention of dementia and associated adverse health outcomes.

Introduction

Frailty, commonly defined as the physical phenotype criterion of wasting, weakness, slowness, exhaustion, and physical inactivity, is associated with adverse health outcomes, including falls, functional disability, hospitalization, institutionalization, and mortality [1–3]. Interestingly, physical frailty has been shown to be associated with increased prevalence and incidence of cognitive impairment [4], which is independently associated with poorer health outcomes, including lower quality of life (QOL) and depression [5, 6]. Hence, cognitive frailty is proposed as a heterogenous clinical manifestation, characterized by a simultaneous presence of both physical frailty and cognitive impairment in older adults without dementia [7]. Cognitive frailty is associated with considerably higher risk of dementia and other adverse health outcomes, including increased prevalence and incidence of functional disability and poor QOL, and increased mortality than either physical frailty or cognitive impairment alone [8, 9].

Cognitive frailty is a dynamic state, in which transitions are influenced by multidimensional factors, such as social factors and lifestyle behavioral factors [10]. Studies aimed at better understanding the underlying risk factors and predictors of the onset and progression of cognitive frailty among older adults provide important insights and opportunities for early interventions to prevent cognitive frailty and reduce healthcare burden and costs of disability, and service provision in these populations.

Large longitudinal cohort studies on the modifiable risk factors for the transition to cognitive frailty are scarce [10, 11]. While some studies have investigated the associations of depression and life satisfaction with cognitive frailty [11, 12], none of the studies have investigated potential modifiable psychosocial risk factors, such as participation in leisure activities (hobbies, games, social group activities, etc.), which may influence mental and social well-being, in turn contributing to cognitive frailty [13, 14]. Furthermore, the effects of other blood biomarkers such as cholesterol [15] and hemoglobin [16], as well as genetic risk factors [17] and their contribution to the risk of cognitive frailty transition, are not known.

In this study, we investigated risk factors for transition to cognitive frailty from 3 to 5 years (mean 4.4 years) of follow-up of a cohort of non-cognitively frail community-dwelling older adults without dementia and degenerative disorders at baseline and follow-up, from the second wave recruitment cohort of the Singapore Longitudinal Ageing Study (SLAS-2).

Methods

Study Design and Participants

We analyzed data in SLAS-2, an ongoing observational prospective cohort study of aging and health transition among middle-aged and older adults in Singapore. Participants were aged 55+ years at baseline and able to self-ambulate, excluding individuals who were unable to participate due to severe physical or mental disabilities. Residents in South-West Singapore were recruited from 6 March 2009 to 11 June 2013. Follow-up visits and assessments were conducted from 16 January 2013 to 24 August 2018 for SLAS-2. Details of methodology have been described in earlier papers [18]. Each participant underwent structured interviews, clinical evaluation, blood sampling and tests for various demographic, medical, biological, and psychosocial characteristics. Ethics approval was obtained from National University of Singapore Institutional Review Board (Ref:04–140). All participants gave written informed consent to participate in the study.

Among 3,270 participants in the SLAS-2 cohort, we excluded those who reported a history of dementia, stroke, Parkinson’s disease, other neurodegenerative disorders at both baseline and follow-up (N = 263) and participants with cognitive frailty at baseline (N = 512). A total of 1,441 (57.8%) participants were lost to follow-up due to loss to contact or refusals. In this study, we analyzed complete baseline and follow-up data of 1,054 participants who were free of cognitive frailty at baseline.

Measurements

Physical Frailty

Frailty at baseline and follow-up was assessed based on the criteria used in the Cardiovascular Health Study [1], with operational modifications [4, 8, 19]. Body mass index (BMI) of less than 18.5 kg/m² or unintentional weight loss of ≥4.5 kg in the past 6 months was used to define shrinking. Weakness was determined as lowest quintile of knee extension strength by sex and BMI strata. Mean reading of 3 trials in the dominant leg for knee extension strength, measured with participant seated at 90° angle of the hip and knee, with the Lord’s strap and strain gauge component of the Physiological Profile Assessment, was used. Average fast gait speed of <0.8 m/s over 6 m over two trials was used to determine slowness. Exhaustion was measured using the vitality domain in the Medical Outcomes Study SF-12 [20]: “Did you feel worn out?”, “Did you feel tired?”, “Did you have a lot of energy?” with total summed scores ranging from 3 to 15. A score of <10 was used to denote exhaustion. Physical activity was assessed based on self-reported time (in hours) spent doing light (strolling, standing with little motion, etc.), moderate-to-vigorous activities (gardening, brisk walking, jogging, swimming, strenuous sports, etc.) on weekdays and weekend. Total time below sex-specific lowest quintile, spent engaging in moderate-to-vigorous activities per week, was used to define low physical activity. Participants were classified as prefrail/frail if they met ≥1 of the above criteria [1].

Cognitive Impairment and Cognitive Frailty

Cognitive impairment was determined using locally modified and validated Mini-Mental State Examination (MMSE) translated versions (English, Chinese, and Malay) [21]. Scores ranged from 0 to 30 and score of <26 denotes mild or greater degrees of cognitive impairment [8]. Cognitive frailty was defined as a combination of prefrailty/frailty and cognitive impairment.

Baseline Risk Factors

Baseline data collected on risk factors included sex, age, education (primary or below vs. secondary or tertiary), marital status (non-married vs. married), housing type (1–3 room public housing apartment vs. higher end housing). Smoking history was self-reported (current smokers vs. nonsmokers). Participants were asked about whether they were living alone, whether they could obtain help when in need (none, sometimes, often), and having someone to confide in.

Participants also self-reported leisure activities including participation in senior activity centers, social group activities, hobbies, playing mahjong, using computer or playing games, singing karaoke, listening to radio, engaging in cognitively stimulating activities (reading books/newspapers/magazine, doing crossword/puzzles, writing, painting, drawing, attending courses, etc.) and physical exercise (calisthenics, jogging, aerobic, bicycle, etc.), brisk walking or other active sports (cycling, swimming, tennis, badminton, bowling, golfing, etc.), according to the following frequencies: 1 – none or less than once a month (never), 2 – once a month to less than once a week (sometimes), and 3 – once a week or more (often). Physical activity was determined by summed scores of frequencies of physical exercise, walking, and active sports. Lowest quartile of summed physical activity scores was used to determine low physical activity. The Nutritional Screening Initiative (NSI) Determine Your Nutritional Health Questionnaire was used to assess nutritional risk (scores ranging from 0 to 11) and identify moderate-to-high nutritional risk (NSI score ≥3) [22].

Life satisfaction was determined using a self-reported Life Satisfaction Scale comprising questions that assessed the subjects’ interest in life, happiness, and general ease of living, which has been shown to predict mortality [23]. Respondents rated whether they find life “interesting or boring,” “happy or sad,” or “easy or hard” on a 5-point Likert scale (such as 1 “very interesting” to 5 “very boring”). Due to small number of respondents for “very boring,” “very sad,” and “very hard,” the responses were combined with “fairly boring,” “fairly sad,” and “fairly hard,” respectively, for analyses.

Self-reports of doctor’s diagnosis and treatment based on inspection of medication packages used, clinical and biochemical measures, were used to determine chronic diseases [19]. These included cardiovascular diseases (myocardial infarction, heart failure, stroke), atrial fibrillation, eye problem, hearing impairment, osteoporosis, arthritis, neurodegenerative disorders, cancer, diabetes (fasting blood glucose ≥5.6 mmol/L), chronic kidney disease (eGFR <60 mL/kg), and chronic obstructive pulmonary disease (FEV1/FVC <0.70). Presence of ≥5 chronic illnesses was used to define multimorbidity. High blood pressure (BP) was determined by systolic BP of ≥130 mm Hg or diastolic BP of ≥85 mm Hg or using BP-lowering medication [24]. Self-reported history of hospitalizations and falls in the previous year, as well as self-rated health status using EQ VAS, a vertical visual analogue scale of values between 100 (best imaginable health) and 0 (worst imaginable health), were determined. The Geriatric Depression Scale 15-items (GDS-15) score (0–15) was used to identify the presence of depressive symptom score of ≥5. Weight and height measurements were used to calculate BMI and obesity (BMI ≥27.5 kg/m²) [25]. QOL was measured using the Medical Outcomes Study 12-Item Short Form Health Survey (SF-12), and low physical and mental QOLs were determined as values below the lowest quartile of physical and mental component scores, respectively.

Polymerase chain reaction or polymerase chain reaction amplification with restriction endonuclease digestion of the product was used for apolipoprotein E genotyping. Carriers (ϵ2/4, ϵ3/4, and ϵ4/4) and noncarriers (ϵ2/2, ϵ2/3, and ϵ3/3) of apolipoprotein E-ϵ4 allele were classified. High total cholesterol levels (≥5.2 mmol/L), high triglyceride levels (≥1.7 mmol/L), low high-density lipoprotein (HDL) cholesterol levels (<1 mmol/L in men and <1.3 mmol/L in women) and high low-density lipoprotein cholesterol levels (≥2.6 mmol/L), low hemoglobin levels (<13 g/dL in men and <12 g/dL in women) were determined [19, 24, 26].

Statistical Analyses

Categorical variables are presented as numbers (percentages), and continuous variables are presented as mean (SD). Binomial logistic regression was used for univariate analysis to determine associations of baseline risk factors with transitions to cognitive frailty. Estimated odds ratios (ORs) were further adjusted for age, sex, education, baseline physical frailty, and MMSE scores in univariate analyses. The “glmnet” package was used to run least absolute shrinkage and selection operator (LASSO) regression analysis [27]. For variable selection, LASSO regression analysis shrinks the regression coefficient of certain variables to zero, through imposing constraints on model parameters. Variables with nonzero regression coefficients were used for preliminary screening of risk factor variables for cognitive frailty, while variables with zero regression coefficients were excluded from the model after contraction. Using k-fold (10-fold) cross-validation, the minimum mean cross-validated error for lambda value was selected. The variables selected in the LASSO regression model were used for multivariate logistic regression analysis to construct the prediction model. We used the “rms” package to carry out logistic regression [28]. Statistically significant predictors from the LASSO regression were selected to establish the cognitive frailty risk prediction model. Bootstrap samples were drawn for internal validation of the final model, and the corrected R-square and area under the ROC curve (AUC) were determined. Baseline sociodemographic comparison of participants included in this study and those lost to follow-up were examined using Mann-Whitney U test for continuous variables and χ² test for categorical variables.

We also used multivariate imputation by chained equation (“mice”) package in R [29], to create five imputed datasets for values assumed missing at random. We used logistic regression analyses to examine univariate associations between predictors and outcomes. The full multivariate logistic regression models were reduced using backward stepwise selection, with an exclusion criterion of alpha = 0.05. We used the “psfmi” package in R, in which variable selection was based on Rubin’s rules for pooled estimates [30]. The regression coefficients in the full model were adjusted for optimality using the “pool_intadj” function in the “psfmi” package, with calibration slope used to determine the need for shrinkage of coefficients due to overfitting. The “psfmi_validate” function evaluated the performance of the models in the multiply imputed datasets. Bootstrap samples were drawn in each imputed dataset prior to combining the results. The overall predictive performance of models was measured using the optimism-corrected Nagelkerke’s R-square. The AUC was corrected for optimism, and the pooled estimate was presented. R software (version 4.2.1) was used to perform all statistical analysis.

Results

Baseline Profiles

The mean (SD) age of 1,054 participants with follow-up data on cognitive frailty included in the analyses was 65.2 (6.5), 665 (63.1%) were women, 507 (48.1%) had secondary or tertiary education, 614 (58.3%) lived in high-end housing, 313 (29.7%) were not married, 80 (7.6%) were current smokers, and 245 (23.2%) had moderate-to-high risk of malnutrition (Table 1). In all, 40.8% (N = 430) of participants at baseline were only prefrail/frail and 2.1% (N = 22) had only MMSE score <26, and the remaining 57.1% (N = 602) were cognitively normal and physically robust (Table 1). A total of 51 (4.8%) participants transited to cognitive frailty during follow-up. Among them, 21 (3.5%) of the cognitively normal and physically robust participants, 20 (4.7%) of the only physically prefrail/frail participants, and 10 (45.4%) of the only cognitively impaired participants transited to cognitive frailty. The sex- and age-adjusted odds likelihood of progression to cognitive frailty was 1.05 (95% CI 0.55–2.01) for baseline physical prefrailty/frailty only, and 16.8 (95% CI 6.07–46.0) for baseline cognitive impairment only.

Table 1.

Baseline characteristics of participants without cognitive frailty (N = 1,054)

	N (%)	Missing, N (%)
Sociodemographic and lifestyle
Age, mean (SD), years	65.2 (6.5)	0 (0)
Sex, F	665 (63.1)	0 (0)
Education, secondary/tertiary	507 (48.1)	2 (0.2)
Housing, higher end	614 (58.3)	4 (0.4)
Marital status, not married	313 (29.7)	1 (0.1)
Smoking, current	80 (7.6)	3 (0.3)
Malnutrition	245 (23.2)	0 (0)
Low physical activity	208 (19.7)	6 (0.6)
Physical frailty score, mean (SD)	0.6 (0.8)	0 (0)
Robust, N (%)	624 (59.2)
Prefrail, N (%)	397 (37.7)
Frail, N (%)	33 (3.1)
MMSE score, mean (SD)	28.8 (1.4)	0 (0)
≥26, N (%)	1,032 (97.9)
<26, N (%)	22 (2.1)
Living alone	145 (13.8)	7 (0.7)
Someone to confide in	1,020 (96.8)	4 (0.4)
Obtain help when in need
None	166 (15.7)	4 (0.4)
Sometimes	313 (29.7)
Often	571 (54.2)
Medical history
Multimorbidity	58 (5.5)	0 (0)
Hospitalization in previous year	37 (3.5)	0 (0)
Falls in the previous year	93 (8.8)	2 (0.2)
Depression	15 (1.4)	1 (0.1)
Obesity	152(14.4)	0 (0)
Self-rated health EQ VAS score, mean (SD)	80.0 (13.2)	5 (0.5)
Diabetes	297 (28.2)	0 (0)
Cardiovascular disease	38 (3.6)	0 (0)
High BP	688 (65.3)	0 (0)
Eye problem	275 (26.1)	0 (0)
Low mental QOL	235 (22.3)	28 (2.7)
Low physical QOL	221 (21.0)	28 (2.7)
Biochemical parameters
Apolipoprotein E (APOE)-ϵ4	164 (15.6)	105 (10.0)
High total cholesterol	497 (47.2)	51(4.8)
Low HDL cholesterol	251 (23.8)	47 (4.5)
High low-density lipoprotein (LDL) cholesterol	746 (70.8)	59 (5.6)
High triglyceride	207 (19.6)	52(4.9)
Low hemoglobin	144 (13.7)	55(5.2)
Activities
Participate in senior activity centers		7 (0.7)
Never	693 (65.7)
Sometimes	117 (11.1)
Often	237 (22.5)
Participate in social group activities		2 (0.2)
Never	808 (76.7)
Sometimes	79 (7.5)
Often	165 (15.7)
Play mahjong
Never	898 (85.2)	2 (0.2)
Sometimes	73 (6.9)
Often	81 (7.7)
Sing karaoke
Never	822 (78.0)	4 (0.4)
Sometimes	86 (8.2)
Often	142 (13.5)
Hobbies
Never	621 (58.9)	3 (0.3)
Sometimes	99 (9.4)
Often	331 (31.4)
Listens radio
Never	331 (31.4)	2 (0.2)
Sometimes	85 (8.1)
Often	636 (60.3)
Computer, games
Never	789 (74.9)	5 (0.5)
Sometimes	54 (5.1)
Often	206 (19.5)
Cognitive stimulating		3 (0.3)
Never	146 (13.9)
Sometimes	70 (6.6)
Often	835 (79.2)
Life satisfaction
Interesting		2 (0.2)
Very interesting	177 (16.8)
Fairly interesting	692 (65.7)
Neither interesting nor boring	165 (15.7)
Fairly boring/very boring	18 (1.7)
Happy
Very happy	192 (18.2)	1 (0.1)
Fairly happy	662 (62.8)
Neither happy nor sad	189 (17.9)
Fairly sad/very sad	10 (0.9)
Easy
Very easy	187 (17.7)	2 (0.2)
Fairly easy	652 (61.9)
Neither easy nor hard	194 (18.4)
Fairly hard/very hard	19 (1.8)

Univariate Analysis

Advancing age (OR = 1.1, 95% CI 1.07–1.16) and female sex (OR = 2.5, 95% CI 1.28–5.31) were associated with increased odds of incident cognitive frailty, while higher education (secondary or tertiary) (OR = 0.1, 95% CI 0.05–0.29) was associated with decreased incidence of cognitive frailty (Table 2). Multimorbidity, diabetes, eye problem, low HDL cholesterol, and low hemoglobin were independently associated with 2–4-fold increased odds of incident cognitive frailty (Table 2). Playing games or on the computer, and engaging in cognitively stimulating activities were associated with 42–56% lower incidence of cognitive frailty (Table 2). Adjusted for age, sex, education, baseline physical frailty, and MMSE scores, variables including multimorbidity, diabetes, eye problem, and low HDL cholesterol remained significantly associated with transition to incident cognitive frailty (Table 2).

Table 2.

Univariate associations of sociodemographic and lifestyle factors, medical history, biochemical parameters, activities, and life satisfaction with incident cognitive frailty

Variables	Model 1	p value	Model 2	p value
	OR (95% CI)		OR (95% CI)
Sociodemographic and lifestyle
Age	1.12 (1.07–1.16)	<0.001*
Sex, F	2.49 (1.28–5.31)	0.011*
Education, secondary/tertiary	0.13 (0.05–0.29)	<0.001*
Housing, 4-5rm/private	0.67 (0.38–1.18)	0.162	1.00 (0.53–1.89)	0.998
Marital status, non-married	1.56 (0.86–2.76)	0.131	0.76 (0.37–1.49)	0.428
Smoking, current	0.75 (0.18–2.11)	0.634	1.59 (0.35–5.37)	0.492
Malnutrition	1.40 (0.73–2.55)	0.287	1.05 (0.51–2.07)	0.888
Low physical activity	0.88 (0.40–1.76)	0.737	0.73 (0.31–1.56)	0.437
Living alone	1.20 (0.51–2.47)	0.652	0.91 (0.36–2.05)	0.827
Someone to confide in	1.49 (0.31–26.90)	0.695	1.50 (0.26–29.05)	0.711
Obtain help when in need	1.01 (0.70–1.50)	0.949	0.99 (0.65–1.56)	0.978
Medical history
Multimorbidity	3.55 (1.48–7.58)	0.002*	2.85 (1.05–7.12)	0.030*
Hospitalization in previous year	1.13 (0.18–3.85)	0.870	0.79 (0.09–3.66))	0.797
Falls in the previous year	1.69 (0.68–3.65)	0.213	1.09 (0.39–2.67)	0.855
Depression	NA†	NA	NA†	NA
Obesity	1.48 (0.69–2.91)	0.283	1.45 (0.62–3.12)	0.364
Self-rated health EQ VAS	0.99 (0.97–1.01)	0.139	0.99 (0.97–1.01)	0.208
Diabetes	2.38 (1.34–4.19)	0.003*	1.94 (1.01–3.69)	0.043*
Cardiovascular disease	1.73 (0.41–5.03)	0.377	1.38 (0.30–4.70)	0.638
Hypertension	1.59 (0.86–3.13)	0.159	0.93 (0.46–1.98)	0.849
Eye problem	3.42 (1.94–6.08)	<0.001*	2.36 (1.20–4.65)	0.012*
Low mental QOL	1.50 (0.76–2.81)	0.216	1.25 (0.59–2.56)	0.543
Low physical QOL	0.88 (0.39–1.78)	0.739	0.88 (0.37–1.92)	0.753
Biochemical parameters
Apolipoprotein E (APOE)-ϵ4	0.98 (0.42–2.03)	0.961	1.12 (0.44–2.55)	0.803
High total cholesterol	0.56 (0.30–1.02)	0.063	0.51 (0.25–1.01)	0.059
Low HDL cholesterol	4.44 (2.45–8.16)	<0.001*	3.54 (1.81–7.03)	<0.001*
High low-density lipoprotein (LDL) cholesterol	0.70 (0.38–1.35)	0.266	0.90 (0.45–1.89)	0.775
High triglyceride	0.91 (0.40–1.82)	0.793	0.99 (0.40–2.25)	0.988
Low hemoglobin	2.13 (1.04–4.10)	0.030*	1.53 (0.66–3.31)	0.298
Activities
Participate in senior activity centers	0.92 (0.64–1.28)	0.632	0.83 (0.55–1.22)	0.365
Participate in social group activities	0.79 (0.49–1.18)	0.292	0.78 (0.46–1.23)	0.320
Play mahjong	0.99 (0.56–1.56)	0.966	0.97 (0.54–1.60)	0.916
Sing karaoke	0.83 (0.51–1.25)	0.421	0.91 (0.54–1.42)	0.694
Hobbies	0.79 (0.56–1.09)	0.162	0.90 (0.61–1.28)	0.561
Listens radio	0.87 (0.65–1.18)	0.365	0.98 (0.70–1.40)	0.922
Computer, games	0.58 (0.33–0.90)	0.029*	1.25 (0.68–2.09)	0.429
Cognitive stimulating	0.44 (0.32–0.60)	<0.001*	0.74 (0.52–1.08)	0.111
Life satisfaction
Interesting	1.48 (0.96–2.26)	0.073	1.42 (0.85–2.36)	0.177
Happy	1.30 (0.84–2.01)	0.240	1.13 (0.67–1.87)	0.639
Easy	0.94 (0.61–1.44)	0.796	0.96 (0.58–1.59)	0.889

Model 1: unadjusted.

Model 2: adjusted for age, sex, education, and baseline physical frailty and MMSE scores.

*Significant at p < 0.05.

^†Unable to estimate OR and 95% CI.

Multivariate Analysis

LASSO regression analysis selected predictors for incident cognitive frailty, indicated in Table 3. In multivariate logistic regression, 4 of the original 8 variables were included in the model, specifically education (OR = 0.3, 95% CI 0.10–0.74), eye problem (OR = 2.6, 95% CI 1.24–5.43), low HDL cholesterol (OR = 4.1, 95% CI 2.03–8.40), and participation in cognitive stimulating activities (OR = 0.4, 95% CI 0.17–0.82) (Table 3). The final model explained 23.6% of the variance in cognitive frailty after adjustment for optimism, and the AUC was 0.83.

Table 3.

LASSO multivariate regression model for incident cognitive frailty (N = 860)

Predictors	OR (95% CI)	p value
Age	1.69 (0.96–2.97)	0.071
Sex, F	1.97 (0.80–4.87)	0.142
Education, secondary/tertiary	0.27 (0.10–0.74)	0.011*
Multimorbidity	1.50 (0.54–4.16)	0.431
Diabetes	1.73 (0.84–3.58)	0.137
Eye problem	2.59 (1.24–5.43)	0.012*
Low HDL cholesterol	4.13 (2.03–8.40)	<0.001*
Cognitive stimulating activity	0.38 (0.17–0.82)	0.014*

*Significant at p < 0.05.

Sensitivity Analysis

Comparison of baseline sociodemographic characteristics between participants included in the study and those lost to follow-up is presented in online supplementary eTable 1 (for all online suppl. material, see https://doi.org/10.1159/000531421). Participants lost to follow-up were older and had lower education and lower end housing at baseline. A larger proportion of participants lost to follow-up were not married, were current smokers, had poorer nutritional status and lower physical activity, were prefrail/frail, and had MMSE scores of <26 at baseline (online suppl. eTable 1). The univariate analyses with multiple imputations for missing values are presented in online supplementary eTable 2 (N = 2,495). The final model predicting cognitive frailty included age, sex, education, low physical activity, eye problem, low HDL cholesterol, playing mahjong, participating in cognitive stimulating activities (online suppl. eTable 3). The final model explained 26.1% of the variance in cognitive frailty, after adjustment for optimism. The optimism-adjusted AUC was 0.83, showing good discriminative value.

Discussion

In this cohort of cognitive frailty-free individuals, incident cases of cognitive frailty arose among individuals who were either physically prefrail/frail only, MMSE score <26 only, or neither at baseline. The progression to cognitive frailty was represented by individuals with prefrailty/frailty who become cognitively impaired over time, or those with MMSE score <26 who become physically frail over time, as well as those who were neither prefrail/frail nor cognitively impaired but showed cognitive frailty over the period of follow-up. It is possible that the last group could have become physically frail or cognitively impaired one after the other or concurrently, but could not be determined given the limitations in the present study. Physical prefrailty/frailty was associated with modestly increased likelihood of incident cognitive frailty, but lower MMSE score (<26) was associated with considerably high (>10 times increased) likelihood of incident cognitive frailty. However, prevalence of physical prefrailty/frailty alone at baseline was high (41%), but prevalence of participants with MMSE score <26 was low (2%) in this population. As such, in our study population, approximately twice the number of incident cases of cognitive frailty arose from physically prefrail/frail individuals (20/51 or 39%) than from individuals with MMSE score <26 (10/51 or 20%). This attests to the conceptual validity of cognitive frailty as a subtype of lower MMSE scores (<26) arising from physical frailty. The incidence of cognitive frailty will vary among different populations, depending on the baseline prevalence of physical prefrailty/frailty and MMSE scores of <26. Whether cognitive frailty arises from incipient physical frailty or cognitive impairment, and with yet unknown temporal sequences or patterns, our study suggests that there are common modifiable risk factors that can be identified and targeted for interventions to prevent progression to dementia and associated adverse health outcomes.

We observed that lower education predicted transition to cognitive frailty. In agreement, cognitively frail community-dwelling older adults were more likely to have lower education levels than cognitively normal or MCI individuals [31]. Consistent with the cognitive reserve hypothesis, prior education and cognitive abilities increase brain structure resilience to disease and injury, suggesting that longer formal education in early life contributes to better cognitive function in older age and attenuates cognitive decline and neuropathology [32–35].

Cross-sectional and longitudinal cohort studies have also found that vision impairment was associated with 2–5-fold increased risks of frailty at 3–4-year follow-up, among community-dwelling older adults [36, 37]. A systematic review of 110 studies showed that visual impairment was associated with cognitive decline and impairment among older adults aged 50–93 (mean age 73 years) [38]. Earlier results collectively suggest that cognitive impairment and frailty are independently associated with visual impairment. It is therefore unsurprising that eye problems are associated with transition to cognitive frailty, in our population of community-dwelling older adults. The relation between vision and cognitive frailty is not fully understood, but may share common pathophysiology with cognitive impairment, including amyloid beta deposition or microvascular disease [39, 40]. It is also plausible that visual impairment causes a reduction in physical activity, social activities, and support, which are risk factors for cognitive frailty [41, 42]. For example, vision impairment may be associated with cognitive decline through sensory loss, leading to depression, social isolation, decline in cognitively stimulating activities (reading, etc.), and cognitive load, as greater dedication of cognitive resources to visual processing could affect other cognitive processes [40]. Visual impairment might also have direct effects on indicators of frailty, such as gait speed and mobility disability [43]. Vision is likely an important and understudied risk factor for cognitive frailty, and more studies are needed to better elucidate underlying mechanisms.

We found that low HDL cholesterol levels predicted incident cognitive frailty. Similarly, lower HDL cholesterol levels at midlife were associated with late-life MCI and dementia in a Japanese population [15]. Compared to non-frail participants, frailty was associated with twofold higher odds of low HDL cholesterol in older British men aged 71–92 [44]. Higher levels of HDL cholesterol were also associated with better physical function among older adults aged ∼86 years [45]. These findings highlight that interventions and behavioral modifications that increase HDL cholesterol, such as physical activity, moderate alcohol consumption, and smoking cessation, should be promoted to reduce the risk of cognitive frailty [46].

Participation in cognitively stimulating activities was also associated with lower incidence of cognitive frailty in this present study. A randomized controlled trial in prefrail/frail community-dwelling older adults showed that cognitive training interventions improve physical frailty in older adults [47]. Systematic review on 33 randomized controlled trials also concluded that cognitive training is associated with small-to-moderate positive effects on global cognition and verbal semantic fluency for people with dementia [48]. Playing mahjong is a form of social activity involving visual and mental focus, common in Asia played among four players with win-or-lose gambling-like characteristics. In multiple imputation analyses, we found that mahjong playing also associated lower incidence of cognitive frailty in our study. This suggests that cognitively stimulating activities with social interaction may improve executive, attention, verbal memory, and cognitive functions in older adults [49, 50]. Additionally, sensitivity analyses with multiple imputation also revealed that low physical activity predicted cognitive frailty incidence, suggesting that cognitive frailty is modifiable [51], and future interventions targeted at improving cognitive frailty outcomes should include cognitive, physical, and social interaction domains.

Our study has some limitations. Given the low frequencies of some risk factors in this study cohort, risk estimates should be interpreted with caution given their small sample size and limited statistical power. There was nontrivial loss of participants at follow-up, and complete case analysis could plausibly underestimate actual associations with cognitive frailty incidence, given that the participants who were lost to follow-up were older had higher physical frailty scores and lower MMSE scores at baseline, suggesting potential associations with higher incidence of cognitive frailty. We presented separate sensitivity analysis using multiple imputation techniques, which has been shown to provide unbiased estimates even with large proportions of missing data commonly observed with epidemiological studies [52]. The multiple imputation results were comparable to complete case analysis for predictors of cognitive frailty, with common predictors selected including age, female sex, education status, eye problem, low HDL cholesterol, and participation in cognitive stimulating activities. Nonetheless, future studies should include a larger sample size and participants with heterogenous baseline characteristics to investigate the predictors of cognitive frailty incidence. Because follow-up was performed at 3–5 years after the baseline assessment visit, there might be missed cases of incident physical frailty, MCI, or cognitive frailty which could otherwise be detected at yearly intervals. As we were unable to determine the temporal associations and order of events between cognitive frailty and the presence of dementia or other neurodegenerative disorders at follow-up, we excluded those who reported a history of dementia, stroke, Parkinson’s disease, and other neurodegenerative disorders at both baseline and follow-up, for the analysis of incident cognitive frailty in this study. Future studies with shorter follow-up intervals should include participants with incident cognitive frailty diagnosed prior to dementia, to better provide insights on the risk factors for cognitive frailty. We were not able to externally validate the prediction models. While we validated the models internally using bootstrapping techniques, bootstrap samples are derived from the same dataset. The findings may not be generalizable to other populations or individuals, given the unique socioeconomic, risk factors, and disease background of this community-dwelling Asian study population. Strengths of our study include the use of a representative sample of community-dwelling older adults in Singapore and the inclusion of predictors from multiple domains.

In conclusion, we found that predictors of cognitive frailty transition encompassed several domains, including sociodemographic (education levels), medical history (eye problem), biochemical (low HDL cholesterol), and participation in activities (cognitive stimulating). The readily accessible predictors in our models could aid early identification and targeted interventions for older adults at risk of cognitive frailty. Modifiable risk factors such as participation in various leisure activities should be promoted among community-dwelling older adults.

Acknowledgments

We thank the following voluntary welfare organizations for their support: Geylang East Home for the Aged, Presbyterian Community Services, St Luke’s Eldercare Services, Thye Hua Kwan Moral Society (Moral Neighbourhood Links), Yuhua Neighbourhood Link, Henderson Senior Citizens’ Home, NTUC Eldercare Co-op Ltd, Thong Kheng Seniors Activity Centre (Queenstown Centre), and Redhill Moral Seniors Activity Centre.

Statement of Ethics

This study involves human participants and was approved by an Ethics Committee or institutional board (National University of Singapore IRB Ref: 04–140), in accordance with the relevant guidelines from the Declaration of Helsinki and the ethical principles in the Belmont Report. All participants gave written informed consent.

Conflict of Interest Statement

The authors declare that they have no competing interests.

Funding Sources

This work was supported by research grants from the Agency for Science Technology and Research (A*STAR) Biomedical Research Council (Grant No. BMRC/08/1/21/19/567) and the National Medical Research Council (Grant No. NMRC/1108/2007, NMRC/CIRG/1409/2014).

Author Contributions

S.Y.L. and T.P.N. designed the study, reviewed the literature, and drafted and revised the manuscript. S.Y.L. analyzed the data. S.Z.N., Q.G., X.G., D.Q.L.L., K.B.Y., and S.L.W. contributed to the conduct of the study and data collection. All authors reviewed the results and drafts, and approved the final manuscript.

Funding Statement

This work was supported by research grants from the Agency for Science Technology and Research (A*STAR) Biomedical Research Council (Grant No. BMRC/08/1/21/19/567) and the National Medical Research Council (Grant No. NMRC/1108/2007, NMRC/CIRG/1409/2014).

Data Availability Statement

Data are not publicly available due to ethical reasons. Further inquiries can be directed to the corresponding author.