Frailty represents a state of decline in functional reserves, which increases the risk of adverse health outcomes such as morbidity, disability, and institutionalization, after a stressor event (1). It can be preceded by, but also occurs in the absence of chronic disease (2) and has been suggested as a better predictor of health and well-being than the presence or absence of disease. Although the term frailty is commonly used in clinical practice, there is no consensus on the definition of frailty. A popular approach to the assessment of frailty as proposed by Fried et al. (1) (i.e., the phenotype approach) encompasses the assessment of five criteria-based primarily on physical attributes and capabilities including poor grip strength, slow walking speed, low levels of physical activity, exhaustion, and unintentional weight loss, whereas an individual is considered to be frail if they present with three or more of five criteria. Another notable approach to the assessment of frailty is that of Rockwood and Mitnitski (3, 4) (i.e., the deficit accumulation model) in which frailty is viewed in terms of the number of health deficits (i.e., integration with measures of physical frailty and other domains) that are manifest in the individual, leading to a continuous measure of frailty (frailty index).

More recently, there is general consensus that measures of cognitive function should be added to physical performance for the definition of frailty, in that there is a bidirectional relationship between physical frailty and cognitive impairment. There is also a parallel pathway among frailty discourse, that cognitive vulnerability (or impairment) may be a precursor of mild neurodegenerative disorder [akin to pre-dementia state of Mild Cognitive Impairment (MCI)] (5) and subsequently major neurodegenerative disorder (dementia) (6). Numerous studies have demonstrated that cognitive impairment may lead to increased risk of acquiring individual components of frailty syndrome (e.g., faster gait speed decline)/future frailty (7–9). The reciprocal relationship, which frailty predicts cognitive decline/incident dementia, has also been reported (10–14). Both frailty and cognitive impairment share many common risk factors and underlying mechanisms (6, 15, 16). Although many studies demonstrate close relationship between frailty and cognitive impairment, most of them have characterized frailty and cognitive impairment as two different entities, and the term “cognitive frailty” has been proposed, to characterize the co-existence of both frailty and cognitive impairment. An international consensus group organized by the International Academy on Nutrition and Aging (IANA) and the International Association of Gerontology and Geriatrics (IAGG) proposed the definition as a clinical condition characterized by the simultaneous presence of both physical frailty and MCI (Clinical Dementia Rating = 0.5) (17). Recent studies have reported that cognitive frailty conferred additional greater risk of adverse outcomes including disability, hospitalization, and mortality (6, 18–20). Understanding the temporal relationship between cognitive impairment and frailty is important, in predicting the onset of the other, with implications for screening and intervention programs. For example, in the Baltimore longitudinal study of aging, a bidirectional relationship was noted for usual gait speed and executive function, with each predicting change in the other, while poor fast walking performance predicted future executive function and memory changes but not vice versa (14). Although there is no universal consensus regarding the entity of cognitive frailty and its definition, there is general consensus of the importance of recognizing cognitive impairment, as differentiated from screening for dementia (21).

According to the IANA/IAGG, the primary criterion of cognitive frailty is the presence of physical frailty and MCI, without dementia. However, different states of cognitive vulnerability and frailty may be relevant to identify older persons with cognitive frailty. Furthermore, it is likely that MCI may represent a later stage of cognitive impairment at which multiple domains of cognition have already occurred. Early detection of abnormalities in specific domains of cognitive function (e.g., memory problems, difficulties in word finding) together with identification of the pre-frail state (an intermediate stage between non-frail and frail) may allow opportunities for reversibility through intervention strategies, which is supported by the findings from a home-based program to prevent functional decline in physically frail elderly persons in which the benefit of the program was observed among those with moderate frailty, but not those with severe frailty (22). Using the Mr and MsOs study of older Chinese men and women who were free of dementia and/or Parkinson’s disease and who were non-frail at baseline, we examined how various combinations of cognitive impairment (overall performance as well as two selected a priori domains) and pre-frailty predict risks of adverse outcomes (poor quality of life, physical limitation, increased cumulative hospital stay, and mortality), and to determine whether cognitive frailty may be defined as the combination of cognitive impairment (overall or domain specific) and the presence of pre-frailty.

At baseline, the mean age of the study sample was 72.0 (4.9) years, 48.4% were female, and 72.0% had primary or lower education. In total, 57.5% were robust, 42.5% were pre-frail, and 17.4% had overall cognitive impairment, with their CMMSE total score <21–27 (depending on participants’ educational levels). Of those who were pre-frail (n = 1,483), 20.4% had overall cognitive impairment (CMMSE total score < 21–27), 55.3% had delayed recall impairment (delayed recall score < 3), and 51.4% had language and praxis impairment (language and praxis score < 9) (Table 1).

The prevalence of overall cognitive impairment was higher in the pre-frail group (17.8%) than in the robust group (14.4%). Compared to participants who were robust and cognitively intact at baseline, those who were robust but cognitively impaired were more likely to develop pre-frailty/frailty after 4 years (P < 0.01). Participants who were pre-frail but cognitively intact at baseline were also more likely to develop frailty at the 4-year follow-up than their cognitively impaired counterparts. However, the association was not significant (P = 0.056) (Table 2).

Compared to participants who were robust and cognitively intact at baseline, those who were pre-frail and with overall cognitive impairment had lower grip strength (P < 0.05), lower gait speed (P < 0.01), poorer lower limb strength (P < 0.01), and poorer performance in delayed recall at year 4 (OR, 1.6; 95% CI 1.2–2.3). They were also associated with increased risks of poor quality of life (OR, 1.5; 95% CI, 1.1–2.2) and incident physical limitation at year 4 (OR, 1.8; 95% CI, 1.3–2.5), increased cumulative hospital stay at year 7 (OR, 1.5; 95% CI, 1.1–2.1), and mortality over an average of 12 years (OR, 1.5; 95% CI, 1.0–2.1) after adjustment for covariates. Participants who were pre-frail and cognitively impaired at baseline were also associated with a higher risk of incident physical limitation at year 4 (OR, 1.8; 95% CI, 1.1–2.8) as compared to the robust but cognitively impaired participants, and had poorer cognitive performance at year 4 as compared to their cognitively intact counterparts (P < 0.01). However, there was no significant difference in risks of adverse outcomes between participants who were pre-frail, with or without cognitive impairment at baseline (Table 3).

When a single a priori selected cognitive domain was used to define cognitive impairment, participants with pre-frailty and a delayed recall score <3 at baseline had lower gait speed (P < 0.001), poorer lower limb strength (P < 0.05), poorer cognitive performance in terms of time orientation (OR, 1.7; 95% CI 1.3–2.3), place orientation (OR, 1.7; 95% CI, 1.3–2.2), attention/calculation (OR, 1.5; 95% CI, 1.1–1.9), as well as language and praxis at year 4 (OR, 1.5; 95% CI, 1.1–1.9). They were also associated with increased risks of poor quality of life (OR, 1.7; 95% CI, 1.3–2.3), and incident physical limitation at year 4 (OR, 1.8; 95% CI, 1.4–2.3), and increased cumulative hospital stay at year 7 (OR, 1.4; 95% CI, 1.1–1.9) as compared to participants who were robust and had a delayed recall score = 3 (Table 4). Similar results were obtained when cognitive impairment was redefined by language and praxis score (Table 5).

The risks of having adverse outcomes at follow-up were also compared between those who were robust and cognitively intact and the rest of the groups. As expected, participants who were pre-frail but cognitively intact at baseline were associated with increased risk of poor quality of life (OR, 1.4; 95% CI, 1.1–1.7) and incident physical limitation at year 4 (OR, 1.5; 95% CI, 1.2–1.8) as well as increased cumulative hospital stay at year 7 (OR, 1.4; 95% CI, 1.2–1.8) as compared to participants who were robust and cognitively intact at baseline. However, there was no significant difference in risks of adverse outcomes between participants who were robust, with or without cognitive impairment at baseline (Table S1 in Supplementary Material).

In a cohort of older people free of dementia and/or Parkinson’s disease and/or frailty at baseline, we showed that robust and cognitively impaired participants were more likely to develop pre-frailty/frailty after 4 years than the robust and cognitively intact participants. Furthermore, participants with both pre-frailty and cognitive impairment at baseline had poorer physical and cognitive performances, higher risks of poor quality of life, incident physical limitation, increased cumulative hospital stay, and mortality over follow-up than those with none of these conditions. These findings support a concept of the combination of cognitive impairment (overall or specific domains) and pre-frailty representing cognitive frailty, with subsequent adverse consequences. In view of the reversibility of the frailty continuum (29) and non-pharmacological strategies to improve frailty status and cognitive impairment (30–34), early detection of cognitive frailty has public health implications since participation in group exercises that combines aerobic and resistance elements with or without cognitive training may retard decline or even lead to some improvement (35). This concept of earlier detection of abnormalities is similar to the current thinking in dementia research, where intervention may be more effective if applied at an early stage.

Our finding is in close agreement with some previous studies which consistently show a higher prevalence of cognitive impairment among physically pre-frail/frail elderly (14, 36), supporting the notion that physical and cognitive impairment are closely related and are integral components of frailty. Our findings also extend a previous study examining the association of impaired cognition with frailty (8, 9) by showing the longitudinal relationship between low cognitive scores and higher risk of incident pre-frailty/frailty, which support results of previous studies proposing the inclusion of cognitive function in the assessment of frailty (3, 37, 38). Several mechanisms might explain the association between cognitive impairment and increased risk of frailty. First, poor cognition in robust individuals may be associated with underlying risk factors (e.g., poor nutritional status, physical inactivity) known to affect the development of frailty. Second, the association could reflect the existence of shared factors (e.g., increased inflammatory markers) that may be causing cognitive decline and the onset of frailty (39, 40).

Given the demonstrated increased risk of developing frailty associated with cognitive impairment at baseline, we further examined the physical and cognitive profile at the 4-year follow-up of participants with both cognitive impairment and pre-frailty at baseline. These participants had lower grip strength, lower gait speed, and poorer performance in the chair stand test as compared to robust and cognitively intact participants; and had poorer cognitive performance in the CMMSE test (in terms of CMMSE total score) compared to their cognitively intact counterparts. Furthermore, they had significantly lower delayed recall domain score. These findings concur with findings from a recent study, which demonstrated that individuals with cognitive frailty showed worse performance in cognitive function, as assessed by a battery of neuropsychological tests than their cognitively normal peers (41). However, these participants did not have poorer performance in non-memory function, suggesting that memory function may decline first in the pre-frail state, while non-memory cognitive function such as executive function and attention may be more closely associated with frailty, but not pre-frailty (42).

In the present study, participants with pre-frailty and cognitive impairment at baseline had increased risks of poor quality of life, incident physical limitation, increased cumulative hospital stay, and mortality over follow-up, independent of age, sex, educational levels, and other potential cofounders. These findings are compatible with previous findings that a measure of frailty that combines a range of diverse deficits, including cognitive functioning, is a better predictor of adverse health outcomes. For example, in the Three-City Study and the Singapore Longitudinal Ageing Studies (18, 20), including cognitive impairment to the operational criteria defining the frailty phenotype could increase its predictive validity with regard to adverse health outcomes. In a sample of community-dwelling Koreans aged 65 years and older, frail persons with cognitive impairment had a lower survival rate as compared to those non-frail and not cognitively impaired (43).

Although cognitive impairment improves predictive validity of frailty, there is no consensus on how cognitive impairment should be defined, and numerous different criteria exist (e.g., amnestic and non-amnestic cognitive impairment; single-domain and multiple-domain impairment). Cognitive impairment is a transitional state between normal cognition and dementia; thus, varying the threshold used for defining impairment would results in different rates of cognitive impairment. To capture cognitive impairment at a point at which the decline in multiple systems is still occurring in its earliest stages, the early symptoms of cognitive impairment (e.g., memory problems, difficulties in word finding) were tested against multiple-domain cognitive impairment to be used in the criteria for cognitive impairment in terms of their predictive value of adverse outcomes. Our findings demonstrated that lower scores on the two selected a priori domains (delayed recall as well as language and praxis) in combinations with pre-frailty at baseline were associated with higher risks of incident physical limitation and increased cumulative hospital stay over follow-up, suggesting that single-domain cognitive impairment may be useful in risk prediction. Although evidence has shown that multiple-domain amnestic cognitive impairment may be a better predictor of dementia than single-domain amnestic or non-amnestic cognitive impairment (44, 45), those with single-domain cognitive impairment have a relatively high rate of reversion to normal cognition (46). Furthermore, multiple-domain cognitive impairment possibly represents a heterogeneous group of individuals with different neuropsychological profiles; hence subtyping cognitive impairment according to number and types of domains impaired may improve the characterization of the cognitive impairment construct and be useful for risk prediction in relation to different outcomes (47). From the clinical practice point of view, a short screening tool would be important, followed by interventions. Our findings suggest that the use of single cognitive domain may be effective in characterizing cognitive impairment groups; and the use of pre-frailty also identifies a subset of individuals at risk of progressing to frailty. Taken together, the findings of this study together with current available literature of cross-sectional and longitudinal studies lend support to the concept that cognitive frailty may be defined as the existence of overall cognitive impairment (or an individual domain) together with pre-frailty. This definition obviates the need for a psychiatric diagnosis such as the concomitant diagnosis of MCI (as proposed by Kelaiditi et al.) (17), or the need to consider cognitive frailty as a precursor condition of dementia.

There were some limitations in this study. First, the study participation was voluntary which could result in selection bias. Compared to the general elderly population in Hong Kong, the participants may represent those who are more robust, as they tended to be more health conscious, had a higher educational level and more physical active. The other limitation relates to the use of CMMSE for delineation of cognitive status. Due to ceiling effect, it may under-diagnose individuals with early dementia such that these individuals were included in the sample. Similarly, it may under-diagnose cognitive impairment such that some individuals, in particular highly educated individuals, are classified as “no cognitive impairment,” (48) albeit the expected effect would be a bias toward the null. Another limitation is the use of a priori selected domains and the arbitrary domain scores from CMMSE, which would be expected to be less psychometrically robust compared with domain scores derived from a neuropsychological battery, and may potentially lead to more false-positives among older people with lower educational levels. Another limitation could be represented by the collection of length of hospital stay using the Hong Kong Hospital Authority records that do not cover the 100% of hospitalization in the Hong Kong population. Finally, the incidence of dementia was not available. However, data regarding incident dementia is being collected in an ongoing follow-up study, which allows incident dementia to be related to baseline cognitive frailty.

In conclusion, our results showed that robust and cognitively impaired participants had higher risks of becoming pre-frail/frail over a period of 4 years than their counterparts with normal cognition. Furthermore, cognitive impairment improved the predictive power of pre-frailty for poor quality of life, incident physical limitation, increased cumulative hospital stay, and mortality. The findings of this study support to the concept that cognitive frailty may be defined as the occurrence of both cognitive impairment and pre-frailty [as opposed to established frailty as per the IANA/IAGG definition by Kelaiditi et al. (17)], not necessarily progressing to dementia. Our results also showed that lower scores in delayed recall as well as language and praxis, in combinations with pre-frailty, may also be used as criteria for cognitive impairment in terms of their predictive value of adverse outcomes.

All participants gave written consent in accordance with the Declaration of Helsinki. The study was approved by the Clinical Research Ethics Committee of the Chinese University of Hong Kong.

RY, JM, TK, and JW conceived and designed the study and wrote the paper; RY, JL, and OC carried out the analysis.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer WL declared a past co-authorship with the authors JM, JW to the handling editor.