Risk factors associated with cognitive frailty development through different transition pathways among community-dwelling older adult Japanese individuals: insights from the NILS-LSA project

Abstract

Background

The risk factors on the transition from robustness to cognitive frailty (CF), characterized by concurrent cognitive impairment and physical frailty (PF), may vary through different transition routes.

Methods

Data from the National Institute for Longevity Sciences-Longitudinal Study of Aging (2000−2012) were analyzed, including data for 1,061 older individuals (baseline age: 60–83 years; 51.5% men) who had a Mini-Mental State Examination (MMSE) score ≥ 24, and were dementia- and CF-free at baseline. Mild cognitive impairment (MCI) was defined as a MMSE score 24–27, PF as ≥ 1 Fried criteria, and CF as both conditions. Dietary diversity (assessed using the Quantitative Index for Dietary Diversity score; 0–1) and energy intake were categorized into sex-specific tertiles (T1–T3). Multistate modeling estimated hazard ratios of risk factors for transitions among robustness (no MCI and PF), MCI, PF, and CF states.

Results

At baseline, robustness, MCI, and PF proportions were 36.9%, 14.4%, and 48.7%, respectively. In robust individuals, aging increased (HR = 1.22), and high dietary diversity (HR = 0.62 for T3) and higher energy intake (HR = 0.36 for T2, 0.37 for T3) mitigated MCI risk. Conversely, obesity (HR = 1.37) and smoking (HR = 1.66) elevated, and stroke history (HR = 0.43) and high dietary diversity (HR = 0.63 for T3) lowered PF risk. Among those with MCI, female sex (HR = 1.69) and higher education (HR = 1.75 for 10−12 years, 2.19 for ≥ 13 years) facilitated recovery, while moderate energy intake impeded it (HR = 0.40 for T2). Aging (HR = 1.32), female sex (HR = 2.63), and smoking (HR = 2.91) increased CF risk, whereas moderate energy intake mitigated it (HR = 0.50 for T2). In PF participants, moderate dietary diversity (HR = 1.49 for T2), high energy intake (HR = 1.39 for T3), and alcohol consumption (HR = 1.32) promoted recovery, while depressive symptoms (HR = 0.25) and living alone (HR = 0.58) hindered it. Aging (HR = 1.28) increased CF risk, whereas higher education decreased it (HR = 0.55 for 10−12 years, 0.54 for ≥ 13 years).

Conclusions

Tailoring intervention strategies according to specific health statuses of older adults may prevent and delay CF onset and facilitate recovery from MCI and PF.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12877-025-06295-9.

Background

In 2013, International Academy of Nutrition and Aging and International Association of Gerontology and Geriatrics defined “cognitive frailty” (CF) as a condition marked by concurrent cognitive impairment and physical frailty (PF) [1]. Subsequent research highlighted the association of CF with notably elevated risks of dementia, long-term incident disability, overall mortality, and other adverse outcomes [2–5]. Considering the high prevalence of CF in older adults worldwide (ranging from 2.2 to 16.0% [6–11]), CF prevention in older adults has emerged as a pivotal focus for promoting healthy aging.

However, the current literature on CF risk/protective factors remains limited. While previous studies have identified several potential risk/protective factors for CF development, including demographic characteristics (e.g., advanced age [11–21] and female sex [15–17, 21, 22]), lifestyle factors (e.g., alcohol drinking [16] and physical inactivity [12, 19]), socioeconomic status (e.g., educational level [16, 17, 19, 20, 22, 23] and single status [17, 20]), diet [13, 19] and nutritional status [16, 18–20], and pre-existing health conditions (e.g., comorbidities and depression) [11–13, 15, 16]18– [22, 24], the evidence remains fragmented and inconsistent. Moreover, many studies on this topic have been cross-sectional in nature, and longitudinal studies have typically involved only a single follow-up assessment. Consequently, these investigations have failed to elucidate the impact of risk factors on diverse pathways leading to CF (such as progression from robustness to cognitive impairment and then to CF, or from robustness to PF and then to CF) or the potential for recovery prior to CF onset. Notably, the sole study that has addressed the risk factors associated with various pathways to CF to date [25] did not explore the effects of potential risk factors on recovery from cognitive impairment or PF. Such insights are crucial for developing CF prevention strategies in both clinical and public health settings, especially in Japan which is experiencing rapid population aging.

Hence, the objective of the current study was to examine previously reported risk factors for CF onset by analyzing long-term longitudinal (panel) data among older community-dwelling individuals. Given the complex and dynamic nature of CF development, we employed multistate modeling - a sophisticated statistical approach that allows us to simultaneously examine multiple transitions between different health states, including both progression and recovery. This approach is particularly valuable as it can quantify not only the risk of transitioning from one state to another but also the time spent in each state, providing a more comprehensive understanding of the CF development process [26]. Using this method, we sought to elucidate the diverse pathways connecting robustness and CF, including pathways from robustness to cognitive impairment to CF, from robustness to PF to CF, and potential reversals from cognitive impairment and PF. Furthermore, we aimed to assess the impact of risk factors on cognitive and physical functioning decline as well as recovery within the framework of transition intensity (i.e., considering time). We expect that the insights gained from this study will inform the development of targeted CF prevention strategies in clinical and public health settings, contributing to the promotion of healthy aging in Japan and globally.

Methods

Study cohort

The data for this study were collected as part of the National Institute for Longevity Sciences-Longitudinal Study of Aging (NILS-LSA) project, which is a Japanese population-based prospective cohort study of normal aging and age-related diseases. The study participants were recruited through age- and sex-stratified random sampling from the neighborhoods of the institute, Obu, and Higashiura Town in Aichi Prefecture, Japan. The first wave of the NILS-LSA was conducted from November 1997 to April 2000 and included 2,267 participants aged 40–79 years at baseline. The participants were followed up every 2 years, and those who were unable to participate in the follow-up survey (aged ≤ 79 years) were replaced by new age- and sex-matched participants. In addition, participants aged 40 years were recruited annually. The details of the NILS-LSA have been reported previously [27].

Participants for this study were selected from the second (April 2000 to May 2002) to seventh (July 2010 to July 2012) waves of the NILS-LSA, as data on both PF assessments and cognitive function tests were available for these periods. Among the 3,635 initially selected participants, 1,425 individuals under the age of 60 years were excluded because cognitive function tests were conducted solely among older participants. Additionally, participants with data on only one visit (n = 487); with missing data on dietary assessments (n = 103), cognitive function tests (n = 14), or PF assessments (n = 198); with a Mini-Mental State Examination (MMSE) score of < 24 or a self-reported history of dementia (n = 39); and those with missing covariate data (n = 52) were excluded. This resulted in a sample of 1,317 participants, accounting for a cumulative total of 4,912 study visits. Participants who exhibited CF (defined in the next section) at baseline (n = 248) and those who developed dementia before the onset of CF (n = 8) were excluded from the analysis. Consequently, a total of 1,061 older Japanese community-dwelling adults (aged 60–83 years at baseline; 51.5% men) were included in the current study. These individuals accounted for a cumulative observation count of 3,647 study visits (median [range]: 3 [2 − 6] visits), and with a follow-up duration of 4.3 (1.8 − 10.8) years.

Assessments of risk factors

For each wave, data on age, sex, disease history (stroke, hypertension, diabetes, and dyslipidemia; yes or no for each), smoking status (current, former, or never), alcohol consumption status (current and former/never), living status (alone or with others), and educational level (years) were collected using self-reported questionnaires.

Height (m) and weight (kg) were measured using a digital scale. Body mass index (BMI; kg/m²) was calculated as weight divided by height squared. The 24-hour total physical activity was assessed according to the metabolic equivalent of task (MET) scores (METs-h/day) obtained from participant interviews, which were conducted by trained interviewers using a semiquantitative assessment tool, i.e., the modified Minnesota Leisure-Time Physical Activity (MLTPA) Questionnaire [28]. Depressive symptoms were assessed using the self-administered Center for Epidemiologic Studies of Depression (CES-D) scale [29, 30].

Dietary intake was assessed using a 3-day (2 weekdays and 1 weekend day) dietary record. Participants weighed food portions (in grams) before cooking and used a disposable camera to photograph the meals before and after eating. Dietitians reviewed the photos, resolved discrepancies through phone calls, and calculated the average energy intake (kcal/day) using the 2010 standard tables of food composition in Japan as well as other sources [31]. Dietary diversity was determined using Quantitative Index for Dietary Diversity (QUANTIDD), which ranges from 0 (homogeneous diet) to 1 (diverse diet) [32]. Based on the 2010 Japanese food composition tables, which contains 18 food groups, the QUANTIDD score was calculated for 13 groups [33]: grains, potatoes and starches, pulses, nuts and seeds, non-green-yellow vegetables, green-yellow vegetables, fruits, mushrooms, algae, fish and shellfish, meat, eggs, and dairy products (beverages and seasonings were excluded) [34].

Evaluation of CF

CF was defined based on both cognitive and PF status [35]. Cognitive status was assessed using MMSE [36] (Japanese version [37]). Participants with an MMSE score between 24 and 27 were defined as having mild cognitive impairment (MCI), whereas those with an MMSE score ≥ 28 were defined as cognitively normal.

PF was assessed based on the Cardiovascular Health Study criteria specifically modified for older adult Japanese individuals [38]. The criteria include the following: (1) weight loss (a weight loss ≥ 5% in the previous 2 years; i.e., as compared with participants’ weight in the previous wave); (2) weakness (maximum grip strength within two trials for each hand, < 26 kg for men and < 18 kg for women); (3) exhaustion (a response of “≥ 1 day within the past week” to either of the following two questions from the CES-D scale [30]: “I felt that everything I did was an effort” and “I could not get ‘going’”; 4) slowness (a gait speed < 1.0 m/s in a 10 m walk test using a comfortable gait, or disturbances in gait); and 5) low physical activity (the lowest 20% MET of leisure-time physical activity by sex, assessed based on the modified MLTPA Questionnaire [39]). PF (including pre-PF) was defined as the presence of ≥ 1 criterion.

In each wave of the survey, participants with MCI only, PF only, and both MCI and PF were assigned to the MCI, PF, and CF groups, respectively, whereas participants who had neither MCI nor PF were defined as the robust group.

Statistical analysis

Dietary diversity scores (points) and energy intake (kcal/day) were categorized into three groups based on sex-specific tertiles, considering the well-documented gender differences in dietary patterns and food choices. This approach allows for more appropriate comparisons within each gender group and accounts for the inherent differences in dietary habits between men and women. BMI was classified into three groups: < 18.5, 18.5− < 25.0, or ≥ 25.0 kg/m². Smoking and alcohol consumption statuses were grouped into two categories: current or former/never. Educational level was classified as ≤ 9, 10 − 12, or ≥ 13 years. Depressive symptoms were defined as CES-D scores ≥ 16 [29, 30].

Differences in mean values (for continuous variables) and proportions (for categorical variables) were assessed using a general linear model and the chi-square test, respectively. For comparing variables that showed distribution differences among multiple groups, Dunn’s test [40] was performed for continuous variables (P values were adjusted using the Bonferroni method), and Fisher’s exact test was performed for categorical variables (P values adjusted using the Benjamini-Hochberg method).

Multistate models [41] were specified to analyze the data and characterize the nature of CF development. Four potential states through which participants could transition were identified: robustness, MCI only, PF only, and CF. To keep the model from becoming too complex and not converging and to focus on how risk factors affect the different pathways to CF, we made some assumptions. We assumed that before developing CF, the participants would experience either MCI or PF, but not both simultaneously. We also assumed that no direct transition occurred between the MCI and PF states. Thus, to go from one to the other of these states, participants would first pass through the robustness state. Therefore, transitions between robustness and MCI (bidirectional), robustness and PF (bidirectional), and MCI or PF and CF (unidirectional) were allowed (Fig. 1). CF was defined as an absorbing state that marked the end of the observation once it was reached by a participant. The number of transitions over visit intervals (snapshots) among participants is displayed in Supplementary Table 1.

Fig. 1.

Potential transitions between four states

With four states and time variables (representing follow-up years) alone, multistate models provide estimated transition intensities, that is, the maximum likelihood estimates and 95% confidence intervals (CIs) for predefined transitions. Upon incorporating the input covariates into the model, estimated hazard ratios (HRs) and corresponding 95% CIs were generated to indicate how the instantaneous risk of a specific transition was modified by the covariates [41, 42]. Therefore, to investigate risk factors associated with the development of CF, several models were created: Model 1 included only age (per 5 years; continuous) and sex; Model 2 added medical history (stroke, hypertension, diabetes, dyslipidemia; yes/no for each; categorical) to Model (1) Model 3 further added lifestyle behaviors factors, such as BMI (< 18.5, 18.5−< 25.0, or ≥ 25.0 kg/m²; categorical), dietary diversity score (sex-specific tertiles; categorical), energy intake (kcal/day; sex-specific tertiles; categorical), smoking status (current, former/never; categorical), and alcohol drinking status (current, former/never; categorical) to Model (2) Model 4 added socioeconomic factors, such as educational level (≤ 9, 10 − 12, or ≥ 13 years) and living status (alone/with others), and depressive symptoms (present/absent) to Model (3) Finally, in Model 5, to enhance the predictive performance of the model, we retained risk factors that demonstrated a significant effect or a tendency (i.e., an upper limit of 95% CI < 1.15 for variables with a point estimation value < 1 or a lower limit of 95% CI > 0.85 for variables with a point estimation value > 1) to have a significant effect (HR) on specific transitions, based on the results of Model (4) For instance, if Model 4 indicated that current smoking status did not tend to affect the transition intensity from robustness to MCI, we would not consider current smoking status as a risk factor for this transition. Consequently, we excluded this variable from the robustness to MCI transition, while retaining it for other transitions.

All statistical analyses described above were two-sided and were accompanied by an exact P value. Statistical significance was set at P < 0.05. All analyses were performed using R Statistical Software (v4.3.2; R Core Team 2023; https://www.r-project.org/).

Results

Baseline characteristics of participants

Table 1 shows the baseline characteristics of the participants categorized according to their initial state (i.e., robustness, MCI, and PF). Multiple comparisons suggested that, compared with participants in the robust group, participants with MCI were older, and participants with PF were more likely to live alone. When comparing participants in the robust and MCI groups to those with PF, the latter were less likely to be men or current drinkers and more likely to have a low dietary energy intake and depression. Additionally, participants with MCI were more likely to have an education level of 9 years or less than those who were robust or physically frail at baseline.

Table 1.

Baseline characteristics of participants (n = 1,061)

		States at baseline			P value a
		Robustness	MCI	PF
No. of participants		391	153	517
Age (years), Mean (SD)		65.8 (5.9) b	67.5 (6.3) b	66.6 (6.4)	0.013
Male, %		56.0 c	64.1 d	44.3 c, d	< 0.001
Medical history (Yes), %
	Stroke	3.1	2.0	4.1	0.409
	Hypertension	32.7	29.4	39.1	0.036
	Diabetes	9.5	7.2	9.1	0.698
	Dyslipidemia	26.1	19.0	24.4	0.217
BMI (kg/m2), Mean (SD)		23.1 (2.6)	23.2 (2.7)	22.8 (3.0)	0.117
Dietary diversity score (point), Mean (SD)		0.88 (0.05)	0.87 (0.05)	0.88 (0.05)	0.153
Energy intake (kcal/day), Mean (SD)		2100.1 (416.1) c	2113.5 (407.2) d	2002.0 (410.0) c, d	< 0.001
Current smoker, %		11.5	19.0	16.4	0.040
Current drinker, %		63.4 c	66.7 d	50.9 c, d	< 0.001
Educational level (years), %
	≤ 9	23.3 b	37.3 b, d	26.5 d	0.020
	10−12	45.0	39.2	44.9
	≥ 13	31.7	23.5	28.6
Living alone, %		5.4 c	7.2	11.4 c	0.005
Depressive symptoms (present), %		2.0 c	0.7 d	17.6 c, d	< 0.001

^aDifferences in mean values (for continuous variables) and proportions (for categorical variables) were assessed using a general linear model and the chi-square test, respectively

^{b, c, d}Multiple comparisons were conducted using Dunn's test for continuous variables (P values were adjusted using the Bonferroni method), while pairwise comparisons were performed using Fisher's exact test for categorical variables (P values were adjusted using the Benjamini-Hochberg method). The results revealed significant differences between groups

MCI mild cognitive impairment, PF physical frailty, BMI body mass index

Transitions among robustness, MCI, PF, and CF

The mean (SD) follow-up duration (from first to last visit) of all participants was 5.2 (2.8) years. Robust participants were 1.5 times more likely to develop PF than experience onset MCI. After the onset of MCI, recovery was 3.57 (= [0.535–0.117]/0.117) times more likely than progression to CF, and after PF onset, recovery was 1.21 (= [0.232–0.105]/0.105) times more probable than progression to CF (Table 2). Before transitioning to other states, the mean amount of time participants spent in robustness, MCI, and PF was 2.0 (=|1/−0.490|), 1.5 (=|1/−0.652|), and 3.0 (=|1/−0.337|) years, respectively.

Table 2.

Estimated transition intensities and 95% CIs during follow up (n = 1,061) ^a

Transitions	Estimated transition intensities
	Maximum likelihood estimates	95% CI
Robustness → Robustness	−0.490	(−0.544, −0.441)
Robustness → MCI	0.198	(0.164, 0.240)
Robustness → PF	0.292	(0.258, 0.331)
MCI → Robustness	0.535	(0.441, 0.649)
MCI → MCI	−0.652	(−0.773, −0.550)
MCI → CF	0.117	(0.088, 0.156)
PF → Robustness	0.232	(0.203, 0.264)
PF → PF	−0.337	(−0.373, −0.304)
PF → CF	0.105	(0.092, 0.121)

^aAnalyzed using multistate model

MCI mild cognitive impairment, CF cognitive frailty, PF physical frailty, CI confidence interval

Risk factors affecting transitions among robustness, MCI, PF, and CF

Table 3 shows the HRs (95% CIs) of the predefined risk factors for each transition in the final model (Model 5) (see Supplementary Table 2 for the log-likelihood value and Akaike information criterion for each model and Supplementary Table 3 for the results of Models 1–4). For individuals in a robust state, advancing age increased the risk of developing MCI, whereas high dietary diversity and energy intake decreased this risk. Conversely, obesity and smoking increased the risk of developing PF, whereas a history of stroke and high dietary diversity reduced this risk.

Table 3.

HRs and 95% CIs of associations between risk factors and transitions among robustness, MCI, PF, and CF (n = 1,061) ^{a, b}

	Age	Sex	Stroke	Hypertension	Diabetes	Dyslipidemia	BMI (kg/m2)		Dietary diversity		Energy intake (kcal/day)		Current smoker	Current drinker	Depressive symptoms	Education levels (year)		living alone
	Per 5 years	Female	Presence	Presence	Presence	Presence	< 18.5	≥ 25.0	Sex-specific tertile 2	Sex-specific tertile 3	Sex-specific tertile 2	Sex-specific tertile 3	Yes	Yes	Presence	10−12	≥ 13	Yes
Model 5
Robustness → MCI	1.22 (1.08, 1.38)	-	-	-	-	-	-	-	0.77 (0.55, 1.10)	0.61 (0.42, 0.87)	0.36 (0.17, 0.77)	0.37 (0.17, 0.81)	-	-	-	-	-	-
Robustness → PF	1.06 (0.94, 1.20)	1.26 (0.95, 1.67)	0.43 (0.21, 0.89)	1.20 (0.90, 1.59)	-	1.22 (0.88, 1.69)	1.64 (0.93, 2.89)	1.37 (1.06, 1.77)	0.90 (0.65, 1.26)	0.63 (0.45, 0.89)	-	-	1.66 (1.14, 2.41)	-	-	-	-	-
MCI → Robustness	- c	1.69 (1.23, 2.33)	-	-	-	-	-	-	-	-	0.40 (0.17, 0.90)	0.49 (0.22, 1.12)	0.70 (0.42, 1.17)	-	-	1.75 (1.21, 2.52)	2.19 (1.47, 3.26)	-
MCI → CF	1.32 (1.06, 1.65)	2.63 (1.41, 4.90)	-	-	-	-	0.84 (0.22, 3.24)	1.31 (0.66, 2.59)	-	-	0.50 (0.26, 0.96)	0.63 (0.33, 1.19)	2.91 (1.33, 6.37)	-	-	-	-	-
PF → Robustness	0.91 (0.81, 1.04)	0.81 (0.60, 1.10)	0.51 (0.23, 1.11)	0.83 (0.61, 1.11)	-	1.25 (0.89, 1.77)	-	-	1.49 (1.03, 2.15)	1.32 (0.93, 1.87)	1.06 (0.79, 1.43)	1.39 (1.05, 1.84)	-	1.32 (1.03, 1.69)	0.25 (0.16, 0.39)	0.92 (0.69, 1.22)	0.74 (0.54, 1.02)	0.58 (0.36, 0.93)
PF → CF	1.28 (1.15, 1.44)	0.78 (0.57, 1.06)	-	-	1.49 (0.98, 2.27)	-	0.69 (0.37, 1.28)	1.31 (0.95, 1.79)	1.00 (0.72, 1.37)	0.78 (0.55, 1.11)	-	-	1.40 (0.91, 2.14)	-	-	0.55 (0.40, 0.75)	0.54 (0.37, 0.77)	-

^aAnalyzed using multistate model

^bThe unshown subgroup of each variable, i.e. male in sex, absence of any medical history, BMI between 18.5 and 25.0 kg/m², sex-specific tertile 1 of dietary diversity and energy intake (kcal/day), former/never smoker, former/never drinker, absence of depressive symptoms, education level ≤ 9 years, and not living alone, were used as reference groups in the model

^cThe"-"signifies that the variable was not considered as a risk factor in the particular transition within the model (based on the results of Model 4 in the Supplementary Table 3)

MCI mild cognitive impairment, CF cognitive frailty, PF physical frailty, CI confidence interval, HR hazard ratio

For participants with MCI, being a woman and having a higher level of education contributed to recovery to a robust state, whereas moderate energy intake decreased the likelihood of recovery. Conversely, aging, female sex, and smoking contributed to the deterioration of CF, whereas moderate energy intake reduced this risk.

In participants with PF, moderate dietary diversity, high energy intake, and alcohol consumption were positively correlated with recovery to a robust state, whereas depressive symptoms and living alone were negatively correlated. Similarly, for such individuals, aging increased the risk of developing CF, whereas having a higher level of education decreased this risk.

Discussion

In this longitudinal study spanning up to 10.8 years, we investigated the risk factors affecting transitions among robustness, MCI, PF, and ultimately deterioration to CF in a community-dwelling older adult Japanese population. We observed that various transition pathways not only share certain factors that promote or inhibit their progression, but also possess specific influencing factors. The specificity of these factors may be determined by the physiological and pathological mechanisms of the transition pathways. To the best of our knowledge, no previous study has used long-term follow-up data to consider risk factors influencing CF pathogenesis comprehensively. Notably, our multistate modeling approach revealed that some risk factors (such as aging) showed consistent effects across different transition pathways, while others (such as dietary factors and education) demonstrated pathway-specific impacts, highlighting the complex nature of CF development.

Consistent with those of previous studies, our results suggested that advancing age [43, 44] is a risk factor for developing MCI and that high dietary diversity is associated with a lower risk of MCI [45, 46] in robust individuals. Intriguingly, higher energy intake, which is a risk factor for cognitive impairment [47, 48], decreased the risk of MCI onset in our participants. However, upon examining the baseline mean energy intake by sex-specific tertile groups of participants transitioning from robustness to MCI, we found that the mean energy intake of the 1 st tertile, 2nd tertile, and the 3rd tertile were 1834.0 kcal/day, 2191.4 kcal/day, and 2647.9 kcal/day for men (with corresponding mean ages of 71.2 years, 69.5 years, and 65.2 years), and 1522.2 kcal/day, 1793.2 kcal/day, and 2165.8, respectively, for women (with corresponding mean ages of 67.8 years, 71.3 years, and 66.8 years), respectively. According to Overview of the Dietary Reference Intakes for Japanese (2020) [49], the recommended energy intake for older adult people (aged 65–74 years) is 2,400 kcal/day for men and 1,850 kcal/day for women. Therefore, in our study, participants in the 2nd and 3rd tertiles of the energy intake group had energy intakes that were more closely aligned with the recommended values, which may have had a positive effect on their cognitive function.

Supported by abundant previous evidence, our findings also indicated that for robust participants, obesity [50, 51] and smoking [52, 53] were associated with a higher risk of PF, whereas dietary diversity had an inhibitory effect on the development of PF [54, 55]. Interestingly, we noted a protective effect of prior stroke against transition from robustness to PF, contradicting the common assumption that a history of stroke would promote this transition [56], with stroke patients typically experiencing greater PF [57]. The discrepant results between our study and those of other studies may stem from various factors. First, the community-based recruitment survey setting of the NILS-LSA may have had an impact on the results. Individuals with a history of stroke recruited in this setting may have experienced nonfatal strokes, including cerebral microbleeds, which may not have significantly affected their physical function. For instance, in our study, 34.7% of the participants who initially reported a history of stroke were classified as robust. Additionally, participants with a history of stroke who were still able to participate in the follow-up survey were likely to possess the ability to manage their condition well, and these individuals may thus also have the ability to maintain and enhance their functional abilities.

In the present study, among individuals with MCI, higher levels of education contributed to their return to a robust state, a finding that has been acknowledged previously [58]. Aging and smoking play a role in exacerbating the decline to CF, potentially via mechanisms that facilitate the emergence of PF. Notably, female sex and moderate energy intake concurrently served as both facilitators (in the case of female sex) and inhibitors (in the case of moderate energy intake) of recovery and deterioration; however, the reasons for this are currently unclear.

In individuals with PF, we found a positive correlation between recovery to a robust state and high dietary diversity, energy intake, and alcohol consumption. A previous observational study pointed out that energy intake is negatively associated with CF in pre-frailty [19]. Moreover, randomized controlled trials have demonstrated that increasing energy intake and nutrient diversity may lead to the reversal of PF status [59, 60]. Nonetheless, caution is warranted when interpreting the potentially beneficial effects of alcohol consumption on PF reversal. While studies have indicated that alcohol consumption, particularly in moderation, is linked to a reduced risk of PF as compared with that for abstention from alcohol [61–63], researchers often attribute this association to confounding factors, such as social support or networks and reverse causality. In contrast, our results suggested that individuals with depressive symptoms and those living alone exhibited a diminished likelihood of recovering from PF. Depression is widely acknowledged as a bidirectionally associated risk factor for PF [64]. Additionally, multi-domain interventions aimed at reversing PF among older individuals living in the community have also been shown to reduce depressive symptoms [65], implying that the presence of depressive symptoms may impede the recovery of frail individuals. In addition, a previous study has reported that individuals who recover from PF are characterized by high individual-level social capital components, such as interaction with neighbors and social participation [66]. This underscores the detrimental effect of solitary living arrangements on the reversal of PF.

In our participants with PF, advancing age was associated with an elevated likelihood of developing CF, whereas a higher educational level was correlated with a reduced risk of CF onset. With aging, cognitive function declines concomitantly, which predisposes to CF development. Conversely, not only have higher levels of education been found to attenuate the progression of frailty [67] but lower education levels have also been linked to CF in frail older adult populations [19], emphasizing the importance of cognitive reserve in CF prevention.

When considering CF in the broader context of age-related cognitive-physical syndromes, it is important to note its distinct characteristics compared to related conditions such as Motoric Cognitive Risk Syndrome (MCRS) [68] and Physio-Cognitive Decline Syndrome (PCDS) [69]. CF represents a more advanced stage of decline which is associated with significantly elevated risks of adverse outcomes including falls, disability, dementia, and mortality [3–5]. This advanced state typically requires more intensive, multidimensional intervention strategies, and the potential for recovery may be more limited due to the complexity of pathophysiological changes and severity of functional impairments. In contrast, MCRS and PCDS generally represent earlier stages in the cognitive-physical decline spectrum. MCRS, characterized by slow gait and subjective cognitive complaints without clear cognitive deficits, and PCDS, which captures pre-clinical changes in both domains, may offer greater opportunities for successful intervention. The earlier identification of these conditions provides a critical window for preventive strategies, including targeted physical activity, cognitive training, and nutritional support, potentially delaying or preventing the progression to CF. This hierarchical understanding of cognitive-physical syndromes underscores the importance of early detection and intervention, particularly given the challenges of reversing CF once established.

The present study, unlike studies to date, utilized longitudinal repeated-measurement data for comprehensive exploration of the diverse pathways leading to CF occurrence, originating from varying pre-CF health statuses, along with their associated risk factors. However, this study had several limitations. First, not all potential influencing factors (e.g., sleep duration and quality, medication, social support, and social networks) were incorporated into the analysis model because these variables were not consistently measured across all waves of the NILS-LSA. The absence of these factors in our analysis may limit our understanding of their roles in CF development. Second, due to the small sample size and model complexity, stratified analyses based on age or sex were not performed. Our attempts at sex-stratified analyses revealed that the reduced sample sizes in subgroups led to model convergence issues with multiple risk factors, and even simplified models produced unstable estimates with wide confidence intervals. This limitation potentially masks important age- and sex-specific patterns in transition pathways and risk factors. It is conceivable that different transition intensities and risk factors may exist among younger and older subsets within the older population, or between sexes. Third, individuals who participated in the survey only once were excluded from the study. Relative to the participants analyzed herein, they had less dietary diversity and energy intake, and lower overall education levels. They also had a higher prevalence of stroke, depressive symptoms, and smoking. Notably, the proportion of robust individuals among the single-visit participants was lower, and individuals with dementia and incomplete baseline data were identified. These differences suggest that our findings may not be generalizable to older individuals in poor health. Finally, due to panel data utilization, the precise timing of transitions between states cannot be ascertained, potentially introducing uncertainty into the parameter estimates. Larger, prospective studies with precise event time data and longer follow-up periods are warranted to confirm the findings of this study.

Conclusions

The pathway from robustness to CF may vary among individuals, and the associated risk factors may also differ. These factors can promote recovery or deterioration of physical and cognitive functions in older adults. For prevention of CF, the current physical and cognitive status of the individual should be considered, and targeted intervention strategies should be developed to prevent or delay the onset of CF effectively.

Abbreviations

BMI

body mass index

CES-D

Center for Epidemiologic Studies of Depression

CF

cognitive frailty

PF

physical frailty

CI

confidence interval

HR

hazard ratio

MCI

mild cognitive impairment

MET

metabolic equivalent of task

MLTPA

Minnesota Leisure-Time Physical Activity

MMSE

Mini-Mental State Examination

NILS-LSA

National Institute for Longevity Sciences-Longitudinal Study of Aging

QUANTIDD

Quantitative Index for Dietary Diversity

Supplementary Infotrmation

Supplementary Material 1. ^{(33.9KB, docx)}

Authors’ contributions

RO, FA, and HS contributed to the conception and design of the NILS­LSA. SZ and RO obtained funding for this study. RO, YN, CT, FA, and HS collected data. SZ drafted the plans for data analysis, conducted the data analysis, and drafted the manuscript. All the authors interpreted the results, revised the manuscript, and approved the final version. SZ and RO are the guarantors. The corresponding author attests that all listed authors meet the authorship criteria, and that no others meeting the criteria have been omitted. All the authors have read and agreed to the published version of the manuscript.

Funding

This work was partially supported by the Japan Society for the Promotion of Science (grant number 22K17847) and the National Center for Geriatrics and Gerontology, Japan (grant number 21 − 18, 24 − 10). The funding sources had no role in the study design, data collection, analysis, interpretation, writing of the report, or decision to submit the paper.

Data availability

The datasets analyzed in the current study are not publicly available for privacy reasons, but are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study strictly adhered to the principles outlined in the Declaration of Helsinki. Approval to conduct the study was obtained from the Committee on the Ethics of Human Research at the National Center for Geriatrics and Gerontology (Approval No. 1665-2) before data collection commenced. All the participants provided written informed consent for data collection and analysis before participating in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.