Intake of vegetables and fruits at midlife and the risk of physical frailty in later life

Abstract

Objectives

Our study evaluated the independent and overall associations of vegetable and fruit consumption at midlife with the likelihood of physical frailty in later life. We also investigated whether specific nutrients in these foods could have accounted for these associations, if present.

Design

Prospective cohort study.

Setting

A population-based cohort of Chinese adults followed over a period of 20 years in Singapore.

Participants

We used data from 11,959 subjects who participated in the baseline (1993–1998) and follow-up 3 (2014–2017) interviews of the Singapore Chinese Health Study.

Measurements

At baseline, dietary intake was evaluated using a validated food frequency questionnaire. During the follow-up 3 visits, physical frailty was assessed using a modified Cardiovascular Health Study phenotype that included weakness, slowness, exhaustion and weight loss. Multivariable logistic regression models were used to estimate odds ratios (OR) and 95% confidence intervals (CI) for the associations with physical frailty.

Results

Participants had mean ages of 52 years at baseline, and 72 years at follow-up 3. Baseline intake of vegetables, but not of fruits, showed a dose-dependent inverse relationship with physical frailty at follow-up 3 (P_trend = 0.001). Compared to participants in the lowest quintile of vegetable intake, those in the highest quintile had reduced odds of frailty [OR (95% CI): 0.73 (0.60−0.89)]. Among the components of physical frailty, vegetable intake had the strongest inverse association with weakness defined by handgrip strength [OR (95% CI) between extreme quintiles: 0.62 (0.52−0.73); P_trend < 0.001]. In models that were individually adjusted for nutrients, the vegetable-frailty association was attenuated and no longer statistically significant after adjusting for the intake of β-carotene, lutein, folate, α-carotene, and isothiocyanates.

Conclusion

: Increased midlife intake of vegetables was associated with reduced odds of physical frailty in later life, and the intake of β-carotene, lutein, folate, α-carotene, and isothiocyanates could have accounted for this association.

1. Introduction

Physical frailty is a syndrome characterized by diminished strength, endurance, and reduced physiologic function [1], and it has been associated with a wide variety of adverse health outcomes among older adults [2,3]. As the global population ages, the number of older adults with physical frailty is expected to rise concomitantly, resulting in burden on both individuals and healthcare systems [2,3]. As such, strategies to prevent physical frailty in older adults or slow its progression are of the utmost importance [2,3].

Certain healthy dietary patterns, such as the Mediterranean dietary pattern, have been shown to be associated with reduced risk of physical frailty [[4], [5], [6], [7], [8]]. Although the exact definitions of such dietary patterns vary, they tend to be characterized by greater consumption of vegetables and fruits [[4], [5], [6], [7]]. As such, many studies have sought to investigate the specific effects of vegetable and fruit intake on the development of physical frailty. However, their findings have been inconsistent [[9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31]]. For example, among studies that had assessed the intake of vegetables and fruits as a single food item, some found significant associations with physical frailty [14,26,27], whereas others reported null or inconclusive results [15,18,21,22]. Likewise, among studies that had investigated vegetable and fruits as distinct food items, there were some that found physical frailty to be associated with the consumption of both vegetables and fruits [[9], [10], [11], [12],19,20,28], some that found associations with vegetables but not fruits [13,16,[30], [31], [32]], and yet others that did not find significant associations with intake of either food item [17,[23], [24], [25],29].

Recent meta-analyses have pooled the available evidence, and showed that although a higher intake of vegetables and fruits as a single food item was indeed associated with decreased risk of physical frailty [32,33], it remained unknown if the consumption of fruits and vegetables, when adjusted for each other and assessed independently, was associated with this risk reduction. Furthermore, the number of cohorts included in these analyses were relatively small, and the authors noted that further large-scale prospective studies were still needed [32,33]. In addition, the majority of these studies were done in Western populations. Since the types of fruits and vegetables commonly consumed in Asian populations could be different due to socio-cultural factors, such as religion, traditional beliefs, and food preferences, as well as socio-economic factors, such as cost, availability, and ease of accessibility [34], a comprehensive study in an Asian population, looking at the intake of a variety of fruits and vegetables, assessed together and separately, would be necessary to fill the gap in the literature.

Hence, our study aimed to evaluate the independent associations of vegetable and fruit consumption at midlife with the likelihood of physical frailty in later life within a population-based cohort of Chinese adults living in Singapore over a follow-up period of 20 years. We also investigated if nutrients present in fruits and vegetables could have accounted for the associations, if present.

2. Methods

2.1. Study population

This study was nested within the population-based prospective cohort known as the Singapore Chinese Health Study (SCHS) [35]. Between April 1993 and December 1998, the SCHS enrolled 63,257 participants (27,959 men and 35,298 women) who were 45–74 years old. All participants were Chinese from the two major Chinese dialect groups in Singapore – the Hokkien and the Cantonese – who originated from the contiguous provinces of Fujian and Guangdong, respectively, in the southern part of China. After recruitment, consenting survivors were re-contacted for follow-up interviews every five to six years: follow-up 1 (1999–2004), follow-up 2 (2006–2010), follow-up 3 (2014–2017). The follow-up 3 interviews were conducted in person, by trained interviewers, from July 2014 to December 2017. In this study, we used data from a subset of 12,583 participants who had complete data on all four criteria that defined physical frailty at the follow-up 3 interviews. Supplementary Figure S1 illustrated the overall design of our study, and detailed the data that was used at each time-point. This study was approved by the Institutional Review Board at the National University of Singapore, and written informed consent was obtained from all study participants.

2.2. Assessment of diet at midlife

At the baseline interviews (1993–1998), participants’ usual dietary intake at midlife, including their intake of vegetables and fruits, was assessed through a 165-item semi-quantitative food frequency questionnaire that was developed for, and subsequently validated in, our study population [35]. For each type of food item, including fresh vegetables and fresh fruits, participants denoted their average intake over the past year by selecting from one of eight frequency categories (‘never or hardly ever’ to ‘two or more times a day’) and one of three serving sizes with accompanying photographs (‘small’, ‘medium’, or ‘large’). Daily intakes of each food item were derived by multiplying intake frequency with serving size. In this study, the intake of 25 types of fresh vegetables (excluding white potatoes) (Supplementary Table S1) and 14 types of fresh fruits (Supplementary Table S2) were assessed [36]. Intake of fresh seasonal fruits, such as tangerines, apricots, peaches, and persimmons, were further adjusted for their annual availability [37]. For beverages, including vegetable and fruit juices, participants selected from one of nine intake frequency categories (ranging from ‘never or hardly ever’ to ‘six or more times a day’) with reference to a standard serving size of 1 cup, which was defined as 237 ml. The total daily intake of vegetables and fruits was computed as the sum of the aforementioned fresh food items, plus any vegetable or fruit juices, plus any vegetables mixed into distinct dishes (rice, noodle, and meat dishes, as well as fast foods, sandwiches, dim sum, preserved foods, breads, crackers, and other snacks), and ultimately expressed in grams per day [[36], [37], [38]]. Vegetables were further categorized into overlapping groups of light-green, dark-green, yellow, and cruciferous vegetables, as well as tomatoes and mushrooms, as several of these subgroups had previously been associated with diabetes [38] and cognitive impairment [36] within our cohort.

Data from the food frequency questionnaire was also used to assess participants’ frequency of alcohol consumption, use of vitamin or mineral supplements (vitamin A, β-carotene, vitamin C, vitamin E, calcium, selenium, and zinc), and usual dietary intake. Participants’ total daily intakes of nutrients and energy were computed using values from the Singapore Food Composition Database, which was developed specifically for the SCHS cohort [35]. Intake of all food items and nutrients were adjusted for total daily energy consumption using the residual method [39].

In a previous validation study that involved 810 randomly selected participants, the food frequency questionnaire was compared against two 24-h dietary recalls, and correlation coefficients were reported for the following nutrients for men and women separately in the two dialect groups: dietary fiber (ranging from 0.65 to 0.72), β-carotene (ranging from 0.38 to 0.61), folate (ranging from 0.50 to 0.69), vitamin C (ranging from 0.63 to 0.67), and vitamin E (ranging from 0.53 to 0.61) [35].

2.3. Assessment of other covariates at midlife

At recruitment, interviewers utilized a structured questionnaire to obtain information on participants’ sociodemographic characteristics, history of physician-diagnosed comorbidities (hypertension, angina or heart attack, stroke, diabetes, and cancer), lifestyle factors (history of smoking, and amount of sleep per day), height, weight, and physical activity (hours of strenuous sports, vigorous work, and moderate activity per week) at midlife. Height and weight were self-reported, and body mass index (BMI) was calculated as weight divided by squared height.

2.4. Assessment of physical frailty in late life

During the follow-up 3 interviews (2014–2017), physical frailty was assessed using a modified version of the Cardiovascular Health Study (CHS) frailty phenotype [1,40,41] that included weight loss, exhaustion, slowness, and weakness [42]. The criterion of low activity was not included as we did not collect relevant data for this measure at the follow-up 3 interviews. Body weight was self-reported at the follow-up 2 (2006–2010) and follow-up 3 interviews, and participants met the criterion for weight loss at follow-up 3 if they had lost 10% or more of their weight since follow-up 2 [mean (SD) duration between interviews: 7.3 (1.0) years]. Participants met the criterion of exhaustion if they answered ‘No’ to the question “Do you feel full of energy?”. The criterion of slowness was met if participants’ timing for the timed up-and-go (TUG) test [43] fell within the slowest sex-specific quintile, and they met the criterion of weakness if their handgrip strength was in the weakest sex-specific quintile [44]. Participants who met two or more of these criteria were classified as being physically frail [42].

2.5. Statistical analyses

Continuous variables that were normally distributed were described with means [standard deviations (SD)] and compared using one-way analyses of variance (ANOVA), whereas continuous variables with skewed distributions were described with medians [interquartile ranges (IQR)] and compared using Kruskal-Wallis equality-of-populations rank tests. Categorical variables were described with N (%) and compared using Pearson’s chi-squared tests. We used multivariable logistic regression models to compute the odds ratios (OR) and 95% confidence intervals (CI) for the associations between quintiles of vegetable and fruit intake at midlife and physical frailty in later life. To test for a linear trend across quintiles of intake, the median values of exposure for each quintile were assessed as a continuous variable. In Model 1, we adjusted for basic sociodemographic characteristics [age at physical tests (years); gender; dialect group (Hokkien, Cantonese); level of education (no formal education, primary, secondary, diploma/university)]. In Model 2, we additionally adjusted for the following covariates assessed at the baseline interviews: history of physician-diagnosed comorbidities (hypertension, angina or heart attack, stroke, diabetes, cancer), history of smoking (never smoker, former smoker, current smoker), frequency of alcohol consumption (never, monthly, weekly, daily), body mass index group (<18.5, 18.5–22.9, 23.0–27.4, 27.5+ kg/m²), amount of sleep per day (≤5, 6, 7, 8, 9+ hours), physical activity at midlife (hours of strenuous sports, vigorous work, and moderate activity per week), total daily energy intake (kcal), total daily protein intake (g) [45], total daily caffeine intake (mg) [46], and use of any vitamin or mineral supplements at least once a week. In Model 3, vegetable and fruit intake at midlife were mutually adjusted for (quintiles as categorical variables). We then conducted separate analyses stratified by gender and by total daily protein intake. For the latter, participants were categorized into two groups based on the median daily intake of total protein in our study population; participants whose daily protein intake fell below the median were categorized as having low protein intake, while those whose daily protein intake were above the median were categorized as having high protein intake. In these stratified analyses, we tested for interactions by adding a product term between the exposure of interest and the factor of potential interaction into our fully adjusted models.

Next, we investigated the associations between vegetable intake and each criterion in our physical frailty phenotype (weight loss, exhaustion, slowness, and weakness), as well as the associations between physical frailty and different subgroups of vegetables (light-green vegetables, dark-green vegetables, yellow vegetables, cruciferous vegetables, tomatoes, and mushrooms) separately. Subsequently, the overall association between vegetable intake and physical frailty was adjusted for the total daily intake of several nutrients found in vegetables that might have been associated with physical frailty: dietary fiber, isothiocyanates, and carotenoids (α-carotene, β-carotene, lycopene, β-cryptoxanthin, lutein), as well as vitamins B9, C, and E [47,48]. Intakes of these nutrients were assessed as continuous variables (μmol/day, μg/day, mg/day, or g/day), and they were each added individually into our fully adjusted models. For these analyses involving multiple vegetable subgroups and nutrients, we corrected for multiple comparisons using the Holm-Bonferroni method [49].

Lastly, we conducted a sensitivity analysis by excluding participants who met either of two criteria at the baseline interviews: (1) 60 years old and above; (2) a history of hypertension, cardiovascular disease, diabetes, or cancer. All statistical analyses were conducted using Stata/SE 14.2 (StataCorp LLC, College Station, TX, USA). All P-values presented were two-sided, and P < 0.05 was considered statistically significant.

3. Results

A total of 12,583 participants had complete data on all four criteria that were used to define physical frailty at the follow-up 3 interviews. Of these, 490 participants were excluded for being 65 years or older at the baseline interviews, and a further 134 participants (52 men and 82 women) were excluded for having implausible daily energy intakes (<700 or >3700 kcal for men, <600 or >3000 kcal for women), thus leaving a total of 11,959 participants for this study. At the baseline interviews (1993–1998), the 11,959 participants included in this study had a mean (SD) age of 52 (5.2) years and ranged from 45 to 64 years old. At the follow-up 3 interviews (2014–2017), after a mean (SD) follow-up time of 20 (1.9) years, these participants had a mean (SD) age of 72 (5.4) years, and ranged from 61 to 87 years old. Of the 11,959 participants included in this study, 1,609 (13.5%) subjects, made up of 674 (13.5%) men and 935 (13.4%) women, were considered to be physically frail.

Participants had a median (IQR) energy-adjusted intake of 105.0 (58.7) grams of vegetables and 197.2 (169.4) grams of fruits per day. As the distributions of both vegetable and fruit intake within our cohort were positively skewed, we categorized participants into quintiles of daily intake for analyses. When compared to participants with lower vegetable and fruit intake, those with higher intake were more likely to be younger, women, non-drinkers, and non-smokers (Table 1). They were also more likely to have had higher levels of education, and more likely to have taken vitamin or mineral supplements at least weekly.

Table 1.

Baseline characteristics of the participants, by quintiles of vegetable and fruit intake at midlife.

	Quintile of vegetable and fruit intake					P-value
	Q1 (Lowest)	Q2	Q3	Q4	Q5 (Highest)
	N = 2,360	N = 2,395	N = 2,379	N = 2,412	N = 2,413
Median vegetable and fruit intake (IQR) [g/day]	145.5 (77.6)	238.7 (36.1)	309.7 (36.2)	391.8 (53.2)	548.8 (152.4)	<0.001
Mean age at baseline (SD) [years]	52.6 (5.4)	52.5 (5.3)	52.4 (5.2)	52.1 (5.0)	51.9 (5.1)	<0.001
Gender						<0.001
Men (%)	1,418 (60.1%)	1,007 (42.0%)	881 (37.0%)	795 (33.0%)	882 (36.6%)
Women (%)	942 (39.9%)	1,388 (58.0%)	1,498 (63.0%)	1,617 (67.0%)	1,531 (63.4%)
Dialect group						<0.001
Hokkien (%)	1,084 (45.9%)	1,132 (47.3%)	1,201 (50.5%)	1,281 (53.1%)	1,277 (52.9%)
Cantonese (%)	1,276 (54.1%)	1,263 (52.7%)	1,178 (49.5%)	1,131 (46.9%)	1,136 (47.1%)
Level of education						<0.001
No formal education (%)	405 (17.2%)	423 (17.7%)	384 (16.1%)	312 (12.9%)	260 (10.8%)
Primary school (%)	1,161 (49.2%)	1,106 (46.2%)	1,027 (43.2%)	1,019 (42.2%)	916 (38.0%)
Secondary school (%)	653 (27.7%)	686 (28.6%)	772 (32.5%)	839 (34.8%)	958 (39.7%)
Diploma/University (%)	141 (6.0%)	180 (7.5%)	196 (8.2%)	242 (10.0%)	279 (11.6%)
Hypertension (%)	383 (16.2%)	442 (18.5%)	433 (18.2%)	444 (18.4%)	483 (20.0%)	0.021
Cardiovascular disease (%)	53 (2.2%)	50 (2.1%)	47 (2.0%)	46 (1.9%)	67 (2.8%)	0.25
Diabetes (%)	102 (4.3%)	127 (5.3%)	95 (4.0%)	118 (4.9%)	95 (3.9%)	0.10
Cancer (%)	26 (1.1%)	45 (1.9%)	35 (1.5%)	49 (2.0%)	42 (1.7%)	0.10
Frequency of alcohol consumption						<0.001
Never / Monthly (%)	1,929 (81.7%)	2,118 (88.4%)	2,161 (90.8%)	2,184 (90.5%)	2,183 (90.5%)
Weekly (%)	291 (12.3%)	216 (9.0%)	176 (7.4%)	188 (7.8%)	201 (8.3%)
Daily (%)	140 (5.9%)	61 (2.5%)	42 (1.8%)	40 (1.7%)	29 (1.2%)
History of smoking						<0.001
Never smoker (%)	1,490 (63.1%)	1,889 (78.9%)	1,966 (82.6%)	2,044 (84.7%)	2,068 (85.7%)
Ever smoker (%)	870 (36.9%)	506 (21.1%)	413 (17.4%)	368 (15.3%)	345 (14.3%)
Mean body mass index (SD) [kg/m2]	22.9 (3.3)	23.0 (3.2)	23.1 (3.2)	23.2 (3.2)	23.4 (3.2)	<0.001
Mean amount of sleep (SD) [hours/day]	7.0 (1.1)	7.0 (1.0)	7.0 (1.0)	7.0 (1.0)	7.0 (1.1)	0.38
Weekly participation in vigorous work or strenuous sports (%)	521 (22.1%)	422 (17.6%)	404 (17.0%)	419 (17.4%)	496 (20.6%)	<0.001
Amount of moderate activity per week						<0.001
None (%)	1,923 (81.5%)	1,927 (80.5%)	1,851 (77.8%)	1,775 (73.6%)	1,696 (70.3%)
0.5 to 3 h (%)	294 (12.5%)	334 (13.9%)	354 (14.9%)	427 (17.7%)	470 (19.5%)
≥4 h (%)	143 (6.1%)	134 (5.6%)	174 (7.3%)	210 (8.7%)	247 (10.2%)
Median total energy intake (IQR) [kcal/day]	1748.2 (751.3)	1445.3 (639.2)	1407.8 (572.4)	1439.5 (595.9)	1584.4 (659.9)	<0.001
Median total protein intake (IQR) [g/day]	57.6 (13.9)	59.7 (10.8)	60.0 (10.6)	60.5 (11.3)	59.0 (13.0)	<0.001
Median caffeine intake (IQR) [mg/day]	146.8 (150.5)	115.0 (156.9)	106.1 (159.4)	104.9 (148.1)	107.8 (149.7)	<0.001
Weekly use of vitamin/mineral supplements (%)	130 (5.5%)	147 (6.1%)	192 (8.1%)	231 (9.6%)	293 (12.1%)	<0.001

Continuous variables were presented as mean (SD) or median (IQR), while categorical variables were presented as N (%).

The associations between intake of vegetables and fruits at midlife and physical frailty in later life were first assessed separately (Table 2). In our fully adjusted models, only vegetable intake at midlife showed a dose-dependent inverse relationship with likelihood of physical frailty in later life (P_trend = 0.001). Compared to participants in the lowest quintile of vegetable intake at midlife [median (IQR): 56.8 (22.2) g/day], those in the highest quintile of intake [median (IQR): 179.8 (48.9) g/day] had significantly reduced odds of frailty in later life [OR (95% CI): 0.73 (0.60, 0.89)]. In contrast, although fruit intake at midlife was associated with physical frailty in later life in our basic model (Model 1), this association was attenuated and no longer statistically significant after adjustment for other covariates (Models 2 and 3). As the women in our study, on average, had a higher intake of vegetables and fruits than the men, we also examined the associations between vegetable/fruit intake and physical frailty in men and women separately (Supplementary Table S3); in these analyses, the associations between intake of vegetables and fruits with physical frailty did not differ significantly across gender. To further investigate if vegetable intake was only associated with lower odds of frailty in participants who consumed sufficient protein, we also stratified our results by the median daily intake of total protein in our study (59.4 g/day) (Supplementary Table S4). In these analyses, the associations between vegetable intake and physical frailty also did not differ significantly by total daily protein intake.

Table 2.

Associations between intake of vegetables and fruits at midlife and physical frailty in later life.

	Quintile of intake					Ptrend
	Q1 (Lowest)	Q2	Q3	Q4	Q5 (Highest)
Vegetables
Median intake (IQR) [g/day]	56.8 (22.2)	84.0 (10.6)	104.8 (10.6)	128.9 (15.3)	179.8 (48.9)
Cases/N	353 / 2382	348 / 2382	340 / 2392	314 / 2407	254 / 2396
Model 1 [OR (95% CI)] a	Ref.	0.95 (0.80, 1.12)	0.97 (0.82, 1.15)	0.88 (0.74, 1.05)	0.75 (0.63, 0.90)	0.001
Model 2 [OR (95% CI)] b	Ref.	0.94 (0.79, 1.11)	0.94 (0.79, 1.13)	0.84 (0.70, 1.01)	0.72 (0.59, 0.87)	<0.001
Model 3 [OR (95% CI)] c	Ref.	0.94 (0.79, 1.12)	0.95 (0.80, 1.14)	0.86 (0.71, 1.03)	0.73 (0.60, 0.89)	0.001
Fruits
Median intake (IQR) [g/day]	61.8 (60.1)	137.9 (30.4)	196.4 (30.3)	268.7 (46.0)	413.6 (139.3)
Cases/N	363 / 2361	345 / 2387	305 / 2394	318 / 2407	278 / 2410
Model 1 [OR (95% CI)] a	Ref.	0.94 (0.79, 1.11)	0.87 (0.73, 1.03)	0.92 (0.78, 1.10)	0.81 (0.67, 0.96)	0.023
Model 2 [OR (95% CI)] b	Ref.	0.97 (0.81, 1.15)	0.90 (0.75, 1.08)	0.95 (0.79, 1.13)	0.84 (0.70, 1.01)	0.07
Model 3 [OR (95% CI)] c	Ref.	0.98 (0.83, 1.17)	0.93 (0.78, 1.11)	0.99 (0.83, 1.19)	0.90 (0.75, 1.08)	0.31

^aModel 1: adjusted for age at physical tests (years), gender, dialect group (Hokkien, Cantonese), level of education (none, primary, secondary, diploma/university).

^bModel 2: adjusted for Model 1 and the following covariates assessed at the baseline interviews: hypertension, angina or heart attack, stroke, diabetes, cancer, alcohol consumption (none, monthly, weekly, daily), smoking history (never, former, current), body mass index (<18.5, 18.5–22.9, 23.0–27.4, 27.5+ kg/m²), amount of sleep per day (≤5, 6, 7, 8, 9+ hours), amount of strenuous sports per week (0, 0.5–1, 2–3, 4+ hours), amount of vigorous work per week (0, 0.5–3, 4–6, 7+ hours), amount of moderate activity per week (0, 0.5–1, 2–3, 4–6, 7+ hours), daily energy intake (kcal), daily protein intake (g), daily caffeine intake (mg), weekly use of vitamin/mineral supplements.

^cModel 3: adjusted for Model 2 plus vegetable or fruit intake (quintiles) [where applicable].

We then examined the associations between vegetable intake and each criterion in our physical frailty phenotype in order to investigate which components were primarily responsible for the observed association (Table 3). Vegetable intake at midlife showed a strong dose-dependent inverse relationship with likelihood of weakness (defined by weak handgrip strength) in later life (P_trend < 0.001). Compared to participants in the lowest quintile of vegetable intake at midlife, participants in the third quintile and above had significantly reduced odds of weakness in later life [OR (95% CI) for the highest quintile: 0.62 (0.52, 0.73)]. Furthermore, there was an inverse association of borderline significance between vegetable intake at midlife and self-reported exhaustion in later life (P_trend = 0.055). In contrast, the seemingly inverse associations between vegetable intake and the other two criteria in our physical frailty phenotype – slowness and weight loss – did not achieve statistical significance. After adjustment for multiple comparisons, only the trend between increasing vegetable intake and reduced likelihood of weakness remained statistically significant (adjusted P_trend < 0.001).

Table 3.

Associations between vegetable intake at midlife and individual criteria of physical frailty in later life.

	Quintile of vegetable intake					Ptrend	Adj Ptrenda
	Q1 (Lowest)	Q2	Q3	Q4	Q5 (Highest)
Median intake (IQR) [g/day]	56.8 (22.2)	84.0 (10.6)	104.8 (10.6)	128.9 (15.3)	179.8 (48.9)
Frailty criteria
Weakness (handgrip strength in the weakest sex-specific quintile)
Cases/N	502 / 2382	463 / 2382	425 / 2392	370 / 2407	328 / 2396
OR (95% CI) b	Ref.	0.87 (0.74, 1.01)	0.79 (0.67, 0.92)	0.66 (0.56, 0.78)	0.62 (0.52, 0.73)	<0.001	<0.001
Exhaustion (answered “No” to the question: “Do you feel full of energy?”)
Cases/N	415 / 2382	440 / 2382	428 / 2392	441 / 2407	379 / 2396
OR (95% CI) b	Ref.	0.99 (0.85, 1.16)	0.96 (0.82, 1.12)	0.98 (0.84, 1.15)	0.86 (0.72, 1.01)	0.055	0.16
Slowness (TUG test time in the slowest sex-specific quintile)
Cases/N	378 / 2382	366 / 2382	322 / 2392	311 / 2407	298 / 2396
OR (95% CI) b	Ref.	0.99 (0.83, 1.19)	0.90 (0.75, 1.08)	0.86 (0.71, 1.04)	0.91 (0.75, 1.11)	0.21	0.42
Weight loss (lost ≥10% of self-reported body weight since the previous follow-up interview)
Cases/N	271 / 2382	245 / 2382	286 / 2392	258 / 2407	227 / 2396
OR (95% CI) b	Ref.	0.88 (0.73, 1.07)	1.08 (0.89, 1.30)	0.95 (0.78, 1.15)	0.87 (0.71, 1.07)	0.26	0.42

TUG: timed up-and-go.

^aP_trend values adjusted for multiple comparisons using the Holm-Bonferroni method.

^bAdjusted for age at physical tests (years), gender, dialect group (Hokkien, Cantonese), level of education (none, primary, secondary, diploma/university), and the following covariates assessed at the baseline interviews: hypertension, angina or heart attack, stroke, diabetes, cancer, alcohol consumption (none, monthly, weekly, daily), smoking history (never, former, current), body mass index (<18.5, 18.5–22.9, 23.0–27.4, 27.5+ kg/m²), amount of sleep per day (≤5, 6, 7, 8, 9+ hours), amount of strenuous sports per week (0, 0.5–1, 2–3, 4+ hours), amount of vigorous work per week (0, 0.5–3, 4–6, 7+ hours), amount of moderate activity per week (0, 0.5–1, 2–3, 4–6, 7+ hours), daily energy intake (kcal), daily protein intake (g), daily caffeine intake (mg), weekly use of vitamin/mineral supplements, fruit intake (quintiles).

The associations between physical frailty and specific subgroups of vegetables were then evaluated (Table 4). Across all types of vegetables, there was a general trend linking greater intake with lower odds of physical frailty. Notably, the associations for dark-green vegetables (P_trend = 0.001), cruciferous vegetables (P_trend = 0.022), and mushrooms (P_trend = 0.023) crossed the threshold for statistical significance. For these subgroups of vegetables, participants in the highest quintile of intake had a significant reduction in odds of physical frailty compared to those in the lowest quintile. Nonetheless, after adjustment for multiple comparisons, only the trend between increasing dark-green vegetable intake and reduced odds of physical frailty remained statistically significant (adjusted P_trend = 0.003).

Table 4.

Associations between vegetable intake at midlife and physical frailty in later life, by subgroups of vegetables.

	Quintile of intake					Ptrend	Adj Ptrenda
	Q1 (Lowest)	Q2	Q3	Q4	Q5 (Highest)
Dark-green vegetables
Median intake (IQR) [g/day]	14.7 (8.5)	24.9 (4.2)	33.4 (4.6)	44.4 (7.1)	65.9 (21.9)
Cases/N	352 / 2380	340 / 2379	336 / 2397	316 / 2412	265 / 2391
OR (95% CI) b	Ref.	0.95 (0.80, 1.13)	0.95 (0.80, 1.14)	0.86 (0.72, 1.03)	0.74 (0.61, 0.89)	0.001	0.003
Cruciferous vegetables
Median intake (IQR) [g/day]	20.0 (9.7)	31.9 (5.0)	41.9 (5.3)	54.2 (7.7)	80.4 (26.5)
Cases/N	345 / 2363	329 / 2388	344 / 2395	323 / 2406	268 / 2407
OR (95% CI) b	Ref.	0.96 (0.80, 1.14)	1.01 (0.85, 1.21)	0.93 (0.78, 1.12)	0.81 (0.67, 0.98)	0.022	0.11
Mushrooms
Median intake (IQR) [g/day]	0.4 (0.7)	1.5 (0.4)	2.4 (0.4)	3.6 (1.5)	6.9 (4.1)
Cases/N	334 / 2398	330 / 2400	319 / 2391	343 / 2380	283 / 2390
OR (95% CI) b	Ref.	0.90 (0.75, 1.07)	0.85 (0.71, 1.02)	0.94 (0.79, 1.13)	0.78 (0.65, 0.94)	0.023	0.11
Yellow vegetables
Median intake (IQR) [g/day]	1.2 (2.2)	4.1 (1.2)	6.6 (1.4)	10.1 (2.9)	19.6 (9.9)
Cases/N	361 / 2391	363 / 2394	289 / 2396	309 / 2377	287 / 2401
OR (95% CI) b	Ref.	0.95 (0.80, 1.13)	0.75 (0.62, 0.90)	0.88 (0.73, 1.05)	0.84 (0.70, 1.01)	0.11	0.32
Light-green vegetables
Median intake (IQR) [g/day]	14.7 (7.3)	23.6 (3.2)	30.1 (3.5)	38.5 (5.3)	56.2 (18.0)
Cases/N	336 / 2364	341 / 2403	324 / 2390	339 / 2399	269 / 2403
OR (95% CI) b	Ref.	0.99 (0.83, 1.18)	0.97 (0.81, 1.17)	1.04 (0.87, 1.24)	0.85 (0.71, 1.03)	0.12	0.32
Tomatoes
Median intake (IQR) [g/day]	0.8 (2.1)	3.4 (1.1)	5.6 (1.2)	8.8 (2.1)	19.3 (10.6)
Cases/N	325 / 2389	367 / 2415	342 / 2389	295 / 2402	280 / 2364
OR (95% CI) b	Ref.	1.05 (0.88, 1.26)	0.99 (0.83, 1.19)	0.91 (0.75, 1.09)	0.92 (0.77, 1.11)	0.18	0.32

^aP_trend values adjusted for multiple comparisons using the Holm-Bonferroni method.

^bAdjusted for age at physical tests (years), gender, dialect group (Hokkien, Cantonese), level of education (none, primary, secondary, diploma/university), and the following covariates assessed at the baseline interviews: hypertension, angina or heart attack, stroke, diabetes, cancer, alcohol consumption (none, monthly, weekly, daily), smoking history (never, former, current), body mass index (<18.5, 18.5–22.9, 23.0–27.4, 27.5+ kg/m²), amount of sleep per day (≤5, 6, 7, 8, 9+ hours), amount of strenuous sports per week (0, 0.5–1, 2–3, 4+ hours), amount of vigorous work per week (0, 0.5–3, 4–6, 7+ hours), amount of moderate activity per week (0, 0.5–1, 2–3, 4–6, 7+ hours), daily energy intake (kcal), daily protein intake (g), daily caffeine intake (mg), weekly use of vitamin/mineral supplements, fruit intake (quintiles).

Finally, we examined if nutrients present in vegetables could have accounted for the association between vegetable intake and physical frailty (Table 5). The association between vegetable intake and physical frailty was attenuated and no longer reached statistical significance in models that were additionally adjusted for the total daily intake of β-carotene (P_trend = 0.25) and lutein (P_trend = 0.24), respectively. After accounting for multiple comparisons, the vegetable-frailty association was not statistically significant in models that were adjusted for the intake of β-carotene (adjusted P_trend = 0.47), lutein (adjusted P_trend = 0.47), folate (adjusted P_trend = 0.07), α-carotene (adjusted P_trend = 0.07), and isothiocyanates (adjusted P_trend = 0.07). In contrast, the vegetable-frailty association remained statistically significant in models that were adjusted for dietary fiber, vitamin E, vitamin C, β-cryptoxanthin, and lycopene.

Table 5.

Associations between vegetable intake and physical frailty, after adjustments for the total daily intake of various nutrients.

	Correlation between vegetable and nutrient intake	Association between quintile of vegetable intake and physical frailty
		Q1 (Lowest)	Q5 (Highest)	Ptrend	Adj Ptrendb
	Spearman’s ρ a	OR (95% CI)	OR (95% CI)
Model plus β-carotene [μg/day] c	0.84	Ref.	0.86 (0.66, 1.13)	0.25	0.47
Model plus lutein [μg/day] c	0.85	Ref.	0.85 (0.63, 1.13)	0.24	0.47
Model plus folate (vitamin B9) [μg/day] c	0.59	Ref.	0.78 (0.62, 0.97)	0.017	0.07
Model plus α-carotene [μg/day] c	0.58	Ref.	0.78 (0.62, 0.96)	0.015	0.07
Model plus isothiocyanates [μmol/day] c	0.80	Ref.	0.73 (0.57, 0.95)	0.014	0.07
Model plus dietary fiber [g/day] c	0.49	Ref.	0.73 (0.60, 0.90)	0.002	0.009
Model plus vitamin E [mg/day] c	0.60	Ref.	0.74 (0.61, 0.90)	0.001	0.009
Model plus vitamin C [mg/day] c	0.43	Ref.	0.74 (0.61, 0.90)	0.001	0.009
Model plus β-cryptoxanthin [μg/day] c	0.20	Ref.	0.73 (0.60, 0.89)	0.001	0.006
Model plus lycopene [μg/day] c	0.30	Ref.	0.72 (0.60, 0.88)	<0.001	0.005

^aTotal daily intake of vegetables (g/day) and nutrients (respective units in table) assessed as continuous variables.

^bP_trend values adjusted for multiple comparisons using the Holm-Bonferroni method.

^cAdjusted for age at physical tests (years), gender, dialect group (Hokkien, Cantonese), level of education (none, primary, secondary, diploma/university), and the following covariates assessed at the baseline interviews: hypertension, angina or heart attack, stroke, diabetes, cancer, alcohol consumption (none, monthly, weekly, daily), smoking history (never, former, current), body mass index (<18.5, 18.5–22.9, 23.0–27.4, 27.5+ kg/m²), amount of sleep per day (≤5, 6, 7, 8, 9+ hours), amount of strenuous sports per week (0, 0.5–1, 2–3, 4+ hours), amount of vigorous work per week (0, 0.5–3, 4–6, 7+ hours), amount of moderate activity per week (0, 0.5–1, 2–3, 4–6, 7+ hours), daily energy intake (kcal), daily protein intake (g), daily caffeine intake (mg), weekly use of vitamin/mineral supplements, fruit intake (quintiles).

At the baseline interviews, we did not collect any other data, beyond participants’ age and their history of physician-diagnosed comorbidities, that would have allowed us to identify subjects who might already have been physically frail at the time. As such, to exclude participants who were likely to have been physically at the time of recruitment, we conducted a sensitivity analysis that removed 3,582 (30.0%) participants [1,532 (30.7%) men and 2,050 (29.4%) women] who were either over 60 years of age or had a history of chronic disease at the baseline interviews. Due to the reduced sample size in this sensitivity analysis (N = 8,377), there was a reduction in statistical power, and participants in the highest quintile of vegetable intake no longer had odds of frailty that were significantly lower than their counterparts with the lowest quintile of intake (Supplementary Table S5). Despite this, vegetable intake continued to show a dose-dependent inverse relationship with likelihood of physical frailty (P_trend = 0.020). In contrast, there was no significant association between fruit intake and physical frailty.

4. Discussion

Our study found that higher intake of vegetables at midlife was associated with lower likelihood of physical frailty in later life. Among the components of physical frailty, vegetable intake had the strongest inverse association with weakness defined by handgrip strength. Moreover, we found that the association between vegetable intake and physical frailty was attenuated after adjustment for β-carotene, lutein, folate, α-carotene, and isothiocyanate intake, suggesting that the associations between these nutrients and physical frailty could have explained the association between vegetable intake and physical frailty.

Our findings concur with prior studies that have also found significant associations between intake of vegetables, but not of fruits, and physical frailty [13,16,[30], [31], [32]]. Nevertheless, there is still much conflicting evidence within the literature, as there are studies that have found physical frailty to be associated with intake of both vegetables and fruits [[9], [10], [11], [12],19,20,28], and several other studies that did not report significant associations with either vegetables or fruits [17,[23], [24], [25],29]. Even among studies which had assessed vegetables and fruits in combination as a single intake, some had found significant associations with physical frailty [14,26,27], whereas others had not [15,18,21,22]. A meta-analysis by Ghoreishy et al. found high heterogeneity across studies, and showed that potential sources of such heterogeneity included geographical region, methods of exposure assessment, and adjustment for physical activity and BMI [33], which could explain the conflict in results from different studies.

In our cohort, higher vegetable intake was strongly associated with reduced likelihood of weakness (weak handgrip strength) during the assessment for physical frailty. This agreed with prior studies that had shown that higher vegetable intake was associated with greater handgrip strength [13,50,51] and a lower risk for sarcopenia [[52], [53], [54], [55], [56]] in older adults. Nevertheless, there was one study that had found vegetable intake to be associated with the elements of exhaustion and weight loss in the physical frailty phenotype [10]. It is plausible that the associations between vegetable intake and the individual components of physical frailty might differ across populations, depending upon the prevalence of each component.

We found that higher intake of dark-green vegetables, cruciferous vegetables, and mushrooms were associated with reduced likelihood of physical frailty, although only the trend with intake of dark-green vegetables remained statistically significant after adjusting for multiple comparisons. To our best knowledge, only one prior study had evaluated the associations between specific types of vegetables and physical frailty. Although this study had categorized their vegetables differently, the authors similarly showed that higher intake of leafy vegetables were significantly associated with reduced risk of physical frailty [9]. Nevertheless, we noted that the subgroups of vegetables in our analyses overlapped and were not mutually exclusive, as some leafy vegetables could have been in more than one subgroup. Hence, at this stage, we could not be certain about which subgroups of leafy vegetables were more likely to reduce the risk of physical frailty than others.

Our findings were not unexpected. Greater vegetable consumption has been shown to be associated with reduced risk for many adverse health outcomes that could lead to the development of physical frailty in aging: cardiovascular disease, chronic kidney disease, diabetes, hip fracture, and cancer [33,[57], [58], [59]]. Vegetables are rich sources of vitamins (e.g., vitamins C and E) and carotenoids (e.g., α-carotene, β-carotene, β-cryptoxanthin, lutein, and lycopene), and these nutrients are thought to reduce the risk of physical frailty via their anti-inflammatory and anti-oxidative effects [4,5,8,33,47,48]. Our findings suggest that intake of β-carotene, lutein, folate, α-carotene, and isothiocyanates could explain the association between vegetable intake and physical frailty.

Strengths of our study included its prospective design, long follow-up period, and comprehensive, validated assessment of dietary intake. However, several limitations should be noted. First, as the intake of vegetables and fruits were only assessed at the baseline interviews, this study was unable to examine the effects of subsequent changes in dietary intake. Second, at the baseline interviews, we did not collect any other data, beyond participants’ age and their history of physician-diagnosed comorbidities, that would have allowed us to identify subjects who might already have been physically frail at the time. As such, we conducted a sensitivity analysis that removed participants who met either of the following criteria at recruitment: (1) over 60 years of age (the United Nations’ definition of an older person [60]), or (2) had a history of chronic disease (hypertension, cardiovascular disease, diabetes, or cancer). We believed that such participants were more likely to already have been physically frail at the time, as advanced age and the presence of chronic diseases are strong risk factors for the onset and progression of physical frailty [2,[61], [62], [63], [64], [65], [66]]. Nevertheless, we recognize that such criteria would not necessarily have identified subjects who were physically frail, and acknowledge that this sensitivity analysis was not a substitute for the assessment of physical frailty. As such, we accept that some of the participants included in our study could already have been physically frail at recruitment, and thus acknowledge that we could not have been absolutely certain of the temporal relationship between vegetable/fruit intake and physical frailty. Hence, despite a follow-up period of 20 years, we could not rule out the possibility that our risk estimates could have been overestimated due to reverse causality. Third, we acknowledge that our study involved only a proportion of those who were recruited at the baseline interviews. Compared to the 51,298 subjects who were not included in the present study due to death, disability, lost to contact, or refusal, the 11,959 participants included in this study were generally younger, more likely to have had higher levels of education, and less likely to have had a history of comorbidities at the baseline interviews (Supplementary Table S6). These participants also generally led healthier lifestyles; they were more likely to have been never-smokers and more physically active at midlife. Thus, the selective bias towards studying participants who were less likely to develop physical frailty due to these other factors could have led to an underestimation of the true association between vegetable intake and the risk of physical frailty.

Fourth, as we did not have data on participants’ level of physical activity at follow-up 3, we adopted a conservative definition of frailty as meeting at least two of the four available criteria, and did not have a category for pre-frailty. However, in doing so, we acknowledge that we could have misclassified pre-frail subjects as frail, and thus underestimated the true association between vegetable intake and risk of physical frailty. Nevertheless, we noted in a previous report that the prevalence of physical frailty among women aged 70–79 years in our cohort (14.4% frail) was comparable to that of three other cohorts that had used the original definitions of the CHS frailty phenotype [42]. Furthermore, we noted that there were two earlier studies that had also omitted the component of physical activity in their modified definitions of physical frailty [67], and classified participants into categories of either ‘frail’ or ‘non-frail’ based on only four criteria [68]. Fifth, there were some limitations in the weight loss and slowness criteria of our modified frailty phenotype. For the criterion of weight loss, we used weight at follow-up 2, which was reported on average 7 years prior, to calculate participants’ change in body weight at follow-up 3. This might not, as in the original definition of the CHS frailty phenotype [1,40,41], have strictly reflected unintentional weight loss in the year prior to the assessment of frailty. Nevertheless, studies in older adults have shown that self-reported weight loss of 5% or more, even when compared to weight 10 years ago, was associated with an increased risk of frailty [69]. Furthermore, in consideration of this 7-year timeframe, we defined the threshold for weight loss as 10% or more, which is both clinically significant [70], and greater than the original threshold of 5% [40]. Meanwhile, for the criterion of slowness, we used the TUG test in place of gait speed. Compared to gait speed, the TUG test involves additional functional movements such as standing, turning, and sitting. Nevertheless, studies have observed high correlations between performance on both tests [43,71], and the TUG test alone has been shown to be a sensitive and specific discriminator for frailty in older adults [72]. Lastly, in our study, participants were only assessed with a single definition of physical frailty: a modified version of the CHS frailty phenotype. However, it should be noted that there are many different definitions of frailty in common use [[1], [2], [3],[73], [74], [75], [76]]. Although these different definitions carry the same nomenclature and share common ground [2,73,76], they nonetheless vary in their biological bases as well as their included components, and often identify different populations and targets for intervention [[1], [2], [3],[73], [74], [75], [76]]. Hence, in this study, we could only discuss the specific components of physical frailty in relation to others that had also used the CHS frailty phenotype. Further studies which utilize other definitions or instruments of physical frailty are needed, as their findings could provide further insight into the specific nature of the association between vegetable intake and physical frailty defined using other criteria.

5. Conclusions

In conclusion, our study found that higher intake of vegetables at midlife was associated with lower likelihood of physical frailty in later life. Our findings also suggested that the intake of β-carotene, lutein, folate, α-carotene, and isothiocyanates could have accounted for the association between vegetable intake and physical frailty. Our study supports existing public health recommendations that encourage greater consumption of vegetables in the general population, and shows that it could also reduce the risk of physical frailty in later life.

Author contributions

KY Chua: Data curation, Formal analysis, Writing – Original Draft. H Li: Data curation, Validation, Writing – Review & Editing. L-T Sheng: Data curation, Validation, Writing – Review & Editing. W-S Lim: Methodology, Writing – Review & Editing. W-P Koh: Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing – Review & Editing.

Funding

This work was supported by the Singapore National Medical Research Council[NMRC/CSA/0055/2013]; the United States National Cancer Institute, National Institutes of Health[UM1 CA182876, R01 CA144034]; and the Saw Swee Hock School of Public Health, National University of Singapore. W-P Koh is supported by the National Medical Research Council, Singapore [CSA-SI (MOH-000434)]. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

We thank Siew-Hong Low of the National University of Singapore for supervising the fieldwork in the Singapore Chinese Health Study.

Appendix A. Supplementary data

The following is Supplementary data to this article: