Patterns of circulating fat-soluble vitamins and carotenoids and risk of frailty in four European cohorts of older adults

Abstract

Purpose

To investigate the cross-sectional and prospective associations between patterns of serum fat-soluble micronutrients and frailty in four European cohorts of older adults 65 years of age and older.

Methods

Participants from the Three-City (Bordeaux, France), AMI (Gironde, France), TSHA (Toledo, Spain) and InCHIANTI (Tuscany, Italy) cohorts with available data on serum α-carotene, β-carotene, lycopene, cryptoxanthin, lutein +zeaxanthin, retinol, α-tocopherol, γ-tocopherol and 25-hydroxyvitamin D3 (25(OH)D) were included. A principal component (PC) analysis was used to derive micronutrient patterns. Frailty was defined using Fried’s criteria. Multivariate logistic regression models adjusted for socio-demographic and health-related covariates were performed to assess the association between micronutrient patterns and prevalent frailty in 1324 participants, and the risk of frailty in 915 initially non-frail participants.

Results

Three different patterns were identified: the first pattern was characterized by higher serum carotenoids and α-tocopherol levels; the second was characterized by high loadings for serum vitamins A and E levels and low loadings for carotenes level; the third one had the highest loading for serum 25(OH)D and cryptoxanthin level and the lowest loading for vitamin A and E. A significant cross-sectional association was only observed between the seconnd PC and prevalent frailty p = 0.02). Compared to the highest quartile, participants in the lowest quartile—i.e., high carotenes and low vitamins E and A levels—had higher odds of frailty (Odds ratio = 2.2; 95% confidence interval 1.3–3.8). No association with the risk of frailty was observed.

Conclusions

These findings suggest that some specific micronutrient patterns are markers but not predictors of frailty in these European cohorts of older adults.

Introduction

Due to worldwide population aging, the prevention of geriatric syndromes is of great importance so as to increase the number of healthy life years in older populations. In particular, frailty, which is one of the reversible geriatric syndromes, is associated with a higher risk of falls, disability, morbidity, institutionalization and death [1]. To date there is not a consensus on the definition of frailty, although it is widely defined as a decreased physiological reserve of adaptive capacities when faced with stressors, even minor ones [2, 3]. Based on the frailty phenotype defined by Fried et al. (i.e., mainly focused on the physical dimension), 10% of community-dwelling adults aged 65 years and older are frail [4]. Among prevention strategies on modifiable factors, nutrition appears to be a promising approach.

Currently, nutritional literature on frailty was mainly restricted to the role of a single nutrient or food [5, 6]. In particular, when macronutrients and foods were assessed, a lower intake of proteins was associated with a higher prevalence of frailty [7], while a higher consumption of fruits and vegetables was associated with a reduced risk of frailty in a dose–response manner [8]. For micronutrients, a vast part of the literature was focused on vitamin D (25(OH) D) [9], as hypovitaminosis D is a condition highly prevalent in older adults [10]. Although observational studies on 25(OH)D and frailty have not definitively established an independent relationship, interventional studies of 25(OH) D supplementation have yielded a positive effect on frailty status, mainly via improvements in physical performances [9]. Other fat-soluble micronutrients (vitamins A and E, carotenoids) were less often investigated and results varied in effect [11–13]. For instance, a cross-sectional analysis of 754 women from the Women’s Health and Aging Studies (WHAS) I and II revealed no association between frailty and lower levels of retinol [11], while higher levels of vitamin E were associated with lower odds of frailty in 827 older adults from the InCHIANTI cohort [12].

All aforementioned studies explored isolated fat-soluble micronutrients as potential predictors of frailty, but nutrients/foods are eaten as part of meals. Therefore, it seems more relevant to study patterns when exploring the complex nature of nutritional exposures in relation to health outcomes [14]. Besides, circulating micronutrient status is affected by dietary intake, but also by inter-individual differences in bioavailability and metabolism (i.e., matrix in which the nutrients are incorporated, effectors of absorption, nutrient status of the host, genetic factors, host-related factors, and interactions between these factors) and is used as an objective marker of micronutrient status [15].

To the best of our knowledge, no study has investigated the relationship between patterns of circulating fat-soluble micronutrients and frailty phenotype to date. Thus, we explored the cross-sectional and prospective associations between patterns of plasma fat-soluble micronutrients and frailty in four European cohorts of older adults aged 65 years or older.

Methods

Study population and cohorts

In this study, we included the following four European cohorts: AMI, (Gironde, France), Three-City (Bordeaux, France), InCHIANTI (Chianti area, Italy) and TSHA (Toledo, Spain) cohorts. All these cohorts are part of the exploratory phase of the FRAILOMIC initiative, which aims to develop validated measures including both classic and ‘omics’-based laboratory biomarkers to predict the risk of frailty, improve the diagnostic accuracy of frailty in day-to-day clinical practice and assess the positive effect of a prognostic forecast of frailty on the onset of disability and other adverse outcomes [16]. These cohorts were notably selected in the FRAILOMIC initiative for their available data on frailty status, as defined by Fried et al. [1], at the time of blood sampling and 2–5 years later [17]. In AMI, InCHIANTI and TSHA, participant selection was carried out as follows: (1) all frail but non-disabled participants were selected; (2) a 1:3 random sample from non-frail participants was then selected; (3) all frail and disabled participants were finally selected. In the Three-City cohort, all participants with blood sample and frailty data were included. More details for each cohort are presented below, and supplemental Figure 1 presents the flowchart of the studied samples.

AMI

The Approche Multidisciplinaire Intégrée (AMI) cohort is an epidemiological prospective study on health and aging among 1002 farmers aged 65 years or more and living in rural areas in Gironde in southwestern France. Its methodology has been described in detail elsewhere [18]. At baseline in 2007 and at follow-ups every 2 years, home visits were conducted to obtain information on socio-demographic factors, lifestyle behaviors, medications, material and social living environment, neuropsychological tests, disability and frailty. AMI was approved by the Ethics Committee of the University Hospital of Bordeaux according to the principles of the Declaration of Helsinki. All participants signed a written consent.

Among the 695 participants with available blood samples in 2007, a total of 320 AMI participants, including 80 frail participants, were then included in the FRAILOMIC initiative. Among these individuals, 307 had fat-soluble micronutrients sampled. We excluded 29 participants with missing covariates, leaving 278 participants for the cross-sectional analysis. The prospective analysis was carried out on 221 participants after exclusion of 49 prevalent cases of frailty and 8 participants who died before the follow-up visit.

TSHA

The Toledo Study for Healthy Aging (TSHA) is a population-based study conducted in 2006–2009 on 2488 subjects aged 65 years or more, and covers both institutionalized and community-dwelling persons from rural and urban settings of Toledo (Spain). Its methodology has been described in detail elsewhere [19]. Individuals were evaluated for frailty at baseline and 5 years later. The questionnaires included information about socio-demography, social support, activities of daily living, health-related quality of life, comorbidity, physical activity, diet, tobacco and alcohol consumption, depressive symptoms and an extensive neuropsychology assessment. Furthermore, after the home interview, subjects underwent a physical performance evaluation where blood and urine samples were taken. TSHA was approved by the Ethics Committee of the University Hospital of Toledo according to the principles of the Declaration of Helsinki. All participants signed a written consent.

A total of 474 participants included in TSHA in 2006–2009 were retained to be part of the FRAILOMIC initiative. Among them, two participants were excluded due to missing data for frailty, 57 were excluded due to lack of fat-soluble micronutrients sampling and 111 were excluded due to missing data for covariates; after all exclusions, 306 participants remained for the cross-sectional analysis. After excluding 66 prevalent cases of frailty, 13 participants who died before the follow-up visit 5 years later and 30 participants lost to follow-up, the prospective analysis sample included 197 participants.

InCHIANTI

The Invecchiare in Chianti study (InCHIANTI) is a longitudinal study designed to explore the causal pathway that leads to mobility decline with age. Since 1998, a representative sample of older people aged 65 years or older at baseline living in the Chianti area nearby Florence (Italy) has been evaluated (N = 1260) with four subsequent follow-ups at 3-year intervals. All participants were interviewed at home and came to the study clinic within 3 weeks of the home interview to provide a blood sample. Within 2 weeks, participants returned to the clinic for a structured medical examination and assessment of physical function including gait speed and grip strength. The design of the study and the complete data collection methods have been described in detail elsewhere [20]. Among the 1260 participants at baseline, a total of 317 participants, including 117 frail participants, were retained to be a part of the FRAILOMIC initiative. Among them, 3 were excluded due to missing data for frailty, 5 were excluded due to lack of fat-soluble micronutrients sampling and 31 were excluded due to missing data for covariates; after all exclusions, 278 participants were included in the cross-sectional analysis. For the prospective analysis sample, 170 participants were included after excluding 79 prevalent cases of frailty, 11 participants that died during the 3-year follow-up and 18 participants lost to follow-up.

Three-City (3C)—Bordeaux cohort

The Three-City (3C) study is a prospective cohort of vascular risk factors of dementia among adults aged 65 years and older living in Bordeaux in southwest of France. Its methodology has been described in detail elsewhere [21]. At baseline (1999–2000) and at each 2-year follow-up visit, information was collected on socio-demographic factors, lifestyle behaviors and medical history at participants’ homes. A physical examination also included anthropometric data, blood pressure, information on frailty and disability, and neuropsychological testing. In 2009–2010, a blood sample was taken on the same day as the interview. All participants provided written informed consent, and the Consultative Committee for the Protection of Persons participating in Biomedical Research of the Kremlin-Bicêtre University Hospital (Paris) approved the study.

Among the 1214 participants seen in 2009–2010 (i.e., 75 years or more), a total of 525 participants with available blood sample and data for frailty were included to be part of the FRAILOMIC initiative. Among them, 63 participants were excluded: one had missing data for frailty, one had no fat-soluble micronutrients sampling and 61 had missing data for one covariate or more. As a result, a total of 462 participants were included in the cross-sectional analysis. Of these, we excluded 96 frail participants, 2 participants that died during before the follow-up 2 years later and 37 participants lost to follow-up for the prospective analysis sample, resulting in 327 participants being included in this analysis.

Frailty

At baseline and follow-up visit, participants were classified as frail if they met three or more of the following criteria [1]: (i) shrinking; (ii) self-reported exhaustion; (iii) low energy expenditure; (iv) slowness and (v) weakness. Some frailty criteria were assessed using a slight modification of the phenotypic criteria proposed by Fried et al. to ensure a comparative assessment between the baseline and the follow-up. Shrinking (i) was defined as a recent self-reported unintentional loss of ≥ 3 kg in the 3C, AMI and InCHIANTI cohorts, and an unintentional loss of ≥ 4.5 kg in the preceding year in TSHA. For all cohorts, self-reported exhaustion (ii) was defined using two items from the Center for Epidemiologic Studies Depression (CES-D) scale [22]. Respondents were considered as exhausted if they answered “yes” to at least one of the following: “during the past week, I felt that everything I did was an effort” and “during the past week, I could not get going”. Low energy expenditure (iii) was defined as reporting no engagement in physical activities (strenuous leisure activities or sport) in AMI, 3C and InCHIANTI, whereas in the TSHA cohort, it is based on the lowest quintile of the Physical Activities Scale for the Elderly (PASE) score (25.1 in men and 50.0 in women). Slowness (iv) was ascertained using the Rosow-Breslau Health scale [23] in 3C and AMI. Respondents were considered as slow if they answered “no” to two Rosow-Breslau Health scale items: “walking between 1.6 and 3.3 feet” and “going up and down a flight of stairs”. The Rosow-Breslau test has been shown to be strongly associated with walking ability [24]. In the InCHIANTI and TSHA cohorts, the gait speed was measured at usual pace for a distance of 4 m in InCHIANTI and 3 m in TSHA. Slowness was then defined by the highest quintile stratified by height and sex. Weakness (v) was defined as the lowest quintile of grip strength adjusted for sex and body mass index in TSHA and InCHIANTI, and as having difficulty rising from a chair without using the armrests in 3C and AMI. The chair stand test was shown to be a good proxy for handgrip strength [25, 26].

Determination of carotenoids, α-, γ-tocopherols and retinol

Fasting plasma/serum samples collected at baseline and stored at − 80 °C were transferred (on dry ice) to the laboratory responsible for analyses of carotenoids and fat-soluble vitamins in Germany. Plasma carotenoids, retinol and tocopherols were analyzed by reversed-phase HPLC coupled with UV–Vis and fluorescence detection as previously described [27]. Methodology validation was carried out in a previous larger study in which inter-batch coefficients of variations (CVs) were < 8% for carotenoids, < 7% for tocopherols and 3.7% for retinol [27].

25(OH)D analysis

Vitamin D (25(OH)D) was determined by HPLC as described by Turpeinen et al. with modifications [28]. In brief, 150 μl of methanol/iso-propanol (80:20 by volume) containing 0.5% pyrogallol and internal standard dodecanophenone (3.5 mg/L) were vigorously mixed with 200 μl of serum/plasma for 30 s. Samples were left in the dark for 30 s, followed by the addition of 800 μl of hexane and were then vortexed vigorously. Samples were centrifuged and supernatants were transferred into fresh tubes. Another 800 μl of hexane were added to the samples, mixed and centrifuged (5 min, 13,000 rpm), now combining supernatants of both extraction steps. Hexane supernatants were evaporated under a constant flow of N₂ at room temperature in the dark. Samples were then reconstituted in 100 μl of methanol in water (80:20 v/v), vortexed, centrifuged, and 50 μl of clear supernatant was analyzed on a Shimadzu Prominence HPLC with UV detection (SPD 20AV set at 265 nm). The separation of 25(OH)D was achieved by use of Reprosil Pur Basic column (C8-2; 5 μm; 250 × 4.6 mm; Dr. Maisch GmbH, Ammerbuch-Entringen, Germany) and an eluent of methanol–water (80:20 V/V) at a flow rate of 1 ml/min. Pure standards of 25(OH)D were diluted to physiological concentrations and used for quantification. Based on previous findings, the expected CV for 25(OH)D analysis was < 6% [28].

Analyses of all biomarkers described here were carried out in a blind manner with respect to frailty status in one single laboratory, which significantly reduces inter-laboratory variations.

Other data

Baseline socio-demographic information included age, sex, and education [i.e., low (no schooling or primary education), medium (secondary and/or vocational education) and high level (higher education)]. Height and weight were measured, and body mass index (BMI) was computed as weight/height squared ratio (kg/m²). Polypharmacy was defined as participants taking 5 or more drugs at the time of the baseline survey. Multimorbidity at baseline is the combination of at least two of the following self-reported chronic diseases: hypertension, diabetes, angina pectoris, cardiac failure, myocardial infarction, stroke and cancer. Global cognitive performance was assessed using the Mini-Mental State Examination (MMSE) [29]. Depressive symptomatology was defined using different scales according to the cohort. In 3C and AMI cohorts, depression was defined as a CES-D score ≥ 23 for women and ≥ 17 for men. In the InCHIANTI cohort, depression was defined as a CES-D score ≥ 20 for women and ≥ 16 for men, while in TSHA, depression was defined as GDS SCORE ≥ 5. Baseline plasma lipid levels [total cholesterol (TC) and triglycerides (TG)] were evaluated by routine enzymatic methods. C-reactive protein (CRP) level was assayed by nephelometry. All data were harmonized to ensure comparability across cohorts.

Statistical analysis

We first compared baseline characteristics of each cohort included in this study. We then described the fat-soluble micronutrients concentrations based on frailty status at baseline and incident frailty. Fat-soluble micronutrients have a non-normal distribution, and thus the median and interquartile ranges were presented for the summary statistics. Comparisons were made using ANOVA, Wilcoxon–Mann–Whitney test or Kruskal–Wallis test for continuous variables, and χ² test for categorical variables, as appropriate.

First, the principal component analysis (PCA) was used to derive the fat-soluble micronutrient patterns based on the plasma concentration of the nine fat-soluble vitamins (retinol, α-tocopherol, γ-tocopherol and 25(OH)D) and carotenoids (α-carotene, β-carotene, lycopene, cryptoxanthin, lutein + zeaxanthin) among all participants with complete data for fat-soluble micronutrients (N = 1556). Since micronutrients were not expressed in the same unit, the mean concentrations of micronutrients were transformed into z-scores. We retained the three principal components (PC) with an eigenvalue superior to the mean eigenvalue observed in 20 simulated random normal data (Table 1).

Table 1.

Eigenvectors of the three dietary patterns of fat-soluble micronutrients derived from principal components (n = 1556)

Fat-soluble micronutrients	Patterns
	Component 1	Component 2	Component 3
	Carotenoids + α-tocopherol	Vitamins E and A – Carotenes	Cryptoxanthin + 25(OH)D – vitamins E and A
α-carotene	0.41	− 0.45	− 0.36
β-carotene	0.50	− 0.37	− 0.13
Lycopene	0.36	− 0.02	0.34
Cryptoxanthin	0.30	0.05	0.47
Lutein + zeaxanthin	0.41	0.01	0.05
α-tocopherol	0.33	0.50	0.05
γ-tocopherol	0.18	0.45	− 0.46
Retinol	0.20	0.41	− 0.31
25(OH)D	0.13	0.18	0.46
% variance explained	29.8	15.6	12.8

The first component explains 29.8% of the total variance and scores high for higher plasma carotenoids and α-tocopherol concentrations. The second component is positive and scores high for higher plasma concentrations in vitamins A and E and negative for higher concentrations in carotenes. This second component explains 15.6% of the total variance. Finally, the third component explains 12.8% of the total variance and is positive and scores high for higher plasma concentrations in 25(OH)D and cryptoxanthin and negative for vitamins A and E.

Second, multivariate logistic regression models were used to assess the cross-sectional and prospective associations between the patterns of fat-soluble micronutrients and frailty. For the cross-sectional and prospective analyses, we categorized each principal component (i.e., our independent variables) based on quartiles. The highest quartile was used as the reference category. We adjusted the analyses for baseline socio-demographic variables (age, sex, and education level), and health-related covariates (total cholesterol, total triglycerides, season of the blood drawing, polypharmacy, BMI, depressive symptomatology, and MMSE). A parameter for cohort was included in all models to account for differences between cohorts.

For secondary analyses, we further adjusted for CRP, except for AMI in which CRP was not measured, since inflammation may affect micronutrient concentrations, particularly of retinol, vitamin E and 25(OH)D [30].

To gain a better understanding of our main results, we ran models with each micronutrient as the main exposure variable separately (Supplementary data). For the lack of established cut-offs, participants were categorized into quartiles based on micronutrients concentration with the highest quartiles as the reference category. Based on the literature, 25(OH)D concentrations can be categorized into three classes: ≥50, 25–49, and < 25 nmol/L [31, 32]. The highest category was selected as the reference.

Statistical analyses were performed with SAS Statistical package release 9.3 (SAS institute Inc., Cary, NC, USA).

Results

Sample characteristics

Table 2 displays the baseline socio-demographic, lifestyle, and clinical characteristics as well as the plasma concentrations of fat-soluble micronutrients and the frailty status of participants included in the four cohorts (3C, AMI, TSHA and InCHIANTI). There are 1324 participants with complete data (i.e., for frailty, micronutrients and covariates) at baseline with mean age of 77.6 (SD 6.5) years old, and 55.8% females. On average, participants of the 3C study were older, less obese, more likely to be multi-medicated and were more educated than participants in the three other cohorts. Participants in the AMI study had the highest percentage of men, and participants in the TSHA cohort had the highest percentage of obesity. Participants from the InCHIANTI cohort exhibited the lowest plasma carotenoids concentration.

Table 2.

Baseline characteristics of participants from the four European cohorts

	Total	COHORT				p
		3C	AMI	TSHA	InCHIANTI
N (%)	1324	462 (34.9)	278 (21.0)	306 (23.1)	278 (21.0)
Socio-demography
Age [mean (SD)]	77.6 (6.5)	82.0 (4.3)	75.2 (6.2)	75.1 (5.7)	75.5 (6.9)	< 0.001
Sex [n (%)]						< 0.001
Males	585 (44.2)	175 (37.9)	173 (62.2)	121 (39.5)	116 (41.7)
Females	739 (55.8)	287 (62.1)	105 (37.8)	185 (60.5)	162 (58.3)
Education level [n (%)]						< 0.001
Low	890 (67.2)	115 (24.9)	220 (79.1)	286 (93.5)	269 (96.8)
Intermediate	306 (23.1)	234 (50.7)	52 (18.7)	11 (3.6)	9 (3.2)
High	128 (9.7)	113 (24.5)	6 (2.2)	9 (2.9)	0 (0.0)
Lifestyle
Smoking status [n (%)]						< 0.001
Non-smoker	852 (64.4)	295 (63.9)	184 (66.2)	200 (65.4)	173 (62.2)
Ex-smoker	404 (30.5)	145 (31.4)	79 (28.4)	75 (24.5)	105 (37.8)
Current smoker	68 (5.1)	22 (4.8)	15 (5.4)	31 (10.1)	0.0 (0.0)
BMI [n (%)]						< 0.001
< 25 kg/m2	416 (31.4)	216 (46.8)	67 (24.1)	54 (17.7)	79 (28.4)
[25–30] kg/m2	563 (42.5)	180 (39.0)	128 (46.0)	138 (45.1)	117 (42.1)
≥ 30 kg/m2	345 (26.1)	66 (14.3)	83 (29.9)	114 (37.3)	82 (29.5)
Health
Polypharmacy [n (%)]	622 (47.0)	279 (60.4)	142 (51.1)	137 (44.8)	64 (23.0)	< 0.001
Depressive symptomatology [n (%)]	185 (14.0)	42 (9.1)	12 (4.3)	61 (19.9)	70 (25.2)	< 0.001
MMSE [mean (SD)]	25.7 (3.7)	27.8 (2.1)	25.6 (3.4)	23.4 (4.3)	24.9 (3.4)	< 0.001
Biology median (IQR)
Total triglycerides (mg/L)	105.0 (64.0)	100.5 (58.0)	108.5 (66.0)	98.5 (64.0)	114.0 (75.0)	< 0.001
Total cholesterol (mg/L)	207.1 (56.0)	212.8 (61.1)	210.0 (56.0)	190.0 (47.0)	220.0 (54.0)	< 0.001
CRP (μg/mL)	2.4 (3.8)	1.6 (2.6)	–	3.1 (4.1)	3.2 (4.2)	< 0.001
Missing values	322	1	278	42	1
Micronutrients median (IQR)
Carotenoids (mmol/L)
Total carotenoids	1678.0 (1335.5)	2094.0 (1302.0)	1983.5 (1514.0)	1795.5 (1084.0)	883.5 (716.0)	< 0.001
Total carotenes	1041.5 (1025.5)	1366.5 (919.0)	1311.0 (1166.0)	1041.5 (770.0)	436.0 (399.0)	< 0.001
α-carotene	127.0 (204.0)	237.0 (227.0)	229.5 (262.0)	77.0 (66.0)	43.0 (39.0)	< 0.001
β-carotene	436.5 (514.0)	679.5 (570.0)	588.5 (581.0)	367.0 (291.0)	190.5 (182.0)	< 0.001
Lycopene	369.5 (417.5)	376.5 (316.0)	422.5 (454.0)	620.0 (510.0)	176.0 (197.0)	< 0.001
Total xanthophylls	591.5 (428.0)	626.0 (442.0)	624.0 (365.0)	643.0 (448.0)	420.0 (352.0)	< 0.001
Cryptoxanthin	206.5 (266.0)	248.5 (285.0)	152.5 (164.0)	318.0 (346.0)	117.5 (162.0)	< 0.001
Lutein + zeaxanthin	335.0 (219.0)	339.0 (206.0)	429.0 (244.0)	306.0 (177.0)	283.5 (203.0)	< 0.001
Vitamin E (μmol/L)
Total vitamin E	29.8 (10.1)	31.5 (9.3)	26.2 (9.1)	32.1 (10.2)	28.2 (9.4)	< 0.001
α-tocopherol	28.4 (9.8)	29.6 (8.4)	24.8 (8.8)	31 (9.9)	26.7 (9.1)	< 0.001
γ-tocopherol	1.2 (0.8)	1.4 (0.8)	1.2 (0.9)	1.0 (0.6)	1.1 (0.6)	< 0.001
Retinol (μmol/L)	1.8 (0.7)	2.0 (0.6)	1.9 (0.7)	1.7 (0.6)	1.7 (0.5)	< 0.001
25(OH)D (nmol/L)	42.6 (33.7)	41.7 (37.0)	38.9 (30.2)	50.5 (36.4)	42.8 (31.0)	< 0.001
Deficiency in 25(OH)D						< 0.001
< 25 nmol/L	242 (18.3)	88 (19.1)	66 (23.7)	37 (12.1)	51 (18.4)
[25–50] (nmol/L)	543 (41.0)	195 (42.2)	116 (41.7)	114 (37.3)	118 (42.5)
≥ 50 nmol/L	539 (40.7)	179 (38.7)	96 (34.5)	155 (50.7)	109 (39.2)
Frailty
Frail [n (%)]	290 (21.9)	96 (20.8)	49 (17.6)	66 (21.6)	79 (28.4)	0.017
Sedentarity [n (%)]	494 (37.4)	265 (57.4)	64 (23.2)	86 (28.1)	79 (28.4)	< 0.001
Missing values	2	0	2	0	0
Weight loss [n (%)]	169 (12.8)	64 (13.9)	26 (9.4)	52 (17.0)	27 (10.0)	0.018
Missing values	8	0	0	0	8
Weakness [n (%)]	339 (26.1)	102 (23.2)	62 (22.3)	85 (27.9)	90 (32.4)	0.017
Missing values	23	22	0	1	0
Slowness [n (%)]	403 (30.6)	139 (30.1)	61 (22.2)	99 (32.4)	104 (37.7)	0.001
Missing values	5	0	3	0	2
Exhaustion [n (%)]	248 (18.8)	74 (16.2)	38 (13.7)	54 (17.7)	82 (29.5)	< 0.001
Missing values	5	4	1	0	0

IQR interquartile range, MMSE mini-mental state examination, SD standard deviation

At baseline, 290 (21.9%) participants were considered frail. The AMI participants had the lowest percentage of frail participants (17.6%) and the InCHIANTI participants had the highest percentage (28.4%), which is explained by the high percentages of weakness, slowness and exhaustion. Among 915 non-frail participants at baseline, 84 became frail during the follow-up.

Table 3 presents the median concentrations for each fat-soluble micronutrient according to prevalent frailty and incident frailty. At baseline, frail older adults had significantly lower concentrations of fat-soluble micronutrients, except retinol. At follow-up, all micronutrients were lower in frail participants. Nevertheless, only the plasma concentration of lutein and zeaxanthin differed significantly from their non-frail counterparts.

Table 3.

Median fat-soluble micronutrient concentrations (IQR) based on prevalent and incident frailty status

	Baseline			Follow-up
	Non-frail	Frail	p	Non-frail	Frail	p
N	1034	290		831	84
Carotenoids (mmol/L)
Total carotenoids	1766.5 (1386.0)	1434.0 (1230.0)	< 0.001	1831.0 (1410.0)	1461.0 (1345)	< 0.02
Total carotenes	1088.5 (1062.0)	895.0 (901.0)	< 0.001	1136.0 (1072.0)	1007.0 (997.5)	0.06
α-carotene	132.5 (205.0)	115.0 (187.0)	< 0.008	142.0 (218.0)	112.5 (214.0)	0.23
β-carotene	466.5 (535.0)	387.5 (486.0)	< 0.001	484.0 (547.0)	417.0 (571.0)	0.07
Lycopene	389.0 (423.0)	305.5 (380.0)	< 0.001	399.0 (433.0)	339.5 (370.0)	0.09
Total xanthophylls	623.5 (442.0)	468.5 (370.0)	< 0.001	638.0 (450.0)	533.5 (399.0)	< 0.009
Cryptoxanthin	219.0 (279.0)	143.0 (234.0)	< 0.001	227.0 (290.0)	173.5 (272.5)	0.10
Lutein + zeaxanthin	352.0 (228.0)	267.5 (183.0)	< 0.001	361.0 (236.0)	315.0 (219.5)	< 0.02
Vitamin E (μmol/L)
Total vitamin E	30.4 (10.3)	28.7 (8.7)	< 0.001	30.5 (10.4)	29.8 (7.1)	0.79
α-tocopherol	29.0 (10.1)	27.6 (8.3)	< 0.001	29.1 (10.2)	28.5 (6.5)	0.86
β-tocopherol	1.2 (0.8)	1.1 (0.6)	< 0.001	1.2 (0.8)	1.2 (0.9)	0.66
Retinol (μmol/L)	1.8 (0.6)	1.8 (0.8)	0.273	1.8 (0.6)	1.9 (0.7)	1.00
25(OH)D (nmol/L)	44.3 (33.7)	39.1 (33.3)	< 0.001	44.4 (34)	44.9 (28.2)	0.58

IQR interquartile range

Cross-sectional and prospective associations between fat-soluble micronutrients and frailty

Table 4 shows the cross-sectional and prospective associations between patterns of fat-soluble micronutrients and frailty. At baseline, multivariate analysis revealed that only the principal component 2 was significantly associated with frailty status (p = 0.02). Indeed, compared with the highest quartile, participants in the lowest quartile, characterized by low concentrations in vitamins E and retinol but high concentrations in carotenes, were more likely to be frail (Odds ratio = 2.2; 95% confidence interval 1.2–3.7 in model adjusted for cohort, socio-demographic, lifestyle, and clinical covariates). However, no significant association was observed in principal components 1 and 3 (p = 0.10 and p = 0.10, respectively).

Table 4.

Cross-sectional and prospective associations between patterns of fat-soluble micronutrients considered as quartiles and frailty in older adults

	Cross-sectional association (n = 1324)				Prospective association (n = 915)
	OR*	95% CI		p	OR*	95% CI		p
Component 1				0.10				0.40
Q1	1.98	1.10	3.56		1.59	0.67	3.82
Q2	1.83	1.07	3.13		1.48	0.69	3.17
Q3	1.62	0.98	2.66		0.86	0.41	1.84
Q4	1				1
Component 2				< 0.02				0.93
Q1	2.15	1.24	3.72		1.19	0.51	2.78
Q2	1.10	0.65	1.86		1.29	0.61	2.76
Q3	1.29	0.79	2.11		1.21	0.58	2.55
Q4	1				1
Component 3				0.10				1.00
Q1	1.76	0.98	3.14		0.97	0.42	2.25
Q2	1.96	1.14	3.39		0.94	0.42	2.13
Q3	1.37	0.81	2.31		0.98	0.45	2.13
Q4	1				1

^*Adjusted for age, sex, education level, cohort, total cholesterol, total triglycerides,season of blood drawing, polypharmacy, body mass index, depressive symptomatology and Mini-Mental State Examination

At follow-up, among 915 non-frail participants at baseline, multivariate analysis did not show any significant association between patterns of fat-soluble micronutrients and incident frailty (p = 0.40 for principal component 1; p = 0.93 for principal component 2 and p = 1.00 for principal component 3).

In complementary analyses, further adjustment for CRP as marker of inflammation did not alter the cross-sectional and prospective results (Supplemental Table 1).

The cross-sectional and prospective associations between individual fat-soluble micronutrients and frailty were also examined (Supplemental Table 2). At baseline, lower concentrations in lycopene, lutein and zeaxanthin, α-tocopherol and retinol were significantly associated with higher odds of being frail in fully adjusted models (p = 0.008, p < 0.001, p = 0.008 and p = 0.003, respectively). However, there were no observed associations between isolated micronutrients and the risk of frailty.

Discussion

Using four European cohorts of older adults 65 years of age or older, our findings revealed a significant association between a fat-soluble micronutrient pattern with low concentrations of vitamins E and A and prevalent frailty, while no association with incident frailty was observed. Surprisingly, the micronutrient patterns that are mainly driven by high concentrations of carotenoids and one characterized by high concentrations of 25(OH)D were not significantly associated with prevalent or incident frailty in our study.

To the best of our knowledge, the present analysis is the first study to investigate the association between various patterns of plasma fat-soluble micronutrients with both prevalent and incident frailty. Previous studies examined isolated micronutrients [11–13, 32] or number of micronutrient deficiencies [13]. A cross-sectional analysis among 754 women aged between 70 and 80 years old, from the WHAS I and II, showed a significantly higher odds ratio of being frail in participants whose micronutrient concentrations for β-carotene, lutein + zeaxanthin and total carotenoids were in the lowest quartile compared to the top three quartiles adjusted for age, socioeconomic score, smoking status and BMI [11]. Retinol, α-tocopherol, 25(OH)D and other carotenoids were not associated with prevalent frailty in this study. In 827 non-disabled, non-demented participants from the InCHIANTI cohort, with a mean age of 73.6 years, Ble et al. showed that participants in the highest tertile of serum vitamin E had a significantly lower odds of being frail than those in the lowest tertile [12]. In 451 non-frail women, age 65 years of age or older, from the WHAS I, Semba et al. showed that women in the lowest quartile of carotenoid concentration were at a higher 3-year risk of frailty after adjusting for age, smoking status and chronic pulmonary disease, while α-tocopherol, retinol, 25(OH) D concentrations were not associated with incident frailty [13]. The authors have also reported that the number of deficiencies in serum vitamins A, D, E, B6 and B12, carotenoids, folates, zinc and selenium were associated with an increased risk of becoming frail.

Contrary to aforementioned studies, we analyzed the patterns of fat-soluble micronutrients to take into account the complexity of nutritional exposure. We used a classical technique of data reduction to explain the maximum amount of variation for the nine micronutrients with a minimum number of variables, avoiding issues of collinearity and multiple testing. Our analysis strategy limits the comparability of our findings with previous studies. However, we observed that the fat-soluble micronutrient pattern characterized by low levels of vitamin E (α- and γ-tocopherol) and retinol, and high levels of carotenes, was associated with a higher odd of being frail in the cross-sectional analysis but did not predict the occurrence of frailty at short term. We thus hypothesized that this pattern might represent a result of frailty rather than a cause of frailty, although randomized controlled trials are the only means to infer causality. Results from the isolated micronutrients analyses suggested that the association between the component 2 and prevalent frailty was mainly driven by low concentrations of α-tocopherol. Low vitamin E status was previously associated with poorer health status and decline of physical function over 3 years of follow-up [34, 35]. The regulation of oxidative stress by this vitamin may potentially explain the association observed in the study. Several previous studies showed that high oxidative stress and subsequent inflammation were associated with prevalent frailty [32–38]. The principal component 2 reflects the consumption of foods from different sources, vitamin A being exclusively provided by animal food products, while carotenes being mainly from vegetable origins. This could be equatable with a “westernized” dietary pattern, which has been inconsistently associated with frailty risk [36, 37].

A growing body of interest was dedicated on 25(OH)D [9]. Indeed, 25(OH)D is a key component in the maintenance of calcium homeostasis as well as bone health, and has been associated with several non-skeletal biologic systems [9]. Observational studies on the association between 25(OH)D status and frailty were not conclusive, although interventional studies of the effect of supplemental 25(OH)D have yielded a positive influence on the frailty status, mainly via improvements in the physical performances [9]. The lack of association with patterns characterized by high concentrations of 25(OH)D in our study may be explained by our method which produced patterns that may have given more importance to retinol and tocopherols status. Furthermore, we also did not observe any association between 25(OH)D and frailty in complementary analyses. This therefore warrants further research.

Our study has several limitations. Studies involving older adults often face a selection bias due to survival effect. Moreover, several frailty criteria were self-reported, which might have led to the underestimation of the number of cases of frailty. In 3C and AMI, slowness was defined using the Rosow–Breslau scale that has been shown to be strongly associated with walking [24]. Furthermore, 3C and AMI assessed weakness using chair stand test that was shown to be a good proxy for handgrip strength [25, 26]. Multiple definitions for shrinking and low energy expenditure were used, and both were self-reported and comparable. However, we cannot preclude a persisting misclassification bias that may lead to an under or overestimation of frailty prevalence/incidence. Another limitation is the lack of availability of other markers that could have explained the mechanisms involved in the association between the identified pattern and frailty (such as markers of redox status).

The utilization of both cross-sectional and prospective analyses is a major strength. The study included participants from four different European cohorts, which made the findings more generalizable. In addition, we considered pattern of micronutrients instead of isolated micronutrients. Other strengths included a large sample size with a population-based design, and the ability to adjust for major confounders, notably factors influencing fat-soluble micronutrients bioavailability (i.e., TC, TG, polypharmacy, season of blood drawing), or frailty status, such as global cognitive performances [38].

Conclusion

In these four cohorts of older adults 65 years of age or older from the south of Europe, the fat-soluble micronutrient pattern characterized by low levels of vitamin E and retinol was associated with higher prevalence of frailty but not with its occurrence. The findings in our study encourage a better nutritional status assessment in the older population.