Association Between Mediterranean Diet and Functional Status in Older Adults: A Longitudinal Study Based on the Washington Heights–Inwood Columbia Aging Project

Abstract

Background

Current evidence on the association between Mediterranean diet (MeDi) intake and activities of daily living (ADL) is limited and inconsistent in older adults.

Methods

This study included 1 696 participants aged ≥65 years in the Washington Heights–Inwood Community Aging Project study. The MeDi score was calculated based on data collected from the Willett’s semiquantitative food frequency questionnaire. The multivariable-adjusted Cox regression model was applied to examine the association of MeDi score with risks of disability in basic (BADL) and instrumental ADL (IADL), as well as the overall ADL (B-IADL).

Results

Eight hundred and thirty-two participants with incident ADL disability were identified over a median follow-up of 5.39 years. The continuous MeDi score was significantly associated with decreased risk of disability in B-IADL (hazard ratio = 0.95, 95% confidence interval = 0.91–0.99, p = .018) in a model adjusted for age, sex, race/ethnicity, educational level, and dietary calories intake but was no longer significant after additionally adjusted for multiple comorbidities and physical activities (0.97 [0.93, 1.01], p = .121). The continuous MeDi score was significantly associated with decreased risk of disability in B-IADL (0.92 [0.85, 1.00], p = .043) and BADL (0.90 [0.82, 0.99], p = .030) in non-Hispanic Whites, but not in non-Hispanic Blacks and Hispanics (p > .05 for all).

Conclusions

Higher MeDi score was associated with decreased risk of ADL disability, particularly in non-Hispanic Whites.

ADL, Disability, Epidemiology, Mediterranean diet

The activities of daily living (ADL) scale is widely applied to evaluate functional status for older adults (1). The inability to accomplish essential ADL may result in poor quality of life, increased health care costs, and unsafe conditions (2). ADL is mainly classified into basic (BADL) and instrumental (IADL) ADL, of which BADL refers to the skills required to manage the basic physical needs (eg, eating, toileting, bathing, dressing, grooming, transferring, mobility), and IADL includes complex activities to live in the community independently (eg, cooking, housekeeping, telephone using, shopping, handling finance, managing medication) (1). IADL ability is reported to be sensitive to early cognitive decline and IADL impairment often occurs in mild cognitive decline, whereas the BADL performance is usually driven by physical functioning and BADL impairment is often not present until the late stage of dementia (2). It has been reported that 14% and 25% of the U.S. noninstitutional population aged 65 years and older has limitations in BADL and IADL, respectively, and needs assistance with ADL from other persons (3).

Mediterranean diet (MeDi) is characterized by a high consumption of olive oil, legumes, nuts, cereals, fruits, and leafy green vegetables; a moderate to high consumptions of fish; a low to moderate intake of meat, poultry, and dairy products; and a moderate intake of wine (4). Some nutrients and foods in MeDi such as fruits, vegetables (5), and dietary diversity (6) have been proposed as protective factors against the ADL impairment.

To our knowledge, only 2 cross-sectional (7,8) and 2 prospective cohort (9,10) studies have been conducted to examine the relationships between MeDi and ADL, but with limited and inconsistent results. The MeDi food pattern has been reported to be linked to decreased risk of disability in BADL (8,9), but not in IADL (9). Among older adults with diabetes, those with higher MeDi score had better B-IADL performance (7). In another study of community dwellers aged 65 and older, MeDi was found to be associated with reduced risks of B-IADL disability for females, but not for males (10).

Both diet quality (11) and prevalence of ADL disability (3) vary by sex and race in the U.S. population. The sex-specific association between MeDi and ADL has been examined in only 1 study (10). However, there are no data focusing on the modification effects by race/ethnicity. In the present study, we aimed to examine the association between MeDi and risks of ADL disability and its modification by sex and by race/ethnicity among participants of the Washington Heights–Inwood Community Aging Project (WHICAP) study.

Method

Participants and Study Design

The WHICAP study is an ongoing prospective, community-based, multiethnic longitudinal study of Medicare beneficiaries 65 years and older residing in northern Manhattan (Washington Heights, Hamilton Heights, and Inwood). The study has been described in detail elsewhere (12). Briefly, at study entry, participants underwent a comprehensive examination including the assessment of general health and function, standardized physical and neurological examination, and a neuropsychological battery of tests. Follow-up visits were performed every 1.5–2 years, repeating similar examinations. The WHICAP study was approved by the Institutional Review Board at Columbia University Medical Center. All participants provided written informed consent.

Among 5 138 participants at baseline, we excluded those who had no data on ADL (n = 183), had ADL disability at baseline (B-IADL scores < 13, n = 1 953), had no repeated ADL assessment during the follow-up visits (n = 718), or had prevalent dementia (n = 59; Supplementary Figure S1). Participants were further excluded for missing data on MeDi (n = 325), years of education (n = 8), body mass index (BMI, n = 154), or leisure-time physical activity (n = 42). A total of 1 696 participants who had no ADL disability at baseline and had complete data on variables of interest were included in the data analysis.

MeDi Quantification

The dietary information was collected by the trained personnel using Willett’s semiquantitative food frequency questionnaire (FFQ) at baseline. A good validity and reliability of components in the FFQ has been reported in the WHICAP study (13).

The degree of adherence to MeDi pattern was quantified with the MeDi scale consisting of 9 components (14). The sex-specific median of caloric-adjusted intake was used as the cutoff value for each of the 9 food components. Individuals were assigned 1 point for intake ≥ cutoff value for the beneficial components (including fruits, vegetables, legumes, cereals, fish, and a ratio of monounsaturated fats to saturated fats [MUFA:SFA]), or for intake < cutoff value for the detrimental components (including meat and dairy products), or for a mild-to-moderate alcohol consumption (>0 to <30 g/day). The total MeDi scores (ranging from 0 to 9) were calculated by summing the scores of 9 food components, with higher scores indicating a better adherence to MeDi pattern.

ADL Assessment

The information on ADL performance was collected at each visit in the WHICAP study. Based on the ADL quantification in the previous WHICAP study (15) and another study (10), similar methods were used to assess the BADL and IADL performance with 7 items (bathing, dressing, toileting, combing hair, getting in and out of bed/chair, walking, and feeding) for BADL and 6 items (telephone using, cooking, shopping, doing light chores, managing own medication, and handling money) for IADL (Supplementary Table S1). Participants were assigned 1 point when they had no difficulties/problems in each ADL function. The ADL scores were calculated by summing the scores of the 7 items in BADL (ranging from 0 to 7) and 6 items in IADL (ranging from 0 to 6), with higher scores indicating better ADL performance. The sum of BADL and IADL (B-IADL) scores (ranging from 0 to 13) was used to quantify the overall ADL performance. Participants with disability in BADL (scores < 7), IADL (scores < 6), and B-IADL (scores < 13) were defined as those who had difficulties/problems in any item for the corresponding ADL.

Covariates

Information on date of birth, sex, race/ethnicity, educational level, and histories of comorbidities were collected at baseline interviews. Age in years at each visit was calculated as the time intervals between survey date and date of birth. Participants were assigned into 1 of the 4 race/ethnicity groups (non-Hispanic White, non-Hispanic Black, Hispanic, or other) by using the format of the 2000 U.S. Census. Education duration in years and smoking history were self-reported by standard questionnaires. For assessment of smoking habit at baseline, participants who had ever smoked ≥1 cigarette per day for a period of ≥1 year were regarded as smokers, otherwise nonsmokers. Daily calories intake were estimated based on the FFQ. Histories of comorbidities of hypertension, diabetes, and heart diseases were obtained from self-reported information as well as the use of medication. Depression status was assessed with the 10-item Center for Epidemiological Studies—Depression scale (CES-D) at each visit, and a participant was identified with major depression if he or she had the CES-D scores ≥ 4 (16). The Godin leisure-time exercise questionnaire (17) was applied to measure the frequency, duration, and intensity of leisure-time physical activity during the most recent 2 weeks, which were calculated into metabolic equivalent minutes. With the participant standing, height was measured to the nearest 0.5 cm without shoes, and the weight was measured to the nearest 0.1 kg with a balance scale. BMI was calculated as weight divided by height squared (kg/m²) at baseline and each follow-up. A composite cognitive score at baseline was developed to assess the overall performance on different cognitive domains including language, processing speed/executive functioning, memory, and visuospatial abilities (18,19).

Statistical Methods

The MeDi score was divided into 3 categories (low: 0–3, medium: 4–5, high: 6–9) following previous WHICAP studies (14,20,21). Differences in demographics across MeDi categories were tested with analysis of variance and Chi-square test for the continuous and categorical variables, respectively.

The association of MeDi with risks of ADL disability was estimated using Cox regression models. The survival-time metrics were years of follow-up from the baseline through the last visit or the first occurrence of incident ADL disability (whichever came first). Both categorical (setting the low levels as reference) and continuous MeDi scores were used as the predictors. Model 1 was adjusted for age at baseline, sex, race/ethnicity, years of education, and dietary calories intake (log-transformed to approximate to normal distribution), and Model 2 was further adjusted for depression, leisure-time physical activity, BMI, smoking history, and comorbidities of hypertension, diabetes, and heart diseases. The method of scaled Schoenfeld residuals was applied to test the proportional hazards assumption for Cox regression models.

In addition, the association between MeDi components and risks of ADL disability was estimated with the single- and multiple-component models. The single-component model was conducted when separately including each MeDi component in the Cox regression model adjusting for all the covariates. All the 9 MeDi components were simultaneously included in the multiple-component model.

Due to the race- and sex-disparity in diet quality (11) and ADL disability (3) in the U.S. population, stratified analyses by race/ethnicity and sex were conducted. An interaction term between MeDi score and race/ethnicity was included in the fully adjusted model and tested by using the likelihood-ratio test. The interaction between MeDi score and sex was tested with the same method. Because interaction tests are generally underpowered (22), the p value of interaction < .10 was regarded as significant (23,24). In the secondary analysis, we examined the association between MeDi score and the ADL by further adjusting the baseline cognition.

All data analyses were conducted with R version 4.0.3. Two-sided p values less than .05 were statistically significant.

Results

Missing Data Analysis

Compared with the participants who were excluded due to missing data or prevalent dementia at baseline, the eligible individuals included in the analytical sample had significantly more years of education and higher baseline ADL score (Supplementary Table S2). In addition, the included participants had a higher proportion of females, and non-Hispanic Whites, and higher proportions of participants with history of hypertension and heart disease.

Baseline Characteristics

The 1 696 participants were followed up over a median follow-up of 5.39 (range, 1.13–19.47; mean, 6.34; SD, 4.09) years, for a total of 41 119.93 person-years. A total of 832 participants with incident ADL disability were identified. The mean age of participants was 75.04 (SD = 5.92) years at baseline, and one-third of participants were males (Table 1). The proportions were 30.54%, 29.72%, 38.09%, and 1.65% for non-Hispanic White, non-Hispanic Black, Hispanic, and other race/ethnicities, respectively. Participants who had higher MeDi score were more likely to be Hispanic, to have higher levels of leisure-time physical activity, and lower levels of BMI.

Table 1.

Baseline Characteristics Across MeDi Categories

	Categories of MeDi*
Characteristics	Low (n = 504)	Medium (n = 722)	High (n = 470)	Total (n = 1696)	p Values
Age at baseline (years), mean (SD)	75.30 (6.00)	74.93 (5.97)	74.94 (5.77)	75.04 (5.92)	.521
Gender					.764
Male	169 (33.53%)	228 (31.58%)	154 (32.77%)	551 (32.49%)
Female	335 (66.47%)	494 (68.42%)	316 (67.23%)	1145 (67.51%)
Race/ethnicity					<.001
Non-Hispanic White	160 (31.75%)	215 (29.78%)	143 (30.43%)	518 (30.54%)
Non-Hispanic Black	177 (35.12%)	218 (30.19%)	109 (23.19%)	504 (29.72%)
Hispanic	160 (31.75%)	278 (38.50%)	208 (44.26%)	646 (38.09%)
Others	7 (1.39%)	11 (1.52%)	10 (2.13%)	28 (1.65%)
Education duration (years), mean (SD)	11.38 (4.66)	11.58 (4.70)	11.10 (4.92)	11.38 (4.75)	.233
Dietary calories (kcal/day), mean (SD)	1410.64 (524.75)	1406.77 (460.39)	1424.07 (455.44)	1412.71 (478.88)	.825
Depression					.071
No	388 (76.98%)	559 (77.42%)	387 (82.34%)	1334 (78.66%)
Yes	116 (23.02%)	163 (22.58%)	83 (17.66%)	362 (21.34%)
Leisure-time physical activity (MET-min/2 wk)					.009
Low (0)	151 (29.96%)	186 (25.76%)	98 (20.85%)	435 (25.65%)
Middle (>0 and <1 260)	172 (34.13%)	234 (32.41%)	160 (34.04%)	566 (33.37%)
High (≥1 260)	181 (35.91%)	302 (41.83%)	212 (45.11%)	695 (40.98%)
BMI (kg/m2), mean (SD)	27.25 (4.25)	27.13 (4.17)	26.60 (4.32)	27.02 (4.24)	.038
Smoking history					.106
No	266 (52.78%)	425 (58.86%)	266 (56.60%)	957 (56.43%)
Yes	238 (47.22%)	297 (41.14%)	204 (43.40%)	739 (43.57%)
History of hypertension					.350
No	77 (15.28%)	115 (15.93%)	87 (18.51%)	279 (16.45%)
Yes	427 (84.72%)	607 (84.07%)	383 (81.49%)	1417 (83.55%)
History of diabetes					.515
No	373 (74.01%)	539 (74.65%)	362 (77.02%)	1 274 (75.12%)
Yes	131 (25.99%)	183 (25.35%)	108 (22.98%)	422 (24.88%)
History of heart diseases					.078
No	270 (53.57%)	431 (59.70%)	278 (59.15%)	979 (57.72%)
Yes	234 (46.43%)	291 (40.30%)	192 (40.85%)	717 (42.28%)
MeDi score, mean (SD)	2.44 (0.74)	4.46 (0.50)	6.49 (0.73)	4.42 (1.66)	<.001

Notes: BMI = body mass index (kg/m²); MeDi = Mediterranean diet scores; MET = metabolic equivalent; SD = standard deviation. Values in bold text mean statistically significant (p < .05).

*MeDi scores were categorized into low (0–3), medium (4–5), and high levels (6–9).

Association Between MeDi and ADL Disability in Overall Sample

The hazard ratio (HR) for risks of ADL disability associated with categorical and continuous MeDi scores among all the participants is shown in Table 2. The proportional hazards assumption was not statistically violated for MeDi score in Cox regression models (B-IADL, p = .494; BADL, p = .738; and IADL, p = .372). Compared with participants at the low levels of MeDi score, those at the high levels had decreased risk of disability in B-IADL (HR = 0.92, 95% confidence interval [CI] = 0.76–1.10), BADL (HR = 0.98, 95% CI = 0.80–1.20), and IADL (HR = 0.89, 95% CI = 0.72–1.10), but without statistical significance. The inverse association between continuous MeDi score and risks of B-IADL disability was statistically significant in Model 1 (HR = 0.95, 95% CI = 0.91–0.99) but not significant in Model 2 (HR = 0.97, 95% CI = 0.93–1.01).

Table 2.

HR (95% CI) for ADL Disability Associated With MeDi by Using the Cox Regression Model (n = 1 696)

	Categories of MeDi
ADL Disability	Low (0–3)	Medium (4–5)	High (6–9)	ptrend	Continuous MeDi
B-IADL disability
No. of cases/total	256/504	354/722	222/470
Model 1*	Reference	0.98 (0.84, 1.16)	0.86 (0.72, 1.03)	.110	0.95 (0.91, 0.99)
Model 2†	Reference	1.00 (0.85, 1.17)	0.92 (0.76, 1.10)	.371	0.97 (0.93, 1.01)
BADL disability
No. of cases/total	209/504	296/722	179/470
Model 1*	Reference	1.01 (0.85, 1.21)	0.90 (0.74, 1.10)	.324	0.96 (0.92, 1.00)
Model 2†	Reference	1.04 (0.87, 1.24)	0.98 (0.80, 1.20)	.849	0.98 (0.93, 1.03)
IADL disability
No. of cases/total	189/504	262/722	170/470
Model 1*	Reference	0.96 (0.80, 1.16)	0.86 (0.70, 1.06)	.167	0.96 (0.92, 1.01)
Model 2†	Reference	0.98 (0.81, 1.19)	0.89 (0.72, 1.10)	.292	0.97 (0.93, 1.02)

Notes: ADL = activities of daily living; BADL = basic activities of daily living; B-IADL = sum of BADL and IADL; BMI = body mass index; CI = confidence interval; HR = hazard ratio; IADL = instrumental activities of daily living; MeDi = Mediterranean diet scores. Values in bold text mean statistically significant (p < .05).

*Model 1 was adjusted for age, sex, race/ethnicity, years of education, and dietary calories intake.

^†Model 2 was adjusted for covariates from Model 1 plus depression, leisure-time physical activity, BMI, smoking history, and comorbidities of hypertension, diabetes, and heart disease.

Association Between MeDi Components and ADL Disability

In the single-component model, a high intake of fruits was associated with decreased risk of disability in B-IADL (high vs low, HR = 0.78, 95% CI = 0.68–0.90), BADL (high vs low, HR = 0.82, 95% CI = 0.70–0.96), and IADL (high vs low, HR = 0.80, 95% CI = 0.68–0.94) (Supplementary Figure S2). Lower consumption of dairy products was associated with increased risks of IADL disability (low vs high, HR = 1.20, 95% CI = 1.02–1.41) (Supplementary Figure S2). Similar results were found when all the MeDi components were mutually adjusted in the multiple-component model (data not shown). In addition, we found significant interactions of “fruit × race/ethnicity” (p_interaction = .018) and “MUFA:SFA ratio × race/ethnicity” (p_interaction = .068) for the B-IADL disability, and significant interactions of “fruit × race/ethnicity” (p_interaction = .030) and “dairy products × race/ethnicity” (p_interaction = .054) for the BADL disability. The results of race-specific association indicated that higher intake of fruits was significantly associated with lower risks of disability in B-IADL (HR = 0.57, 95% CI = 0.44–0.74) and BADL (HR = 0.58, 95% CI = 0.43–0.77) in non-Hispanic Whites but not in non-Hispanic Blacks or in Hispanics.

Association Between MeDi and ADL Disability Stratified by Race/Ethnicity and by Sex

Compared with non-Hispanic Blacks, non-Hispanic Whites and Hispanics had significantly higher MeDi score (Supplementary Figure S3). There was no difference of MeDi score between males and females.

Among the non-Hispanic Whites, higher MeDi was associated with decreased risk of disability in B-IADL (high vs low, HR = 0.66, 95% CI = 0.46–0.94), BADL (high vs low, HR = 0.63, 95% CI = 0.42–0.94), and IADL (high vs low, HR = 0.61, 95% CI = 0.38–0.98) in Model 1 but no longer significant in the fully adjusted Model 2 (Table 3). Higher continuous MeDi score was significantly associated with lower risks of disability in B-IADL (HR = 0.92, 95% CI = 0.85–1.00, p = .043) and BADL (HR = 0.90, 95% CI = 0.82–0.99) in Model 2. We found a significantly inverse association between categorical MeDi and risks of disability in B-IADL (high vs low, HR = 0.69, 95% CI = 0.48–0.99) and BADL (high vs low, HR = 0.65, 95% CI = 0.43–0.98) in Model 1 but not in Model 2 among the non-Hispanic Blacks. No significant association was found in Hispanics. There was significant interaction between MeDi and race/ethnicity for BADL disability (p_interaction = .054).

Table 3.

Race-Specific HR (95% CI) for ADL Disability Associated With MeDi by Using the Cox Regression Model (n = 1 696)

	Categories of MeDi
ADL Disability	Low (0–3)	Medium (4–5)	High (6–9)	ptrend	Continuous MeDi	pinteraction‡
B-IADL disability						.210
Non-Hispanic White, no. of cases/total	77/160	115/215	49/143
Model 1*	Reference	1.09 (0.81, 1.47)	0.66 (0.46, 0.94)	.031	0.89 (0.83, 0.97)
Model 2†	Reference	1.10 (0.81, 1.48)	0.72 (0.50, 1.05)	.120	0.92 (0.85, 1.00)
Non-Hispanic Black, no. of cases/total	86/177	92/218	46/109
Model 1*	Reference	0.93 (0.69, 1.25)	0.69 (0.48, 0.99)	.050	0.93 (0.86, 1.00)
Model 2†	Reference	0.92 (0.68, 1.24)	0.76 (0.52, 1.10)	.158	0.94 (0.87, 1.03)
Hispanic, no. of cases/total	92/160	142/278	123/208
Model 1*	Reference	0.94 (0.72, 1.23)	1.09 (0.83, 1.44)	.482	1.00 (0.94, 1.06)
Model 2†	Reference	0.99 (0.76, 1.29)	1.12 (0.85, 1.48)	.393	1.00 (0.94, 1.07)
BADL disability						.054
Non-Hispanic White, no. of cases/total	67/160	96/215	40/143
Model 1*	Reference	1.02 (0.74, 1.41)	0.63 (0.42, 0.94)	.030	0.88 (0.80, 0.96)
Model 2†	Reference	0.98 (0.71, 1.36)	0.71 (0.47, 1.06)	.112	0.90 (0.82, 0.99)
Non-Hispanic Black, no. of cases/total	71/177	83/218	33/109
Model 1*	Reference	1.06 (0.77, 1.47)	0.65 (0.43, 0.98)	.069	0.92 (0.84, 1.00)
Model 2†	Reference	1.05 (0.76, 1.47)	0.74 (0.48, 1.13)	.244	0.95 (0.86, 1.04)
Hispanic, no. of cases/total	70/160	114/278	102/208
Model 1*	Reference	1.02 (0.75, 1.37)	1.28 (0.93, 1.75)	.114	1.04 (0.97, 1.11)
Model 2†	Reference	1.09 (0.80, 1.47)	1.29 (0.94, 1.78)	.106	1.04 (0.97, 1.12)
IADL disability						.721
Non-Hispanic White, no. of cases/total	49/160	83/215	28/143
Model 1*	Reference	1.04 (0.73, 1.50)	0.61 (0.38, 0.98)	.056	0.91 (0.82, 1.00)
Model 2†	Reference	1.06 (0.73, 1.54)	0.67 (0.41, 1.07)	.128	0.92 (0.83, 1.03)
Non-Hispanic Black, no. of cases/total	57/177	60/218	31/109
Model 1*	Reference	0.92 (0.64, 1.32)	0.78 (0.50, 1.21)	.274	0.98 (0.89, 1.07)
Model 2†	Reference	0.93 (0.65, 1.35)	0.80 (0.51, 1.26)	.355	0.98 (0.89, 1.08)
Hispanic, no. of cases/total	82/160	115/278	107/208
Model 1*	Reference	0.85 (0.64, 1.13)	0.95 (0.71, 1.27)	.781	0.97 (0.91, 1.04)
Model 2†	Reference	0.91 (0.68, 1.21)	0.99 (0.74, 1.34)	.985	0.98 (0.92, 1.05)

Notes: ADL = activities of daily living; BADL = basic activities of daily living; B-IADL = sum of BADL and IADL; BMI = body mass index; CI = confidence interval; HR = hazard ratio; IADL = instrumental activities of daily living; MeDi = Mediterranean diet scores. Values in bold text mean statistically significant (p < .05).

*Model 1 was adjusted for age, sex, years of education, and dietary calories intake.

^†Model 2 was adjusted for covariates from Model 1 plus depression, leisure-time physical activity, BMI, smoking history, and comorbidities of hypertension, diabetes, and heart disease.

^‡The interaction term “MeDi × race” was included in Model 2 and tested by using the likelihood-ratio test.

Among females, the association of continuous MeDi score with decreased risk of B-IADL disability was significant in Model 1 (HR = 0.94, 95% CI = 0.89–0.99) and marginally significant in Model 2 (HR = 0.95, 95% CI = 0.90–1.00, p = .052) (Table 4). We also found a significant association between continuous MeDi score and BADL disability (HR = 0.95, 95% CI = 0.90–1.00, p = .041) in Model 1 but not in Model 2 among females. There was no significant association between MeDi and risks of ADL disability in males. The interaction between MeDi and sex was not significant (p_interaction > .10).

Table 4.

Sex-Specific HR (95% CI) for ADL Disability Associated With MeDi by Using the Cox Regression Model (n = 1 696)

	Categories of MeDi
ADL Disability	Low (0–3)	Medium (4–5)	High (6–9)	ptrend	Continuous MeDi	pinteraction‡
B-IADL disability						.232
Male, no. of cases/total	83/169	113/228	66/154
Model 1*	Reference	1.06 (0.79, 1.41)	0.98 (0.71, 1.36)	.952	0.98 (0.91, 1.05)
Model 2†	Reference	1.08 (0.80, 1.45)	1.10 (0.79, 1.53)	.562	1.01 (0.94, 1.09)
Female, no. of cases/total	173/335	241/494	156/316
Model 1*	Reference	0.95 (0.78, 1.16)	0.81 (0.65, 1.01)	.059	0.94 (0.89, 0.99)
Model 2†	Reference	0.95 (0.78, 1.16)	0.85 (0.68, 1.06)	.158	0.95 (0.90, 1.00)
BADL disability						.312
Male, no. of cases/total	63/169	85/228	48/154
Model 1*	Reference	1.04 (0.75, 1.45)	1.05 (0.72, 1.53)	.798	0.99 (0.91, 1.08)
Model 2†	Reference	1.04 (0.74, 1.47)	1.19 (0.81, 1.76)	.382	1.03 (0.94, 1.13)
Female, no. of cases/total	146/335	211/494	131/316
Model 1*	Reference	1.00 (0.81, 1.24)	0.83 (0.66, 1.06)	.143	0.95 (0.90, 1.00)
Model 2†	Reference	1.02 (0.82, 1.26)	0.88 (0.69, 1.13)	.340	0.96 (0.91, 1.01)
IADL disability						.566
Male, no. of cases/total	62/169	85/228	46/154
Model 1*	Reference	1.01 (0.73, 1.41)	0.90 (0.61, 1.32)	.613	0.97 (0.89, 1.06)
Model 2†	Reference	1.08 (0.77, 1.51)	0.97 (0.65, 1.44)	.914	0.99 (0.91, 1.09)
Female, no. of cases/total	127/335	177/494	124/316
Model 1*	Reference	0.93 (0.74, 1.17)	0.84 (0.65, 1.08)	.163	0.96 (0.91, 1.02)
Model 2†	Reference	0.93 (0.74, 1.18)	0.86 (0.66, 1.11)	.232	0.97 (0.91, 1.02)

Notes: ADL = activities of daily living; BADL = basic activities of daily living; B-IADL = sum of BADL and IADL; BMI = body mass index; CI = confidence interval; HR = hazard ratio; IADL = instrumental activities of daily living; MeDi = Mediterranean diet scores. Values in bold text mean statistically significant (p < .05).

*Model 1 was adjusted for age, race/ethnicity, years of education, and dietary calories intake.

^†Model 2 was adjusted for covariates from Model 1 plus depression, leisure-time physical activity, BMI, smoking history, and comorbidities of hypertension, diabetes, and heart disease.

^‡The interaction term “MeDi × sex” was included in Model 2 and tested by using the likelihood-ratio test.

In the secondary analysis additionally adjusting for baseline cognition, we found the associations between continuous MeDi score and risk of ADL disability were attenuated but generally remained similar to the main analyses (data not shown). For example, when additionally adjusting for the composite cognitive score, the association [HR (95% CI; p)] between continuous MeDi score and risk of B-IADL disability was 0.97 (0.93–1.01; .114) in all participants and was 0.93 (0.86–1.01; .087) in non-Hispanic Whites; the association between continuous MeDi score and BADL disability was 0.91 (0.83–1.00; .050) in non-Hispanic Whites.

Discussion

In the current study, we found MeDi score was associated with decreased risk of functional disability in non-Hispanic Whites but not in non-Hispanic Blacks or Hispanics. Higher consumption of fruits and dairy, as 2 components of MeDi food pattern, was associated with lower risks of disability.

Loss of ADL is associated with multifaceted causes including aging (25), cardiovascular (26,27), neurological (28), musculoskeletal, and sensory conditions (1). The aging process is accompanied by a chronic increase in reactive oxygen species and accumulation of a low-grade pro-inflammation (29). The MeDi is rich in antioxidants and anti-inflammatory nutrients (30) which may be associated with antiaging effects and protection for ADL maintenance. In addition, the MeDi is suggested to have beneficial effects on gut microbiota and metabolic health (30) which affect risk for some diseases of older adults (eg, cardiovascular diseases [CVD], diabetes, osteoporosis, and stroke) and improves ADL performance (31). Epidemiological evidence has shown that the MeDi food pattern has protective effects against CVD, cancers, neurodegenerative diseases (30), frailty (32), and physical mobility (33) which are associated with subsequent ADL performance (1).

In the present study of older adults (mean age = 75.04 years), MeDi score was associated with decreased risk of B-IADL disability when adjusting for age, sex, race/ethnicity, years of education, and dietary calories intake, but the magnitude of inverse association was attenuated to be not significant after the adjustment for physical activity and comorbidities among the overall sample. It has been reported that Israeli older adults with higher MeDi score had better B-IADL performance (7), and had decreased likelihoods of poor BADL performance which was defined as below the median of BADL scores (8) in the cross-sectional studies. However, results from a cohort study of the U.S. older adults (mean age = 80.7 years) indicated that MeDi score was significantly inversely associated with disability in BADL but not in IADL (9). Overall, it seems that the inconsistent association between MeDi and ADL performance can be due to multiple reasons, including difference in the study design, population, sample size, assessments of MeDi score and ADL performance, analytical approaches, and covariates adjusted in the models.

Among MeDi components, we found that a high consumption of fruits was associated with decreased risk of disability in B-IADL, BADL, and IADL. Our findings were similar to the results from a previous study which focused on the health effects of single foods on ADL (34). Significantly inverse association between fruits intake and likelihoods of ADL disability was reported among the African American women aged 45–64 years (34). Compared with the French older adults who consumed fruits and vegetables less than once a day, those with at least once a day consumption had decreased risk of ADL disability (5). In the present study, lower intake of dairy products was associated with higher risks of IADL disability, but such an association was not found for B-IADL and BADL disability. Similarly, it has been reported that higher dairy intake was associated with lower likelihoods of IADL disability in the African American women (34) and Korean men (35). However, results from other prospective studies did not find significant association of dairy intake with risks of IADL disability in French women (36) and BADL disability in Japanese adults (37). Overall, results from these epidemiological studies were inconsistent, and more prospective studies consisting of different race/ethnicity as well as the intervention studies are recommended to obtain solid evidences.

We found that the inverse association between MeDi and ADL disability was stronger in non-Hispanic Whites than in other racial/ethnic groups. To our knowledge, this is the first study to examine the race-specific association between MeDi and ADL disability. Our findings were supported by the following evidence. Adherence to MeDi is culturally relevant and varies by race/ethnicity (38). In the present study, we found that non-Hispanic Whites consumed more alcohol and fruits, non-Hispanic Blacks consumed more meats, and Hispanics consumed more cereals, legumes, and dairy products (Supplementary Figure S3). Thus, a high MeDi score might be obtained by fulfilling the consumption of different foods in individuals with different races/ethnicities, which can contribute to the race-disparity in MeDi–ADL association. In addition, the leukocyte telomere length was suggested as a biological aging biomarker which is inversely associated with ADL disability among the older U.S. population (39). Results from a previous WHICAP study showed that MeDi-associated increased leukocyte telomere length was only found in non-Hispanic Whites but not in Hispanics or non-Hispanic Blacks (40). Furthermore, older adults with higher cognition levels were reported to have a better ADL performance (28). Data from a nationally representative population implied that the MeDi-associated improved cognitive function was stronger in non-Hispanic Whites compared with other race/ethnicity (non-Hispanic Black, Hispanic, other) (41).

Finally, it has been reported that the inverse association between MeDi and CVD risks was confined to participants at high socioeconomic levels (42). In our data, the non-Hispanic Whites had the highest levels of education (p < .001) and income (p < .001) compared with other race/ethnicity (data not shown). As a result, it could be presumed that different dietary habits, biological and aging processes, and socioeconomic status might play important roles in modifying the pathway from MeDi to the ADL-related health conditions, contributing to the disparities in the MeDi–ADL association across racial/ethnic groups. However, inconsistent results have also been reported that MeDi–CVD association was similar across race/ethnicity (43). In addition, the associations of continuous MeDi score with disability in B-IADL and BADL were attenuated to be not significant when the composite cognitive score was additionally adjusted. Similarly, results from another cohort study showed that the MeDi–ADL association was significantly inverse in the fully adjusted model but became to be not significant when additionally adjusting for cognitive performance among older adults (age range = 58–97 years) (9). More studies with multiethnic population are needed to confirm the findings in this study and to elucidate the underlying mechanisms for the race-disparity.

Results of the sex-stratified analysis indicated that the MeDi score was marginally significantly associated with decreased risk of B-IADL disability in females but not in males. A significant inverse association between MeDi and B-IADL disability was found for females but not for males in a French cohort study (10). It has been well documented that dietary quality (11) varies by sex in the U.S. population. Our data showed that females consumed more fruits, vegetables, and dairy products, but less alcohol, meat, and cereals compared with males (Supplementary Figure S3). In addition, cardiovascular conditions (26,27) and obesity (44,45) were found to be associated with poor ADL performance. Epidemiological studies have shown that compared with males, females had stronger inverse associations of MeDi with obesity (46), CVD (47,48), and a clustering of cardiovascular risk factors (diabetes, hypertension, obesity, and dyslipidemia) (46), suggesting potential modification effects by sex. In the present study, the sample size of males (n = 551) was about half of females (n = 1 145). The weaker association in males compared with females should be interpreted with caution because the statistical analysis might be underpowered due to the limited number of males with incident ADL disability within each MeDi category.

There are some limitations in the present study. A single measure of MeDi score might not capture the long-term levels of exposure. But a high stability in MeDi score over intervals greater than 7 years has been reported in a previous WHICAP study (14). Self-reported ADL performance might lead to biased estimation. In addition, the information on some other measures of IADL, such as public transportation or driving (1,10), was not available in the present study. Compared with the participants who were included, the excluded ones were more likely to be males and Hispanics, to have lower educational levels, and lower proportions of hypertension and heart disease at baseline, which might induce selection bias and biased estimation. The magnitude of association between MeDi and disability in B-IADL and BADL seemed to be stronger in non-Hispanic Whites than in other race/ethnicity groups, but p values of interactions between MeDi and race/ethnicity were .210 for B-IADL disability and .054 for BADL disability. The relaxation of type I error with higher values of α at 0.10 or 0.15 is increasingly employed to explore potential interactions in the epidemiological studies (23). Thus, the interaction between MeDi and race/ethnicity for BADL disability could not be ruled out.

This study had several advantages. The multiethnic participants in WHICAP study allow us to analyze race- and sex-specific association between MeDi and ADL performance. Our findings suggested that the potential modification effects by race need to be considered for the studies on this issue in the future. The prospective study design reduced the possibility of reverse causality. The association analyses were adjusted for multiple covariates. Furthermore, our findings can be well generalized by the multiethnic and community-based participants.

Conclusion

Higher MeDi score is associated with decreased risk of functional disability in the older adults, especially in non-Hispanic Whites. Due to the large number of older adults that need help with ADL, our findings have substantial public health implications.

Acknowledgments

This manuscript has been reviewed by WHICAP investigators for scientific content and consistency of data interpretation with previous WHICAP study publications. We acknowledge the WHICAP study participants and the WHICAP research and support staff for their contributions to this study.

Funding

Data collection and sharing for this project was supported by the Washington Heights–Inwood Columbia Aging Project (WHICAP, PO1AG007232, R01AG037212, RF1AG054023, R01AG059013, R56AG060156, AG061008) funded by the National Institute on Aging (NIA) and by the National Center for Advancing Translational Sciences, National Institutes of Health, through grant number UL1TR001873.

Conflict of Interest

None declared.

Author Contributions

Conceptualization: Y.G.; Data acquisition: N.S., R.P.M., Y.G; Data analysis: J.G.; Data interpretation: J.G., Y.G.; Drafting manuscript: J.G.; Revising manuscript: N.S., E.C., Y.S., R.P.M., Y.G.; Approving final version of the manuscript: all authors.