**3. Results**

#### *3.1. Sample Characteristics*

Among 1597 participants who answered the dietary survey at baseline, 107 were excluded from all analyses for the following reasons: 20 were ADL dependent at baseline, 67 could not be classified for frailty, 9 had missing information about DP consumption and 11 participants had missing information for covariates. Therefore, the final sample for the cross-sectional analysis comprised 1490 participants (including 1427 non-frail). Among those participants, 979 (69%) were followed up at 10 years (during the follow-up, 355 participants died). An additional 156 participants were excluded from longitudinal analyses (*n* = 79 participants were identified as dependent and *n* = 77 with missing frailty status at 10 years). Thus, 823 participants were prospectively analyzed (Figure 1).

**Figure 1.** Flow chart of the cross-sectional and the prospective studied samples. Three-City Bordeaux Study, 2000–2010.

In cross-sectional analyses, the studied sample (*n* = 1490) constituted mainly of females (*n* = 906, 60.8%) and had an average age of 74.1 ± 4.9 (standard deviation) years (Table 1). Half the sample was married (57%) and reported multimorbidity (48%), and a third (32%) was taking 6 medications/d or more. Prevalence of frailty was 4.2% (*n* = 63). The most prevalent frailty criterion was low physical activity (*n* = 234, 20.1%) followed by slow walking speed (*n* = 281, 19%) while the least prevalent frailty criterion was muscle weakness (*n* = 77, 5.3%) followed by weight loss (*n* = 82, 5.5%). Those included in prospective analyses (*n* = 823) were non-frail participants at baseline, mainly females (65.0%) and were on average 72.8 ± 4.4 years old. A total of 150 participants (18.2%) exhibited frailty at the 10-year follow-up and the most incident frailty criterion was low physical activity (*n* = 473, 58.2%) followed by muscle weakness (*n* = 199, 26.6%). The least incident frailty criterion was weight loss (*n* = 69, 8.4%) followed by exhaustion (*n* = 138, 17.6%).

**Table 1.** Socio-demographic, clinical, dietary, and lifestyle characteristics according to the frailty prevalence (cross-sectional sample, *n* = 1490 in 2000) and incidence (prospective sample, *n* = 823 between 2000–2010) of older adults from the Three-City study, Bordeaux (France).


\* All data are presented as *n* (%) except for age, MMSE, BMI, and protein intake where the mean (SD) is presented; a Baseline differences between prevalent frail and non-frail (*n* = 1490) and incident frail and non-frail (*n* = 823) participants tested by *t*-tests or chi square tests depending on the type of the variable; b Mini-Mental State Examination; m = missing.

> Prevalent and incident frail older adults were significantly older, were more likely to be depressed, to take 6 medications/day or more, and to have comorbidities at baseline compared with prevalent non-frail participants, and with participants free from frailty over time, respectively (Table 1). Moreover, prevalent frail participants exhibited a significantly higher BMI on average than non-frail participants, and the daily consumption of proteins was not significantly different between the frail and non-frail participants at baseline (i.e., cross-sectional sample). Regarding the sample enrolled in prospective analyses, incident frail participants had a similar BMI compared to those who remained free from frailty, while a higher percentage of incident frail participants had a lower diet quality score compared with participants who remained free from frailty (53% vs. 44%, *p* = 0.045).

Frequencies of consumption of DP (total DP, milk, fresh DP, and cheese) are presented in Table 2 for both cross-sectional and prospective samples. No significant differences were observed between prevalent frail and non-frail or incident frail and non-frail participants regarding the frequency of total DP and DP-subtypes consumption at baseline.

**Table 2.** Frequency of consumption of dairy products (total and sub-types) according to the frailty prevalence (cross-sectional sample, *n* = 1490 in 2000) and incidence (prospective sample, *n* = 823 between 2000–2010) of older adults from the Three-City study, Bordeaux (France).


All data are presented as *n* (%); a Baseline differences between prevalent frail and non-frail (*n* = 1490) and incident frail and non-frail (*n* = 823) participants tested by *t*-tests or chi square tests depending on the type of the variable; m = missing.

#### *3.2. Associations between Spectrum of DP Exposure and Prevalence of Frailty*

In models adjusted for age, sex, marital status and education, we did not observe any significant association between total DP and DP sub-types and frailty prevalence, when comparing the highest frequency to the lowest frequency consumption of DP (Table 3). In models additionally adjusted for smoking status, multimorbidity, polypharmacy, protein intake, diet quality and global cognitive score, all associations with the prevalence of frailty remained not significant for all DP exposures: total DP (OR = 1.08, 95% CI = 0.54–2.17 and 1.40, 95% CI = 0.65–3.01 for intermediate and high consumption vs. low, respectively), milk (OR = 1.13, 95% CI = 0.56–2.31), fresh DP (OR = 1.13, 95% CI= 0.54–2.33), and cheese (OR = 0.89; 95% CI = 0.43–1.88) for high vs. low frequency of consumption.

#### *3.3. Associations between Spectrum of DP Exposure and Incidence of Frailty*

When focusing on the 10-year risk for frailty, we observed that baseline frequencies of consumption of total DP and DP sub-types were not significantly associated with the frailty risk when we compared the lowest frequency to the highest frequency of consumption of total DP (OR = 0.74, 95% CI = 0.42–1.30), milk (OR = 0.80, 95% CI = 0.48–1.35), fresh DP (OR = 0.68, 95% CI = 0.38–1.20) and cheese consumption (OR = 1.19, 05% CI = 0.68–2.10) in fully adjusted models (Table 4).

#### *3.4. Sensitivity Analyses*

The FRAIL scale was also implemented to alternatively identify prevalent and incident frail participants. Sixty out of 1552 participants (3.9%) were considered as frail at baseline according to this scale. In fully adjusted models (i.e., model 2), all associations with the prevalence of frailty were not significant for all DP exposures: total DP (OR = 1.42, 95% CI = 0.64–3.13), milk (OR= 1.11, 95% CI = 0.54–2.32), fresh DP (OR = 0.96, 95% CI= 0.46–1.98), and cheese (OR = 0.86; 95% CI = 0.39–1.88) for high vs. low frequency of consumption. Among 1492 non-frail non-dependent, 1006 were followed at

the 10-year follow-up (lost to follow-up *n* = 486). Among those, an additional 87 were excluded from the analysis because they were ADL dependent and another 23 were excluded because they were unclassified for the FRAIL scale, leading to a final sample size of 896, with 45 (5.0%) classified as frail on the FRAIL scale. Regarding the spectrum of DP exposures, we only observed that the highest, compared with the lowest, frequency of consumption of fresh DP was associated with lower frailty risk in the fully adjusted model (OR = 0.35, 95% CI = 0.13–0.97, *p* = 0.04, *p* global = 0.13) while all other associations were non-significant, for instance, total DP (OR = 1.66, 95% CI = 0.26–1.67), milk (OR= 1.21, 95% CI = 0.49–2.98), and cheese (OR = 1.25; 95% CI = 0.49–3.21) for high vs. low frequency of consumption.

Second, when all ADL dependent individuals were maintained in analytic samples, 1501 and 885 participants were included for the cross-sectional and prospective analyses, respectively. Among those 1501 participants, 68 (4.5%) were identified as frail. None of the total DP or DP sub-types exposures were associated with frailty prevalence in the fully adjusted models: total DP (OR = 1.44, 95% CI = 0.68–3.04), milk (OR= 1.04, 95% CI = 0.52–2.1), fresh DP (OR = 1.21, 95% CI= 0.59–2.48), and cheese (OR = 0.85; 95% CI = 0.42–1.74) for high vs. low frequency of consumption. Among those 1433 non-frail at baseline, 449 were lost at the 10-year follow-up and 99 were unclassified for frailty incidence, leading to a final sample size of 885 for the prospective analyses. At the 10-year follow-up, 195 participants (22%) were identified as frail. Regarding the spectrum of DP exposures, none of the frequency of consumption of total DP or DP sub-types were significantly associated with frailty risk in the fully adjusted models: total DP (OR = 0.65, 95% CI = 0.39–1.09), milk (OR = 0.61, 95% CI = 0.38–1.00), fresh DP (OR = 0.64, 95% CI = 0.39–1.08), and cheese (OR = 1.4; 95% CI = 0.83–2.41) for high vs. low frequency of consumption.

**Table 3.** Multivariate association between baseline frequencies of consumption of total DP, milk, fresh DP, and cheese and frailty prevalence among older adults in the Three-City Study, Bordeaux (*n* = 1490, 2000).


OR: Odds ratio, CI: Confidence Intervals; Model 1: Model adjusted for age, sex, marital status and education; Model 2: Model 1 + additional adjustment for smoking status, multimorbidity, polypharmacy, total protein, diet quality score, and Mini-Mental State Examination.

**Table 4.** Multivariate association between baseline frequencies of consumption of total DP, milk, fresh DP, and cheese and the 10-year frailty risk among older adults in the Three-City Study, Bordeaux (*n* = 823, 2000–2010).


OR: Odds ratio, CI: Confidence Intervals; Model 1: Model adjusted for age, sex, marital status and education; Model 2: Model 1 + additional adjustment for smoking status, multimorbidity, polypharmacy, total protein, diet quality score, and Mini-Mental State Examination.
