Abstract
Background
Frailty is a dynamic condition in which individuals may deteriorate, recover, remain stable, or die. Although healthier dietary patterns have been associated with a lower risk of frailty, less is known about their associations with specific frailty-state transitions. We examined whether a study-specific score emphasizing protein density and carbohydrate quality was associated with transitions between frailty states.
Methods
We linked the 2013 Health and Retirement Study nutrition assessment to repeated frailty assessments from 2014 to 2021 among participants aged 50 years or older. The study-specific score combined higher densities of protein, fiber, and whole-grain densities with lower densities of added-sugar and glycemic-load densities. Frailty was assessed using a published 28-item deficit-accumulation index. The primary continuous-time multistate model allowed baseline transition intensities to change at 2 and 4 years and adjusted for age, sex, education, body mass index (BMI), ethnicity, Hispanic ethnicity, smoking, alcohol consumption, and moderate and light physical activity. Dietary components were also examined individually and jointly.
Results
The primary analysis included 6,982 participants aged 50–102 years (mean, 69.0 years; 60.3% aged 65 years or older), with 25,108 person-state records and 1,487 deaths. Including the score improved model fit (likelihood-ratio [LR], 24.08; degrees of freedom [df], 7; p = 0.001). Each 1-standard deviation (SD) higher score was associated with lower robust-to-pre-frail intensity (hazard ratio [HR], 0.933; nominal 95% confidence interval [CI], 0.875–0.995) and pre-frail-to-death intensity (HR, 0.809; 95% CI, 0.662–0.989). Confidence intervals included 1 for the remaining worsening, recovery, and death transitions. No individual component association remained statistically significant after mutual adjustment and false-discovery-rate correction.
Conclusion
After comprehensive adjustment, a higher study-specific score was associated with lower robust-to-pre-frail and pre-frail-to-death intensities, but evidence for other transitions was inconclusive. The score remains unvalidated, and these findings do not establish a causal effect of diet on frailty transitions.
Keywords: carbohydrate quality, continuous-time multistate Markov model, frailty, health and retirement study, nutrition, protein density
Introduction
Frailty reflects diminished physiologic reserve and heightened vulnerability to stressors beyond that attributable to chronological age, individual diseases, or disability alone (1, 2). Two complementary conceptual approaches have shaped the field. The physical phenotype emphasizes weakness, slowness, exhaustion, low activity, and weight loss, whereas deficit-accumulation indices quantify health deficits across multiple systems (3, 4). In longitudinal population studies, the deficit-accumulation approach provides a graded measure of vulnerability that can be repeatedly assessed over time.
Frailty is clinically and epidemiologically important not only because it predicts adverse outcomes, but also because its severity may change over time. Although greater frailty burden predicts mortality and other adverse outcomes across cohorts and meta-analyses, frailty does not invariably progress: older adults may deteriorate, recover, remain stable, or die between assessments (5–10). A transition-based perspective is particularly relevant to nutritional epidemiology because associations with deterioration, recovery, and death may differ across these distinct outcomes (9, 11, 12).
Diet is a modifiable factor that may shape frailty trajectories. Diet quality is inherently multidimensional. Established indices, including the Mediterranean, dietary approaches to stop hypertension (DASH), and Healthy Eating Index measures, integrate multiple foods, nutrients, and adherence targets rather than relying on a single dietary component. Prospective studies and meta-analyses have linked Mediterranean-style and other healthier dietary patterns, protein intake, fruit and vegetable consumption, and nutrient adequacy to frailty-related outcomes (13–20). However, most studies have examined prevalent or incident frailty at subsequent assessment points, making it difficult to determine whether diet is associated with specific transitions, such as slower deterioration, greater recovery, or lower mortality.
We therefore focused on a targeted dietary construct rather than overall diet quality. Protein density may support muscle maintenance, whereas fiber, whole grains, added sugars, and glycemic load capture complementary aspects of carbohydrate quality and glycemic burden (21–29). A composite score can be used to assess whether this specific dietary profile is associated with frailty dynamics, but it is not equivalent to a validated Mediterranean, DASH, or Healthy Eating Index score.
We used the Health and Retirement Study (HRS), a longitudinal study of aging in the United States. We examined whether a study-specific protein-density and carbohydrate-quality score was associated with frailty-state transitions over up to 7.1 years (30, 31). Frailty states were defined using a 28-item deficit-accumulation frailty index constructed according to published HRS methods (32). We hypothesized that higher scores would be associated with lower transition intensities for frailty worsening and death and higher transition intensities for recovery.
Methods
Study design and analytic sample
This secondary longitudinal cohort analysis used data from the HRS. We linked the 2013 nutrition assessment to frailty assessments at waves 12–15, corresponding nominally to 2014, 2016, 2018, and 2020. Actual interview dates ranged from March 2014 to May 2021. Because HRS may include spouses or partners younger than the age-eligible respondents, eligibility was restricted to participants aged 50 years or older at wave 12 (30, 31).
Of 8,035 nutrition records, 8,021 were successfully linked to RAND HRS data. We excluded 445 participants with missing baseline age and 142 participants aged younger than 50 years, leaving 7,434 age-eligible participants. We then sequentially excluded 3 participants without a valid wave 12 frailty state, 124 without complete core covariate data, and 246 with fewer than two valid frailty-state observations. An additional 79 participants with missing values for covariates in the fully adjusted model were excluded. The primary complete-case sample comprised 6,982 participants (Figure 1).
Figure 1.
Sequential participant selection for the fully adjusted primary analysis. FI, frailty index; HRS, Health and Retirement Study.
The primary dataset contained 25,108 person-state records and 1,487 deaths. Participants’ baseline ages ranged from 50 to 102 years; 4,213 participants (60.3%) were aged 65 years or older. The maximum follow-up was 7.13 years. Death was modeled as an absorbing state.
Dietary exposure
The primary exposure was a study-specific, hypothesis-driven score of protein density and carbohydrate quality, derived from the 2013 HRS nutrition assessment. This score represents a targeted dietary profile rather than a validated instrument or a comprehensive measure of overall diet quality. The five components were selected from prior evidence linking protein adequacy and carbohydrate quality to physical and metabolic reserve (21–29).
Protein, fiber, whole-grain, added-sugar, and glycemic-load variables were expressed per 1,000 kcal. Each component was standardized across all 8,035 nutrition records. The equally weighted score was calculated as the mean of the z scores for protein, fiber, and whole-grain intake and the reverse-coded z scores for added sugar intake and glycemic load. All five components were complete. Component weights were independent of frailty outcomes. The composite was standardized using the mean and SD among the 7,061 participants eligible for core-adjusted analyses; one SD corresponded to 0.616 score units. This scale was used in all model comparisons. Quartiles and prediction percentiles were calculated among the 6,982 participants in the primary sample (Supplementary material (Methods); Supplementary Table S1).
Outcome definition: frailty index and frailty states
Frailty was assessed using a 28-item deficit-accumulation frailty index constructed from HRS variables according to published methods (32). The same multidomain deficits were available consistently across the four waves, whereas complete measures required to construct Fried phenotype performance were unavailable for every participant at every wave. The index covered chronic disease, memory-related disease, sensory impairment, general health, activities of daily living, mobility, depressive symptoms, and cognition. Because it includes ADL and IADL limitations, the index captures multidomain vulnerability that partially overlaps with disability. Supplementary Table S2 provides detailed coding definitions.
At each wave, the frailty index was calculated as the sum of observed deficits divided by the number of observed items. Calculation required at least 22 of 28 observed items, thereby allowing up to six missing items; otherwise, the index was set to missing. Supplementary Table S3 reports wave-specific availability and distributions.
Following the published HRS frailty-index implementation, we defined three living frailty states (32). Participants were classified as robust for values below 0.10, pre-frail for values from 0.10 to below 0.25, and frail for values of 0.25 or higher. Death constituted the fourth, absorbing state. Although categorization results in some loss of continuous information, it defines a finite and clinically interpretable state space for deterioration, recovery, and death.
Continuous-time multistate Markov model
We fitted continuous-time multistate Markov models using the msm package in R. The allowed transitions were robust to pre-frail, pre-frail to robust, pre-frail to frail, frail to pre-frail, and transitions from each living state to death.
Direct transitions between robust and frail states were not parameterized. Accordingly, observed non-adjacent changes between biennial visits were treated as interval changes that could have occurred through the unobserved intermediate state.
Follow-up was measured in years from each participant’s wave-12 interview. Interview dates were obtained from RAND monthly midpoint variables. Deaths were modeled as exact events using the recorded RAND death dates. The primary model allowed baseline transition intensities to vary across three follow-up periods: 0 to <2, 2 to <4, and at least 4 years. Dietary associations were estimated separately for each transition and assumed to remain constant across the three follow-up periods.
The primary model adjusted for age, sex, education, baseline BMI, ethnicity, Hispanic ethnicity, current smoking, alcohol consumption, and moderate and light physical activity. These covariates were included as potential demographic, socioeconomic, and behavioral confounders, with separate coefficients for each allowed transition. Physical-activity frequencies were modeled as standardized ordinal variables. Core-adjusted models included age, sex, education, and baseline BMI. Core-adjusted and time-homogeneous models were fitted to the same complete-case sample for comparison. A sensitivity analysis omitted BMI because it may mediate long-term dietary associations. Models were conditional on the initial frailty state; continuous baseline FI was not included as an additional covariate.
We estimated state-occupation probabilities at 6 years for the 10th and 90th percentiles of the dietary profile score. Other covariates were fixed at their participant-level sample means, including mean indicators for categorical variables. Piecewise-model probabilities were calculated from the time-ordered product of transition matrices across the three periods. These predictions describe a reference covariate profile; they are neither population-standardized estimates nor estimates of causal effects. Supplementary Tables S4, S5 describe participant selection, ascertainment of study dates, and model diagnostics.
Model checking and sensitivity analyses
Likelihood-ratio tests compared nested models fitted to the same participants and observations. We compared piecewise and time-homogeneous models under full adjustment, and assessed models with continuous or quartile dietary scores against the corresponding model without the dietary score. Comparisons of Akaike information criterion (AIC) values also used the same sample. Transition-specific 95% CIs for the composite score were nominal and were not adjusted for multiplicity. Numerical convergence was assessed using optimizer status, Hessian and gradient diagnostics, and finite confidence intervals.
Sensitivity analyses used time-homogeneous or core-adjusted models, required at least 25 of 28 observed FI items, omitted BMI, or applied delayed entry from year 2. For delayed entry, follow-up began at the first available living-state assessment at or after year 2, with at least one subsequent state observation required. We also examined each dietary component separately and all five components jointly, using the primary model’s time periods and covariate adjustment. Component HRs were expressed per 1-SD higher intake density in the primary sample; added sugar and glycemic load were not reverse-coded. Benjamini-Hochberg correction was applied to 35 component-transition tests within each exploratory family: the separate-component models and the mutually adjusted model.
Data preparation and analysis used R and Python. Continuous-time models were fitted in R version 4.6.0 using msm version 1.8.2. Hazard ratios for transition intensities are reported with 95% confidence intervals and interpreted as associations rather than causal effects.
Results
Analytic sample and baseline characteristics
The primary sample included 6,982 participants, 25,108 person-state records, and 1,487 deaths. Mean baseline age was 69.00 years (SD, 10.49); 58.8% were female and 60.3% were aged 65 years or older. At baseline, 2,196 participants were robust (31.5%), 2,664 were pre-frail (38.2%), and 2,122 were frail (30.4%; Figure 1).
Baseline characteristics differed across dietary profile score quartiles. Mean FI was 0.17 in Q4 and 0.23 in Q1. Table 1 summarizes baseline characteristics; Supplementary Table S6 provides detailed demographic and lifestyle characteristics, including all covariates in the primary model.
Table 1.
Baseline characteristics by quartile of dietary profile score.
| Characteristic | Overall | Q1 | Q2 | Q3 | Q4 | p |
|---|---|---|---|---|---|---|
| Participants | 6,982 | 1,746 | 1,745 | 1,745 | 1,746 | |
| Dietary profile score | 0.00 (0.62) | −0.80 (0.42) | −0.14 (0.11) | 0.22 (0.10) | 0.73 (0.28) | <0.001 |
| Age, years | 69.00 (10.49) | 68.89 (10.84) | 69.83 (10.61) | 69.40 (10.30) | 67.89 (10.10) | <0.001 |
| Female | 4,103 (58.8%) | 937 (53.7%) | 1,010 (57.9%) | 1,060 (60.7%) | 1,096 (62.8%) | <0.001 |
| Education, years | 12.89 (3.03) | 12.30 (2.78) | 12.83 (2.88) | 13.26 (2.91) | 13.19 (3.42) | <0.001 |
| BMI, kg/m2 | 28.60 (6.08) | 28.44 (6.30) | 28.58 (5.96) | 28.76 (6.01) | 28.60 (6.05) | 0.493 |
| Current smoking | 771 (11.0%) | 324 (18.6%) | 209 (12.0%) | 127 (7.3%) | 111 (6.4%) | <0.001 |
| Frailty index | 0.20 (0.16) | 0.23 (0.18) | 0.21 (0.16) | 0.19 (0.15) | 0.17 (0.15) | <0.001 |
| Robust | 2,196 (31.5%) | 460 (26.3%) | 506 (29.0%) | 566 (32.4%) | 664 (38.0%) | <0.001 |
| Pre-frail | 2,664 (38.2%) | 624 (35.7%) | 675 (38.7%) | 693 (39.7%) | 672 (38.5%) | |
| Frail | 2,122 (30.4%) | 662 (37.9%) | 564 (32.3%) | 486 (27.9%) | 410 (23.5%) | |
| Death during follow-up | 1,487 (21.3%) | 472 (27.0%) | 411 (23.6%) | 349 (20.0%) | 255 (14.6%) | <0.001 |
Observed frailty-state transitions
Both deterioration and recovery were observed between successive assessments. Table 2 presents the observed state changes, including unchanged states and deaths, in the primary sample.
Table 2.
Observed frailty-state transitions between HRS visits.
| From | To | n | Row % | Model representation |
|---|---|---|---|---|
| robust | robust | 4,002 | 73.7 | |
| robust | pre-frail | 1,213 | 22.3 | |
| robust | frail | 90 | 1.7 | Latent intermediate state |
| robust | death | 126 | 2.3 | |
| pre-frail | robust | 858 | 12.3 | |
| pre-frail | pre-frail | 4,473 | 63.9 | |
| pre-frail | frail | 1,302 | 18.6 | |
| pre-frail | death | 371 | 5.3 | |
| frail | robust | 20 | 0.4 | Latent intermediate state |
| frail | pre-frail | 806 | 14.2 | |
| frail | frail | 3,875 | 68.1 | |
| frail | death | 990 | 17.4 |
The model represented non-adjacent living-state changes as passage through the unobserved intermediate state between visits (Table 2; Supplementary Table S7).
Dietary profile score in the primary piecewise multistate model
In the primary piecewise model, each 1-SD higher dietary profile score was associated with lower robust-to-pre-frail intensity (HR, 0.933; 95% CI, 0.875–0.995). Higher scores were also associated with lower pre-frail-to-death intensity (HR, 0.809; 95% CI, 0.662–0.989; Figure 2; Table 3).
Figure 2.
Transition-specific hazard ratios per 1-SD higher dietary profile score in the fully adjusted piecewise continuous-time multistate model. Points show HRs, and error bars show 95% confidence intervals.
Table 3.
Primary fully adjusted piecewise model estimates per 1-SD higher dietary profile score.
| Transition | HR | 95% CI |
|---|---|---|
| Robust to pre-frail | 0.933 | 0.875–0.995 |
| Robust to death | 0.848 | 0.625–1.150 |
| Pre-frail to robust | 1.024 | 0.950–1.105 |
| Pre-frail to frail | 0.978 | 0.922–1.038 |
| Pre-frail to death | 0.809 | 0.662–0.989 |
| Frail to pre-frail | 1.050 | 0.979–1.127 |
| Frail to death | 0.939 | 0.880–1.002 |
The nominal 95% CIs included 1 for pre-frail-to-frail progression, both recovery pathways, and transitions from the robust and frail states to death, indicating inconclusive evidence regarding these transitions.
Predicted 6-year state probabilities
For an initially robust reference profile, the high-score scenario had a 6-year probability of robust status 6.7 percentage points higher, a probability of frail status 2.3 percentage points lower, and a probability of death 3.0 percentage points lower than the low-score scenario.
For an initially pre-frail reference profile, the high-score scenario had a probability of robust status 2.9 percentage points higher and a probability of death 3.7 percentage points lower. For an initially frail profile, the probability of pre-frail status was 3.4 percentage points higher and the probability of death was 4.3 percentage points lower. These are descriptive, model-based contrasts rather than intervention effects (Figure 3; Table 4).
Figure 3.
Six-year state-occupation probabilities from the fully adjusted piecewise model for low and high dietary scores (10th and 90th percentiles). Other covariates are fixed at their participant-level means. Connected points compare the two score profiles and do not denote confidence intervals.
Table 4.
Six-year predicted state-occupation probabilities from the fully adjusted piecewise model.
| Initial state | State at 6 years | Low score | High score | Difference, percentage points |
|---|---|---|---|---|
| robust | robust | 31.9% | 38.6% | +6.7 |
| robust | pre-frail | 41.6% | 40.2% | −1.4 |
| robust | frail | 17.2% | 15.0% | −2.3 |
| robust | death | 9.3% | 6.2% | −3.0 |
| pre-frail | robust | 13.6% | 16.5% | +2.9 |
| pre-frail | pre-frail | 43.0% | 45.1% | +2.1 |
| pre-frail | frail | 29.1% | 27.9% | −1.2 |
| pre-frail | death | 14.3% | 10.6% | −3.7 |
| frail | robust | 5.0% | 6.4% | +1.4 |
| frail | pre-frail | 25.2% | 28.6% | +3.4 |
| frail | frail | 44.9% | 44.3% | −0.6 |
| frail | death | 25.0% | 20.7% | −4.3 |
Dietary profile specification and model fit
Table 5 reports Q4 vs. Q1 estimates under the same fully adjusted piecewise specification. Using the same sample, the fully adjusted piecewise model fitted better than the time-homogeneous model (LR, 354.02; df, 14; p < 0.001). Corresponding AIC values were 31,763.53 and 32,089.56. Including the continuous score improved fit relative to the piecewise model without the dietary score (LR, 24.08; df, 7; p = 0.001). The quartile comparison yielded LR 32.31 (df, 21; p = 0.054). Table 6 and Supplementary Table S8 provide the model comparisons. Supplementary Table S9 reports estimates and confidence intervals for all quartiles.
Table 5.
Highest vs. lowest quartile of the dietary profile score.
| Transition (Q4 vs. Q1) | HR | 95% CI | p |
|---|---|---|---|
| Robust to pre-frail | 0.807 | 0.680–0.958 | 0.014 |
| Robust to death | 0.527 | 0.213–1.303 | 0.166 |
| Pre-frail to robust | 1.085 | 0.883–1.335 | 0.437 |
| Pre-frail to frail | 0.905 | 0.766–1.070 | 0.244 |
| Pre-frail to death | 0.693 | 0.401–1.197 | 0.189 |
| Frail to pre-frail | 1.142 | 0.926–1.408 | 0.214 |
| Frail to death | 0.844 | 0.693–1.028 | 0.091 |
Table 6.
Global model comparison tests.
| Comparison | LR | df | p |
|---|---|---|---|
| Continuous score vs. score omitted | 24.08 | 7 | 0.001 |
| Score quartiles vs. score omitted | 32.31 | 21 | 0.054 |
| Piecewise vs. time-homogeneous, full adjustment | 354.02 | 14 | <0.001 |
Sensitivity and component-specific analyses
With stricter FI completeness, nominal CIs remained below 1 for the robust-to-pre-frail and pre-frail-to-death transitions. When BMI was omitted, CIs for these two associations included 1, whereas the frail-to-death CI excluded 1 (Table 7). Supplementary Table S9 also presents core-adjusted and time-homogeneous estimates for the same complete-case sample.
Table 7.
Sensitivity analyses of transition-specific dietary profile associations.
| Transition | Primary piecewise | Time-homogeneous | FI > = 25 items | Entry from year 2 | BMI omitted |
|---|---|---|---|---|---|
| Robust to pre-frail | 0.933 (0.875–0.995) | 0.934 (0.876–0.996) | 0.935 (0.876–0.997) | 0.940 (0.859–1.028) | 0.939 (0.880–1.000) |
| Robust to death | 0.848 (0.625–1.150) | 0.856 (0.617–1.187) | 0.847 (0.629–1.141) | 0.849 (0.563–1.281) | 0.847 (0.616–1.164) |
| Pre-frail to robust | 1.024 (0.950–1.105) | 1.025 (0.950–1.105) | 1.030 (0.954–1.111) | 1.039 (0.940–1.148) | 1.009 (0.935–1.088) |
| Pre-frail to frail | 0.978 (0.922–1.038) | 0.976 (0.920–1.035) | 0.985 (0.928–1.045) | 1.024 (0.945–1.110) | 0.980 (0.924–1.040) |
| Pre-frail to death | 0.809 (0.662–0.989) | 0.831 (0.670–1.030) | 0.798 (0.654–0.973) | 0.787 (0.602–1.029) | 0.827 (0.676–1.012) |
| Frail to pre-frail | 1.050 (0.979–1.127) | 1.049 (0.978–1.126) | 1.055 (0.983–1.132) | 1.059 (0.964–1.164) | 1.039 (0.968–1.114) |
| Frail to death | 0.939 (0.880–1.002) | 0.944 (0.885–1.008) | 0.942 (0.882–1.005) | 0.939 (0.867–1.018) | 0.920 (0.862–0.982) |
Except for the homogeneous comparator, the models allow piecewise-constant baseline intensities and retain full adjustment unless explicitly stated.
The delayed-entry analysis included 5,920 participants and 932 deaths. Estimates were less precise, and all seven transition-specific CIs included 1. Overall, the associations were not consistently supported across sensitivity analyses.
In separate component models, only the association between higher fiber density and lower robust-to-pre-frail intensity remained statistically significant after false-discovery-rate correction (Supplementary Table S10). Joint inclusion of the five components improved model fit (LR, 96.94; df, 35; p < 0.001). However, no individual component association remained statistically significant after mutual adjustment and false-discovery-rate correction (Supplementary Table S11). Supplementary Table S12 presents period-specific baseline intensities, and Supplementary Table S13 reports component correlations and variance inflation factors.
Discussion
Among adults aged 50 years or older, higher dietary profile scores were associated with lower robust-to-pre-frail and pre-frail-to-death intensities in the fully adjusted piecewise model. Evidence for the remaining progression, recovery, and death transitions was inconclusive. Several associations weakened with broader covariate adjustment, and the delayed-entry analysis did not provide consistent support for the primary findings.
This study extends prior work by treating frailty as a dynamic state process rather than a single endpoint. Previous studies have shown that frailty may improve or worsen and that these transitions carry prognostic information (9–11). By separating deterioration, recovery, and death, our analysis shows that the strength and robustness of dietary associations differ across transition pathways.
Prior studies have linked Mediterranean-style and other healthier dietary patterns to frailty risk (13–15, 18–20), and have examined protein intake and diet quality in relation to frailty outcomes (16, 17, 23). Our analysis examines a targeted dietary profile and distinguishes associations with deterioration, recovery, and death. Differences in exposure definitions, covariate adjustment, and frailty measurement limit direct comparisons with studies of incident frailty or established diet-quality indices.
The score represents a targeted dietary profile rather than a comprehensive measure of diet quality. Protein density relates to muscle maintenance (27–29), whereas the other components capture aspects of carbohydrate quality and glycemic burden (24, 25). Although joint inclusion of the components improved model fit, no individual component association remained statistically significant after mutual adjustment and false-discovery-rate correction. These findings do not establish which components account for the composite association, nor do they demonstrate synergy. External validation and comparison with established dietary indices remain necessary.
Deficit-accumulation indices and multistate models are well suited to repeated measures of aging. The frailty index provides a multidomain measure with transparent coding and missingness rules (3, 4). The continuous-time model estimates transition-specific intensities, accommodates irregular spacing between interviews, and treats death as an absorbing state (33–36). The modeled states therefore represent multidomain vulnerability rather than a purely physical frailty phenotype.
Strengths include repeated assessments in a large longitudinal cohort, a published 28-item frailty index, and explicit modeling of death as an absorbing state. Interview midpoint and recorded death dates defined follow-up intervals. The primary piecewise model accommodated time-varying baseline transition intensities.
Several limitations warrant consideration. The observational design precludes causal inference, and residual confounding remains possible. Baseline BMI and physical activity may reflect pre-existing frailty or consequences of earlier diet; the estimates are conditional associations rather than total causal effects. Diet was measured only once, in 2013, leaving dietary change, measurement error, and reverse causation unresolved. Delayed-entry analyses were conditional on survival and continued observation and do not exclude reverse causation. The equally weighted score is study-specific and unvalidated, and the component analyses do not establish its external validity. Only 60.3% of participants were aged 65 years or older, so the cohort represents middle-aged and older adults rather than an exclusively geriatric population. Muscle failure and sarcopenia were not evaluated. Because these conditions may precede or contribute to frailty, we could not determine whether dietary associations were independent of prior changes in muscle health (37). The FI includes ADL and IADL limitations and therefore partly overlaps with disability. Panel observations also require inference regarding unobserved transitions. The piecewise model relaxes time homogeneity but retains the Markov assumption and assumes constant transition-specific dietary HRs within each period. Complete-case selection and the absence of survey weighting may further limit generalizability.
In conclusion, higher dietary profile scores were associated with lower robust-to-pre-frail and pre-frail-to-death transition intensities, but these associations were not consistently supported across sensitivity analyses. Evidence for other transitions was inconclusive. Replication using validated, repeated dietary measures and direct assessments of muscle health is needed before considering implications for dietary interventions.
Acknowledgments
I thank the Gateway to Global Aging Data for providing harmonized data information. I also appreciate the work done by staff and investigators from the Health and Retirement Study, and I appreciate the participation of respondents in these studies.
Funding Statement
The author(s) declared that financial support was not received for this work and/or its publication.
Footnotes
Edited by: Hélio José Coelho Júnior, Catholic University of the Sacred Heart, Italy
Reviewed by: Miriam Teresa López Teros, National Institute of Medical Sciences and Nutrition Salvador Zubirán, Mexico
Lincheng Duan, Chengdu University of Traditional Chinese Medicine, China
Data availability statement
The original contributions presented in the study are included in the article/Supplementary material, further inquiries can be directed to the corresponding author.
Ethics statement
The studies involving humans were approved by the Institutional Review Board at the University of Michigan and the National Institute on Aging (HUM00061128). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.
Author contributions
JW: Formal analysis, Writing – review & editing, Writing – original draft, Supervision, Data curation, Conceptualization, Methodology.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that Generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnut.2026.1926576/full#supplementary-material
References
- 1.Clegg A, Young J, Iliffe S, Rikkert MO, Rockwood K. Frailty in elderly people. Lancet. (2013) 381:752–62. doi: 10.1016/S0140-6736(12)62167-9, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Fried LP, Tangen CM, Walston J, Newman AB, Hirsch C, Gottdiener J, et al. Frailty in older adults: evidence for a phenotype. J Gerontol A Biol Sci Med Sci. (2001) 56:M146–56. doi: 10.1093/gerona/56.3.m146, [DOI] [PubMed] [Google Scholar]
- 3.Rockwood K, Mitnitski A. Frailty in relation to the accumulation of deficits. J Gerontol A Biol Sci Med Sci. (2007) 62:722–7. doi: 10.1093/gerona/62.7.722, [DOI] [PubMed] [Google Scholar]
- 4.Searle SD, Mitnitski A, Gahbauer EA, Gill TM, Rockwood K. A standard procedure for creating a frailty index. BMC Geriatr. (2008) 8:24. doi: 10.1186/1471-2318-8-24, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Song X, Mitnitski A, Rockwood K. Prevalence and 10-year outcomes of frailty in older adults in relation to deficit accumulation. J Am Geriatr Soc. (2010) 58:681–7. doi: 10.1111/j.1532-5415.2010.02764.x, [DOI] [PubMed] [Google Scholar]
- 6.Kojima G, Iliffe S, Walters K. Frailty index as a predictor of mortality: a systematic review and meta-analysis. Age Ageing. (2017) 47:193–200. doi: 10.1093/ageing/afx162 [DOI] [PubMed] [Google Scholar]
- 7.Vermeiren S, Vella-Azzopardi R, Beckwée D, Habbig A-K, Scafoglieri A, Jansen B, et al. Frailty and the prediction of negative health outcomes: a meta-analysis. J Am Med Dir Assoc. (2016) 17:1163.e1–1163.e17. doi: 10.1016/j.jamda.2016.09.010 [DOI] [PubMed] [Google Scholar]
- 8.Hoogendijk EO, Afilalo J, Ensrud KE, Kowal P, Onder G, Fried LP. Frailty: implications for clinical practice and public health. Lancet. (2019) 394:1365–75. doi: 10.1016/S0140-6736(19)31786-6, [DOI] [PubMed] [Google Scholar]
- 9.Gill TM, Gahbauer EA, Allore HG, Han L. Transitions between frailty states among community-living older persons. Arch Intern Med. (2006) 166:418–23. doi: 10.1001/archinte.166.4.418, [DOI] [PubMed] [Google Scholar]
- 10.Xue Q-L. The frailty syndrome: definition and natural history. Clin Geriatr Med. (2011) 27:1–15. doi: 10.1016/j.cger.2010.08.009, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Cesari M, Gambassi G, van Kan GA, Vellas B. The frailty phenotype and the frailty index: different instruments for different purposes. Age Ageing. (2013) 43:10–2. doi: 10.1093/ageing/aft160, [DOI] [PubMed] [Google Scholar]
- 12.Hanlon P, Nicholl BI, Jani BD, Lee D, McQueenie R, Mair FS. Frailty and pre-frailty in middle-aged and older adults and its association with multimorbidity and mortality: a prospective analysis of 493 737 UK biobank participants. Lancet Public Health. (2018) 3:e323–32. doi: 10.1016/S2468-2667(18)30091-4, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Talegawkar SA, Bandinelli S, Bandeen-Roche K, Chen P, Milaneschi Y, Tanaka T, et al. A higher adherence to a Mediterranean-style diet is inversely associated with the development of frailty in community-dwelling elderly men and women. J Nutr. (2012) 142:2161–6. doi: 10.3945/jn.112.165498, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.León-Muñoz LM, García-Esquinas E, López-García E, Banegas JR, Rodríguez-Artalejo F. Major dietary patterns and risk of frailty in older adults: a prospective cohort study. BMC Med. (2015) 13:11. doi: 10.1186/s12916-014-0255-6, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Kojima G, Avgerinou C, Iliffe S, Walters K. Adherence to Mediterranean diet reduces incident frailty risk: systematic review and Meta-analysis. J Am Geriatr Soc. (2018) 66:783–8. doi: 10.1111/jgs.15251, [DOI] [PubMed] [Google Scholar]
- 16.Beasley JM, LaCroix AZ, Neuhouser ML, Huang Y, Tinker L, Woods N, et al. Protein intake and incident frailty in the women’s health initiative observational study. J Am Geriatr Soc. (2010) 58:1063–71. doi: 10.1111/j.1532-5415.2010.02866.x, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Shikany JM, Barrett-Connor E, Ensrud KE, Cawthon PM, Lewis CE, Dam T-TL, et al. Macronutrients, diet quality, and frailty in older men. J Gerontol A Biol Sci Med Sci. (2013) 69:695–701. doi: 10.1093/gerona/glt196, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Chan R, Leung J, Woo J. Dietary patterns and risk of frailty in Chinese community-dwelling older people in Hong Kong: a prospective cohort study. Nutrients. (2015) 7:7070–84. doi: 10.3390/nu7085326, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Lorenzo-López L, Maseda A, de Labra C, Regueiro-Folgueira L, Rodríguez-Villamil JL, Millán-Calenti JC. Nutritional determinants of frailty in older adults: a systematic review. BMC Geriatr. (2017) 17:108. doi: 10.1186/s12877-017-0496-2, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Ni Lochlainn M, Cox NJ, Wilson T, Hayhoe RPG, Ramsay SE, Granic A, et al. Nutrition and frailty: opportunities for prevention and treatment. Nutrients. (2021) 13:2349. doi: 10.3390/nu13072349, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.García-Esquinas E, Rahi B, Peres K, Colpo M, Dartigues J-F, Bandinelli S, et al. Consumption of fruit and vegetables and risk of frailty: a dose-response analysis of 3 prospective cohorts of community-dwelling older adults. Am J Clin Nutr. (2016) 104:132–42. doi: 10.3945/ajcn.115.125781, [DOI] [PubMed] [Google Scholar]
- 22.Sandoval-Insausti H, Blanco-Rojo R, Graciani A, López-García E, Moreno-Franco B, Laclaustra M, et al. Ultra-processed food consumption and incident frailty: a prospective cohort study of older adults. J Gerontol A Biol Sci Med Sci. (2019) 75:1126–33. doi: 10.1093/gerona/glz140, [DOI] [PubMed] [Google Scholar]
- 23.Struijk EA, Fung TT, Rodríguez-Artalejo F, Bischoff-Ferrari HA, Hu FB, Willett WC, et al. Protein intake and risk of frailty among older women in the nurses’ health study. J Cachexia Sarcopenia Muscle. (2022) 13:1752–61. doi: 10.1002/jcsm.12972, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Reynolds A, Mann J, Cummings J, Winter N, Mete E, Te Morenga L. Carbohydrate quality and human health: a series of systematic reviews and meta-analyses. Lancet. (2019) 393:434–45. doi: 10.1016/S0140-6736(18)31809-9, [DOI] [PubMed] [Google Scholar]
- 25.Ludwig DS. The glycemic index: physiological mechanisms relating to obesity, diabetes, and cardiovascular disease. JAMA. (2002) 287:2414. doi: 10.1001/jama.287.18.2414, [DOI] [PubMed] [Google Scholar]
- 26.Micha R, Peñalvo JL, Cudhea F, Imamura F, Rehm CD, Mozaffarian D. Association between dietary factors and mortality from heart disease, stroke, and type 2 diabetes in the United States. JAMA. (2017) 317:912–24. doi: 10.1001/jama.2017.0947, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Volkert D, Beck AM, Cederholm T, Cruz-Jentoft A, Goisser S, Hooper L, et al. ESPEN guideline on clinical nutrition and hydration in geriatrics. Clin Nutr. (2019) 38:10–47. doi: 10.1016/j.clnu.2018.05.024, [DOI] [PubMed] [Google Scholar]
- 28.Bauer J, Biolo G, Cederholm T, Cesari M, Cruz-Jentoft AJ, Morley JE, et al. Evidence-based recommendations for optimal dietary protein intake in older people: a position paper from the PROT-AGE study group. J Am Med Dir Assoc. (2013) 14:542–59. doi: 10.1016/j.jamda.2013.05.021, [DOI] [PubMed] [Google Scholar]
- 29.Cruz-Jentoft AJ, Bahat G, Bauer J, Boirie Y, Bruyère O, Cederholm T, et al. Sarcopenia: revised european consensus on definition and diagnosis. Age Ageing. (2018) 48:16–31. doi: 10.1093/ageing/afy169, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Sonnega A, Faul J, Ofstedal M, Langa K, Phillips J, Weir D. Cohort profile: the health and retirement study (HRS). Int J Epidemiol. (2014) 43:576–85. doi: 10.1093/ije/dyu067, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Fisher GG, Ryan LH. Overview of the health and retirement study and introduction to the special issue. Work Aging Retire. (2017) 4:1–9. doi: 10.1093/workar/wax032, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Wang H, Sun R, Yan B, Wei Y, Wu Y, Ling F, et al. Digital exclusion and frailty among older adults: findings from four longitudinal cohort studies. Age Ageing. (2026) 55:afag133. doi: 10.1093/ageing/afag133, [DOI] [PubMed] [Google Scholar]
- 33.Jackson C. Multi-state models for panel data: the msm package for R. J Stat Softw. (2011) 38:1–29. doi: 10.18637/jss.v038.i08 [DOI] [Google Scholar]
- 34.Putter H, Fiocco M, Geskus R. Tutorial in biostatistics: competing risks and multi-state models. Stat Med. (2006) 26:2389–430. doi: 10.1002/sim.2712, [DOI] [PubMed] [Google Scholar]
- 35.Andersen PK, Keiding N. Multi-state models for event history analysis. Stat Methods Med Res. (2002) 11:91–115. doi: 10.1191/0962280202SM276ra [DOI] [PubMed] [Google Scholar]
- 36.Meira-Machado L, de Uña-Alvarez J, Cadarso-Suárez C, Andersen PK. Multi-state models for the analysis of time-to-event data. Stat Methods Med Res. (2008) 18:195–222. doi: 10.1177/0962280208092301, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Coelho-Júnior HJ, Álvarez-Bustos A, Rodríguez-Sánchez I, Landi F, Marzetti E. Muscle failure and risk of adverse outcomes in older adults: a derivation and multicohort validation study. Lancet Healthy Longev. (2026) 7:100843. doi: 10.1016/j.lanhl.2026.100843, [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The original contributions presented in the study are included in the article/Supplementary material, further inquiries can be directed to the corresponding author.