Biomarker-Calibrated Protein Intake and Physical Function
The WHI Observational Study (OS) of 93,676 women, along with 68,132 women enrolled in the Clinical Trials (CT), contained women aged 50 to 79 year when recruited between 1993 and 1998 from 40 clinical centers across the United States. More than 80% of WHI CT and OS women also enrolled in an extension study (WHI Extension) to ascertain additional health outcomes beginning in 2005. Women were eligible for study inclusion if they were postmenopausal, of stable residence, and unlikely to die within 3 years. Additional eligibility criteria were assigned to each CT for reasons of safety, competing risk, adherence, and retention. Further details regarding the design, recruitment strategy, and data collection methods have been published. Institutional review boards at each participating institution reviewed and approved the study, and all participants provided written, informed consent.
The study population for this analysis included all WHI OS and CT women with at least one physical function measure (self-reported or objective). Women must also have completed a food frequency questionnaire (FFQ) at baseline with reported energy intake between 600 and 5,000 kcal/d and have had available information necessary to compute calibrated protein intake: age, race and ethnicity, body mass index (BMI), recreational physical activity, and smoking status (N = 134,961). Self-reported physical function scores were available for at least two visits for 75.0% of participants and three visits for 70.2%.
Information on objective physical function measurements was available for a subset of the study population. A 25% random sample of CT participants aged 65 and older completed measures of grip strength (n = 5,331), chair stands (n = 5,294), and timed walk (n = 5,335), for a total of 5,346 women completing at least one of the three measures at one or more time points. For the subset of CT women who completed physical performance measures, data were available for at least two visits for 94.8%, 93.6%, and 95.1% and three visits for 84.0%, 81.4%, and 84.6% of participants for grip strength, chair stands, and timed walk, respectively. The subset of women completing physical performance measures were generally older, were more likely to be obese, had lower income and education, were more likely to live alone, and reported more comorbidities (e.g., arthritis, diabetes mellitus) than the overall study population (data not shown).
Self-reported physical function was assessed using a summary measure—a subscale from the RAND Short Form-36 (SF-36)—that includes 10 items related to ability to engage in activities (vigorous and moderate activities, lifting and carrying groceries, climbing stairs, bending, walking, and bathing and dressing oneself) and is scored from 0 to 100, with a higher score indicating greater physical function capacity. The SF-36 was assessed at baseline and annually beginning in 2005 for all WHI participants and at closeout for CT participants (average ~7 years after baseline). For this analysis, women were followed for a mean 11.5 ± 3.1 years.
Trained, certified study staff assessed physical performance measures (grip strength, chair stands, timed walk) at baseline, at follow-up Years 1, 3, and 6 in the CT, and at follow-up Year 3 in the OS using standard protocols. Hand grip strength was measured using a handheld dynamometer (Jamar hand dynamometer; Lafayette Instruments, Lafayette, IN). Two measurements were taken using the dominant hand with staff coaching for maximal performance, and the higher score was used in the analysis.
Repeated chair stands and the timed walk, which represent two of three items of the Short Physical Performance Battery (SPPB) commonly employed to assess function in aging populations, correlate with disability and mortality in older adults. The chair-stand test was conducted if the participant was able to stand at least once without using hands or arms from a straight-backed, nonpadded, flat-seated, armless chair. Two 15-second trials of repeated chair stands were performed with arms folded across the chest, with a 1- to 2-minute rest between trials, and the score with the greater number of chair stands was included in the analysis. The 6-m timed walk was performed with women walking at usual speed, using ambulatory aids as needed. The test was repeated, and the faster of the two measured times was included in this analysis.
The 122-item WHI FFQ was self-administered at baseline for all participants. Daily energy (kcal) and protein (g) intake were estimated for individual foods and food groups, 19 adjustment items, and summary questions (uncalibrated energy and protein). The WHI Nutritional Biomarkers Study (NBS; n = 544) was conducted to evaluate accuracy of self-reported energy and protein consumption from the FFQ using biomarkers (doubly labeled water for energy and urinary nitrogen for protein) of true intake.
Uncalibrated protein intake refers to the amount calculated directly on the FFQ, whereas biomarker-calibrated protein intake refers to the adjusted levels obtained after using regression equations developed based upon the NBS. These regression equations incorporate participant characteristics such as BMI, age, race and ethnicity, and smoking status. The biomarker-calibrated values are more accurate than the uncalibrated values, because they account for differential reporting errors in dietary data based on participant characteristics. Protein intake measures were categorized into quintiles separately for the self-reported (median calibrated protein intake of 14.3% energy) and objective physical function (median calibrated protein intake of 13.7% energy) measures based on distributions among the women included in each analysis.
Demographic characteristics (age, family income, education, race and ethnicity), medical history (hip fracture, emphysema, treated diabetes mellitus, hypertension (taking medication or blood pressure >140/90 mmHg), arthritis, cancer), and other health-related factors (having a current healthcare provider, number of falls, living alone, ability to perform activities of daily living, depression) were self-reported at baseline. BMI was computed using measured height (m) and weight (kg) at baseline (kg/m). Smoking status was self-reported and classified as current, past, or never. Depressive symptoms were assessed using the six-item short form of the Center for Epidemiologic Studies Depression (CESD) Scale24. Postmenopausal hormone therapy was self-reported and categorized as current, past, or never used, with separate indicator variables for unopposed estrogen and estrogen plus progestin.
Baseline characteristics were examined according to quintile of calibrated protein intake (% energy). Multivariate adjusted linear generalized estimating equations (GEEs) with an exchangeable covariance matrix were used to examine the longitudinal association between baseline protein intake and each outcome (self-reported physical function and three physical performance measures), including a protein-by-time interaction to allow for varying slopes, with time modeled as a continuous variable. The models test whether the mean scores on these outcome measures differ at baseline (intercept) or with respect to annual change over time (slope) according to quintile of protein intake. The reasonableness of linear fit was confirmed by comparing these estimates with results obtained by treating time as a categorical variable. Models were adjusted for previously identified independent predictors of frailty: age, income, education, race and ethnicity, BMI, smoking status, alcohol consumption, physical activity, hormone therapy use, whether the participant lived alone, having a healthcare provider, number of falls, disability, depression, and self-reported history of medical conditions (emphysema, diabetes mellitus, hypertension, arthritis, and cancer). Models were also adjusted for calibrated total energy intake and clinical trial arm. To evaluate potential effect modification, the relationship between calibrated protein intake and each of the four outcome measures was stratified according to median age, recreational physical activity, BMI, or protein source (animal:total protein ratio).
Bootstrapping procedures were conducted to approximate variance around the estimated physical function outcomes for levels of calibrated protein intake. First, variance in the coefficients for the calibration regression equations in the NBS was determined by constructing 1,000 bootstrap replicates using their original dataset. Next, using the data set for the current study, 1,000 bootstrap replicates were constructed, and each new randomly selected sample of participants was paired with a new set of regression coefficients from the first bootstrapping procedure to recalculate calibrated protein and energy intake. Then, the previously described GEE models were repeated for each sample, and 95% confidence intervals were determined from the distribution of 1,000 new estimates. Tolerance (relative change in the coefficient vector from one iteration to the next) was set to 10 during bootstrapping to ensure consistent model convergence. A trend in the relationship between protein quintile (ordinal) and each estimated physical function score at baseline (or annualized change) was assessed by calculating the slope between these variables for each bootstrap replicate. These slopes were used to construct a distribution to determine whether there was a significant trend in physical function across levels of protein intake. Statistical analyses were conducted using Stata 12.1 (Stata Corp, College Station, TX) and SAS 9.3 (SAS Institute, Inc., Cary, NC); all reported P-values are two-sided.
Methods
Study Population
The WHI Observational Study (OS) of 93,676 women, along with 68,132 women enrolled in the Clinical Trials (CT), contained women aged 50 to 79 year when recruited between 1993 and 1998 from 40 clinical centers across the United States. More than 80% of WHI CT and OS women also enrolled in an extension study (WHI Extension) to ascertain additional health outcomes beginning in 2005. Women were eligible for study inclusion if they were postmenopausal, of stable residence, and unlikely to die within 3 years. Additional eligibility criteria were assigned to each CT for reasons of safety, competing risk, adherence, and retention. Further details regarding the design, recruitment strategy, and data collection methods have been published. Institutional review boards at each participating institution reviewed and approved the study, and all participants provided written, informed consent.
The study population for this analysis included all WHI OS and CT women with at least one physical function measure (self-reported or objective). Women must also have completed a food frequency questionnaire (FFQ) at baseline with reported energy intake between 600 and 5,000 kcal/d and have had available information necessary to compute calibrated protein intake: age, race and ethnicity, body mass index (BMI), recreational physical activity, and smoking status (N = 134,961). Self-reported physical function scores were available for at least two visits for 75.0% of participants and three visits for 70.2%.
Information on objective physical function measurements was available for a subset of the study population. A 25% random sample of CT participants aged 65 and older completed measures of grip strength (n = 5,331), chair stands (n = 5,294), and timed walk (n = 5,335), for a total of 5,346 women completing at least one of the three measures at one or more time points. For the subset of CT women who completed physical performance measures, data were available for at least two visits for 94.8%, 93.6%, and 95.1% and three visits for 84.0%, 81.4%, and 84.6% of participants for grip strength, chair stands, and timed walk, respectively. The subset of women completing physical performance measures were generally older, were more likely to be obese, had lower income and education, were more likely to live alone, and reported more comorbidities (e.g., arthritis, diabetes mellitus) than the overall study population (data not shown).
Physical Function Measures
Self-reported physical function was assessed using a summary measure—a subscale from the RAND Short Form-36 (SF-36)—that includes 10 items related to ability to engage in activities (vigorous and moderate activities, lifting and carrying groceries, climbing stairs, bending, walking, and bathing and dressing oneself) and is scored from 0 to 100, with a higher score indicating greater physical function capacity. The SF-36 was assessed at baseline and annually beginning in 2005 for all WHI participants and at closeout for CT participants (average ~7 years after baseline). For this analysis, women were followed for a mean 11.5 ± 3.1 years.
Trained, certified study staff assessed physical performance measures (grip strength, chair stands, timed walk) at baseline, at follow-up Years 1, 3, and 6 in the CT, and at follow-up Year 3 in the OS using standard protocols. Hand grip strength was measured using a handheld dynamometer (Jamar hand dynamometer; Lafayette Instruments, Lafayette, IN). Two measurements were taken using the dominant hand with staff coaching for maximal performance, and the higher score was used in the analysis.
Repeated chair stands and the timed walk, which represent two of three items of the Short Physical Performance Battery (SPPB) commonly employed to assess function in aging populations, correlate with disability and mortality in older adults. The chair-stand test was conducted if the participant was able to stand at least once without using hands or arms from a straight-backed, nonpadded, flat-seated, armless chair. Two 15-second trials of repeated chair stands were performed with arms folded across the chest, with a 1- to 2-minute rest between trials, and the score with the greater number of chair stands was included in the analysis. The 6-m timed walk was performed with women walking at usual speed, using ambulatory aids as needed. The test was repeated, and the faster of the two measured times was included in this analysis.
Dietary Protein
The 122-item WHI FFQ was self-administered at baseline for all participants. Daily energy (kcal) and protein (g) intake were estimated for individual foods and food groups, 19 adjustment items, and summary questions (uncalibrated energy and protein). The WHI Nutritional Biomarkers Study (NBS; n = 544) was conducted to evaluate accuracy of self-reported energy and protein consumption from the FFQ using biomarkers (doubly labeled water for energy and urinary nitrogen for protein) of true intake.
Uncalibrated protein intake refers to the amount calculated directly on the FFQ, whereas biomarker-calibrated protein intake refers to the adjusted levels obtained after using regression equations developed based upon the NBS. These regression equations incorporate participant characteristics such as BMI, age, race and ethnicity, and smoking status. The biomarker-calibrated values are more accurate than the uncalibrated values, because they account for differential reporting errors in dietary data based on participant characteristics. Protein intake measures were categorized into quintiles separately for the self-reported (median calibrated protein intake of 14.3% energy) and objective physical function (median calibrated protein intake of 13.7% energy) measures based on distributions among the women included in each analysis.
Potential Confounders
Demographic characteristics (age, family income, education, race and ethnicity), medical history (hip fracture, emphysema, treated diabetes mellitus, hypertension (taking medication or blood pressure >140/90 mmHg), arthritis, cancer), and other health-related factors (having a current healthcare provider, number of falls, living alone, ability to perform activities of daily living, depression) were self-reported at baseline. BMI was computed using measured height (m) and weight (kg) at baseline (kg/m). Smoking status was self-reported and classified as current, past, or never. Depressive symptoms were assessed using the six-item short form of the Center for Epidemiologic Studies Depression (CESD) Scale24. Postmenopausal hormone therapy was self-reported and categorized as current, past, or never used, with separate indicator variables for unopposed estrogen and estrogen plus progestin.
Analytic Approach
Baseline characteristics were examined according to quintile of calibrated protein intake (% energy). Multivariate adjusted linear generalized estimating equations (GEEs) with an exchangeable covariance matrix were used to examine the longitudinal association between baseline protein intake and each outcome (self-reported physical function and three physical performance measures), including a protein-by-time interaction to allow for varying slopes, with time modeled as a continuous variable. The models test whether the mean scores on these outcome measures differ at baseline (intercept) or with respect to annual change over time (slope) according to quintile of protein intake. The reasonableness of linear fit was confirmed by comparing these estimates with results obtained by treating time as a categorical variable. Models were adjusted for previously identified independent predictors of frailty: age, income, education, race and ethnicity, BMI, smoking status, alcohol consumption, physical activity, hormone therapy use, whether the participant lived alone, having a healthcare provider, number of falls, disability, depression, and self-reported history of medical conditions (emphysema, diabetes mellitus, hypertension, arthritis, and cancer). Models were also adjusted for calibrated total energy intake and clinical trial arm. To evaluate potential effect modification, the relationship between calibrated protein intake and each of the four outcome measures was stratified according to median age, recreational physical activity, BMI, or protein source (animal:total protein ratio).
Bootstrapping procedures were conducted to approximate variance around the estimated physical function outcomes for levels of calibrated protein intake. First, variance in the coefficients for the calibration regression equations in the NBS was determined by constructing 1,000 bootstrap replicates using their original dataset. Next, using the data set for the current study, 1,000 bootstrap replicates were constructed, and each new randomly selected sample of participants was paired with a new set of regression coefficients from the first bootstrapping procedure to recalculate calibrated protein and energy intake. Then, the previously described GEE models were repeated for each sample, and 95% confidence intervals were determined from the distribution of 1,000 new estimates. Tolerance (relative change in the coefficient vector from one iteration to the next) was set to 10 during bootstrapping to ensure consistent model convergence. A trend in the relationship between protein quintile (ordinal) and each estimated physical function score at baseline (or annualized change) was assessed by calculating the slope between these variables for each bootstrap replicate. These slopes were used to construct a distribution to determine whether there was a significant trend in physical function across levels of protein intake. Statistical analyses were conducted using Stata 12.1 (Stata Corp, College Station, TX) and SAS 9.3 (SAS Institute, Inc., Cary, NC); all reported P-values are two-sided.
Source...