The authors state that they have no conflicts of interest.
The association between bone and renal function in healthy seniors is not well studied. In this cross-sectional and longitudinal study in 1713 older men and women, creatinine clearance was significantly associated with hip BMD. If confirmed, this may warrant adding mild to moderate renal dysfunction as an indication for osteoporosis screening.
Introduction: This study determined the cross-sectional and longitudinal association between measures of renal function and BMD, bone loss, and osteoporotic fracture in older adults. It determined which measure of renal function'creatinine clearance by the Cockcroft-Gault (CG) equation, estimated glomerular filtration rate by the Modification of Diet in Renal Disease (MDRD) equation, or serum creatinine'is most strongly associated with BMD and osteoporotic fracture.
Materials and Methods: This was a cross-sectional and prospective study in older community-dwelling men and women. Between 1992 and 1995, 1713 participants (average age, 71.3 ± 11.1 years) completed standardized questionnaires, physical examinations, laboratory testing, and bone densitometry; 1023 participants returned for a follow-up visit in 1997–1999, an average of 4.1 ± 0.9 years later.
Results: Calculated renal function declined with age (p < 0.001). Renal function was categorized by Kidney Disease Outcomes Quality Initiative (K/DOQI) chronic kidney disease (CKD) stage. By the CG equation, at baseline, 5.5% of participants had stage 1 CKD (glomerular filtration rate ≤ 90 ml/min/1.73 m2), 43.0% had stage 2 CKD (60–89 ml/min/1.73 m2), 48.8% had stage 3 CKD (30–59 ml/min/1.73 m2), and 2.7% had stages 4 and 5 CKD (<30 ml/min/1.73 m2). Using the MDRD equation, these percents were 7.0%, 61.7%, 30.9%, and 0.5%, respectively. In cross-sectional analyses, there was a significant linear association between creatinine clearance by CG or glomerular filtration rate by MDRD and hip BMD. In prospective analyses, there was an average annual bone loss of 0.6% and a significant association between baseline CG and 4-year hip bone loss. There was no association between baseline MDRD or serum creatinine and bone loss. At baseline, 180 of 1713 participants (11%) reported at least one clinical fracture of the hip, femur, forearm, or wrist; 79 (8%) reported new clinical fractures during follow-up. Baseline renal function by any measure was not significantly associated with prevalent or incident clinical fractures.
Conclusions: Although renal function measured by both CG and MDRD was associated with BMD in cross-sectional analyses, only creatinine clearance by CG predicted 4-year bone loss. If confirmed, this should be the preferred method for assessing the association between renal function and BMD. Cross-sectional associations between renal function and BMD were strongest at higher CKD stage. None of the baseline renal function estimates was associated with prevalent or incident fractures, perhaps reflecting the multifactorial etiology of fractures beyond BMD. If further studies in the elderly confirm renal function as an important predictor of bone loss and fracture, this may warrant adding mild to moderate renal dysfunction as an indication for osteoporosis screening.
It is well established that patients with end-stage renal disease (ESRD) are at increased risk of osteopenia and hip fracture.(1,2) Hip fracture in dialysis patients is associated with increased mortality.(3) Renal function declines with age. Approximately 20% of U.S. adults ≤65 years of age have moderately or severely decreased renal function (chronic kidney disease [CKD] stage ≤ 3).(4) As the glomerular filtration rate (GFR) falls below 60 ml/min/1.73 m2, phosphate is retained, inducing a rise in PTH and a decline in 1,25-dihydroxyvitamin D levels; both changes negatively affect BMD.(5)
Although the association between renal function and bone in the ESRD population has been studied extensively, there are little data in community-dwelling seniors with subclinical renal dysfunction. Most prior studies of renal function and bone have been small, cross-sectional, or have not focused on the elderly; these studies have shown conflicting results.(6–12) Only two have assessed fractures as an outcome.(13,14) Most used the Cockcroft-Gault (CG) equation, which includes age, sex, weight, and serum creatinine, to estimate renal function.(15) However, some studies have questioned the use of CG in the elderly(16–19) and in women, because 96% of study subjects were men.(15) Another prediction equation including age, sex, race, and serum creatinine was developed using data from the Modification of Diet in Renal Disease (MDRD) study(20); the MDRD equation is thought by some to be a more accurate estimate of actual GFR.(20–24) However, because the MDRD study population consisted almost entirely of dialysis patients, others have criticized the MDRD equation for underestimating GFR in healthy people.(19,25,26) The validity of MDRD in community-dwelling seniors is unknown, but one recent study has documented a clear discrepancy between the CG and MDRD equations in predicting GFR in the elderly.(27)
Subclinical renal dysfunction is common and is often not detected by routine clinical chemistry such as blood urea nitrogen (BUN) or creatinine. Although WHO guidelines recommend screening with DXA in all women >65 years of age, there are no age only criteria for men, and screening is recommended only in those with risk factors. The Kidney Disease Outcomes Quality Initiative (K/DOQI) guidelines similarly recommend screening with DXA in those with known fractures or other risk factors, but recognize that bone disease caused by secondary hyperparathyroidism begins with stage 3 CKD(28); renal disease of this degree or greater is estimated to be present in >20% of the U.S. population.(4) If subclinical renal disease is associated with increased bone loss, appropriate measures of renal function would have important implications for identifying older men at increased risk of osteoporosis, for whom BMD testing and intervention would be appropriate.
We tested the hypothesis that renal function in community-dwelling seniors is associated with BMD and prevalent fracture and will predict bone loss and future fracture. We also hypothesized that the MDRD equation would be more strongly associated with BMD and fracture than the CG equation, based on current expert opinion that the former is a more accurate estimate of actual renal function.
MATERIALS AND METHODS
The Rancho Bernardo Study, a cohort of white, middle to upper middle class, community-dwelling adults in southern California, was established in 1972. Between 1992 and 1995, −80% (n = 1778) of all surviving locally dwelling residents attended a clinic visit designed to study chronic diseases including osteoporosis; 1713 had both blood tests necessary to estimate renal function and BMD measured and were included in this study. Seventy-seven percent of surviving community-dwelling participants returned to the research clinic in 1997–1999 (n = 1096), when BMD and fracture history were reassessed; 1021 had both blood tests necessary to estimate renal function and BMD measured. The University of California, San Diego institutional review board approved the research protocol; all participants gave written informed consent.
At baseline, participants completed standardized questionnaires about medical history including current medication and dietary supplement use, which was validated by examination of pills and prescriptions brought to the clinic for that purpose. Information about cigarette smoking (yes/no), alcohol consumption (number of drinks per day during the last 2 weeks), and exercise three or more times per week (yes/no) was assessed using standard questionnaires. History of reported clinical fractures was validated in a subset of 36%, with validation of 71% by chart and X-ray review.
Height and weight were measured using a calibrated stadiometer and balance beam scale with participants wearing light clothing and no shoes. Blood pressure was measured twice in seated resting subjects using the Hypertension Detection and Follow-up Program protocol.(29) Body mass index (BMI) was calculated as kilograms per square meter. Baseline laboratory tests on morning fasting blood included glucose, serum electrolytes, BUN, and serum creatinine. Diabetes was defined (yes/no) by patient report of physician diagnosis of diabetes and by WHO 1999 criteria.(30) Although information about falls was not obtained at this visit, tandem gait (number of heel to toe steps) and speed of gait (normal, moderately slow, or very slow) by clinic nurse observation were recorded.
Serum creatinine was measured, at both visits in the same commercial laboratory, using a variation of the Jaffe enzymatic method on a Hitachi 911 analyzer (Roche Diagnostics, Indianapolis, IN, USA), with inter- and intra-assay CVs of 4.0%. CKD stage was categorized based on K/DOQI guidelines: stage 1, normal or increased GFR (≤90 ml/min/1.73 m2); stage 2, mild decreased GFR (60–89 ml/min/1.73 m2); stage 3, moderately decreased GFR (30–59 ml/min/1.73 m2); stage 4, severely decreased GFR (15–29 ml/min/1.73 m2); stage 5, kidney failure (<15 ml/min/ 1.73 m2).(21)
BMD (g/cm2) at the hip was measured by licensed trained technicians using DXA and the same Hologic 1000. Daily calibration was used to maintain the manufacturer's precision standards of <1.5% for the total hip. Osteoporosis was defined by the WHO criterion of a T score < −2.5.(31)
Urine samples were collected and stored at −70°C until analysis. Urinary cross-linked N-telopeptide of type I collagen (NTX) levels were measured for 355 men and 553 women at the baseline visit only, using an ELISA (Ostex International). For normalization of urinary NTX, urinary creatinine was measured by an enzymatic assay (Ostex International). The assay was calibrated using standard amounts of human bone collagen digested with bacterial collagenase, and the results were expressed as nanomoles of bone collagen equivalents per millimoles of creatinine (nmolBCE/mmol CR).(32)
Creatinine clearance was calculated using the CG equation:
GFR was calculated using the abbreviated MDRD study equation(20,21):
where serum creatinine is serum creatinine concentration in milligrams per deciliter, age is in years, weight is in kilograms, and BSA is body surface area, estimated using the Dubois equation(33):
where height is in meters and weight in kilograms.
For the longitudinal analyses, annual percent bone change was calculated as:
Fracture prevalence by renal function was compared using the χ2 statistic. Differences in mean renal function or BMD from the baseline to follow-up visits were compared using the paired t-test. Linear regression was used to test separately the association of creatinine clearance by CG, GFR by MDRD, and serum creatinine with BMD or annual percent change in BMD. Logistic regression was used to generate odds ratios to test the association between creatinine clearance, GFR, and serum creatinine and clinical fractures. ANOVA was used to study the association between renal function and measures of gait. Statistical tests were two-tailed; statistical significance was defined as p < 0.05. SPSS (SPSS Base 11.5 for Windows) was used for analysis. Analyses using creatinine clearance by CG were not adjusted for age, sex, or BMI because measures of each of these variables are included in the CG equation. Similarly, analyses using GFR by MDRD were adjusted for BMI but not for age or sex because both of these variables are included in the MDRD equation.
At the baseline visit, the 1713 participants (678 men and 1035 women) were 40–97 years of age. Table 1 shows the baseline characteristics for all participants. More than 80% were >60 years of age and two thirds were women. Most were not overweight, exercised regularly, and did not smoke. Their average systolic blood pressure (SBP) was 136.6 ± 22.2 mmHg. Mean baseline creatinine clearance by CG was 60.4 ± 17.8 ml/min/1.73 m2. Mean baseline GFR by MDRD equation was 67.7 ± 16.2 ml/min/1.73 m2. Mean serum creatinine was 1.0 ± 0.25 mg/dl. All measures of body composition, BMD, and renal function were normally distributed; urine NTX was skewed, and the log transformation of the variable, which was normally distributed, was used for all analyses.
Table Table 1.. Baseline Characteristics of 1713 Participants of the Rancho Bernardo Study (1992–1995)
Figure 1 shows the baseline prevalence of CKD stage by the CG and MDRD equations. By CG, 5.5% of participants were CKD stage 1, 43.0% were CKD stage 2, 48.8% were CKD stage 3, and 2.7% were stage 4 or 5. Using the MDRD equation, these numbers were 7.0%, 61.7%, 30.9%, and 0.5%, respectively.
Figure 2 shows that calculated renal function (creatinine clearance by CG or GFR by MDRD) significantly decreased and serum creatinine significantly increased with age (p < 0.001). The rate of decline in creatinine clearance by CG was significantly steeper than the estimated decline of GFR by MDRD (p < 0.001).
At baseline, mean total hip BMD at the total hip was 0.864 ± 0.164 g/cm2. Cross-sectional analyses are shown in Table 2. Table 2 shows linear regression models of total hip BMD by measures of renal function, adjusted for SBP, diabetes, exercise, smoking, and alcohol and variably adjusted for age, sex, and BMI. There was a significant linear association between creatinine clearance by CG or GFR by MDRD and hip BMD at the same visit, but not between serum creatinine and BMD. Figures 3 and 4 show that, in stratified cross-sectional analyses, this association persisted among those with CKD stages 2 and greater, but not in those with stage 1 CKD, and was strongest in those with CKD stage ≤3. When estimated by MDRD, only renal function in CKD stage ≤3 was significantly associated with BMD.
Table Table 2.. Linear Regression of Hip BMD and Renal Function in 1713 Participants of the Rancho Bernardo Study (1992–1995)
In prospective analyses, the mean BMD among those who had BMD measured at both visits was 0.876 ± 0.156 g/cm2 at the baseline visit and 0.856 ± 0.163 g/cm2 at the follow-up visit (p < 0.001); creatinine clearance by CG was 62.1 ± 17.0 ml/min/1.73 m2 at baseline and 62.7 ± 19.6 ml/min/1.73 m2 at follow-up (p = 0.11), GFR by MDRD was 68.7 ± 16.0 ml/min/1.73 m2 at baseline and 73.5 ± 19.0 ml/min/1.73 m2 at follow-up (p < 0.001), and serum creatinine was 1.00 ± 0.22 mg/dl at baseline and 0.96 ± 0.36 mg/dl at follow-up (p < 0.001). The mean annual percent change in total hip BMD, measured an average of 4.1 ± 0.9 years apart, was −0.62 ± 1.23%/year. As shown in Table 3, there was a significant (p < 0.01) association between baseline creatinine clearance by CG and annual percent bone change, but neither baseline GFR by MDRD nor baseline serum creatinine was associated with change in BMD. In analyses stratified by CKD stage (Table 4), there was an association between baseline CKD stages 1 or ≤3 by CG and baseline CKD stages 1 or ≤3 by MDRD and annual percent bone change, but there was an unexplained significant, negative association between baseline renal function and annual percent bone change in those with more normal renal function (stage 1) (i.e., as renal function increased in this subgroup, annual percent bone change declined).
Table Table 3.. Linear Regression of Annual Percent Change in Hip BMD and Baseline Renal Function in 1021 Participants of the Rancho Bernardo Study (1992–1999)
Table Table 4.. Linear Regression of Annual Percent Change in Hip BMD and Renal Function by K/DOQI CKD Stage in 1021 Participants of the Rancho Bernardo Study (1992–1999)
To further study a possible association between renal function and biochemical markers of bone turnover, additional cross-sectional analyses were performed to look for an association between all three measures of renal function and urine NTX or serum alkaline phosphatase. These analyses showed a pattern consistent with that seen with hip BMD, with a significant inverse association between creatinine clearance by CG and both urine NTX or serum alkaline phosphatase; in stratified analyses, this association was significant only in those with CKD stage ≤3. GFR by MDRD was significantly, inversely associated with serum alkaline phosphatase (but not urine NTX), and in analyses stratified by CKD stage, only in those with CKD stage ≤3 (data not shown).
Thirty-seven percent of participants reported current use of calcium supplements and 15% reported current use of vitamin D supplements at the baseline visit. Calcium and vitamin D supplementation was not associated with renal function by the CG or MDRD equation, but both were inversely associated with serum creatinine (age-adjusted β = −0.184, p < 0.001 for calcium, β = −0.095, p < 0.001 for vitamin D). Both calcium and vitamin D supplementations were also inversely associated with hip BMD in age-adjusted analyses (β = −0.188, p < 0.001 for calcium, β = −0.117, p < 0.001 for vitamin D). However, adjusting for baseline calcium or vitamin D supplementation did not materially change results of cross-sectional and longitudinal analyses exploring the association between renal function and BMD.
At baseline, 180 of 1713 participants (11%) reported at least one clinical fracture of the hip, femur, forearm, or wrist. Four years later, 79 of 1021 participants (8%) reported 83 new clinical fractures. We compared the prevalence of fracture at the baseline visit among all those who came to the baseline visit and in only those who also returned to the follow-up visit; there was no significant difference in prevalence (p = 0.30). There was no significant association between any baseline measure of renal function and the odds of prevalent or incident clinical fracture at all sites combined or at the hip, forearm, or wrist. Cross-sectional analyses showed a significant association between all three measures of baseline renal function and measures of gait. Mean renal function improved as the number of steps the participant was able to take increased (p < 0.001) or as speed of gait category increased from very slow to normal (p < 0.001).
When all of the above analyses were repeated using femoral neck BMD instead of total hip BMD, results were not materially changed. Analyses were also repeated after excluding persons with diabetes (by history or WHO criteria(30)); again results were not materially changed. Adding serum calcium, use of thiazide diuretics, or current estrogen replacement (for women) to multivariable analyses did not change these results (data not shown).
To our knowledge, this is the first large prospective study to assess the association of renal function assessed by three different measures (CG, MDRD, and serum creatinine) with BMD, bone loss, and fracture risk. In this cohort, renal function assessed either as creatinine clearance by CG or GFR by MDRD was associated with hip BMD in cross-sectional analyses, but only creatinine clearance by CG was consistently associated with annual percent bone loss assessed longitudinally. Baseline renal function assessed by any method was not associated with prevalent or incident clinical fractures.
In these older, community-dwelling adults, 52% had CKD stage ≤3 by CG and 32% by MDRD, higher than the 20% reported for other U.S. studies, made up of younger subjects.(4,34) Most older individuals have normal serum creatinine levels despite loss of renal function(35,36); consequently, impaired renal function is underdiagnosed. This disconnect between serum creatinine and renal function reflects the decline in muscle mass and altered creatinine metabolism that occurs with aging.(35,37) Measurement of creatinine clearance using a 24-h urine sample has been thought to be a superior marker for impaired kidney function; however, collection of a 24-h urine specimen is impractical in primary care patients and may be unsuccessful in older adults who have residual urine, incontinence, and forgetfulness.(36) Because of the effect of secreted creatinine, creatinine clearance by 24-h collection is particularly poor at low levels of glomerular function, with values 20% higher than those obtained using direct techniques such as inulin or radioisotope clearance.(38) This discrepancy is one reason the use of equations has been advocated to estimate renal function. There is, however, considerable debate regarding which equation to use.(23–25,36,39) In a study of healthy adults, the MDRD equation was found to consistently underestimate and the CG to consistently overestimate GFR when compared with 125I-iothalamate or 99mTc-diethylenetriamine-pentaacetic acid renal clearance studies.(26) In a large, retrospective study of people 65 and older, Garg et al.(27) found that CG consistently predicted a lower estimate of GFR than MDRD, and this discrepancy was most pronounced in the oldest participants. Because of the known discrepancy,(27) and because the validity of both the CG and MDRD equations in healthy, community-dwelling seniors has been questioned,(19) we compared both equations, as well as serum creatinine, the measure most commonly used by primary care physicians as an estimate of renal function.
The cross-sectional renal function and BMD results reported here are similar to results from several cross-sectional studies. In two small studies, Yendt et al.(6,7) showed a strong, positive correlation between creatinine clearance (based on three successive 2-h urine specimens for creatinine) and bone mass at the radius and lumbar spine. In a second study, they compared 77 nonosteoporotic women to 37 women with primary osteoporosis and 25 osteoporotic patients with new vertebral crush fractures and found a significant association between creatinine clearance and BMD at the radius and lumbar spine, independent of age and body stature(7); creatinine clearance was significantly lower in women with vertebral fractures than in age-matched women without fractures. Similarly, in a third study of 19 postmenopausal women, Buchanan et al.(12) found BMD decreased in direct proportion to the decrement in creatinine clearance, but this association was no longer significant after controlling for the effects of 1,25-hydroxyvitamin D and PTH; they concluded that occult renal insufficiency may contribute to bone loss in aging women and that this effect may be mediated by vitamin D and PTH, which were not measured in our study at this visit. Our results are also compatible with those from a large (n = 13,831) cross-sectional study of the NHANES III cohort that showed the substantial prevalence of renal insufficiency (estimated as creatinine clearance by CG) in participants with osteopenia and osteoporosis, particularly with advancing age and in women.(9)
However, another cross-sectional report from the same NHANES III cohort(8) showed no association between poor renal function and lower femoral neck BMD after controlling for sex, age, and weight, and concluded that renal function was not independently associated with BMD. These differences may be because the second NHANES study adjusted for sex, age, and weight, although these variables are already included in the CG equation. Another probable reason for differences between NHANES and Rancho Bernardo is the much younger age of NHANES (average age, 48 years) versus Rancho Bernardo (average age, 71 years) participants. Our results also differed from those of a cross-sectional study that found no association between creatinine clearance and BMD in 311 healthy men and women; they also adjusted for age.(11)
Only one other study has investigated the cross-sectional association between renal function and fracture.(13) In a study of 5313 community-dwelling elderly men and women treated for osteoporosis, Dukas et al.(13) found that a creatinine clearance <65 ml/minute significantly increased both the risk of hip, vertebral, and radial fractures and the risk of falls reported within the past year.(13) Although information about falling was not obtained in our study, renal function by all three measures was associated with the measures of balance and gait that were assessed, supporting the hypothesis that renal function may be a measure of frailty and contribute to fracture risk by mechanisms beyond its association with BMD.
In a posthoc analysis of pooled data from nine drug trials, investigators found patients with the most severe renal impairment also had more severe osteoporosis and a higher incidence of new vertebral fractures.(14) Although others have prospectively studied the risk of bone loss or fracture among patients with ESRD,(2,40) to our knowledge, there are no prospective studies assessing the ability of renal function to predict bone loss or incident fracture in community-dwelling older adults.
Limitations of this study should be noted. The Rancho Bernardo Study population is almost entirely white, middle to upper middle class, and relatively healthy. These results may not be generalizable to other populations or places. Vitamin D deficiency is uncommon in this southern California community, with a study of women from the cohort at a subsequent visit reporting 2% had 25(OH)D <50 nM.(41) Serum vitamin D and PTH levels are important mediators in the association between renal function and bone; unfortunately, vitamin D and PTH were not measured at the baseline visit and thus we could not assess the role of these mediators. Another limitation common to all studies of the elderly is survival bias; if those with more severe renal disease were more likely to have died or were too ill to return to the follow-up visit, this would be expected to attenuate any association. In fact, renal function in the cohort seemed to increase significantly (p < 0.001) from the baseline to follow-up visit by the MDRD or serum creatinine, but not by CG. We have no clear explanation for this unexpected increase in renal function. Furthermore, a gold standard measure of GFR, such as inulin or radioisotope clearance or 24-h urine collection, was not performed. Serum creatinine was measured on a single blood specimen for each visit. Because it may vary significantly based on a number of factors (including hydration status or medication exposure) and is a critical component to each of the equations used to estimate renal function, it is possible that within-person measurement errors in serum creatinine resulted in imprecise measurement of the participants' renal function.(42) Furthermore, between-laboratory differences in creatinine assay may reduce the accuracy of renal function estimated by equations, particularly when estimating renal function in individuals with normal or mildly impaired renal function, because small changes in serum creatinine may result in large changes in calculated creatinine clearance and GFR.(26,42,43) Inaccurate estimation of renal function may be responsible for misclassification of some participants in the analyses by K/DOQI CKD stage and may explain the inconsistent findings in the longitudinal analyses stratified by K/DOQI CKD stage; however, serum creatinine was measured at both visits in the same laboratory, using the same technique, with good inter- and intra-assay CVs of 4.0%.
The failure to find an association between renal function and fracture likely reflects both the multiple causes of fracture beyond BMD (including fall risk, vision, and many others), and the small number of fractures. Incident fractures were reported by only 8%. A posthoc power analysis based on this small number of new fractures showed that those with abnormal renal function would have to have had a 230% increase (OR = 2.3) in fracture risk compared with those with normal renal function to have 80% power to detect a significant p < 0.05 difference. Thus, this study was underpowered to detect a smaller difference in fracture risk.
In summary, although renal function measured by both CG and MDRD was associated with BMD in cross-sectional analyses, only creatinine clearance by CG prospectively predicted bone loss. If confirmed, this should be the preferred method for assessing the association between renal function and BMD. Furthermore, given that serum creatinine was not associated with BMD or fracture in cross-sectional or longitudinal analyses, these findings provide further support for the K/DOQI recommendation that renal function be estimated by an equation to calculate GFR, rather than simply by the serum creatinine itself.(21) We also note that adjusting for age, sex, and weight is inappropriate in any equation that includes these variables. Cross-sectional associations between renal function and BMD were strongest at higher CKD stage, confirming that those with lower renal function are at higher risk for osteoporosis. This supports K/DOQI guidelines that recognize that bone disease caused by secondary hyperparathyroidism begins with stage 3 CKD(28); renal disease of this degree or greater was present in 52% of our population by the CG equation and in 32% by MDRD. None of the baseline renal function estimates were associated with prevalent or incident fractures, a finding that may have resulted from inadequate power and that requires further study. Further studies in the elderly are necessary to investigate the association between different measures of renal function, bone loss, and fracture, and the mechanisms for this association, including alterations in vitamin D and PTH. If such studies confirm renal function as an important predictor of bone loss and fracture, this may warrant adding mild to moderate renal dysfunction as an indication for osteoporosis screening, particularly for older men who may not meet other criteria for screening.
The authors thank Dr Ravindra Mehta (UCSD, Medical Center, Division of Nephrology, San Diego, CA) for expert advice on this manuscript. This study was supported by National Institute on Aging Grant AG07181, National Institute of Diabetes and Digestive and Kidney Diseases Grant DK31801, and a Procter & Gamble Pharmaceuticals Unrestricted Educational Grant.