Kreijkamp-Kaspers S, Kok L, Grobbee DE, de Haan EHF, Aleman A, Lampe JW, van der Schouw YT. Effect of Soy Protein Containing Isoflavones on Cognitive Function,
Bone Mineral Density, and Plasma Lipids in Postmenopausal WomenA Randomized Controlled Trial. JAMA. 2004;292(1):65–74. doi:10.1001/jama.292.1.65
Author Affiliations: The Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht (Drs Kreijkamp-Kaspers, Kok, Grobbee, and van der Schouw), Division of Psychonomics, Helmholz Research Institute, Utrecht University (Dr de Haan), and Department of Neuroscience, University Medical Center Utrecht (Dr Aleman), Utrecht, the Netherlands; and Fred Hutchinson Cancer Research Center, Seattle, Wash (Dr Lampe).
Context Postmenopausal estrogen therapy has been posited to have some beneficial
effects on aging processes, but its use has risks. Isoflavones, estrogenlike
compounds naturally occurring in plant foods, might confer positive effects
with fewer adverse effects.
Objective To investigate whether soy protein with isoflavones improves cognitive
function, bone mineral density, and plasma lipids in postmenopausal women.
Design, Setting, and Participants Double-blind, randomized, placebo-controlled trial of 202 healthy postmenopausal
women aged 60 to 75 years, recruited from a population-based sample in the
Netherlands, conducted between April 2000 and September 2001.
Intervention Participants were randomly assigned to receive 25.6 g of soy protein
containing 99 mg of isoflavones (52 mg genistein, 41 mg daidzein, and 6 mg
glycetein or total milk protein as a powder on a daily basis for 12 months.
Main Outcome Measures Cognitive function was assessed using the following instruments: dementia,
Mini-Mental State Examination; memory, Rey Auditory Verbal Learning Test,
immediate recall, delayed recall, and recognition, the Digit Span forward
and reversed, and the Doors test; complex attention tasks, Digit Symbol Substitution
and Trailmaking, A1, A2, and B; and verbal skills, Verbal Fluency A and N,
animals and occupations, and the Boston Naming Task. Bone mineral density
of the hip and lumbar spine was assessed using dual-energy x-ray absorptiometry
scanning. Lipid assessment included lipoprotein(a), total cholesterol, low-density
lipoprotein, high-density lipoprotein, and triglycerides.
Results A total of 175 women completed the baseline and at least 1 postintervention
analysis and were included in the modified intent-to-treat analysis. Adherence
was good (median plasma genistein levels, 17.2 and 615.1 nmol/L for placebo
and soy group, respectively). Cognitive function, bone mineral density, or
plasma lipids did not differ significantly between the groups after a year.
Conclusion This double-blind randomized trial does not support the hypothesis that
the use of soy protein supplement containing isoflavones improves cognitive
function, bone mineral density, or plasma lipids in healthy postmenopausal
women when started at the age of 60 years or later.
The sudden decline in estrogen levels after menopause coincides with
acceleration of several aging processes.1 On
average, bone mineral density (BMD) decreases and cognitive function declines,
whereas total cholesterol and low-density lipoprotein cholesterol (LDL-C)
increase. It has been suggested that postmenopausal estrogen therapy might
counteract some of these changes. However, short-term estrogen use is associated
with the recurrence of vaginal bleeding,2 and
long-term use has been associated with an increased risk of breast cancer,3,4 stroke,5 and
Phytoestrogens, including isoflavones and lignans, are estrogenlike
compounds naturally occurring in plant foods such as soy, beans and peas,
fruits, vegetables, and nuts and grains.6 These
compounds can activate the estrogen receptor and cause messenger RNA transcription.7 Depending on the situation, binding to the receptor
in the presence of endogenous estrogen in premenopausal women could result
in an antagonist action by competitive binding, whereas in the postmenopausal
state, phytoestrogens have been hypothesized to act as an agonist. If that
is the case, they could provide an alternative for traditional estrogen therapy.
The normal consumption of phytoestrogens in western populations is very low,
less than 5 mg/d,6,8 but the consumption
in Asian populations is 10- to 40-fold higher; the estimated intake for women
in Shanghai is 40 mg/d.9 A meta-analysis on
the effects of soy protein supplementation on plasma lipids comprising 38
studies reported a decrease in total cholesterol levels by 9.3% and LDL-C
levels by 12.9%.10 Both animal and human research
has suggested a preventive effect of isoflavones on bone loss11 and
recent trials with isoflavones reported an improvement in cognitive function
in both college students and postmenopausal women.12- 14 Their
main limitations were that the studies were small and had methodological issues,
in particular the absence of blinding.
The aim of our study was to determine the effects of supplementation
with soy protein, which naturally contains large amounts of the isoflavones
genistein and daidzein,15 on cognitive function,
BMD, and plasma lipids in older postmenopausal women.
The Institutional Review Board of the University Medical Center Utrecht
approved the study protocol and all participants gave written informed consent.
Participants were identified via the database of a breast cancer screening
program in Utrecht. We invited by mail women aged 60 to 75 years to participate.
Details of the study have been published previously.16 We
excluded women with conditions for which estrogens are contraindicated (active
liver disease, impaired renal function, history of breast cancer or other
malignancy, history of thromboembolism or deep venous thrombosis); women with
an endometrial thickness of more than 4 mm, current and recent (within past
6 months) estrogen users; and women with a known allergy or hypersensitivity
to soy or cow's milk. The study was conducted between April 2000 and September
After completing the baseline tests, participants were randomly assigned
to the intervention or the placebo group in blocks of 10. A list of randomization
numbers was computer-generated. Each randomization number corresponded to
1 of the 2 possible interventions, and personnel not involved in the trial
attached a label with the number to the identical boxes containing soy or
total milk protein. To assess the efficacy of blinding, at the end of intervention
the participants were asked whether they thought they had been assigned to
the placebo or the soy group.
The intervention consisted of 25.6 g of isoflavone-rich soy protein
containing 52 mg genistein, 41 mg daidzein, and 6 mg glycitein (aglycone weights)
in 36.5 g of powder (Solae, Solae Co, St Louis, Mo) that could be mixed with
food or beverages. The placebo (25.6 g of total milk protein) looked and tasted
identical to the soy and contained the same nutrients other than isoflavones.
Extra vitamins and minerals were added to the supplement for both groups (riboflavin,
pyridoxine hydrochloride, cyanocobalamin, folic acid, cholecalciferol, and
calcium). One supplement was taken per day for a total of 12 months.
A certified dietitian assessed the usual dietary pattern using a food
frequency questionnaire at baseline, at 3 and 6 months, and at the final visit.17 This questionnaire has been validated and was modified
slightly to capture dietary phytoestrogen intake. The dietitian counseled
the participants on incorporating the supplement into their diets by providing
recipes and making other suggestions. The participants were individually advised
from which sources to decrease their protein intake to compensate for the
extra protein intake from the supplement.
For all participants, we measured vital signs (pulse rate, blood pressure,
and weight) and recorded adverse events every 3 months during the visits.
A fasting blood sample was taken at baseline and at 12 months, the final visit.
Because the primary goal of this study was to investigate the effects of phytoestrogens
on clinical end points, we did not plan to assess the effects of the intervention
on sex hormone levels beforehand, but we collected and stored additional blood
samples during the study to explore biological mechanisms for potential soy
effects. Adherence was checked by assessing plasma genistein levels in the
final-visit blood sample. Blood levels reflect intake from the preceding 24
to 36 hours. Participants did not know that adherence was determined by blood
analysis. During the trial, women kept diaries to record the time and amount
they took in. Nonused supplements were also counted.
The participants were tested during a morning visit in a quiet room
by neuropsychologically trained personnel. Cognitive testing was performed
at baseline and at the final visit. Since the performance on cognitive tests
can be influenced by concomitant depression,18 we
assessed the presence of depression using the self-rated Geriatric Depression
Scale (GDS).19 Depression was defined as a
score of at least 11.
Cognitive tests were selected that have been documented to be sensitive
to the effects of aging and that have been included in previous trials of
estrogen treatment. More specifically, due to its receptor affinity in brain
structures subserving memory (ie, hippocampal formation), estrogen has been
associated with memory performance.20,21 We
therefore included measures of short-term and long-term verbal and visual
memory. In addition, estrogen has been related to verbal processing.22 Hence, we included measures of naming and verbal
fluency. Finally, estrogen could have a more general beneficial effect on
brain metabolism, which potentially could benefit complex attention functions
that are universally compromised with increasing age. Therefore, we included
tests of complex attention that have been widely documented to be sensitive
to cognitive aging.
The Mini-Mental State Examination23 was
used as a global test for Alzheimer disease or dementia from other causes
(maximum score, 30). The Rey Auditory Verbal Learning Test was used as a measure
of verbal episodic memory.24 In this test,
the participants are asked to recall a 15-word list immediately (immediate
recall) for 5 times consecutively (maximum score, 75), and, after 25 to 30
minutes (delayed recall; maximum score, 15). Furthermore, the participants
were asked to recognize the words out of a list of 30 (recognition, maximum
score, 30). The Doors test was used to assess visual memory.25 Participants
are shown 2 series of 12 photographs of doors, which they subsequently have
to recognize from arrays of 4 pictures of doors (maximum score, 24). In the
Digit Span test, a subtest of the Wechsler Adult Intelligence Scale,26 participants are asked to repeat a string of digits
in the original order (digit span forward, maximum score, 8) and in the reverse
order (digit span reversed, maximum score, 7) to give an impression of short-term
memory and working memory. To test verbal fluency, the participant is asked
to list as many nouns as possible beginning with the letters N and A and to name as many animals and occupations,
each in 1 minute (score, number of nouns named). In the Boston naming task
for verbal competence and semantic retrieval, the participant is shown line
drawings, which have to be properly named27 (maximum
score, 200). The digit symbol substitution test, also from the Wechsler Adult
Intelligence Scale,26 measures cognitive and
perceptual speed. The participant is given a code that pairs symbols with
digits. The test consists of pairing as many digits to their corresponding
symbols as possible in 90 seconds (score, number of paired digits). The Trailmaking
test A1, A2, and B is a complex attention and mental flexibility task. In
the Trailmaking test, pseudorandomly placed circles with numbers (Trailmaking
A1), with letters (Trailmaking A2), and with both letters and numbers (Trailmaking
B) have to be connected with a line as fast as possible in a fixed order28 (score, seconds needed to complete the task). At
baseline, we also assessed the verbal intelligence quotient using the Dutch
Adult Reading Test (a Dutch version of the National Adult Reading Test29,30) in which the participants have to
read out loud a list of words with irregular pronunciation. Completion of
the entire test battery took 1 hour on average.
Bone Mineral Density. At baseline and at the
end of the 12-month intervention, BMD of the left proximal femur and the lumbar
spine (L1-L4) was measured by dual-energy x-ray absorptiometry (DXA) using
a Hologic QDR 1000 densitometer (Hologic Inc, Waltham, Mass). All scans were
analyzed according to written manufacturer procedures.
At baseline, 201 participants underwent a measurement of BMD of the
hip. In one participant scanning was not possible because of hip prostheses
on both sides. In 198 participants we were able to scan the left hip, but
in 3 we had to scan the right hip because of a prosthesis on the left side.
Because of this small number, we analyzed the results of the right-sided scan
together with those on the left side. At the end of participation the hip
was scanned in 174 participants.
Dual-energy x-ray absorptiometry of the lumbar spine (L1-L4) was obtained
in 202 participants at baseline and in 175 at the end of participation. Vertebral
measurements had to be excluded in 12 scans for several reasons, eg, insufficient
scanning of L1 or L4 in 7, projection of a corpus alienum over 1 of the 4
vertebrae with subsequent overestimation of BMD in 3, and exclusion by the
machine of vertebrae during the analysis procedure in 2 patients. When we
excluded a vertebra we also excluded the result of the total spine, because
this is calculated from the results of the 4 vertebrae and consequently also
Bone Parameters. Ostase immunoradiometric assay
(Beckman Coulter, Inc, Fullerton, Calif) is a quantitative measurement of
the bone formation marker bone-specific alkaline phosphatase. We used a timed-rate
method to determine inorganic phosphorus concentration and indirect potentiometry
to determine calcium concentration (Synchron LX System and Access2, Beckman
Coulter). All bone parameters were measured in blood plasma.
Plasma lipid levels were assessed at baseline and the final visit. Cholesterol
was assessed by an enzymatic method with the Cholesterol Reagent (Synchron
LX Systems, Beckman Coulter). High-density lipoprotein-cholesterol was assessed
using a direct HDL-C assay with a timed–end point method. Triglycerides
were also determined with a timed–end point method (Synchron LX Systems;
Plasma genistein levels were measured with blood obtained at the final
visit using Labmaster TR-FIA kits (Turku, Finland). Fluorescence was measured
on the Wallac Victor 2 model 1420 spectrofluorimeter (Labmaster, Turku, Finland).
Data were analyzed using GraphPad Prism software (GraphPad Software Inc, San
Diego, Calif). Intraassay and interassay coefficients of variation were 2.2%
and 14.8%, respectively. Equol is a highly active metabolite produced by the
intestinal flora from the isoflavone daidzein. This metabolite is only produced
in a proportion of the population. It has been suggested that possibly only
this subgroup benefits from the intervention. Equol producer status was defined
as equol of higher than 83 nmol/L in plasma.31 The
proportion of equol producers was 29.9%.
We performed power calculations for 3 primary end points: the Rey Auditory
Verbal Learning Test for cognitive function, density of the lumbar spine for
BMD, and total cholesterol for lipids. We planned to recruit a total of 200
participants, 100 for each group. This number was based on conventional assumptions
of α = .05 and β = .20 and a 25% rate of withdrawal from the intervention
group. Assuming that soy isoflavones are as effective as conventional hormone
therapy, we would be able to demonstrate an improvement of 13% on the Rey
Auditory Verbal Learning Test,20 detect a total
cholesterol decrease of 7.4%, and a BMD increase of 6.7%. Similar changes
in total cholesterol and BMD have been reported for soy isoflavone supplementation.10,32
Data were analyzed according to a modified intention-to-treat principle,
including all those who had 2 measurements, including baseline, in the groups
to which they were randomized. Linear regression analysis was used with baseline-to-final
visit changes as the dependent variable and group allocation as the independent
variable. This procedure could result in slight rounding effects in the tables.
We performed a closeout visit when a participant had remained for at least
1 month. Fourteen percent of participants did not complete 1 month of treatment
or were unable or unwilling to participate in a final visit. The 175 participants
(86%) who completed a closeout or final visit were included in the primary,
modified intention-to-treat analysis. Secondary analyses comprised a per-protocol
analysis, including only the 153 participants (76%) who had completed the
whole treatment protocol, and an analysis of the cognitive function results
excluding 17 participants who were depressed. Furthermore, we studied whether
the effect of soy differed across subgroups of postmenopausal years (<14,
14-22, >22 years), equol-producer status, body mass index (<24, 24-28,
>28), smoking history (ever/never), and history of estrogen use was studied
by looking at effects of intervention in the specific subgroups. We tested
this effect modification by adding interaction terms between intervention
and group variables to the basic model containing the 2 individual variables.
To assess whether baseline differences in smoking and BMD influenced our results,
we repeated the analysis adjusting for the baseline values. At the outset
of the study, we decided not to adjust for multiple comparisons.33,34 SPSS
11.0 statistical software was used to perform all the analyses (SPSS Inc,
Chicago, Ill). P<.05 was considered statistically
Participant recruitment and enrollment are shown in Figure 1. Between March 2000 and September 2000, we randomly assigned
202 women to the 2 treatments.
Forty-nine participants (24%) did not complete the trial for various
reasons, the most important being gastrointestinal tract complaints and aversion
to the taste of the supplement. Median duration of participation for the dropouts
was 79 days (range, 4-285 days). There was no difference in dropout rate between
the 2 groups (24 placebo, 25 soy).
Table 1 shows the baseline
characteristics of the participants by intervention group included in the
modified intent-to-treat analysis. There were no major differences between
the 2 groups. Characteristics of these 2 groups also are very similar to all
those who enrolled, except that those who had at least 2 assessments were
somewhat less likely to smoke, were better educated overall, and were more
likely to be taking statins than those who did not. (A table showing all randomized
participants is available from the author on request.) Table 2 shows the adverse events reported during the trial. The
mean number of events per participant was 2.54 in the soy group and 2.56 events
in the placebo group. There were also no differences in types of adverse events.
The nutrient intake, as calculated from the food frequency questionnaire,
showed similar dietary patterns for the 2 groups. Both groups reduced their
intake of protein and total calories as a result of the nutrients provided
by the supplement (Table 3). Table 4 shows the efficacy of the blinding,
confirming that blinding had been effective. Genistein levels during the trial
were markedly different between the intervention group and the placebo group
(median [interquartile range]) was 17.2 nmol/L (10.29-54.0 nmol/L) for the
placebo group and 615.1 nmol/L (234.3-1634.6 nmol/L) for the soy group; P value for difference <.001), demonstrating that adherence
was good. In addition, 90% of participants used at least 80% of their supplements.
Analyses evaluating whether genistein concentrations were associated with
treatment effects showed no significant differences from the main effects.
Baseline performance on the cognitive function tests was similar for
the 2 groups (Table 5). On most
of the tests focusing on memory the soy group scored slightly higher, but
the differences were clinically small and not statistically significant. The
tests for verbal skills and the more complex tasks requiring concentration
and visual attention showed no significant between-group differences. When
participants with depression11 were excluded
from the analysis, the results were virtually identical. Years since menopause,
equol production, body mass index, smoking status, and history of estrogen
therapy did not affect the results, and adjustment for smoking to control
for baseline differences also did not alter the results. (Results are available
from the author on request.)
Baseline BMD and plasma bone-specific alkaline phosphatase, calcium,
and phosphorus levels are shown in Table
6. Both groups showed a decrease in BMD after a year. At 1 year,
the placebo group had a mean BMD decrease of 0.009 g/cm2 in the
intertrochanter region of the hip, whereas the soy group had a mean BMD increase
of 0.004 g/cm2, resulting in a difference in BMD change of 1.31%
(P = .02). In the other regions, both in hip and
lumbar spine, there were no major differences between the 2 groups. No significant
differences were seen for bone-specific alkaline phosphatase, calcium, and
phosphorus measurements. Adjustment for smoking history and baseline BMD did
not change the results. (Results are available on request from the author.)
Subgroup analysis by the number of postmenopausal years showed that
women in the lowest tertile, ie, who had the most recent start of menopause,
had better results after a year of soy intervention, while women in the highest
tertile did slightly worse compared with the placebo group (Table 7). This was the case for both hip and lumbar spine BMD measurements,
but the interaction term did not reach statistical significance. Only the
intertrochanter region of the hip showed a statistically significant interaction
(P = .04) with years since menopause. Again, there
was no significant interaction with equol production, body mass index, smoking
history, or history of estrogen therapy.
At baseline, the levels of plasma lipids were similar between the 2
groups (Table 8). In the soy group,
the LDL-C and total cholesterol levels remained constant while the placebo
group showed a small decrease, but the differences were not statistically
significant. There was no significant interaction with age, equol production,
baseline cholesterol level, smoking history, body mass index, and history
of estrogen use. Adjustment for smoking to control for baseline differences
did not change the results.
In this longer-term, relatively large double-blind, placebo-controlled,
randomized trial, we did not find any effect of soy protein supplementation,
which naturally contains large amounts of isoflavones, on cognitive function,
BMD, or plasma lipids in the relevant population of aging women. It should
be acknowledged that although BMD of the intertrochanter region of the hip
was significantly higher in the soy group, it was only 1 comparison among
13 BMD measurements and may well be a chance finding.
To fully appreciate these results, some issues need to be addressed.
First, a null finding in a trial could be caused by insufficient statistical
power. However, the number of participants needed according to the power calculation
was reached and the withdrawal from treatment did not exceed the predefined
25%. Furthermore, previous positive trials for isoflavones and cognitive function
enrolled a maximum of 56 participants and the first trial to find an effect
on BMD comprised only 66 participants. Since our trial had 202 participants,
it seems unlikely that lack of power is a major concern.
Second, adherence is always a concern, especially with a relatively
unattractive intervention like a dietary supplement. However, serum genistein
levels were markedly different between the intervention group and the placebo
group, demonstrating good adherence.
The null finding for the Mini-Mental State Examination, the digit span,
and the Rey Recognition could be caused by a ceiling effect because the average
score in this relatively healthy population was close to the maximum score.
However, our participants had a wide range of scores for the other cognitive
function tests, making a ceiling effect unlikely.
The lack of an effect on cognitive function contrasts with findings
in animal experiments and a few recent trials. Animal studies using a rat
model showed a clear effect of an isoflavone-rich diet on cognitive function.35,36 In humans, 3 intervention studies
have been published. The first37 reported improvements
in both long-term and short-term memory and better performance on a mental
flexibility task after a 10-week soy diet. However, in addition to the small
number of participants, participants were not blinded for the intervention,
and comprised male and female college students, which limits the comparability
with our study. In postmenopausal women, one study showed improvement on picture
recall, learning rule reversals, and a planning task while another reported
improvements on verbal fluency only. These studies were small and used isolated
isoflavones in pills.
For BMD, several studies have indicated encouraging effects of isoflavones
on osteoporosis. A recent review11 lists the
available data. Both in vitro cell cultures and in vivo animal studies almost
invariably show positive effects of isoflavones on parameters related to BMD.
Clinical studies are less consistent, with results varying from an improvement
of 2.2% in lumbar BMD after 6 months of treatment32 to
no effect.38 In another study, soy milk containing
85 mg isoflavones was effective in preserving bone.39 Interestingly,
the interim analyses of the second study after 1 year did not disclose a significant
effect on BMD, suggesting that the duration of the intervention in our 1-year
trial may still have been too short. In several clinical studies the bone
formation marker bone-specific alkaline phosphatase was not affected by the
intervention,40- 42 which
agrees with our findings.
For plasma lipids, a large body of evidence points to a favorable effect
of soy protein. A meta-analysis of 38 studies reported a decrease in total
cholesterol of 9.3% and LDL-C of 12.9%.10 However,
in contrast to the meta-analysis, several subsequent studies aimed at postmenopausal
women did not demonstrate any effect on cholesterol, LDL-C, or HDL-C levels43- 48 except
for 249,50 that showed effects
on LDL-C and HDL-C but not on total cholesterol. Possibly, the findings in
the meta-analyses, based mainly on trials performed in men, do not hold for
A possible explanation for the striking discrepancy between the promising
findings in animal research and subsequent lack of confirmation in human trials,
especially for BMD and cognitive function, may be found in species differences
in the metabolism of isoflavones. One of the main metabolites of the isoflavone
daidzein in rodents is equol. In humans the production is dependent on the
individual's intestinal flora, and research shows that only about one third
of people will produce equol when exposed to high amounts of daidzein.51,52 Equol may be an important modifier
of the effects of isoflavones.53 However, equol
production was assessed in our trial and the effects of soy on the different
end points were not different from equol producers when compared with the
nonproducers although statistical power was limited.
Another important factor may be the timing of the supplementation. In
our trial women were on average 18 years menopausal. In a rat model, isoflavones
were very effective in preventing bone loss shortly after ovariectomy, but
late supplementation may not restore bone or prevent further losses.54 Similarly, the most pronounced effects of estrogen
on cognitive function have been reported in perimenopausal women, and not
in late postmenopausal women.1,55 With
respect to bone, it has been suggested that it is easier to prevent changes
or losses after menopause than reverse them when they have already taken place.11 Our findings in subgroups according to years since
menopause appear to support this hypothesis; in the women who were recently
menopausal, our intervention seemed to improve BMD while in the late menopausal
women such effect was absent. However, only the intertrochanter region of
the hip showed a statistically significant interaction (P = .04). The influence of the timing of supplementation needs to be
elucidated in further research.
In conclusion, the results of this large, double-blind, 1-year randomized
trial do not support the hypothesis that isoflavones from soy protein have
beneficial effects on cognitive function, BMD, or plasma lipids in older postmenopausal