The association of bone mineral density with HIV infection and antiretroviral treatment in women.

Abstract

Low bone mineral density (BMD) has been reported in HIV-infected women and men.
We analysed cross-sectional BMD measured by regional dual X-ray absorptiometry at the lumbar spine (LS) and femoral neck (FN) in 152 HIV-negative and 274 HIV-positive (HIV+) women, adjusted for traditional low BMD risk factors.
BMD was significantly lower in protease inhibitor (PI) users than in all other groups, and highest in HIV-negative women. In multivariate analyses the prevalence of T-score < -1.0 was significantly higher in the HIV+ women naive to antiretroviral therapy (ART; odds ratio [OR] 4.36, 95% confidence interval [CI] 1.61, 11.8) and the women receiving PI-containing HAART (OR 3.72, CI 1.43, 9.68), with a non-significant difference in non-PI HAART users (OR 2.43, CI 0.92, 6.45), compared with HIV-negative women. In pair-wise adjusted comparisons, BMD was lower in ART-naive than in HIV-negative women (1.22 versus 1.30 g/cm2 at LS; P = 0.004), in PI compared with non-PI HAART users (1.00 versus 1.05 g/cm2 at FN; P = 0.014) and with those ART-naive (1.00 versus 1.03 g/cm2 at FN; P = 0.146). Potential confounders, including duration of ART, prior treatment regimens and traditional risk factors for low BMD did not explain these differences. Longer lopinavir use was significantly correlated with lower BMD (r2 = -0.39, P = 0.024 and r2 = -0.46, P = 0.006 at LS and FN, respectively) and longer efavirenz use with higher BMD (r2 = +0.32, P = 0.004 at FN).
HIV infection was associated with lower BMD in women, independent of the traditional risk factors for low BMD. PI-containing HAART compared with non-PI-containing HAART, and longer lopinavir use, were both associated with lower BMD, and efavirenz use was associated with higher BMD.

Full text

Page 1

Background: Low bone mineral density (BMD) has been
reported in HIV-infected women and men.
Methods: We analysed cross-sectional BMD measured by
regional dual X-ray absorptiometry at the lumbar spine
(LS) and femoral neck (FN) in 152 HIV-negative and 274
HIV-positive (HIV+) women, adjusted for traditional low
BMD risk factors.
Results: BMD was significantly lower in protease inhibitor
(PI) users than in all other groups, and highest in HIV-
negative women. In multivariate analyses the prevalence
of T-score <-1.0 was significantly higher in the HIV+
women naive to antiretroviral therapy (ART; odds ratio
[OR] 4.36, 95% confidence interval [CI] 1.61, 11.8) and
the women receiving PI-containing HAART (OR 3.72, CI
1.43, 9.68), with a non-significant difference in non-PI
HAART users (OR 2.43, CI 0.92, 6.45), compared with HIV-
negative women. In pair-wise adjusted comparisons, BMD
was lower in ART-naive than in HIV-negative women
(1.22 versus 1.30 g/cm2at LS; P=0.004), in PI compared
with non-PI HAART users (1.00 versus 1.05 g/cm2at FN;
P=0.014) and with those ART-naive (1.00 versus 1.03
g/cm2at FN; P=0.146). Potential confounders, including
duration of ART, prior treatment regimens and traditional
risk factors for low BMD did not explain these differences.
Longer lopinavir use was significantly correlated with
lower BMD (r2=-0.39, P=0.024 and r2=-0.46, P=0.006 at
LS and FN, respectively) and longer efavirenz use with
higher BMD (r2=+0.32, P=0.004 at FN).
Conclusions: HIV infection was associated with lower
BMD in women, independent of the traditional risk
factors for low BMD. PI-containing HAART compared
with non-PI-containing HAART, and longer lopinavir use,
were both associated with lower BMD, and efavirenz use
was associated with higher BMD.
The association of bone mineral density with HIV
infection and antiretroviral treatment in women
Kathryn Anastos1*, Dalian Lu2, Qiuhu Shi3, Kathleen Mulligan4, Phyllis C Tien4,5, Ruth Freeman6,
Mardge H Cohen7, Jessica Justman8and Nancy A Hessol4
1Montefiore Medical Centre, Bronx, NY, USA
2Data Solutions LLC, Bronx, NY, USA
3New York Medical College, Valhalla, NY, USA
4University of California at San Francisco, CA, USA
5Veterans Affairs Medical Centre, San Francisco, CA, USA
6Albert Einstein College of Medicine, Bronx, NY, USA
7CORE Centre and Storger (formerly Cook County) Hospital, Chicago, IL, USA
8Mailman School of Public Health, Columbia University, New York, NY, USA
*Corresponding author: Tel: +1 718 515 2593; Fax: +1 718 547 0584; E-mail: kanastos@verizon.net
Antiviral Therapy 12:1049–1058
Low bone mineral density (BMD) in HIV-infected
individuals was first reported before the era of highly
active antiretroviral therapy (HAART) [1] and later
observed in men receiving HAART containing
protease inhibitors (PI) [2]. Several subsequent reports
have suggested that low BMD is associated with HIV
infection itself in women [3–13] and men [7–15]. Some
authors [2,7,8] have found an association of HAART
and PIs with even lower BMD, whereas others have
found either no association [3,9–12] or that PIs were
associated with higher BMD than in untreated HIV-
infected persons [12]. Others have suggested [6,14,16]
that the differences in BMD are secondary to tradi-
tional risk factors for low BMD, for example smoking
or lower body mass index (BMI). One report found a
higher prevalence of osteoporosis in HIV-infected
men than in HIV-infected women [9], reversing the
relationship found in the general population.
In all populations, BMD decreases after the age of
30 years in both women and men, with a much more
rapid rate of bone loss in women during the peri-
menopause and early menopause [17]. Women of
African descent and women with higher body mass
index (BMI) have higher BMD [18]. Thus, because
HIV-infected US women have a median BMI in the
overweight to obese range [19,20], and as more than
half are of African descent [21], they might be at higher
risk of osteoporosis than HIV-infected men because of
Introduction
© 2007 International Medical Press 1359-6535
1049

Page 2

their sex, but at lower risk because of their African
ancestry and higher BMI. To help delineate the influence
of these factors in women, we conducted a cross-
sectional study of the association of BMD with HIV
infection and HIV therapy in 426 women enrolled in
the Women’s Interagency HIV Study (WIHS).
Methods
Study population
WIHS is a multicentre prospective study of the natural
history of HIV-1 infection in women, conducted in five
locations within the United States: New York City (two
sites), Washington D.C., Chicago, Southern California
and the San Francisco Bay area. The WIHS methods
and baseline cohort characteristics have been described
previously [22,23]. Briefly, 3,770 women (2,794 HIV-1
seropositive and 976 seronegative) were enrolled
during two periods: October 1994–November 1995
(69.6% of the cohort) and October 2001–September
2002. For participants enrolled in 2001–2002, medical
record abstraction was performed for all those
reporting HAART use at enrolment, to verify their pre-
HAART CD4+T-cell and HIV RNA counts, date of
HAART initiation and regimen. Every 6 months partic-
ipants were interviewed using a structured question-
naire and received a physical examination. Multiple
gynaecological and blood specimens were collected at
each visit. At each study visit, self-reported antiretro-
viral therapy (ART) use in the period since the previous
study visit was assessed by interviewers, stating both
the generic and brand name of each drug, and showing
participants photo-medication cards. Informed consent
was obtained from the participants in accordance with
procedures and consent materials reviewed and
approved by the committee on human research at each
of the collaborating institutions.
From April 2001 to October 2005, women who met
the eligibility criteria at the Bronx, Manhattan, San
Francisco and Chicago WIHS consortia were invited to
participate in this cross-sectional substudy. Exclusion
criteria included type I diabetes, weight >264 pounds (the
weight limit of the dual X-ray absorptiometry [DXA]
scanners) and use of corticosteroids, drugs used to treat
osteoporosis or exogenous hormones including hormonal
contraceptives. BMD of the lumbar spine (LS) and the
femoral neck (FN) was measured using DXA (Lunar
Prodigy, Madison, WI, USA) in the Bronx and Chicago
throughout the study period, and in San Francisco after
September 2003. Before September 2003 a Lunar DPX
(Lunar Prodigy) was used in San Francisco, which was
cross-calibrated with a live person scanned at both the
Bronx and San Francisco sites. The machine-based
control data are derived from the National Health and
Nutrition Study (NHANES) population [24].
Laboratory methods
Plasma HIV-1 RNA was quantified using the
isothermal nucleic-acid-sequence-based amplification
(NASBA/Nuclisens) method (Organon Teknika Corp.,
Durham, NC, USA) in laboratories participating in the
NIH/NIAID Virology Quality Assurance Laboratory
proficiency testing program. The lower limit of quan-
tification was 80 copies/ml. Lymphocyte subsets were
quantified using standard flow cytometric methods in
laboratories participating in the NIH/NIAID Flow
Cytometry Quality Assessment Program [25].
Outcome variables
The dependent variables were DXA-measured BMD
and calculated T-scores, obtained by comparing the
women’s measured BMD with a specified standard (the
average BMD of a young white woman) [24].
Exposure variables
The primary independent variables were HIV infection
and use of ART. HAART was defined as follows: (i)
two or more nucleoside reverse transcriptase inhibitors
(NRTIs) in combination with at least one PI or non-
nucleoside reverse transcriptase inhibitor (NNRTI); (ii)
one NRTI in combination with at least one PI and at
least one NNRTI; (iii) a regimen containing ritonavir
and saquinavir in combination with one NRTI and no
NNRTIs; or (iv) an abacavir- or tenofovir-containing
regimen of three or more NRTIs in the absence of PIs
and NNRTIs. Combinations of zidovudine and stavu-
dine with a PI or NNRTI were not considered HAART.
For women enrolled in 1994–1995, duration of use of
specific antiretroviral agents was measured by deter-
mining the number of 6-month intervals on therapy for
each drug, as reported at the semiannual WIHS visits.
For women enrolled in 2001–2002, duration of
therapy was calculated as the time from first HAART
use to WIHS enrolment, plus the reported number of
6-month intervals on therapy after WIHS enrolment
prior to the DXA scan.
Secondary independent variables included BMI
(nadir and at the time of DXA scanning) calculated as
weight in kilograms divided by the square of height in
metres; CD4+T-cell count (nadir and at most recent
6-monthly WIHS visit; nadir was used as a proxy for
disease severity because WIHS does not have informa-
tion on duration of infection); age; seropositivity for
hepatitis C virus (yes/no); self-reported menopausal
status; amenorrhoea (no menses for two consecutive
6-month periods before the DXA); history of hysterec-
tomy with and without bilateral oophorectomy; cumula-
tive years of use during WIHS participation of calcium
supplements; lifetime history of smoking, opiate use and
cocaine use determined by self report during 6-month
follow-up visits (vitamin D supplementation was not
K Anastos et al.
© 2007 International Medical Press
1050

Page 3

examined because of scanty relevant data); current
tobacco smoking (yes/no); alcohol consumption
reported at most recent WIHS visit (more or less than
16 g/day); and race. Race was based on self-categoriza-
tion, and was stratified as: African-American (Hispanic
and non-Hispanic), white (Hispanic and non-
Hispanic), and Latina (women identifying as Hispanic
but neither black nor white). Nadir BMI was more
strongly associated than most recent BMI with HIV
serostatus and treatment category and was included as
a covariate in multivariate analyses.
Statistical methods
Analysis of variance (ANOVA) or χ2was used to
compare mean or percentage for continuous and
categorical variables, respectively, among four groups
categorized by HIV serostatus and treatment at the
time of the DXA: HIV-negative, HIV-positive (HIV+)
ART-naive, HIV+on non-PI HAART, and HIV+on PI-
containing HAART. Variables with a P-value <0.10
were included in the multivariate analyses of covari-
ance (ANCOVA) model analysing BMD among
groups. ANCOVA adjusting for race, menopausal
status, nadir BMI and CD4+T-cell count was used to
compare BMD among groups. A closed test procedure
was used to assess pair-wise comparisons only if the
overall group test was significant. Odds ratios (ORs) of
low BMD were computed for subjects who were HIV+
versus HIV-negative, and for treatment type. Pearson
correlations were used to assess associations of BMD
with duration of therapy class measured as cumulative
exposure during WIHS participation and categorized
into mutually exclusive groups: ever exposed to NRTIs
only; to NNRTIs with NRTIs only; to PIs with NRTIs
only; or to all three classes of antiretroviral agents.
Similar analyses were performed for individual drugs
for which there were ≥30 users.
In subanalyses to assess potential confounding, we
restricted the two treatment groups to those women
reporting only PI HAART use ever, and only non-PI
HAART use ever, to eliminate possible confounding by
prior therapy with the other class of HAART. Menopause
was examined more closely by examining self-report of
oophorectomy in women with a history of hysterectomy.
We further assessed the prevalence of amenorrhoea of
≥1 year in women not reporting post-menopausal status.
We also performed sub-analyses of age-matched pre-
menopausal women with pair-wise comparisons among
the four groups, adjusted for nadir BMI, race and years
of smoking. In the two groups being compared, one
subject was randomly selected from the group with fewer
subjects, and then matched with one subject randomly
selected from those of the same age in the second group.
All analyses were performed using the SAS statistical
package (Version 8, SAS Institute, Cary, NC, USA).
Results
The demographic and clinical characteristics of the 426
participants are shown in Table 1. The HIV-negative
women were younger, less likely to have hepatitis C
infection or to report menopause, more likely to be of
African descent and to report fewer years of tobacco
smoking and greater alcohol use, and had higher mean
BMI, both nadir and most recent, which was in the
overweight range (25.0–30.0 kg/m2) in all groups.
Among the HIV+women, the nadir CD4+T-cell count
was significantly lower, and the peak HIV RNA signif-
icantly higher in the women on HAART. Women on
non-PI-containing HAART were half as likely as PI
users to report being post-menopausal (P<0.001).
The use of each specific antiretroviral agent ever and
at the time of the study visit, and median duration of
use, are shown in Table 2.
Association of treatment regimen with BMD
Results of univariate analyses are shown in Table 3.
In multivariate analyses (Table 4) adjusted for race,
nadir BMI, menopausal status and years of smoking,
all of which were associated with BMD in the
univariate models, BMD was lower in PI than in non-
PI HAART users (1.22 versus 1.26 g/cm2, P=0.086,
and 1.00 versus 1.05 g/cm2, P=0.014 at LS and FN,
respectively) and in HIV-negative women (1.22
versus 1.30g/cm2, P=0.004 and 1.00 versus
1.08 g/cm2, P=0.0008 at LS and FN, respectively).
HIV+women naive to ART had lower BMD than
HIV-negative women both at LS (P=0.004) and at FN
(P=0.04). Age and menopause were highly collinear
and when both were included in multivariate models
age was not significant. For this reason, and because
epidemiological data demonstrate that the greatest
age-related drop in BMD in women occurs at
menopause [17], we included only menopausal status
in adjusted models. To assess possible confounding
from prior treatment regimens, we repeated the
multivariate analysis restricting the two groups of
treated women to those with a lifetime exposure only
to non-PI HAART (n=54) or only to PI HAART
(n=65). The adjusted LS BMDs in this restricted
analysis were 1.26 and 1.19 g/cm2in the non-PI and
PI HAART users respectively, similar to that found
with the unrestricted treatment groups (1.25 and
1.20 g/cm2, respectively). With the smaller sample
size the strength of the associations was weaker:
P=0.004 and P=0.099 for PI users compared with
HIV-negative women and non-PI HAART users,
respectively. Similarly, the FN BMD in the PI users
was 0.98 g/cm2compared with 1.10 g/cm2(P=0.011)
and 1.03 g/cm2(P=0.031) in the HIV-negative
women and the non-PI HAART users, respectively.
Antiviral Therapy 12:7
1051
HIV, HAART and bone mineral density in women

Page 4

We further investigated possible confounding by
duration of cumulative exposure ever to PIs, NNRTIs
and NRTIs. Participants were categorized by exposure
to drug class (n, median duration of exposure): NRTI
exposure only (10, 1.25 years); NNRTI plus NRTI
exposure only (47, 2.5 years); PI plus NRTI exposure
only (55, 3.5 years); or exposure to all three classes of
ARVs (86, 6.0 years). The Pearson correlation of BMD
with duration of therapy was not statistically significant
in any of the groups: all P-values >0.156.
Associations of BMD with specific antiretroviral drugs
In analyses of BMD with duration of use among users
of specific ARVs, most agents demonstrated a small,
statistically and clinically (r2>-0.10) insignificant inverse
correlation. However, longer lopinavir use demon-
strated a moderately strong, statistically significant
inverse correlation with lower BMD: r2=-0.39, P=0.024
and r2=-0.46, P=0.006 at LS and FN respectively.
Longer efavirenz use was associated with a higher
BMD: r2=+0.32, P=0.004 and r2=+0.12, P=0.32 at FN
and LS, respectively. This favourable association was
not seen with nevirapine (r2=+0.05 and +0.07 at FN
and LS, P>0.5 for both). Longer tenofovir use was
inversely but not significantly correlated with BMD at
FN (r2=-0.21, P=0.118) but not at LS (r2=+0.01,
P=0.95). Longer abacavir (ABC) and stavudine (d4T)
use were associated with small, inverse, non-significant
correlations with BMD: at FN, r2=-0.19, P=0.11 and
r2=-0.16, P=0.073, for ABC and d4T, respectively; and
at LS, r2=-0.06, P=0.63 and r2=-0.13, P=0.15 for ABC
and d4T, respectively. For perspective, we provide the
correlations of age and BMI with BMD: at FN r2=-0.33
and r2=+0.31 for age and BMI, respectively, P<0.0001
for both; and at LS r2=-0.21 and r2=+0.16, P<0.0001
and P=0.001, respectively.
Associations with T-score
The mean T-scores (Figure 1) were highest in the
HIV-negative women. Figure 2 shows the proportion
K Anastos et al.
© 2007 International Medical Press
1052
HIV-negative,
n=152
HIV + ARV-
naive, n=76
HIV + on non-PI
HAART, n=101
HIV + on 1 PI
HAART, n=97
P-value
Cohort, n
Original
New
Race, n (%)
Black
White
Latina
Age, years
Mean (SD)
Median
Time smoking, median years (95% CI)
Body mass index, mean (SD)
Most recent
Nadir
Post-menopausal, %
Current alcohol use ≥16 g/day, n (%)
Hepatitis C seropositive, n (%)
Ever used opiates, %
Ever used cocaine, %
CD4+T-cell count, cells/μl
At last routine visit, mean (SD)
At last routine visit, median
Nadir, mean (SD)
Nadir, median
HIV-1 RNA, copies/ml
At last routine visit, mean (SD)
At last routine visit, median
Peak, mean (SD)
58
94
38
38
66
35
75
22
–
<0.0001
99 (65.1)
36 (23.7)
17 (11.2)
57 (75.0)
11 (14.5)
8 (10.5)
56 (55.4)
22 (21.8)
23 (22.8)
55 (56.7)
21 (21.6)
21 (21.6)
0.032
–
–
36.4 (8.78)
36.7
15.0 (11.2, 17.1)
42.5 (7.16)
42.1
18.4 (15.0, 21.2)
41.6 (7.26)
40.5
18.0 (15.0, 21.5)
44.2 (7.82)
43.8
20.6 (16.8, 23.9)
<0.0001
–
0.053
29.9 (6.72)
28.5 (6.28)
6.6
64 (42.1)
29 (19.1)
14.5
17.1
28.9 (6.24)
27.7 (6.19)
26.3
18 (23.7)
31 (40.8)
9.2
15.8
26.8 (5.81)
25.4 (5.54)
16.8
24 (23.8)
33 (32.7)
3.0
11.9
28.4 (6.87)
26.3 (6.36)
37.1
10 (10.3)
48 (49.5)
10.3
14.4
0.003
0.0007
<0.0001
<0.0001
<0.0001
0.028
0.715
–
–
–
–
490 (268)
431
420.0 (237)
373
462 (281)
448
327.2 (207)
324
435 (246)
428
312.1 (238)
243
0.400
–
0.005
–
–
–
–
40,909 (219,000)
2150
47,057 (221,000)
9,885 (52,972)
<80
76,016 (338,000)
4,586 (17743)
<80
88,004 (281,000)
0.112
–
0.644
Smoking, opiate and cocaine exposure calculated as cumulative lifetime use. ARV, antiretroviral; CI, confidence interval; HAART, highly active antiretroviral therapy;
PI, protease inhibitor.
Table 1. Demographic and clinical characteristics

Page 5

of women in each group having a T-score <-1.0 and <-2.5
at either anatomical site. The proportion of women
with osteoporosis (T-score <-2.5) was 0.7%, 4.0%,
4.0% and 7.3% in the HIV-negative women, ART-
naive HIV+women, non-PI HAART users and PI
HAART users, respectively, with the difference
between groups significant only for the comparison of
HIV-negative women with PI HAART users.
Table 5 shows unadjusted and adjusted ORs for
factors significantly associated with LS or FN T-scores
<-1.0 and <-2.5. In unadjusted analyses, the HIV+
women were fourfold to eightfold more likely than the
HIV-negative women to have a T-score <-1.0. In multi-
variate analyses that included race, nadir BMI, HIV
and hepatitis C serostatus, use of HAART, lifetime
exposure to smoking and menopause, a T-score <-1.0
was predicted by HIV+serostatus, use of PI-containing
HAART, lower nadir BMI and post-menopausal status.
The ART-naive HIV+women (OR 4.58, P=0.003) and
PI users (OR=3.57, P=0.009) were significantly more
likely than the HIV-negative women to have T-scores
<-1.0.The OR of 2.43 associated with use of non-PI
HAART was not significant (P=0.075). Hepatitis C
was not a significant predictor after adjustment for
HIV infection. The magnitude of the ORs was similar
for factors associated with T-scores <-2.5, but with the
smaller sample size only menopausal status and nadir
BMI were statistically significant (OR 4.81 and 0.76,
P=0.025 and 0.001, respectively).
Traditional risk factors for low BMD
Because of the differences by group in the proportion
of women reporting post-menopausal status
(Table 1), and its profound effect on BMD, we
performed age-matched analysis among pre-
menopausal women by pair-wise comparisons in the
four groups defined by HIV serostatus and treatment
category, adjusted for nadir BMI, race and years of
smoking. The differences in BMD were slightly miti-
gated in these bivariate analyses. ART-naive women
had lower BMD than HIV-negative women: 1.26
versus 1.32 g/cm2, P=0.045 at LS, and 1.05 versus
1.11, P=0.059 at FN. PI users had BMD that was
7–9% lower compared with HIV-negative women
(1.23 versus 1.32 g/cm2, P=0.019, and 1.00 versus
1.10 g/cm2, P=0.001 at LS and FN, respectively), and
compared with the non-PI HAART users at FN (1.00
versus 1.07 g/cm2, respectively, P=0.007), and 5%
lower than in ART-naive women (1.00 versus
1.06 g/cm2, P=0.060). We also conducted an age-
matched analysis of amenorrhoeic women, with
similar findings (data not shown).
We assessed as possible confounders the presence of
amenorrhoea, the misclassification of pre-menopausal
women as post-menopausal because of hysterectomy
without bilateral oophorectomy, and the use of calcium
supplements. Only five women with amenorrhoea for
>1 year prior to the DXA had not reported themselves
to be post-menopausal. Twenty of the thirty-five
Antiviral Therapy 12:7
1053
HIV, HAART and bone mineral density in women
Median duration of use among
participants who had use, months DrugCurrent use, n (%)Past use, n (%)
Nucleoside RTIs
Zidovudine
Lamivudine
Stavudine
Abacavir
Didanosine
Nucleotide RTI
Tenofovir
Non-nucleoside RTIs
Nevirapine
Efavirenz
Delavirdine
Protease inhibitors
Saquinavir
Ritonavir
Indinavir
Nelfinavir
Amprenavir
Lopinavir*
Atazanavir
88 (44.4)
147 (74.2)
62 (31.3)
54 (27.3)
27 (13.6)
146 (73.7)
187 (94.4)
124 (62.6)
74 (37.4)
78 (39.4)
30
36
36
18
18
57 (28.8) 59 (29.8)12
34 (17.2)
53 (26.8)
2 (1.0)
72 (36.4)
79 (39.9)
5 (2.5)
21
24
45
6 (3.0)
18 (9.1)
10 (5.1)
32 (16.2)
4 (2.0)
30 (15.2)
12 (6.1)
34 (17.2)
50 (25.3)
57 (28.8)
88 (44.4)
11 (5.6)
34 (17.2)
14 (7.1)
48
18
45
30
27
15
6
Sample was 198 women using HAART at time of dual X-ray absorptiometry. *Boosted with ritonavir. RTI, reverse transcriptase inhibitor.
Table 2. Proportion of HAART users on specific antiretroviral agents and median duration of use