Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2019 Feb 1.
Published in final edited form as: Pharmacoepidemiol Drug Saf. 2017 Dec 12;27(2):182–190. doi: 10.1002/pds.4362

Effects of Analgesics on Bone Mineral Density: a Longitudinal Analysis of the Prospective SWAN Cohort with Three-group Matching Weights

Kazuki Yoshida 1,2,3, Zhi Yu 3,4, Gail A Greendale 4, Kristine Ruppert 5, Yinjuan Lian 5, Sara K Tedeschi 3, Tzu-Chieh Lin 3, Sebastien Haneuse 2, Robert J Glynn 2,6, Sonia Hernandez-Diaz 1, Daniel H Solomon 3,6
PMCID: PMC5799005  NIHMSID: NIHMS932150  PMID: 29230890

Abstract

PURPOSE

To examine the effects of analgesics on bone mineral density (BMD), which have not been examined in a longitudinal study with multiple measurements.

METHODS

We investigated changes in BMD associated with new use of analgesics in a prospective longitudinal cohort of mid-life women. BMD and medication use were measured annually. We compared BMD among new users of acetaminophen, NSAIDs, and opioids. Adjustment for baseline covariates was conducted through propensity score matching weights. On-treatment analysis was conducted with inverse probability of censoring weights. Analysis based on the initial treatment group was also conducted to provide insights into selection bias. Repeated BMD measurements were examined with generalized estimating equations.

RESULTS

We identified 71 acetaminophen new users, 659 NSAID new users, and 84 opioid new users among 2,365 participants. In the on-treatment analysis, the opioid group in comparison to the acetaminophen group had an additional average BMD decline of −0.06% [−1.24, 1.11] per year in the spine and −0.45% [−1.51, 0.61] per year in the femoral neck. BMD mean trajectories over time suggested a fifth-year decline in the opioid persistent users compared to other two groups. In the initial treatment group analysis, all three groups showed similar trajectories.

CONCLUSION

The BMD decline over time was similar among the three groups. However, five years of continuous opioid use may be associated with a greater BMD decline than five years on other analgesics. Further studies examining the relationship between very long term persistent opioid use and BMD are warranted.

Keywords: analgesics, bone mineral density, osteoporosis, propensity score, matching weights

INTRODUCTION

Many classes of drugs have been linked to increased risk of fractures or reduced bone mineral density (BMD). Histamine H2 receptor antagonists1, opioids, and anticonvulsants2 were among them. The cross-sectional nature of these studies, however, limit the assessment of temporality. Additionally, long-term effects of analgesics on bone health are not well understood.

Opioids have been associated with increased fracture risks in multiple longitudinal studies3–8. The increased risks occurring soon after initiation4,6 suggest the primary mechanism is through acute neurologic effects, such as gait imbalance. However, chronic opioid use may also have indirect effects via endocrine changes9–11; for example, hypogonadotropic hypogonadism has been found in patients receiving methadone maintenance therapy12. Several studies also suggest lower BMD in opioid users13–15. However, these studies were cross-sectional, had limited control of confounding, and focused on a particular subset of chronic opioid users (i.e., former heroin addicts on methadone maintenance).

NSAID use has been associated with higher BMD in two cross-sectional studies16,17 adjusted for potential confounders such as body weight, although a more recent study found increased fracture risk among NSAID users despite stable BMD5. Both selective and non-selective NSAIDs exhibit anti-inflammatory properties through inhibiting cyclooxygenase(COX)-2, an enzyme that plays a role in prostaglandins synthesis. Prostaglandins, in turn, play important roles in both bone formation and bone resorption18.

The Study of Women’s Health Across the Nation (SWAN)19,20 allows for rigorous assessment of the effects of analgesics on BMD because of its longitudinal design and repeatedly measured BMD. We hypothesized that opioids and NSAIDs are associated with BMD reduction compared to acetaminophen (active control). Presence of three treatment groups of quite different sizes as well as frequent treatment changes posed challenges in analysis. Thus, we used recently proposed matching weights in a multiple group setting21,22 along with inverse probability of censoring weights over time23.

METHODS

Study population and design

SWAN19,20, a prospective longitudinal, community-based cohort study of mid-life women, enrolled participants in their pre-menopause between 1996 and 1997 from 8 U.S. sites to observe the natural history of menopause. Eligibility criteria included age 42–52 years old, at least one menstrual period within the past three months, and no hormonal medication use within the last three months. The SWAN BMD substudy enrolled 2,365 women of four racial/ethnic groups (1,177 Caucasian, 665 African American, 273 Japanese, and 250 Chinese) with approximately annual BMD measurement. Longitudinal follow-up is still ongoing, and SWAN data collection consists of physical measures, fasting morning blood draw, interviewer-administered and self-administered questionnaires (completed at home or in clinic). Participants gave written informed consent and study sites obtained institutional review board approval.

Exposure assessment

The exposure of interest was the type of analgesic–opioid, NSAID including COX-2 selective inhibitors, and/or acetaminophen–that participants took at ≥ 2 consecutive annual visits. The individual-specific baseline visit was defined as the visit immediately before the first of these consecutive visits. Medication use, including both prescription and over the counter (OTC), was ascertained through interviewer-administered questionnaire for medications used twice or more per week during the past month and was then verified by inspection of medication containers. The exposure definition was constructed hierarchically (eTable 1): opioid user if an opioid is used regardless of the other two; NSAID user if an NSAID is used but not opioids regardless of acetaminophen; and acetaminophen user if it is the only analgesic used. Participants who transitioned between these exposure categories were assigned the exposure status at the time when they first met the eligibility criteria.

Outcome assessment

Details of the BMD measurements have been described in previous studies using SWAN24–26. BMD (g/cm2) was measured in the lumbar spine and femoral neck at each study visit. Raw BMD measurements were converted to baseline-normalized %BMD values for interpretability, as regularly done in major osteoporosis clinical trials27–30. That is, for each individual, the outcome was defined as 100% at the individual-specific baseline visit when the covariates were ascertained (year 0), and subsequent values were described in relation to this baseline value (e.g., 96% of the baseline value at year 4). Follow-up was truncated at year 5 because very few people remained in the initial treatment categories beyond that point.

Covariate assessment

Covariates were assessed at the individual-specific baseline visit. Body mass index (BMI) was calculated from height and weight at the study baseline. The demographic variables included age, race/ethnicity, self-reported annual income (low [≤ $19,999], medium [$20,000–49,999], and high [≥ $50,000-]), and college education (yes/no). Alcohol intake (none/low [< 1 drink/month], moderate [up to 1/week], and high [≥ 2/week]), current tobacco use (yes/no), and physical activity measures were available as lifestyle variables. Physical activity was measured using the modified Baecke Physical Activity Questionnaire (range 3–15, with lower scores indicating less exercise)31,32. Self-reported comorbidities included thyroid disease, diabetes, and history of cancer. Self-reported pain-related quality of life (range 0–100, with 100 indicating excellent quality of life33), vasomotor symptoms, and overall perception of health were also reported. Medications included hormone therapy for menopause, bisphosphonates, calcium supplements, vitamin D supplements, and oral glucocorticoids. Menopause transition (MT) stage was defined based on menstrual cycles25 (eTable 2). We created four categories of MT stages for the main analysis: pre- or early perimenopause; late perimenopause; postmenopause; and unknown (eTable 2). We also conducted a subgroup analysis among those who had a known date of the final menstrual period (FMP), using MT stages based on time prior to or after the FMP (eTable 3).24

Statistical analyses

Participant characteristics at the study baseline were summarized within each exposure group. To examine between group imbalance in the unmatched cohort of patients, the standardized mean differences (SMD)34 were calculated in each pairwise treatment contrast and then averaged across all three contrasts. The SMD represents how different groups are for a given covariate. Covariates that have SMD ≤ 0.1 are considered reasonably balanced34. We multiply imputed missing covariates via the mice R package.35,36

Multinomial logistic regression was used for the propensity score (PS) model because the exposure status had three categories (acetaminophen, NSAIDs, or opioids)37, resulting in one PS for each exposure category. All baseline covariates listed in the baseline table (Table 1) were included as explanatory variables. We used the PSs as matching weights (MW; eAppendix Methods), a PS weighting method proposed by Li and Greene21. A recent study generalized MW to multiple treatment group settings22. Compared to 1:1:1 PS matching, MW allows for retention of all subjects, which is a potential advantage when the group sizes are dissimilar. Compared to the conventional inverse probability of treatment weights (IPTW), the target of inference focuses on those who are in clinical equipoise among all drugs (i.e., similar estimand to PS matching). This clinical equipoise estimand was more stably estimated in the settings where baseline covariates were more different among groups22.

Table 1.

Baseline characteristics of analgesics new users before propensity score weighting.

APAP users NSAID users Opioid users SMD
N 71 659 84
Age (mean (SD)) 49.34 (4.33) 49.43 (4.02) 50.63 (4.42) 0.200
Ethnicity (%) 0.493
 Caucasian 33 (46.5) 385 (58.4) 40 (47.6)
 Black 25 (35.2) 191 (29.0) 43 (51.2)
 Asian 13 (18.3) 83 (12.6) 1 (1.2)
Income (%) 0.383
 Low (-19k) 9 (15.5) 46 (7.9) 13 (20.0)
 Middle (20k–49k) 16 (27.6) 203 (34.7) 28 (43.1)
 High (50k-) 33 (56.9) 336 (57.4) 24 (36.9)
College education (%) 26 (36.6) 302 (46.2) 23 (27.7) 0.260
BMI (mean (SD)) 28.43 (8.23) 29.16 (7.19) 32.43 (7.25) 0.354
Physical activity [3–15] (mean (SD)) 7.49 (1.51) 7.87 (1.65) 7.08 (2.08) 0.295
Vasomotor symptoms (%) 34 (49.3) 341 (52.2) 47 (56.0) 0.089
Menopause transition stage (%) 0.318
 Pre/Early Peri 45 (63.4) 458 (69.5) 41 (48.8)
 Late Peri 3 (4.2) 43 (6.5) 6 (7.1)
 Post 16 (22.5) 108 (16.4) 26 (31.0)
 Unknown 7 (9.9) 50 (7.6) 11 (13.1)
Lumbar spine BMD g/cm2 (mean (SD)) 1.04 (0.14) 1.08 (0.15) 1.11 (0.16) 0.291
Femoral neck BMD g/cm2 (mean (SD)) 0.81 (0.13) 0.85 (0.14) 0.88 (0.14) 0.340
Pain-related QoL [0–100] (mean (SD)) 70.44 (18.26) 69.75 (19.74) 48.77 (25.35) 0.647
Overall perception of health (%) 0.386
 Same 28 (42.4) 270 (43.1) 30 (40.0)
 Better 31 (47.0) 297 (47.4) 22 (29.3)
 Worse 7 (10.6) 60 (9.6) 23 (30.7)
Alcohol (%) 0.177
 None/Low 34 (56.7) 285 (47.1) 34 (50.7)
 Moderate 16 (26.7) 168 (27.8) 21 (31.3)
 High 10 (16.7) 152 (25.1) 12 (17.9)
Current smoker (%) 14 (20.3) 96 (14.7) 21 (25.0) 0.173
Thyroid disease (%) 9 (13.4) 66 (10.1) 10 (11.9) 0.069
Diabetes (%) 6 (8.5) 34 (5.2) 16 (19.0) 0.292
Calcium supplement (%) 21 (29.6) 216 (32.8) 18 (21.4) 0.171
Vitamin D supplement (%) 15 (21.1) 105 (15.9) 8 (9.5) 0.218
Oral glucocorticoids (%) 3 (4.2) 14 (2.1) 2 (2.4) 0.080
Open in a new tab

Missing proportions: BMI 7%; Income 13%; College education 1%; Physical activity 52% (not measured at every visit by design); Vasomotor symptoms 1%; BMD 10%; Pain-related QoL 15%; Alcohol 11%; Smoking 1%; Cancers 1%; Thyroid disease 1%.

Abbreviations: APAP: acetaminophen; NSAID: non-steroidal anti-inflammatory drug; SMD: standardized mean difference; BMI: body mass index; Menopausal status: menopausal status define by menstrual cycles (See eTable 2); BMD: bone mineral density in g/cm2; QoL: quality of life.

MW, as it is known currently, is only applicable to time-invariant exposure. However, a drug exposure is typically time-varying. Therefore, we used the on-treatment analysis and initial treatment group analysis to make treatment group assignment effectively time-invariant. The main analysis was on-treatment analysis of those who remained in the initial treatment category (adherers). That is, those who deviated from their initial category were censored at the time of deviation, making the treatment assignment effectively time-invariant among uncensored time points remaining in the analysis dataset. We additionally censored patients at the initiation of hormone therapy for menopause or bisphosphonate or cancer diagnosis. Such censoring of participants who deviate from the initial treatment status or started bone active medications can introduce selection bias –those who are censored and retained may not share the same risks for BMD changes. Thus, we additionally assigned time-varying inverse probability of censoring weights (IPCW)23 to ameliorate this selection bias issue using the same set of covariates as the time-invariant MW model, but updated for each time point. A final weight for a given time point was constructed as the product of the individual-specific time-invariant MW and the individual-specific, time point-specific time-varying IPCW and was normalized to represent the sample size of each treatment group at each time point.38 This approach should estimate the effect of continuous treatment,23,39 assuming both MW for baseline confounding by indication and IPCW for selection bias introduced by artificial censoring are successful. We also conducted an alternative analysis based on the initial treatment category at the study baseline (initial treatment group analysis). Participants remained in their original treatment category regardless of subsequent medication changes in this analysis, also making the treatment variable time-invariant. This approach is an observational analogue of the intention-to-treat analysis used in clinical trials, and should estimate the effect of assigned treatment39, assuming MW for baseline confounding by indication is successful. Censoring also occurred administratively because some subjects started analgesics late in the SWAN study, thus, reaching the latest SWAN visit (visit 13) before having the fifth-year visit after analgesic initiation. This type of administrative censoring was assumed non-informative.

The mean baseline-normalized %BMD over time for the spine and femoral neck were plotted in both the on-treatment analysis and initial treatment group analysis. We used the generalized estimating equation with the auto-regressive correlation structure to account for weighting and the clustering of repeated BMD measurements within each individual during follow-up. Confidence intervals were calculated based on robust sandwich standard error estimates. The time effect on the mean baseline-normalized %BMD was modeled as a linear term to provide average yearly change estimates. The slope differences of interest, NSAIDs versus acetaminophen, and opioids versus acetaminophen, were incorporated into the model as time-group interaction terms. We repeated the analyses in the FMP subgroup. We also repeated the main analysis after excluding an outlying data point as a sensitivity analysis. Another sensitivity analysis for the outcome model further adjusted for variables that had SMD > 0.1 after balancing by MW.

RESULTS

Study population

Among 2,365 participants in the SWAN BMD cohort, 71 acetaminophen new users, 659 NSAID new users, and 84 opioid new users were identified (eFigure 1; break down by generic names in eTable 4). Their unadjusted baseline characteristics are shown in Table 1. The most prominent baseline differences were noted for pain-related quality of life (QoL), ethnic composition, income, overall perception of health, BMI, femoral neck BMD, and physical activity. The pain-related QoL was lower for the opioid users (48.8) compared to the other two groups that had scores around 70. Femoral neck BMD was higher in the opioid group than the other groups likely associated with their higher BMI. Physical activity was highest among NSAID users and was lowest among opioid users. Twenty-six percent of NSAID users were also exposed to acetaminophen. Opioid users also had substantial concurrent exposure (acetaminophen 80% and NSAIDs 66%). Matching weights reduced group imbalance at the baseline (Table 2), even in comparison to other PS methods (eFigure 2)40,41. The mean follow-up durations were similar across treatment groups (eTable 5).

Table 2.

Baseline characteristics of analgesics new users after propensity score weighting.

APAP users NSAID users Opioid users SMD
Age (mean (SD)) 49.25 (3.72) 49.66 (4.23) 49.73 (4.27) 0.086
Ethnicity (%) 0.103
 Caucasian 53.9 54.0 50.6
 Black 41.7 43.4 46.9
 Asian 4.4 2.6 2.5
Income (%) 0.121
 Low (-19k) 16.3 19.4 20.2
 Middle (20k–49k) 33.7 34.2 37.6
 High (50k-) 50.0 46.5 42.2
College education (%) 37.0 32.0 31.1 0.085
BMI (mean (SD)) 30.79 (7.87) 30.56 (7.35) 30.66 (6.54) 0.031
Physical activity [3–15] (mean (SD)) 7.25 (1.63) 7.26 (1.67) 7.45 (1.92) 0.089
Vasomotor symptoms (%) 46.7 54.6 55.7 0.125
Menopause transition stage (%) 0.155
 Pre/Early Peri 61.8 58.8 57.1
 Late Peri 7.2 5.3 9.2
 Post 21.5 22.0 21.2
 Unknown 9.5 13.9 12.6
Lumbar spine BMD g/cm2 (mean (SD)) 1.08 (0.14) 1.07 (0.15) 1.08 (0.14) 0.066
Femoral neck BMD g/cm2 (mean (SD)) 0.85 (0.13) 0.84 (0.13) 0.86 (0.12) 0.074
Pain-related QoL [0–100] (mean (SD)) 65.17 (17.45) 62.66 (19.84) 63.84 (20.99) 0.088
Overall perception of health (%) 0.092
 Same 40.4 44.8 39.7
 Better 42.7 38.5 43.8
 Worse 16.9 16.7 16.5
Alcohol (%) 0.074
 None/Low 52.0 50.0 49.2
 Moderate 32.8 32.2 34.0
 High 15.3 17.8 16.8
Current smoker (%) 19.8 26.1 23.9 0.102
Thyroid disease (%) 10.2 14.6 12.4 0.091
Diabetes (%) 12.8 13.4 12.2 0.030
Calcium supplement (%) 16.5 22.6 21.8 0.106
Vitamin D supplement (%) 8.4 12.7 12.6 0.098
Oral glucocorticoids (%) 2.9 2.3 1.3 0.079
Open in a new tab

Abbreviations: APAP: acetaminophen; NSAID: non-steroidal anti-inflammatory drug; SMD: standardized mean difference; BMI: body mass index; Menopausal status: menopausal status define by menstrual cycles (See eTable 2); BMD: bone mineral density in g/cm2; QoL: quality of life.

Adjusted main analysis using menstrual period-defined stages

Figure 1 shows the mean baseline-normalized BMD over the five-year follow-up period for each treatment group (n = 814) as well as the treatment group contrasts from the generalized estimating equation (see eFigure 3 for unadjusted counterpart). The mean annual change in each treatment group as well as group differences in slopes are shown in Table 3.

Figure 1. Group mean trajectories of baseline-normalized % bone mineral density (BMD).

Figure 1

Open in a new tab

The numbers at the bottom of each panel are number of individuals contributing BMD measurements (Top: Acetaminophen, Middle: NSAIDs, and bottom: Opioids). On-treatment analysis censored patients at the time they changed analgesic categories, whereas initial treatment group analysis retained these patients in the initial treatment groups.

Abbreviations: NSAID: non-steroidal anti-inflammatory drug; N vs A: NSAID group vs Acetaminophen group; O vs A: Opioid group vs Acetaminophen group; Spine: lumbar spine BMD; Time Since Baseline: Time since the baseline visit in years.

Table 3.

Main bone mineral density analysis results from generalized estimating equations.

Analysis Type Site Group Mean Annual Change (%) Group Difference (%)
On Treatment Spine Acetaminophen −0.90 [−1.58, −0.21] Ref.
NSAIDs −0.76 [−0.92, −0.59] 0.14 [−0.56, 0.85]
Opioids −0.96 [−1.92, −0.00] −0.06 [−1.24, 1.11]
Femoral Neck Acetaminophen −0.61 [−1.21, −0.02] Ref.
NSAIDs −0.60 [−0.81, −0.39] 0.02 [−0.61, 0.64]
Opioids −1.07 [−1.95, −0.19] −0.45 [−1.51, 0.61]
Initial Treatment Spine Acetaminophen −0.72 [−1.13, −0.30] Ref.
NSAIDs −0.79 [−0.93, −0.66] −0.07 [−0.51, 0.36]
Opioids −0.66 [−1.25, −0.07] 0.06 [−0.66, 0.78]
Femoral Neck Acetaminophen −0.82 [−1.28, −0.36] Ref.
NSAIDs −0.64 [−0.80, −0.49] 0.18 [−0.30, 0.66]
Opioids −0.74 [−1.31, −0.16] 0.08 [−0.65, 0.82]
Open in a new tab

On-treatment analysis censored patients at the time they changed analgesic categories, whereas initial treatment group analysis retained these patients in the initial treatment groups.

Abbreviations: NSAID: non-steroidal anti-inflammatory drug; Ref.: Reference.

The on-treatment analysis (Figure 1, left panels) was suggestive of a greater decline in BMD in the opioid group compared to the acetaminophen group, principally at the fifth year. The opioid group in comparison to the acetaminophen group had an additional mean BMD decline of −0.06% [−1.24, 1.11] per year in the spine and −0.45% [−1.51, 0.61] per year in the femoral neck. The initial treatment group analysis, on the other hand, demonstrated more similar trajectories for all three groups (Figure 1, right panels). The difference between the opioid group and the acetaminophen group diminished to −0.06% [−0.66, 0.78] in the spine and to 0.08% [−0.65, 0.82] in the femoral neck.

Adjusted final menstrual period-based analysis

eFigure 4 shows the corresponding outcome analysis in the subgroup of women with a known FMP date (n = 471). The adjustment for the menopause transition stages at individual-specific baseline visit was based on the time prior to or after FMP (pre-transmenopause, transmenopause, or postmenopause; eTable 3)24. The baseline characteristics before propensity score weighting are in eTable 6. Propensity score weighing improved covariates balance, but to a lesser extent than in the main cohort (eTable 7). The mean trajectories were less stable due to the smaller sample size, particularly in the on-treatment analyses. The mean annual change in each treatment group is shown in eTable 8. The on-treatment analyses exhibited overlapping mean trajectories (eFigure 4, left panels). The initial treatment group analyses (eFigure 4, right panels) produced trajectories with more separation than the main initial treatment group analyses (Figure 1, right panels).

Sensitivity analysis

As the main on-treatment analysis showed a strong fifth-year deflection in the trajectory, we examined for the presence of outliers. One subject with probable thyroid disease exhibited an outlying decline trajectory. This subject remained in the opioid category for the full five years without meeting any of the censoring criteria, thus, she was gradually up-weighted over time via IPCW, becoming more influential. Reanalysis excluding this subject (eFigure 5) resulted in a less prominent decline in the fifth year, although the opioid group remained the lowest group at the fifth year. Outcome analysis further adjusting for the sub-optimally balanced variables gave similar estimates of group differences in slopes (eTable 9).

DISCUSSION

In the current study, we examined the association between analgesic use and BMD decline over time in a well-established cohort of mid-life women, with a focus on the contrasts between opioids and acetaminophen as well as NSAIDs and acetaminophen. We used three-group MW for baseline covariate balancing and time-varying IPCW to reduce selection bias by artificial censoring over time. To our knowledge, the current study is the first instance of MW used in conjunction with IPCW in the multiple treatment group setting. The average slope differences were not statistically significant in both on-treatment analysis and initial treatment group analysis. However, the on-treatment analysis was suggestive of a potentially greater decline in the BMD in the opioid group compared to the acetaminophen group after five years of continuous use. The trajectory of BMD decline in the NSAID group was similar to the acetaminophen group. Between-group differences were not clearly observed in the initial treatment group analysis.

There is no established gold standard for the clinically meaningful group difference in BMD changes over time, however, several clinical trials were summarized in eTable 10 to give some idea27–30. In the FIT study27, which demonstrated hip fracture reduction, the annual slope difference in the femoral neck BMD was +1.0%/year in the alendronate group compared to the placebo group. Our study found that the annual slope difference was −0.45%/year [−1.51, 0.61] for the femoral neck BMD comparing the opioid new users to the acetaminophen new users (Figure 1), which was not statistically significant, but did not rule out 1.0% difference in annual slopes. The five-year difference in BMD comparing the opioid group to the acetaminophen group was close to −10% in the on-treatment analysis although the difference was negligible in the initial treatment group analysis (Figure 1). The noticeable discrepancy between the on-treatment analysis and the initial treatment group analysis suggests the contribution of residual selection bias that was not fully controlled by IPCW, likely due to the small size of the opioid arm that remained on treatment, in addition to the exposure misclassification in the initial treatment group analysis. However, even in the sensitivity analysis removing an outlying observation, some group difference in the range of −3 to −5% remained, which may suggest a potentially greater decline in BMD among persistent opioid users.

Although the longitudinal association of opioid use and fractures has been well documented in multiple studies3–8, the association of opioid use and lower BMD has been shown only in cross-sectional studies13–15. To our knowledge, only one study has examined the longitudinal effect of opioids on BMD5 and reported no clinically relevant longitudinal association based on BMD measurements ten years apart. Our study provides additional insight into the potential effect of opioids on BMD by providing more granular follow-up, although this study alone is not conclusive. Past cross-sectional studies that demonstrated an association between opioid use and lower BMD were among former opioid abusers undergoing methadone therapy, whereas the current study was among community-dwelling healthy women.

Several studies have suggested potentially beneficial effects of NSAIDs on BMD. Bauer et al. found a cross-sectional association between higher BMD and current frequent NSAID use compared to infrequent use and non-use in their 1996 study on community-dwelling women aged at least 65 years old16. Carbone et al.17 examined the cross-sectional association between NSAID use and BMD in the Health ABC study among community-dwelling men and women 70–79 years of age. They found that current users of COX-2 selective NSAIDs with concurrent aspirin use had higher BMD than non-users. A 10-year longitudinal study by Vestergaard et al.5, which also examined acetaminophen, NSAIDs, and opioids, found a very minor (clinically insignificant) increase in spine and whole body BMD among NSAID users compared to non-users. The current study showed essentially identical BMD trajectories between NSAID users and acetaminophen users in both the on-treatment analysis and the initial treatment group analysis.

SWAN was designed to characterize the biological, symptomatic, and psychosocial changes that occur during the menopausal transition and their effects on women’s health and well-being. Thus, our findings may not generalize to men, or to women in different age ranges. SWAN did not specifically enroll analgesic users, thus, the number of users was small, limiting our ability to draw firm conclusions. Also SWAN does not have reliable medication dosage information. Doses of opioids can be highly variable among opioid users due to the highly individualized nature of these prescriptions42. However, high-dose opioid use is unlikely in this population cohort of generally healthy mid-life women.

Our longitudinal study design has some unique strengths compared to the prior cross-sectional studies on this topic. Use of acetaminophen as a comparator medication–active comparator design43– ensured that all three treatment groups had at least some pain. Non-users – individuals who do not use analgesics – are expected to have much less pain than analgesic users, thus using such a comparator group without pain could induce a spurious association between BMD changes and medication use44, which can be difficult to control for. We also used a new user design43, which examines subjects starting the medication of interest, in an attempt to parallel the design of a hypothetical clinical trial45 and ensures that the baseline covariates were measured before medication initiation.

As a safety outcome study, the primary effect of interest is the on-treatment effect39, that is the effect of medication on the outcome if subjects were made to adhere to the regimen23. However, the naïve on-treatment analysis that simply censors subjects who do not follow the initial regimen of interest often introduces selection bias46. Therefore, we used IPCW to account for selection. The study revealed a difficulty of IPCW in the presence of small number of subjects in each arm. One of the few persistent opioid users happened to have an outlying decline in BMD, thereby exerting increasing influence at later time points because of progressively greater IPCW. Some of the differences in BMD trajectories, however, persisted after excluding this subject. Examination of the very long-term on-treatment effect beyond 5 years was not possible due to the very few adherers, potentially limiting the scope of the study.

In conclusion, the average BMD slope differences over a five-year period were not statistically significant among mid-life female analgesic new users. However, five years of persistent opioid use may be associated with a greater BMD decline. It is important to remember that chronic opioid use, although becoming common, is not a well-justified practice in the setting of non-cancer pain47,48. Further studies examining the relationship between very long term persistent opioid use and BMD as well as their dose response are warranted.

Supplementary Material

Supp info

KEY POINTS.

  1. Three-group matching weights are modified propensity score weights that emulate three-way simultaneous matching.

  2. Matching weights balanced covariates reasonably across three groups with size imbalance.

  3. We observed a potential decrease in bone mineral density among mid-life women who remained on opioids for five years compared to users of acetaminophen.

Acknowledgments

ACKNOWLEDGEMENT (SWAN 12/02/2016 long form):

The Study of Women’s Health Across the Nation (SWAN) has grant support from the National Institutes of Health (NIH), DHHS, through the National Institute on Aging (NIA), the National Institute of Nursing Research (NINR) and the NIH Office of Research on Women’s Health (ORWH) (Grants U01NR004061; U01AG012505, U01AG012535, U01AG012531, U01AG012539, U01AG012546, U01AG012553, U01AG012554, U01AG012495). The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the NIA, NINR, ORWH or the NIH.

Clinical Centers: University of Michigan, Ann Arbor – Siobán Harlow, PI 2011 – present, MaryFran Sowers, PI 1994–2011; Massachusetts General Hospital, Boston, MA – Joel Finkelstein, PI 1999 – present; Robert Neer, PI 1994 – 1999; Rush University, Rush University Medical Center, Chicago, IL – Howard Kravitz, PI 2009 – present; Lynda Powell, PI 1994 – 2009; University of California, Davis/Kaiser – Ellen Gold, PI; University of California, Los Angeles – Gail Greendale, PI; Albert Einstein College of Medicine, Bronx, NY – Carol Derby, PI 2011 – present, Rachel Wildman, PI 2010 – 2011; Nanette Santoro, PI 2004 – 2010; University of Medicine and Dentistry – New Jersey Medical School, Newark – Gerson Weiss, PI 1994 – 2004; and the University of Pittsburgh, Pittsburgh, PA – Karen Matthews, PI.

NIH Program Office: National Institute on Aging, Bethesda, MD – Chhanda Dutta 2016- present; Winifred Rossi 2012–2016; Sherry Sherman 1994 – 2012; Marcia Ory 1994 – 2001; National Institute of Nursing Research, Bethesda, MD – Program Officers.

Central Laboratory: University of Michigan, Ann Arbor – Daniel McConnell (Central Ligand Assay Satellite Services).

Coordinating Center: University of Pittsburgh, Pittsburgh, PA – Maria Mori Brooks, PI 2012 - present; Kim Sutton-Tyrrell, PI 2001 – 2012; New England Research Institutes, Watertown, MA - Sonja McKinlay, PI 1995 – 2001.

Steering Committee: Susan Johnson, Current Chair

Chris Gallagher, Former Chair

We thank the study staff at each site and all the women who participated in SWAN.

FUNDING INFORMATION:

KY receives tuition support jointly from Japan Student Services Organization (JASSO) and Harvard T.H. Chan School of Public Health (partially supported by training grants from Pfizer, Takeda, Bayer and PhRMA).

SKT receives salary support from the Lupus Foundation of America and NIH-L30 AR070514.

DHS receives salary support from NIH-K24AR055989.

RJG received research support in the form of grants to his institution for clinical trial design, monitoring, and analysis from AstraZeneca, Kowa, Novartis, and Pfizer.

DHS receives salary support from institutional research grants from Eli Lilly, Amgen, Pfizer, AstraZeneca, Genentech, and Corrona. He also receives royalties from UpToDate, and serves in unpaid roles in studies funded by Pfizer and Eli Lilly.

Footnotes

CONTRIBUTORSHIP (tentative and subject to change):

Study design: KY, GAG, KR, and DHS;

Statistical analysis planning: KY, KR, SH, RJG, SHD, and DHS;

Data preparation and analysis: KY, ZY, and YL;

Data interpretation: KY, GAG, KR, SKT, TCL, and DHS

Drafting/revising manuscript: All authors

Approving final manuscript: All authors

KY takes responsibility for the integrity of the data analysis.

References

  • 1.Kinjo M, Setoguchi S, Solomon DH. Antihistamine therapy and bone mineral density: analysis in a population-based US sample. Am J Med. 2008;121:1085–1091. doi: 10.1016/j.amjmed.2008.06.036. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Kinjo M, Setoguchi S, Schneeweiss S, Solomon DH. Bone mineral density in subjects using central nervous system-active medications. Am J Med. 2005;118:1414. doi: 10.1016/j.amjmed.2005.07.033. [DOI] [PubMed] [Google Scholar]
  • 3.Teng Z, Zhu Y, Wu F, et al. Opioids contribute to fracture risk: a meta-analysis of 8 cohort studies. PLoS ONE. 2015;10:e0128232. doi: 10.1371/journal.pone.0128232. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Li L, Setoguchi S, Cabral H, Jick S. Opioid use for noncancer pain and risk of fracture in adults: a nested case-control study using the general practice research database. Am J Epidemiol. 2013;178:559–569. doi: 10.1093/aje/kwt013. [DOI] [PubMed] [Google Scholar]
  • 5.Vestergaard P, Hermann P, Jensen J-EB, Eiken P, Mosekilde L. Effects of paracetamol, non-steroidal anti-inflammatory drugs, acetylsalicylic acid, and opioids on bone mineral density and risk of fracture: results of the Danish Osteoporosis Prevention Study (DOPS) Osteoporos Int. 2012;23:1255–1265. doi: 10.1007/s00198-011-1692-0. [DOI] [PubMed] [Google Scholar]
  • 6.Miller M, Stürmer T, Azrael D, Levin R, Solomon DH. Opioid analgesics and the risk of fractures in older adults with arthritis. J Am Geriatr Soc. 2011;59:430–438. doi: 10.1111/j.1532-5415.2011.03318.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Saunders KW, Dunn KM, Merrill JO, et al. Relationship of opioid use and dosage levels to fractures in older chronic pain patients. J Gen Intern Med. 2010;25:310–315. doi: 10.1007/s11606-009-1218-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Solomon DH, Rassen JA, Glynn RJ, Lee J, Levin R, Schneeweiss S. The comparative safety of analgesics in older adults with arthritis. Arch Intern Med. 2010;170:1968–1976. doi: 10.1001/archinternmed.2010.391. [DOI] [PubMed] [Google Scholar]
  • 9.Coluzzi F, Pergolizzi J, Raffa RB, Mattia C. The unsolved case of “bone-impairing analgesics”: the endocrine effects of opioids on bone metabolism. Ther Clin Risk Manag. 2015;11:515–523. doi: 10.2147/TCRM.S79409. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Katz N, Mazer NA. The impact of opioids on the endocrine system. Clin J Pain. 2009;25:170–175. doi: 10.1097/AJP.0b013e3181850df6. [DOI] [PubMed] [Google Scholar]
  • 11.Merza Z. Chronic use of opioids and the endocrine system. Horm Metab Res. 2010;42:621–626. doi: 10.1055/s-0030-1254099. [DOI] [PubMed] [Google Scholar]
  • 12.Hallinan R, Byrne A, Agho K, McMahon CG, Tynan P, Attia J. Hypogonadism in men receiving methadone and buprenorphine maintenance treatment. International Journal of Andrology. 2009;32:131–139. doi: 10.1111/j.1365-2605.2007.00824.x. [DOI] [PubMed] [Google Scholar]
  • 13.Grey A, Rix-Trott K, Horne A, Gamble G, Bolland M, Reid IR. Decreased bone density in men on methadone maintenance therapy. Addiction. 2011;106:349–354. doi: 10.1111/j.1360-0443.2010.03159.x. [DOI] [PubMed] [Google Scholar]
  • 14.Kim TW, Alford DP, Malabanan A, Holick MF, Samet JH. Low bone density in patients receiving methadone maintenance treatment. Drug Alcohol Depend. 2006;85:258–262. doi: 10.1016/j.drugalcdep.2006.05.027. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Milos G, Gallo LM, Sosic B, et al. Bone mineral density in young women on methadone substitution. Calcif Tissue Int. 2011;89:228–233. doi: 10.1007/s00223-011-9510-4. [DOI] [PubMed] [Google Scholar]
  • 16.Bauer DC, Orwoll ES, Fox KM, et al. Aspirin and NSAID use in older women: effect on bone mineral density and fracture risk. Study of Osteoporotic Fractures Research Group. J Bone Miner Res. 1996;11:29–35. doi: 10.1002/jbmr.5650110106. [DOI] [PubMed] [Google Scholar]
  • 17.Carbone LD, Tylavsky FA, Cauley JA, et al. Association between bone mineral density and the use of nonsteroidal anti-inflammatory drugs and aspirin: impact of cyclooxygenase selectivity. J Bone Miner Res. 2003;18:1795–1802. doi: 10.1359/jbmr.2003.18.10.1795. [DOI] [PubMed] [Google Scholar]
  • 18.Blackwell KA, Raisz LG, Pilbeam CC. Prostaglandins in bone: bad cop, good cop? Trends Endocrinol Metab. 2010;21:294–301. doi: 10.1016/j.tem.2009.12.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Sowers M, Crawford SL, Sternfeld B, et al. SWAN: a multi-center, multi-ethnic, community-based cohort study of women and the menopausal transition. In: Lobo RA, Kelsey J, Marcus R, editors. Menopause: Biology and Pathobiology. Academic Press; 2000. pp. 175–188. Available at: https://www.researchgate.net/publication/43196481_SWAN_A_Multicenter_Multiethnic_Community-Based_Cohort_Study_of_Women_and_the_Menopausal_Transition. Accessed January 17, 2017. [Google Scholar]
  • 20.Study of Women’s Health Across the Nation (SWAN) SWAN Study. Available at: http://www.swanstudy.org. Accessed September 22, 2015.
  • 21.Li L, Greene T. A weighting analogue to pair matching in propensity score analysis. Int J Biostat. 2013;9:215–234. doi: 10.1515/ijb-2012-0030. [DOI] [PubMed] [Google Scholar]
  • 22.Yoshida K, Hernandez-Diaz S, Solomon DH, et al. Matching Weights to Simultaneously Compare Three Treatment Groups: Comparison to Three-way Matching. Epidemiology. 2017;28:387–395. doi: 10.1097/EDE.0000000000000627. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Toh S, Hernán MA. Causal inference from longitudinal studies with baseline randomization. Int J Biostat. 2008;4 doi: 10.2202/1557-4679.1117. Article 22. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Greendale GA, Sowers M, Han W, et al. Bone mineral density loss in relation to the final menstrual period in a multiethnic cohort: results from the Study of Women’s Health Across the Nation (SWAN) J Bone Miner Res. 2012;27:111–118. doi: 10.1002/jbmr.534. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Solomon DH, Diem SJ, Ruppert K, et al. Bone mineral density changes among women initiating proton pump inhibitors or H2 receptor antagonists: a SWAN cohort study. J Bone Miner Res. 2015;30:232–239. doi: 10.1002/jbmr.2344. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Solomon DH, Ruppert K, Zhao Z, et al. Bone mineral density changes among women initiating blood pressure lowering drugs: a SWAN cohort study. Osteoporos Int. 2015 doi: 10.1007/s00198-015-3332-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Black DM, Cummings SR, Karpf DB, et al. Randomised trial of effect of alendronate on risk of fracture in women with existing vertebral fractures. Fracture Intervention Trial Research Group. Lancet. 1996;348:1535–1541. doi: 10.1016/s0140-6736(96)07088-2. [DOI] [PubMed] [Google Scholar]
  • 28.Black DM, Delmas PD, Eastell R, et al. Once-yearly zoledronic acid for treatment of postmenopausal osteoporosis. N Engl J Med. 2007;356:1809–1822. doi: 10.1056/NEJMoa067312. [DOI] [PubMed] [Google Scholar]
  • 29.Cummings SR, San Martin J, McClung MR, et al. Denosumab for prevention of fractures in postmenopausal women with osteoporosis. N Engl J Med. 2009;361:756–765. doi: 10.1056/NEJMoa0809493. [DOI] [PubMed] [Google Scholar]
  • 30.Neer RM, Arnaud CD, Zanchetta JR, et al. Effect of parathyroid hormone (1–34) on fractures and bone mineral density in postmenopausal women with osteoporosis. N Engl J Med. 2001;344:1434–1441. doi: 10.1056/NEJM200105103441904. [DOI] [PubMed] [Google Scholar]
  • 31.Baecke JA, Burema J, Frijters JE. A short questionnaire for the measurement of habitual physical activity in epidemiological studies. Am J Clin Nutr. 1982;36:936–942. doi: 10.1093/ajcn/36.5.936. [DOI] [PubMed] [Google Scholar]
  • 32.Sternfeld B, Ainsworth BE, Quesenberry CP. Physical activity patterns in a diverse population of women. Prev Med. 1999;28:313–323. doi: 10.1006/pmed.1998.0470. [DOI] [PubMed] [Google Scholar]
  • 33.Ware JE, Sherbourne CD. The MOS 36-item short-form health survey (SF-36). I. Conceptual framework and item selection. Med Care. 1992;30:473–483. [PubMed] [Google Scholar]
  • 34.Austin PC. An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies. Multivariate Behav Res. 2011;46:399–424. doi: 10.1080/00273171.2011.568786. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.van Buuren S, Groothuis-Oudshoorn K. mice: Multivariate Imputation by Chained Equations in R. Journal of Statistical Software. 2011;45:1–67. [Google Scholar]
  • 36.Schafer JL, Yucel RM. Computational Strategies for Multivariate Linear Mixed-Effects Models With Missing Values. Journal of Computational and Graphical Statistics. 2002;11:437–457. doi: 10.1198/106186002760180608. [DOI] [Google Scholar]
  • 37.Imbens GW. The role of the propensity score in estimating dose-response functions. Biometrika. 2000;87:706–710. doi: 10.1093/biomet/87.3.706. [DOI] [Google Scholar]
  • 38.Xu S, Ross C, Raebel MA, Shetterly S, Blanchette C, Smith D. Use of stabilized inverse propensity scores as weights to directly estimate relative risk and its confidence intervals. Value Health. 2010;13:273–277. doi: 10.1111/j.1524-4733.2009.00671.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Hernán MA, Hernández-Díaz S. Beyond the intention-to-treat in comparative effectiveness research. Clin Trials. 2012;9:48–55. doi: 10.1177/1740774511420743. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Rassen JA, Shelat AA, Franklin JM, Glynn RJ, Solomon DH, Schneeweiss S. Matching by propensity score in cohort studies with three treatment groups. Epidemiology. 2013;24:401–409. doi: 10.1097/EDE.0b013e318289dedf. [DOI] [PubMed] [Google Scholar]
  • 41.Robins JM, Hernán MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology. 2000;11:550–560. doi: 10.1097/00001648-200009000-00011. [DOI] [PubMed] [Google Scholar]
  • 42.Chou R, Fanciullo GJ, Fine PG, et al. Clinical guidelines for the use of chronic opioid therapy in chronic noncancer pain. J Pain. 2009;10:113–130. doi: 10.1016/j.jpain.2008.10.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Yoshida K, Solomon DH, Kim SC. Active-comparator design and new-user design in observational studies. Nat Rev Rheumatol. 2015;11:437–441. doi: 10.1038/nrrheum.2015.30. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Walker AM. Confounding by indication. Epidemiology. 1996;7:335–336. [PubMed] [Google Scholar]
  • 45.Hernán MA, Sauer BC, Hernández-Díaz S, Platt R, Shrier I. Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses. J Clin Epidemiol. 2016 doi: 10.1016/j.jclinepi.2016.04.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Hernán MA, Hernández-Díaz S, Robins JM. A structural approach to selection bias. Epidemiology. 2004;15:615–625. doi: 10.1097/01.ede.0000135174.63482.43. [DOI] [PubMed] [Google Scholar]
  • 47.Franklin GM, American Academy of Neurology Opioids for chronic noncancer pain: a position paper of the American Academy of Neurology. Neurology. 2014;83:1277–1284. doi: 10.1212/WNL.0000000000000839. [DOI] [PubMed] [Google Scholar]
  • 48.Berthelot J-M, Darrieutort-Lafitte C, Le Goff B, Maugars Y. Strong opioids for noncancer pain due to musculoskeletal diseases: Not more effective than acetaminophen or NSAIDs. Joint Bone Spine. 2015;82:397–401. doi: 10.1016/j.jbspin.2015.08.003. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supp info
NIHMS932150-supplement-Supp_info.pdf (27.3KB, pdf)

RESOURCES