Health Outcomes Among Long-term Opioid Users With Testosterone Prescription in the Veterans Health Administration

Key Points español 中文 (chinese)

Question  What are the health outcomes among long-term opioid users who receive testosterone treatment compared with opioid users who do not?

Findings  In this cohort study of 21 272 male long-term opioid users with testosterone deficiency, those who received opioids plus testosterone therapy had significantly lower all-cause mortality and lower incidence of major adverse cardiovascular events, anemia, and femoral or hip fractures than their counterparts who received opioids only in covariate-adjusted and propensity score–matched models.

Meaning  This study’s findings suggest that receiving opioids plus testosterone treatment is associated with lower all-cause mortality and a lower incidence of other adverse health outcomes among men with opioid-induced androgen deficiency.

Importance  Androgen deficiency is common among male opioid users, and opioid use has emerged as a common antecedent of testosterone treatment. The long-term health outcomes associated with testosterone therapy remain unknown, however.

Objective  To compare health outcomes between long-term opioid users with testosterone deficiency who filled testosterone prescriptions and those with the same condition but who did not receive testosterone treatment.

Design, Setting, and Participants  This cohort study focused on men in the care of the Veterans Health Administration (VHA) facilities throughout the United States from October 1, 2008, to September 30, 2014. It included male veterans who were long-term opioid users, had low testosterone levels (<300 ng/dL), and received either a testosterone prescription or any other prescription. It excluded male patients with HIV infection, gender dysphoria, or prostate cancer and those who received testosterone in fiscal year 2008. Data were analyzed from April 1, 2017, to April 30, 2019.

Exposure  Prescription for testosterone.

Main Outcomes and Measures  All-cause mortality and incidence of major adverse cardiovascular events (MACE), vertebral or femoral fractures, and anemia during the 6-year follow-up through September 30, 2015.

Results  After exclusions, 21 272 long-term opioid users (mean [SD] age, 53 [10] years; n = 16 689 [78.5%] white) with low total or free testosterone levels were included for analysis, of whom 14 121 (66.4%) received testosterone and 7151 (33.6%) did not. At baseline, compared with opioid users who did not receive testosterone, long-term opioid users who received testosterone treatment were more likely to have obesity (43.7% vs 49.0%; P < .001), hyperlipidemia (43.0% vs 48.8%; P < .001), and hypertension (53.9% vs 55.2%; P = .07) but had lower prevalence of coronary artery disease (15.9% vs 12.9%; P < .001) and stroke (2.4% vs 1.3%; P < .001). After adjusting for covariates, opioid users who received testosterone had significantly lower all-cause mortality (hazard ratio [HR] = 0.51; 95% CI, 0.42-0.61) and lower incidence of MACE (HR = 0.58; 95% CI, 0.51-0.67), femoral or hip fractures (HR = 0.68; 95% CI, 0.48-0.96), and anemia (HR = 0.73; 95% CI, 0.68-0.79) during the follow-up period of up to 6 years, compared with their counterparts without a testosterone prescription. In covariate-adjusted models, men who received opioids plus testosterone were more likely to have resolved anemia compared with those who received opioids only during the 6-year follow-up (HR = 1.16; 95% CI, 1.02-1.31). Similar results were obtained in propensity score–matched models and when analyses were restricted to opioid users with noncancer pain or those who did not receive glucocorticoids.

Conclusions and Relevance  This study found that, in the VHA system, male long-term opioid users with testosterone deficiency who were treated with opioid and testosterone medications had significantly lower all-cause mortality and significantly lower incidence of MACE, femoral or hip fractures, and anemia after a multiyear follow-up. These results warrant confirmation through a randomized clinical trial to ascertain the efficacy of testosterone in improving health outcomes for opioid users with androgen deficiency.

Opioid use in the United States has reached epidemic proportions.¹ Although deaths from opioid overdose^2,3 have captured the world’s attention, the endocrine complications of chronic opioid use have remained underappreciated. All opioid medications, especially long-acting opioids, suppress testosterone levels, often into the severely hypogonadal range.^4,5 Prescription opioid use has been shown to have an association with testosterone deficiency; nearly 20% of testosterone prescriptions written in the Veterans Health Administration (VHA) system were for men who were opioid users.⁶ Testosterone deficiency among opioid users is associated with increased risk of sexual dysfunction, osteoporosis, and bone fractures.⁷ However, to our knowledge, no data exist on major health outcomes in opioid users who are hypogonadal and either receive or do not receive testosterone treatment.

Testosterone treatment of men with hypogonadism is associated with improved sexual desire, erections, and sexual activity^8,9; self-reported mobility^10-12; volumetric bone density and estimated bone strength^13,14; and corrected anemia.¹⁵ However, the long-term association of testosterone treatment with major adverse cardiovascular events (MACE), mortality, and bone fractures remains unknown.^16-23 Little is known about the effects of testosterone treatment on health outcomes in men with opioid-induced androgen deficiency (OPIAD), who often experience marked suppression of testosterone levels. One short-term randomized clinical trial reported improvements in pain sensitivity, sexual desire, and body composition with testosterone treatment in men with OPIAD.²⁴ However, the long-term implications of testosterone treatment for major health outcomes in opioid users are unknown to date. Patients who receive opioids experience more severe testosterone deficiency, often have multiple comorbid conditions, use multiple prescription drugs, and are at increased risk of mortality,²⁵ which could influence the benefit to risk ratio of testosterone treatment.

In the absence of a long-term randomized clinical trial of testosterone in men with OPIAD, we compared major health outcomes in male long-term opioid users who received testosterone treatment with those who did not receive testosterone. We selected health outcomes, including overall mortality, MACE, fractures (vertebral, femoral, and hip), and incident anemia, that have public health importance and could be ascertained with accuracy. Most randomized testosterone trials¹⁴ have been conducted in men with mild testosterone deficiency. Because opioid use is typically associated with moderate to severe testosterone deficiency, a study of men with OPIAD offers an opportunity to elucidate the association between testosterone treatment and health outcomes among opioid users with moderate to severe testosterone deficiency. Recognizing the limitations of observational studies, we performed propensity matching and several sensitivity analyses to assess the potential effect of confounding owing to observed and unobserved differences between groups.

This nationwide cohort study included men who received prescriptions within the VHA system from October 1, 2008, to September 30, 2014, and were followed up through September 30, 2015. The institutional review board of Bedford VHA Medical Center approved the study and waived the need for informed consent because deidentified data were used. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

For primary analyses, we selected male veterans who were long-term opioid users, had testosterone deficiency, and received either a testosterone prescription (testosterone recipients) or any other prescription (nonrecipients of testosterone) in each of the 2 or more years after filling an opioid prescription. Opioids and their morphine equivalents are listed in eTable 1 in the Supplement. We focused on long-term opioid users, defined as those who received 120 or more days’ supply of opioids during at least 1 continuous 180-day interval²⁶ in 2 or more years as indicated by prescription fills. To identify opioid users with testosterone deficiency, we selected those with testosterone levels measured before they received a testosterone prescription. Those with a low total testosterone level (<300 ng/dL; to convert to nanomoles per liter, multiply by 0.0347) or free testosterone level (<70 pg/mL) were included. These cutoff points for testosterone levels were consistent with thresholds in guidelines available at that time and with published reference ranges.^27,28 We excluded, among others, those with HIV infection, gender dysphoria, and prostate cancer and those who received testosterone in fiscal year (FY) 2008 because these individuals could have been using testosterone before receiving an opioid prescription.

For patients who received testosterone prescriptions, the earliest testosterone prescription fill between FY 2009 and 2014 was used as the index fill. For patients who did not receive testosterone prescriptions, a fill from FY 2009 to 2012 was chosen at random as the index fill in order to match the time distribution of testosterone fills.

To avoid the immortal person time bias,²⁹ the exposure interval for testosterone recipients was calculated as the interval between the date of their first testosterone prescription and a date 3 months after their last testosterone fill. For nonrecipients of testosterone, the duration of exposure was calculated as the interval between the date of their most recent testosterone-level check preceding their index date and a date 3 months after their last opioid prescription. This definition of exposure period ensured that patients were actively in care at least at the start and end of the period and had sufficient care within the VHA system to be included in the present study cohort. The outcome period started at the beginning of the specific exposure period for both recipients and nonrecipients of testosterone.

We selected all-cause mortality and first occurrence of MACE as the primary outcomes of the study and vertebral, femoral, or hip fractures and anemia as the secondary outcomes. These outcomes were selected because they are clinically important and can be ascertained with a high level of accuracy. The date of death was extracted from the VHA Vital Status File, which is compiled by merging data from multiple sources to create a single date of death.³⁰ Major adverse cardiovascular events included myocardial infarction, ischemic stroke, or death (defined with the International Classification of Diseases, Ninth Revision, Clinical Modification, [ICD-9-CM] codes [eTable 2 in the Supplement]). Vertebral, femoral, or hip fractures were identified using ICD-9-CM diagnosis codes in both outpatient and inpatient data from the VHA Corporate Data Warehouse (CDW) (eTable 2 in the Supplement).

Anemia was defined as hemoglobin level less than 12 g/dL (to convert to grams per liter, multiply by 10.0) or a hematocrit reading less than 36%³¹; the measurement closest to the index fill was used as the baseline level.

For anemia analyses, patients were required to have both baseline and follow-up hemoglobin or hematocrit values. We separately analyzed the resolution of anemia among those who had anemia at baseline (n = 1567) and the development of anemia among those who did not have anemia at baseline (n = 17 355).

We adjusted covariate-adjusted and propensity score–matched models for sociodemographic variables, including age, marital status, copayment, race/ethnicity, and poverty level in residential zip code. We also adjusted for a number of physical and mental comorbidities and specific medications (eTable 1 in the Supplement) using data from the VHA CDW. A 1-year look-back period was used to check for comorbidities and medications that occurred before the date of index prescription for testosterone recipients and nonrecipients of testosterone. These conditions were ascertained by the presence of at least 2 ICD-9-CM codes separated by 7 or more days (eTable 2 in the Supplement).

Baseline characteristics of study population and covariates were noted. We generated unadjusted bivariate- and covariate-adjusted Cox proportional hazards and propensity score–matched models. To minimize covariate imbalance between groups, we used a one-to-one propensity score match with a caliper of 0.001. Propensity scores were estimated using a logistic model, including demographic characteristics; coronary artery disease; hypertension; diabetes; hyperlipidemia; heart failure; stroke; chronic kidney disease; cancers; liver disease; dementia; depression; bipolar disease; indications for pain medication; substance use disorder; alcohol dependence; psychosis; and use of glucocorticoid, antidepressant, and antipsychotic medications.

We ran Cox proportional hazard models for up to 6 years of follow-up (FY 2009-2014). Testosterone recipients entered the analyses on the day of their index prescription fill and were followed up until 3 months after the date of their last testosterone fill. Nonrecipients of testosterone entered the analyses at the first date of their testosterone-level check and were followed up until 3 months after the date of their last opioid prescription. Study participants were censored at the earliest of the follow-up date, death, latest date of VHA service use, or September 30, 2015. Baseline comparisons between the recipients and nonrecipients of testosterone were performed with χ² test, test of proportions, or unpaired, 2-tailed t test as appropriate. From the Cox models we determined hazard ratios (HRs) and 95% CIs associated with the testosterone recipients compared with the nonrecipients of testosterone. For models of outcomes, goodness of fit and proportional hazards assumptions were evaluated. We also used Kaplan-Meier curves to compare the unadjusted probability of survival separately for each primary outcome.

All analyses were conducted with SAS, version 9.4 (SAS Institute Inc). A statistical significance threshold of a 2-sided P = .05 was used. Data were analyzed from April 1, 2017, to April 30, 2019.

Because of the lack of randomization, confounding owing to baseline between-group differences was a potential challenge. Therefore, we conducted several sensitivity analyses to assess the robustness of associations with outcomes.

The first sensitivity analysis excluded patients who had a diagnosis of cancer (n = 20 366), given that the treatment of cancer pain can differ from that of noncancer pain.³² The second sensitivity analysis excluded men who received glucocorticoids (n = 15 149), given that these medications can suppress testosterone levels and affect outcomes.³³ The third sensitivity analysis examined all-cause mortality and MACE outcomes as a function of different testosterone formulations (injections, gels, and patches), comparing testosterone recipients with nonrecipients because difference in outcomes have been reported with different formulations.³⁴

The fourth sensitivity analysis used a simulation algorithm to examine the role of a potential unobserved confounder as discussed by Higashi et al.³⁵ We assumed the omitted confounder to be a binary variable and simulated it to correlate with death in patients who received long-term opioid plus testosterone therapy. We estimated the degree of correlation that would be needed between the unmeasured confounder and outcome (death) and the exposure (use of opioid plus testosterone treatment) to explain the association with survival.

Among the 1 437 460 men who received an outpatient prescription in a VHA facility between FY 2008 and FY 2014, we excluded 10 001 men (0.7%) with HIV infection, 508 (0.04%) with gender dysphoria, 13 112 (0.9%) with 1 or more prescriptions filled only in FY 2008 but not before, and 189 857 (13.2%) without any filled prescription in FY 2008 (Figure 1). We excluded 33 694 men (2.3%) who received testosterone in FY 2008; 35 393 (2.5%) who had prostate cancer; and 1 017 323 (70.8%) who received no opioids, received opioids but less than the 120 days’ supply during any continuous 180 days, or received opioids but for less than 1 year from earliest to latest opioid prescription. From the 111 101 long-term opioid users, we selected those who received testosterone prescription fills for 2 or more years (testosterone recipients) or any other prescription fill for 2 or more years (nonrecipients of testosterone). In all, 21 272 men (19.1%) with a total testosterone level less than 300 ng/dL or free testosterone level less than 70 pg/mL were included in the analyses.

Among the 21 272 long-term opioid users in this study, 14 121 (66.4%) received testosterone and 7151 (33.6%) did not. The race/ethnicity of most patients was white (n = 16 689 [78.5%]), and the mean (SD) age of the entire sample was 53 (10) years (Table 1). Compared with opioid users who did not receive testosterone, a slightly higher proportion of opioid users who received testosterone had hypertension (53.9% vs 55.2%; P = .07), hyperlipidemia (43.0% vs 48.8%; P < .001), obesity (43.7% vs 49.0%; P < .001), and posttraumatic stress disorder (24.2% vs 25.6%; P = .02) and were more likely to be receiving an opioid dose greater than 50 morphine milligram equivalents (MME) (29.4% vs 42.8%; P < .001). A slightly lower proportion of testosterone recipients compared with nonrecipients had prevalent coronary artery disease (12.9% vs 15.9%; P < .001) and stroke (1.3% vs 2.4%; P < .001). The distribution of these characteristics in the 2 groups was similar after the propensity score match, as expected.

In unadjusted and covariate-adjusted models, men who received opioid plus testosterone therapy had statistically significantly lower all-cause mortality than men who received opioids only during the follow-up period of up to 6 years (HR = 0.51; 95% CI, 0.42-0.61) (Table 2 and Figure 2). The incidence of MACE was also significantly lower in recipients of opioids plus testosterone treatment compared with recipients of opioids only during follow-up in the covariate-adjusted models (HR = 0.58; 95% CI, 0.51-0.67).

The incidence of femoral and hip fractures was significantly lower in recipients of opioids plus testosterone than in recipients of opioids only in unadjusted and covariate-adjusted models during follow-up (HR = 0.68; 95% CI, 0.48-0.96). When vertebral fractures were considered individually (HR = 0.86; 95% CI, 0.63-1.18) or when all fractures (vertebral plus femoral and hip fractures) were considered (HR = 0.80; 95% CI, 0.63-1.01), the associations with testosterone treatment were not significant.

Among participants who were anemic at baseline, testosterone treatment was significantly associated with the resolution of anemia during the 6-year follow-up (HR = 1.16; 95% CI, 1.02-1.31) in the covariate-adjusted model. Long-term opioid users who received testosterone had a significantly lower risk of incident anemia compared with opioid users who did not receive testosterone (HR = 0.73; 95% CI, 0.68-0.79). The findings of propensity score–adjusted models were similar to those of the covariate-adjusted models (Table 2). Cox proportional hazards models for propensity score–matched samples demonstrated a significantly lower hazard for all-cause mortality (HR = 0.54; 95% CI, 0.44-0.67); significantly lower incidence of MACE (HR = 0.60; 95% CI, 0.52-0.70), femoral or hip fractures (HR = 0.60; 95% CI, 0.40-0.89), and anemia (HR = 0.73; 95% CI, 0.67-0.80); and significantly higher rates of resolved anemia (HR = 1.17; 95% CI, 1.01-1.35) for testosterone recipients compared with nonrecipients.

In a sensitivity analysis that excluded men with cancer pain, testosterone recipients in covariate-adjusted models had significantly lower all-cause mortality (HR, 0.51; 95% CI, 0.42-0.62) and lower incidence of MACE (HR, 0.58; 95% CI, 0.50-0.67), femoral or hip fracture (HR, 0.65; 95% CI, 0.45-0.94), and anemia (HR, 0.74; 95% CI, 0.68-0.80) compared with nonrecipients of testosterone, a finding consistent with that of the primary analysis (Table 3). When the analysis was limited to patients who did not receive glucocorticoid medication, with the exception of femoral or hip fractures, the results were comparable (mortality: HR, 0.56 [95% CI, 0.44-0.71]; MACE: HR, 0.57 [95% CI, 0.48-0.68]; anemia: HR, 0.71 [95% CI, 0.64-0.78]) (eTable 3 in the Supplement).

In sensitivity analyses comparing different testosterone formulations on all-cause mortality and MACE, all testosterone formulations were associated with lower likelihood of mortality (HR, 0.51; 95% CI, 0.42-0.61) and MACE (HR, 0.58; 95% CI, 0.51-0.67) in fully adjusted models comparing testosterone recipients with nonrecipients (eTable 4 in the Supplement).

We examined the potential role of an unobserved confounder and identified the critical levels for correlation of a potential confounder with all-cause mortality and testosterone exposure that would completely explain the observed effects (eTable 5 in the Supplement). The association between treatment with opioid plus testosterone therapy and mortality was moderately sensitive to unmeasured confounding. An unmeasured confounder with 0.3 correlation with opioid plus testosterone treatment would need to have a −0.06 (HR, 1.09; 95% CI, 0.90-1.32; P = .37) correlation or stronger with mortality to eliminate the negative implication of testosterone for mortality.

Among long-term opioid users in the VHA, men who received opioid plus testosterone therapy had significantly lower all-cause mortality and significantly lower incidence of MACE, anemia, and femoral or hip fractures compared with men who received opioid treatment alone. The association between testosterone treatment and MACE and mortality was robust to analyses, which used both covariate-adjusted and propensity score–matched models. Sensitivity analyses in men who had noncancer pain or who did not receive glucocorticoids confirmed the findings of the primary analysis. This cohort study, a first step toward understanding the association of testosterone treatment with health outcomes in long-term opioid users, has clinical implications because of the high prevalence of opioid use among US military veterans and high rates of androgen deficiency and testosterone use among opioid users. These findings need confirmation in a randomized clinical trial.

Because assignment to testosterone treatment was not randomized, the differences in outcomes between those treated with opioids plus testosterone and those who received opioids only cannot be attributed with certainty to testosterone treatment. Although differences in outcomes remained statistically significant even after propensity score matching, baseline differences between patient groups may have been factors in outcomes in ways not captured by propensity score. Compared with men who received opioids only, men who received opioids plus testosterone had a higher mean body mass index and higher prevalence of hyperlipidemia, hypertension, and psychiatric disorders. The men who received opioids only had slightly higher prevalence of some other comorbid conditions, including coronary artery disease and stroke. We attempted to address this concern by propensity score matching and by simulating the implication of a potential unobserved confounder. The simulation analyses showed that an unobserved confounder (eg, better access to health care) would need to be moderately correlated with both the exposure (receipt of testosterone and opioids) and the outcome (mortality) to explain these findings beyond adjustment for observed covariates.

Because we used propensity score matching to account for baseline differences between groups, the existence of such strong, unmeasured confounding was not likely, but it cannot be ruled out entirely. Propensity scores can balance only observed covariates, and findings could be subject to bias from unmeasured confounding variables. Furthermore, establishing the index date for individuals not receiving medications is complex. On the basis of similar previous approaches to identifying an index date for individuals not receiving medications,^36,37 we used the date of documented testosterone level as the beginning of exposure for nonrecipients of testosterone.

To our knowledge, this cohort study is the first to examine the association between testosterone treatment in long-term opioid users and MACE, fractures, anemia, and all-cause mortality. The only short-term randomized clinical trial in men with OPIAD found that testosterone treatment improved sexual function, pain sensitivity, and body composition.²⁴ Similar improvements in sexual function³⁸ have been reported in uncontrolled short-term studies in patients treated with opioid medications.³⁹ Opioid agonists used in the treatment of opioid addiction and treatment have been shown to differently affect testosterone levels and sexual function in opioid-dependent men. Compared with methadone, buprenorphine has been found to be associated with substantially lower suppression of testosterone levels and with lower rates of sexual dysfunction.^40,41

These findings are in contrast to reports that testosterone therapy was associated with adverse cardiovascular events in men with low testosterone, most of whom did not have long-term use of opioids.^16-18 However, not all studies found this association between testosterone and increased risk of death or cardiovascular outcomes, and some studies even reported cardiovascular advantages.^19,20,42

This study has several strengths, one of which is its use of the large and detailed VHA CDW. The CDW included medication dispensing records, laboratory results, ICD-9-CM diagnosis codes, and demographics. The large sample size and long follow-up period enabled the reliable assessment of MACE, mortality, and fracture outcomes, which would not have been possible in smaller studies of shorter duration. The study included hard outcomes that are clinically important. The concern about confounding by indication owing to testosterone recipients having low testosterone levels at baseline was accounted for by requiring control patients to meet that criterion as well. In addition, we used propensity score matching to minimize the factors in baseline differences, and we performed additional sensitivity analyses to assess the effect of unmeasured confounding. We included only patients who received an opioid prescription in the year before receiving a testosterone prescription and who had a confirmed low total or free testosterone level before receiving testosterone. To avoid the immortal person time bias,²⁹ we defined the exposure by the dates of the first and last testosterone prescription fills.

This study’s limitations include its observational design, whereby unmeasured confounding could have affected the findings. Physicians also could have selected healthier patients to receive testosterone therapy. We attempted to control for confounding by adjusting for a wide range of relevant demographics, conditions, and medications; by using propensity score matching; and by simulating the potential effect of an unobserved confounder. However, unmeasured confounding cannot be fully excluded. The quality of the assays for total and free testosterone levels within the VHA varied. Immunoassays for testosterone commonly used in this period were susceptible to inaccuracy in the low range; some patients on testosterone treatment might not have been hypogonadal, whereas some men with hypogonadism may have been misdiagnosed as eugonadal. The protective influence of testosterone on mortality in patients taking opioids remained robust in each of these different analyses. We only had access to VHA pharmacy data; some patients may have filled a testosterone prescription outside the system. Because medication costs in the VHA system were lower than outside the system, it was unlikely that many patients obtained their testosterone prescription externally. Furthermore, this factor would only dilute the treatment effect and drive the results toward null.

Sexual function, quality of life, and well-being outcomes could not be ascertained from the VHA CDW, and mental health outcomes were not assessed. Although ICD-9-CM codes are commonly used to identify conditions, these codes may not have been recorded accurately. Study outcomes were ascertained based on clinical coding and were not adjudicated. The total number of fractures was small, and the study may not have had sufficient statistical power to detect between-group differences in fracture events. Opioid use was defined according to prescription fills, and low testosterone levels were identified by documented laboratory results. On-treatment testosterone levels in men who received testosterone treatment varied widely⁴³; however, testosterone levels during treatment were not consistently monitored. Although we required all patients to have documented opioid prescriptions in 2 or more years, a patient in either group could have stopped using opioids or could have changed their use pattern within the study period. Patients in the VHA system typically have a greater burden of comorbid conditions compared with the general population, which may affect the generalizability of these findings to non-VHA patients.⁴⁴

This cohort study found that, among men who were long-term opioid users, those who received a testosterone prescription had significantly lower all-cause mortality and a significantly lower incidence of MACE, anemia, and femoral or hip fractures in up to 6 years of follow-up compared with male opioid users who did not receive testosterone therapy. Because of the observational nature of this study, confounding owing to known and unknown differences between groups cannot be disregarded. Because of the high prevalence of opioid use among US veterans and the high proportion of opioid users who receive testosterone treatment, we believe a randomized clinical trial is warranted to ascertain whether testosterone treatment is safe and whether it is associated with improved health outcomes among opioid users who have androgen deficiency.

Accepted for Publication: October 21, 2019.

Published: December 11, 2019. doi:10.1001/jamanetworkopen.2019.17141

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2019 Jasuja GK et al. JAMA Network Open.

Corresponding Author: Guneet K. Jasuja, PhD, Center for Healthcare Organization and Implementation Research, Edith Nourse Rogers Memorial Veterans Hospital, Veterans Administration Medical Center, 200 Springs Rd, Bedford, MA 01730 (guneet.jasuja@va.gov).

Author Contributions: Dr Jasuja had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Jasuja, Ameli, Rose, Miller, Berlowitz, Bhasin.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Jasuja, Ameli, Reisman.

Critical revision of the manuscript for important intellectual content: Jasuja, Ameli, Rose, Miller, Berlowitz, Bhasin.

Statistical analysis: Ameli, Reisman, Miller.

Administrative, technical, or material support: Miller.

Supervision: Jasuja, Rose, Berlowitz, Bhasin.

Conflict of Interest Disclosures: Dr Bhasin reported receiving grants from Alivegen, AbbVie, Transition Therapeutics, and Metro International Biotechnology; personal fees from AbbVie and OPKO; and equity interest in Functioning Promoting Therapies outside the submitted work; holding a patent (A Method for Calculating Free Testosterone) issued; and serving as the chair of the Endocrine Society's Expert Panel, which developed guidelines for testosterone treatment of hypogonadal men; these conflicts are overseen by the Partners Healthcare System’s Office of Industry Interactions. No other disclosures were reported.

Funding/Support: This study was funded by the Department of Veterans Affairs, Veterans Health Administration, Health Services Research and Development Service. Dr Jasuja is a VA HSR&D Career Development awardee at the Bedford VA (CDA 13-265).

Role of the Funder/Sponsor: The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: The views expressed herein are those of the authors and do not necessarily represent the views of the US Department of Veterans Affairs.