Comparative Safety of Testosterone Dosage Forms

Importance  Increases in testosterone use and mixed reports of adverse events have raised concerns about the cardiovascular safety of testosterone. Testosterone is available in several delivery mechanisms with varying pharmacokinetics; injections cause spikes in testosterone levels, and transdermal patches and gels cause more subtle but sustained increases. The comparative cardiovascular safety of gels, injections, and patches has not been studied.

Objective  To determine the comparative cardiovascular safety of testosterone injections, patches, and gels.

Design, Setting, and Participants  A retrospective cohort study was conducted using administrative claims from a commercially insured (January 1, 2000, to December 31, 2012) and Medicare (January 1, 2007, to December 31, 2010) population in the United States and general practitioner records from the United Kingdom (January 1, 2000, to June 30, 2012). Participants included men (aged ≥18 years) who initiated use of testosterone patches, gels, or injections following 180 days with no testosterone use. Our analysis was conducted from December 11, 2013, to November 12, 2014.

Exposures  New initiation of a testosterone dosage form, with use monitored for up to 1 year.

Main Outcomes and Measures  Inpatient or outpatient medical records, diagnoses, or claims for cardiovascular and cerebrovascular events including myocardial infarction (MI), unstable angina, stroke, and composite acute event (MI, unstable angina, or stroke); venous thromboembolism (VTE); mortality; and all-cause hospitalization.

Results  We identified 544 115 testosterone initiators between the 3 data sets: 37.4% injection, 6.9% patch, and 55.8% gel. The majority of men in the Medicare cohort were injection initiators (51.2%), most in the US commercially insured population were gel initiators (56.5%), and the UK database included equal proportions of injections and gel users (approximately 41%). With analysis conducted using hazard ratios and 95% CIs, compared with men using gels, injection initiators had higher hazards of cardiovascular events (ie, MI, unstable angina, and stroke) (1.26; 1.18-1.35), hospitalization (1.16; 1.13-1.19), and death (1.34; 1.15-1.56) but not VTE (0.92; 0.76-1.11). Compared with gels, patches did not confer increased hazards of cardiovascular events (1.10; 0.94-1.29), hospitalization (1.04; 1.00-1.08), death (1.02; 0.77-1.33), or VTE (1.08; 0.79-1.47).

Conclusions and Relevance  Testosterone injections were associated with a greater risk of cardiovascular events, hospitalizations, and deaths compared with gels. Patches and gels had similar risk profiles. However, this study did not assess whether patients met criteria for use of testosterone and did not assess the safety of testosterone among users compared with nonusers of the drug.

Testosterone use has increased considerably in the United States, United Kingdom, and other countries,^1-5 and there is a lack of clear documented indications for treatment in many men.^1,4 Ongoing unresolved concerns about cardiovascular safety have been raised by the halting of a randomized trial⁶ of testosterone gels in older men with limited mobility owing to increased cardiovascular events and of nonexperimental studies^7,8 reporting increased cardiovascular risk in older men with cardiovascular disease. Although the recent literature is mixed, with some studies suggesting no harmful effects,^9-11 there has been considerable use of testosterone contrary to recommended guidelines,^1,4 prompting interest and investigation into its use and safety.

Testosterone is available in multiple dosage forms, including intramuscular injections, transdermal patches and gels, implantable pellets, intranasal sprays, and oral/buccal applications. Although gels, injections, and patches all effectively raise testosterone levels, their pharmacokinetics differ; injections create spikes of supernormal testosterone levels that slowly decrease until a subsequent injection¹²; this cycling results in less time within normal ranges than with transdermal systems.¹² Gels and patches result in subtle, short-term (24-48 hours) increases in testosterone levels, and daily reapplication can maintain consistent levels.¹² However, gels provide longer-lasting increases than patches.¹³ Because testosterone levels may influence short-term clotting and polycythemia, differing pharmacokinetics may result in varying safety profiles. We compared the cardiovascular risk of testosterone gels, injections, and patches in cohorts of real-world users drawn from large health care databases.

We conducted a new-user¹⁴ cohort study of testosterone injection, gel, and patch initiators in 3 secondary data sources: 2 from the United States and 1 from the United Kingdom. This project was approved by the institutional review board of The University of North Carolina at Chapel Hill. The protocol was also approved by the Independent Scientific Advisory Committee of the Clinical Practice Research Datalink (CPRD), Medicines & Healthcare Products Regulatory. The University of North Carolina at Chapel Hill institutional review board determined that written informed consent was not required from study participants, and there was no financial compensation. We performed our analyses from December 11, 2013, to November 12, 2014, using SAS, version 9.2 (SAS Institute Inc) and Episheet Spreadsheets for the Analysis of Epidemiologic Data (http://www.drugepi.org/wp-content/uploads/2012/10/Episheet.xls).

The first US cohort consisted of commercially insured men from the Truven MarketScan Commercial Claims and Encounters and Medicare Supplementary and Coordination of Benefit files (Truven Health Analytics, Inc; January 1, 2000-December 31, 2012). This database contains adjudicated insurance claims for inpatient and outpatient procedures and diagnoses and pharmacy-dispensed medications for individuals with employer-sponsored commercial insurance, spouses, dependents, and retirees with employer-sponsored Medicare supplementary plans from large US employers. Supplementary laboratory test results were available for a subset of people whose laboratory tests were processed by a national laboratory testing company during January 1, 2007, to December 31, 2012. We included men 18 years or older.

The Medicare cohort was drawn from a national random 20% sample of the US Medicare fee-for-service population from January 1, 2007, to December 31, 2010. The database contains billing claims for procedures, diagnoses, and dispensed medications for adults 65 years or older throughout the United States. No laboratory test results were available in this cohort.

The UK cohort was drawn from the CPRD, a compilation of general practitioner medical records throughout the United Kingdom from January 1, 2000, to June 30, 2012, which contains outpatient clinical characteristics, diagnoses and procedures, hospital and specialist notes, and prescribed medications as recorded by general practitioners. Laboratory test results were available for most tests performed. We included men 18 years or older.

We identified men newly initiating testosterone therapy following a 180-day washout period free of documented testosterone use. Only the first eligible new-use period per individual was included. Testosterone dosage forms of interest included pharmacy-dispensed transdermal gels and patches, pharmacy-dispensed injections, or in-office injections as determined from procedure and supply codes. Exposure categories were grouped as gel, injection, and patch. Prior use of implanted pellets, oral/buccal testosterone, and oral methyltestosterone during the washout period were considered exclusion criteria. However, because of rare, esoteric use and documented risks of methyltestosterone-induced liver problems, these forms were not considered as exposures for the comparative analysis. Patients with claims for 2 different testosterone forms on the index day were excluded because their exposure could not be accurately categorized.

The date of the first pharmacy prescription or injection procedure code following the washout period was considered the index date for the new-user cohorts.¹⁴ Owing to a potential for differential adherence and discontinuation between the dosage forms (injections are administered every several weeks and patches/gels are administered daily), we used a first-exposure-carried-forward analysis in which the patient was considered to be exposed continuously throughout the follow-up period (eFigure 1 in the Supplement).

We monitored initiators of testosterone therapy for up to 1 year to observe outcomes including myocardial infarction (MI), unstable angina, stroke, composite acute events (ie, MI, unstable angina, or stroke), all-cause hospitalization, mortality, and venous thromboembolism (VTE). The effects were estimated separately for each outcome, and individuals experiencing the outcome of interest during the baseline period were excluded to restrict to new-onset outcomes. Only the first occurrence of each outcome during follow-up was considered.

In the MarketScan and Medicare databases, outcomes were based on International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9) diagnosis codes. Myocardial infarction, unstable angina, and stroke required an inpatient diagnosis of the condition with a hospital stay of at least 1 day.^15,16All-cause hospitalization was defined as hospitalization for any reason. Venous thromboembolism was defined as an inpatient diagnosis claim followed by a prescription for an antithrombotic drug in the following 30 days.^17,18 Mortality data were unavailable in MarketScan.

In the CPRD database, outcomes were assessed using Read codes recorded by the general practitioner. Inpatient and specialist encounters were reported to the general practitioner from hospitals and specialists.^19,20

We identified total serum testosterone tests performed during the baseline period with procedure codes. If a test result was available, the test result was categorized as high, normal, or low according to assay-specific result flags or reference ranges. If flags or ranges were unavailable, we classified the testosterone level results as low, normal, or high (<300, 300-849, or ≥850 ng/dL, respectively; to convert to nanomoles per liter, multiply by 0.0347).

Other covariates were assessed during the baseline period using diagnosis, procedure, and medication codes and included calendar year, age, comorbidity, cardiovascular risk factors, health care utilization, preventive and screening care, indications for testosterone, and other medication use. Owing to the medical records–based structure of the CPRD, body mass index and smoking status were available in the CPRD cohort.

Hazard ratios (HRs) and 95% CIs were estimated with multivariable Cox proportional hazards regression models using days since initiation as the time scale; models were adjusted for a priori–defined confounders. Censoring took place at the first occurrence of the end of the study (MarketScan: December 31, 2012; Medicare: December 31, 2010; and CPRD: June 30, 2012), 1 year after the index date, at disenrollment, or at death (except when death was the outcome of interest). Separate models for injections and patches were run using gels as the referent group since gels were the most commonly and broadly used dosage form.

In addition, we estimated the effects of testosterone dosage forms using propensity score (PS) matching. We estimated the predicted probability (ie, PS) of receiving the dosage form of interest vs gels by modeling treatment prescribed with the measured covariates as predictors in logistic regression models. Gel initiators were matched up to 2:1 to initiators of injections or patches using a greedy matching algorithm that matches 1:1 without replacement to the fifth digit of the PS, if possible.²¹ The algorithm then attempted to match 1:1 the remaining unmatched gel initiators to the treatment group, resulting in 1 or 2 matched gel initiators for each injection or patch initiator. The HRs then estimated the treatment effect among men who received the dosage form of interest in the match data sets.²²

Owing to differences in database structure and covariate availability, analyses were performed and results were presented separately in the 3 databases. Summary estimates were obtained by meta-analyzing the 3 database-specific multivariable-adjusted estimates with fixed-effects models.²³

We repeated the analyses separately in subgroups of men with documented low or normal baseline testosterone levels and those without recent measurements to account for the potential of differential prescribing by testosterone level. In addition, to remove episodes of contraindicated use of testosterone in prostate cancer, we performed the analysis excluding patients with diagnoses of prostate cancer or any other cancer. Because usage patterns changed over time,¹ we performed an analysis restricted to later years of the study after gels had become the predominant therapy option (January 1, 2007, to December 31, 2012). Owing to a study⁷ suggesting that cardiovascular risk increased relatively soon after testosterone initiation, we performed an analysis monitoring patients for only 6 months. We also performed an as-treated analysis in the MarketScan database in which initiators’ follow-up time was censored at treatment discontinuation, when switching from the index testosterone dosage form, or at 1 year for continuous users. Because there are multiple injectable testosterone formulations, we also separately considered injectable testosterone cypionate, enanthate, and propionate formulations compared with gels.

We identified 544 115 testosterone initiators between the 3 data sets: 37.4% injection, 6.9% patch, and 55.8% gel. Of these, there were 515 132 eligible testosterone initiators in MarketScan (56.5% gel, 36.7% injection, and 6.8% patch), 22 376 in Medicare (42.9% gel, 51.2% injection, and 5.9% patch), and 6607 in CPRD databases (42.4% gel, 39.6% injection, and 18.1% patch). In the MarketScan cohort, gel, injection, and patch initiators tended to be similar, with a few notable exceptions. Mean age was similar between users of the 3 dosage forms, but patch initiation was concentrated much earlier in the study period than other dosage forms, and gels were concentrated later. As testing became more commonplace over the time period, gels had more total testosterone laboratory testing before gel initiation, but injections had more recorded ICD-9 or Read code diagnoses of hypogonadism. In addition, gel initiators received more prostate-specific antigen tests and lipid profiles (Table 1 reports selected covariates; full covariate distributions are listed in eTable 1 in the Supplement). Of the testosterone injections used, 83.3% were testosterone cypionate, 9.2% were testosterone enanthate, and 1.3% were testosterone propionate; 55.4% of the injections were administered in offices and 43.7% were pharmacy dispensed (1.0% of the men received both on the same day). In the 6 months after initiation of testosterone, serum total testosterone level tests were performed in 36.4% of gel users, 32.2% of injection users, and 28.1% of patch users.

The mean age of the Medicare cohort was older than that of the MarketScan or CPRD cohorts. Characteristics were similar between the dosage forms since the Medicare cohort was drawn from a narrower time period. However, patch initiators reported more heart failure and psychiatric disorders. Injection initiators had more recorded hypogonadism diagnoses, gel initiators had more prostate-specific antigen screening, and patch initiators had fewer sexual dysfunction diagnoses. In the 6 months after initiation of testosterone therapy, more gel users (49%) had follow-up testosterone measurements than did men receiving injections (39%) or patches (40%). Some characteristics were not displayed owing to small cohort size restrictions in our data use agreement with the Centers for Medicare & Medicaid Services. Complete covariate distributions are reported in eTable 2 in the Supplement.

Among the CPRD cohort, a much larger proportion of injection initiators did not have a recent testosterone measurement. Use of gel tended to occur later in the time period, and gel initiators had more diagnoses of hypogonadism and prostate-specific antigen screening. There was also less use of statins among the injection initiators (eTable 3 in the Supplement).

The 1-year incidence of cardiovascular events was low among the younger MarketScan and CPRD testosterone initiators but more common in the older Medicare sample (Table 2 and Table 3). Hospitalization rates were higher in the US databases than in the CPRD database, and mortality was more frequent in the Medicare database. The VTE rates were low owing to the restrictive definition used to ensure true cases. In all databases, gel initiators tended to have lower crude rates of all outcomes than did injection or patch initiators (eFigure 2 in the Supplement).

For the injection vs gel comparisons, we observed elevated crude HRs that showed higher risks in injections for most outcomes in all data sets. Upon adjustment, most effect measure estimates were attenuated, yet all remained elevated (Table 2 and Figure 1) except for VTE. Propensity score–matched estimates generally closely agreed with the adjusted estimates.

For the patch vs gel comparisons, the crude VTE HR estimates were null in all databases, and the crude CPRD death effect estimate was slightly protective; all other unadjusted estimates were above the null. Owing to small sample sizes and the limited number of events, some adjusted effect measures could not be estimated. Upon adjustment and PS matching, the crude HR estimates attenuated toward the null, suggesting no or slight increases in risk with use of the patch compared with the gel, although smaller sample sizes resulted in less precise effect measure estimates (Table 3 and Figure 2). Propensity score–matched estimates and adjusted estimates were similar.

Results from the 3 databases generally agreed. The MarketScan estimates tended to be closest to the null, followed by Medicare, with the CPRD estimates being the most extreme (Figure 1 and Figure 2).

In analyses within testosterone level subgroups, small samples resulted in imprecise, unstable estimates (eTables 4 and 5 in the Supplement). Although the estimates agreed with overall analyses in direction of effect, they should be interpreted cautiously. When patients with prior cancers were excluded, the results were similar to the overall estimates (eTables 6 and 7 in the Supplement). When considering initiation of testosterone during the years 2007-2012, no meaningful differences were observed from the overall sample (eTable 8 in the Supplement), and when considering a 6-month follow-up period, results were also similar (eTable 9 in the Supplement). The as-treated analysis in the MarketScan cohort generally yielded HRs slightly higher than the estimates from the first-exposure-carried-forward analysis, but the conclusions generally agree with the primary analysis (eTable 10 in the Supplement). Mean (SD) treatment duration was gel, 122 (112) days; injection, 105 (104) days; and patch, 96 (91) days. When considering injection formulations separately, testosterone cypionate and enanthate showed effect estimates similar to the overall results (eTable 11 in the Supplement); testosterone propionate was used rarely, and estimates were imprecise but were consistent with the overall estimates.

In this multicohort comparison of testosterone dosage forms, we observed consistent increases in the risk of cardiovascular and cerebrovascular events, hospitalization, and death among men receiving injections compared with those receiving gels. When comparing patches with gels, we observed a slight increase in MI among patch initiators, but all other outcomes were inconsistent. We did not observe any dosage form carrying a higher risk of VTE than the others. Although the increased risk of outcomes in injection initiators was consistent across databases, absolute incidences were small in this relatively short time period. In Medicare—the cohort with the oldest mean population—the 1-year incidence of the composite MI, angina, and stroke outcome was 23.1 events per 1000 person-years in gels, 36.6 in injections, and 34.9 in patches. In the MarketScan and CPRD cohorts, outcome occurrence was lower, and even consistently increased relative risks translated into low absolute increases.

Although prior safety research^6-8,24,25 has investigated testosterone as a class, the present study directly compared individual dosage forms. A prior CPRD study²⁶ demonstrated that hypertension and polycythemia risk were higher with injection vs oral testosterone, suggesting that the risk profiles of dosage forms differ. A reanalysis of the Testosterone in Older Men with Mobility Limitations trial²⁷ found that gel users who experienced adverse cardiovascular events had greater increases in serum free testosterone levels. Different dosage forms lead to different serum testosterone levels over time—injections result in spikes and periods of super-normal levels¹²—possibly accounting for the observed risk of cardiovascular disease.

This analysis is subject to limitations inherent in the use of secondary health care data: unavailability of important patient characteristics, missing data, the nonrandomized nature of the exposure, and potential outcome misclassification. Some predictors of cardiovascular events, such as smoking status and body mass index, were unavailable in the US cohorts. However, predictors were available in the CPRD, and using a variety of cohorts allowed us to estimate the effect in settings with and without these potential confounders. The CPRD estimates generally agreed with the US-based estimates, and obesity was evenly distributed between treatments (eTable 3 in the Supplement) and did not contribute substantially to the injection vs gel outcome model for the composite event (β = −0.13; P = .80). Smoking status was slightly imbalanced between treatments, and being a current smoker contributed to the outcome model (β = 0.78; P < .001). However, a comparison of models in the CPRD adjusted for obesity and smoking (HR, 0.75; 95% CI, 0.42-1.33) with models not adjusted for those variables (HR, 0.70; 95% CI, 0.39-1.24) demonstrated no substantial variation.

In addition, hypogonadism symptoms can be diffuse and indistinct, and they were infrequently diagnosed and coded, often leaving us without the primary indications for testosterone treatment. However, by comparing individuals initiating treatment with different dosage forms, we restricted analysis to those whom physicians had determined required treatment, and choice of the dosage form is likely not heavily influenced by indication or cardiovascular risk. Important unmeasured behavioral, economic, or social differences between treatment groups may remain. All of the included men had insurance coverage, but injections tend to be less expensive than branded gels and can be administered by health care providers; therefore, in-office injections may be prescribed more frequently for individuals with reduced personal disease-management skills or resources and subsequent higher cardiovascular risk. Gel initiators received more follow-up serum testosterone tests, possibly indicating better health management and physician monitoring. However, patches had lower postinitiation monitoring, similar to the rate with injections, yet similar increased adverse event rates were not observed. In most measured clinical respects, initiators of various dosage forms were not meaningfully different.

In addition, time trends in testosterone testing and treatment exist, with both becoming more common later in the study period.¹ The proportion of men using a gel as their index prescription has increased, with injection and patch use decreasing during the study period.¹ Increased cardiovascular risk with use of injections vs gels may be a function of comparing an older, sicker patient case mix early in the study period with healthier gel initiators. However, we did not see the same patterns when comparing patches with gels, and patches were usually prescribed even earlier, suggesting that the observed effects are not merely time trends. We adjusted for calendar year, and sensitivity analyses considering only the later years of the study did not show different effects.

Many patients initiated testosterone without recorded serum testosterone tests or relevant diagnoses, and we did not require evidence of use according to guidelines²⁸ for inclusion in the cohort. Because of inadequate information on testosterone levels before and after the initiation of therapy, we could not measure the impact of achieved blood levels. We did not adjust for baseline testosterone levels in the primary analyses. Baseline testosterone levels were unavailable in Medicare and for most men in the MarketScan database. In addition, many individuals did not have a serum testosterone level measured in the 6 months before initiation of therapy (MarketScan, 37.3%; Medicare, 32.9%; and CPRD, 53.1%); some baseline tests may be missing because of out-of-pocket payment, use of other insurance coverage, or failure to bill or record the test, although treatment in men without measured levels or with normal levels has been observed.¹ However, treatment choice could not be influenced by baseline testosterone levels in men without measurements, and levels were not strongly associated with the choice of dosage form in men with baseline measurements. Selection bias may be introduced by restricting analysis to individuals with tests performed and results available; therefore, we adjusted for having a baseline test performed but not for the result.

Different routes of testosterone administration may lead to differential nonadherence. Gel and patch use were assessed through pharmacy dispensing (United States) or written prescriptions (United Kingdom); injections were assessed from both pharmacy information and in-office procedure codes. We could not measure whether the written prescriptions (United Kingdom) were dispensed or the dispensed prescriptions (United States) were used, whereas codes for in-office injections have a much lower potential for nonreceipt than pharmacy prescriptions, leading to potential differences in misclassification owing to nonadherence between dosage forms. However, 44.7% of injection users in the MarketScan cohort initiated therapy with pharmacy-dispensed injections, allowing for greater nonadherence than with in-office injections.

In addition, claims-based studies have the potential for outcome misclassification because claims are generated for billing rather than research. In the US cohorts, we used restrictive claims-based definitions for MI, angina, stroke, and VTE to avoid inclusion of rule-out diagnoses. Similar definitions in other studies have demonstrated high specificity^15,18; thus, we estimated relative measures of effect rather than absolute measures, which will be unbiased in the presence of nondifferential outcome misclassification.²⁹ However, differences in diagnosing, reporting, and recording between the databases may have resulted in the different magnitudes of effect estimates observed between databases.

Although limited by the nonrandomized nature of the exposure, this study benefits greatly from the large diverse patient sample representing men across age groups, populations, treatment and practice patterns, and health care systems. The MarketScan cohort is representative of those with employer-sponsored commercial insurance in the United States, and our Medicare cohort came from a random 20% sample of Medicare beneficiaries throughout the United States. The CPRD is widely representative of primary medical practice in the United Kingdom, where testosterone use is more restrained,¹ pharmaceutical advertising is limited, and there is less disparity in health care access. Despite these differences, effect measure estimates generally agreed across cohorts, suggesting robustness of the results.

Our analysis suggests that testosterone injections may increase the short-term risk of cardiovascular events, stroke, death, and hospitalization compared with gels. The risks associated with patches and gels appeared to be similar and lower than the risk with injections. With potential long-term effects of testosterone on lipid levels, further exploration of cardiovascular risk associated with longer-term treatment is warranted. With continuing concern about the safety and effectiveness of testosterone treatment in men with primary and age-related hypogonadism and the trend of treatment in men with normal testosterone levels or without recent baseline testing, it is important to understand the potential hazards of testosterone treatment.

Accepted for Publication: March 19, 2015.

Corresponding Author: J. Bradley Layton, PhD, Department of Epidemiology, The University of North Carolina at Chapel Hill, 725 Martin Luther King Jr Blvd, Ste 234, CB 7590, Chapel Hill, NC 27599 (blayton@unc.edu).

Published Online: May 11, 2015. doi:10.1001/jamainternmed.2015.1573.

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

Study concept and design: Layton, Sharpless, Brookhart.

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

Drafting of the manuscript: Layton, Stürmer, Brookhart.

Critical revision of the manuscript for important intellectual content: Meier, Sharpless, Stürmer, Jick, Brookhart.

Statistical analysis: Layton, Brookhart.

Obtained funding: Layton, Brookhart.

Administrative, technical, or material support: Stürmer.

Study supervision: Layton, Jick, Brookhart.

Conflict of Interest Disclosures: Dr Stürmer receives investigator-initiated research funding and support as principal investigator (grant R01 AG023178) and coinvestigator (grant R01 AG042845) from the National Institute on Aging, as coinvestigator (grant R01 CA174453) from the National Cancer Institute at the National Institutes of Health (NIH), and as principal investigator of a pilot project from the Patient Centered Outcomes Research Institute (PCORI). Dr. Stürmer does not accept personal compensation of any kind from any pharmaceutical company, although he receives salary support from the Center for Pharmacoepidemiology (current members: GlaxoSmithKline, UCB BioSciences, and Merck) and research support from pharmaceutical companies (Amgen and Merck) to the Department of Epidemiology, The University of North Carolina at Chapel Hill (UNC). Dr Brookhart receives investigator-initiated research funding from the NIH (grants R01 AG042845, R21 HD080214, and R01 AG023178) and through contracts with the Agency for Healthcare Research and Quality’s Developing Evidence to Inform Decision Effectiveness program and the PCORI. He has received research support from Amgen for unrelated projects and has served as a scientific advisor for Amgen, Merck, and GSK (honoraria/payment received by the institution). He has received consulting fees from RxAnte and World Health Information Consultants. No other disclosures were reported.

Funding/Support: This work was funded by the US National Institute on Aging (grant 5 R01 AG042845 02). Portions of the database infrastructure used for this project were funded by the Department of Epidemiology, UNC Gillings School of Global Public Health; the Cecil G. Sheps Center for Health Services Research at UNC; the Comparative Effectiveness Research Strategic Initiative of UNC’s Clinical Translational Science Award (1 ULI RR025747); and the UNC School of Medicine.

Role of the Funder/Sponsor: The funding sources 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.

Previous Presentation: Preliminary results of this work were presented at the 31st International Convention on Pharmacoepidemiology and Therapeutic Risk Management; October 26, 2014; Taipei, Taiwan.

Additional Contributions: Dongmei Li, MS (UNC), and Pascal Egger (University of Basel) provided programming and database assistance. There was no financial compensation.

Correction: This article was corrected on May 21, 2015, to fix the abstract.