Lung Cancer–Related Mortality With Inhaled Insulin or a Comparator: Follow-Up Study of patients previously enrolled in Exubera Controlled Clinical Trials (FUSE) Final Results

Exubera (EXU) is an inhaled, fast-acting form of recombinant human insulin that was developed by Pfizer in collaboration with Sanofi and Nektar Therapeutics. EXU was approved by the U.S. Food and Drug Administration (FDA) and European Medicines Agency (EMA) in 2006 for the treatment of adult patients with type 1 or type 2 diabetes, but it was voluntarily withdrawn from the market in 2008 for commercial reasons.

At the time of new drug application filing with FDA and EMA, newly diagnosed lung cancer cases were slightly imbalanced in the EXU controlled clinical program (three among EXU users, one in a user of a comparator, per the December 2006 Investigator’s Brochure) (1); by May 2008 this imbalance had increased to five among EXU-randomized patients and one among comparator-randomized patients (crude hazard ratio [HR] 5.1 [95% CI 0.7, 121.4]). “Comparator-randomized” refers to any patient who was randomized to any non-EXU diabetes treatment across 17 trials. All six of these patients had type 2 diabetes, were over 55 years of age, and had a history of smoking.

While more than 7,700 person-years (PY) composed the combined (EXU and comparator groups) follow-up time in controlled EXU clinical trials, those trials were not powered to exclude a specified imbalance in lung cancer risk. The expected numbers of lung cancer cases in each group were estimated by applying age-, sex-, and smoking status–specific lung cancer incidence rates among people with diabetes who were enrolled in the Northern California Kaiser Permanente Database (data available upon request) to the stratum-specific distribution of person-time in the trial data. The five lung cancer cases observed in patients randomized to EXU, and the one case in the comparator group, were fewer than expected (EXU 7.5 cases expected [95% CI 5.6, 9.7]; comparator 7.4 cases expected [95% CI 5.5, 9.6]).

The sponsor would typically continue to monitor lung cancer cases arising within the clinical program, but EXU was not available for use after August 2008. Large U.S. and European Union electronic health record and claims databases were queried, but they contained very few patients who had been exposed to EXU; thus, previous and ongoing clinical trials contained the largest defined group of such patients. As a commitment to the EMA and FDA, the Follow-Up Study of patients previously enrolled in Exubera controlled clinical trials (FUSE) was designed to evaluate whether EXU-exposed patients died from primary lung cancer at a substantially higher rate than unexposed patients. FUSE aimed to estimate among EXU-treated and comparator-treated patients the rates, and the corresponding rate ratios and rate differences, of 1) primary lung cancer–related mortality (primary objective), 2) primary lung cancer–related mortality among former smokers, 3) all-cause mortality, and 4) primary lung cancer incidence.

FUSE is an international, multicenter, comparative, hybrid, randomized, controlled trial/cohort study assessing primary lung cancer–related mortality in patients previously enrolled in EXU controlled clinical trials, comparing those previously randomized and treated with EXU with those previously randomized and treated with non-EXU comparator treatments. The follow-up period included retrospective and prospective components (Fig. 1). All investigators participating in 1 of 17 EXU controlled clinical trials were invited to join the prospective, observational component of FUSE. Data from all patients in these trials were included in the retrospective component. Patients from sites that agreed to participate were eligible for enrollment in the prospective follow-up study if they provided informed consent, had not participated in an investigational study of an unapproved drug since completing the EXU trial, and had never used another (non-EXU) inhaled insulin. For the duration of the 2-year prospective follow-up, all patients were treated for diabetes according to usual clinical practice.

Because data from the period between the end of a trial and the start of prospective follow-up are most susceptible to reporting bias, two analyses were conducted for each end point. The “continuous” analyses included all person-time and events occurring between the start of the original trial and the end of the prospective FUSE follow-up. The “in-study” analyses were restricted to events occurring during the original trial or during the prospective FUSE follow-up, and they exclude events in which diagnosis, death, or both occurred after the original trial but before the start of the prospective FUSE follow-up.

Patients from each trial contributed retrospective data and person-time from the date of randomization in the original trial until the last trial visit. Patients at sites that chose not to participate in the prospective follow-up (group 1) had no opportunity to contribute additional data and thus were censored at the end of trial participation. Patients at sites that chose to participate in the prospective follow-up but ultimately did not consent to participate (group 2) accrued additional retrospective person-time from the last trial visit date until the screening date for the prospective follow-up. Patients that consented to participate in the prospective follow-up (group 3) accrued additional retrospective person-time from the last trial visit date until the prospective baseline date, and they accrued prospective person-time (up to 2 years) from the prospective baseline to the end of the study period, death, voluntary withdrawal, or date of last known contact—whichever occurred first.

For each patient, most baseline data were derived from the baseline case report form (CRF) from their original trial. At participating sites, investigators attempted to obtain consent from all individuals who had participated in an included trial. If a patient had died since their participation in the clinical trial, the investigator recorded the date of death on the screening CRF and submitted documents needed for adjudicating the cause of death (e.g., death certificate, hospital and medical records), where allowable by law.

Patients enrolled in prospective follow-up had no study-mandated visits, tests, or interventions. The investigator entered baseline data from patient interviews and relevant medical records into the prospective follow-up baseline CRF. At years 1 and 2, the investigator determined and recorded the patient’s vital status on the appropriate follow-up CRF. Additional data collection included current smoking status and any reported lung cancer diagnosis.

When a potential study end point was identified, the investigator obtained any medical records necessary for adjudication, including medical and hospital records, lung cancer information, radiology reports, and date of diagnosis. The investigator forwarded these documents to the designated research partner, who collated, de-identified, and (where necessary) translated the documents into English and redacted the treatment arm before review by the end point committee.

Potential study outcomes were identified from four sources: 1) the original trial database (these outcomes may have been reported during or after the original trial), 2) the screening questionnaire, 3) the prospective follow-up baseline questionnaire, and 4) the prospective follow-up questionnaire (completed with information from scheduled telephone calls, other treating physicians, and hospital records).

In order to minimize potential error when assessing end points, an external expert committee (blinded to exposure group) adjudicated each reported lung cancer death and diagnosis to evaluate whether the lung cancer was incidental (newly diagnosed) and whether death was due to primary lung cancer. Lung cancer–related end points were those adjudicated as “highly likely” or “likely” to meet the definition of a lung cancer end point. When medical records were available, all-cause deaths were adjudicated in order to identify lung cancer deaths not reported as such by the investigators. Two of four end point committee members (including R.A.W. and R.O.C.) independently reviewed each set of records and classified them according to the diagnostic algorithms described in the study end point manual. If the two adjudicators disagreed, a third adjudicator broke the tie.

To estimate minimal detectable HRs, we assumed that 2,400 patients would be enrolled in the prospective follow-up and followed for 2 years; 50% of these patients would have been randomized to EXU and 10% would drop out or be lost to follow-up each year in each group. We would include a 6-month enrollment period. Although expected rates of lung cancer–related mortality were uncertain, FUSE was estimated to have an ∼80% power (with a one-sided upper 95% confidence limit) to exclude a threefold or higher rate of primary lung cancer–related mortality among EXU-exposed patients if the comparator rate approached that expected in the general population of people not currently smoking, or 1.30 per 1,000 PY (based on internal calculations, which are available upon request). Alternatively, the study was estimated to have ∼80% power to exclude a sixfold or higher rate if the FUSE comparator rate was 0.23 per 1,000 PY (i.e., the rate in comparator across pooled trials in 2008).

All primary and secondary analyses were based on the group to which each subject was randomized in his or her first eligible EXU clinical trial (equivalent to the principle of intention to treat). Descriptive statistics summarized data collected at the original trial baseline, during the prospective follow-up screening, and during the annual prospective follow-up. We used multivariate analysis techniques, as appropriate, to account for any baseline differences in the study groups. We reported cumulative incidence rates of the mortality end points (with 95% CIs), as well as corresponding crude and adjusted rate ratios (and two-sided 95% confidence limits), for the EXU and comparator groups.

We derived HRs and 95% CIs from Cox proportional hazards models using the proc phreg function in SAS software. If fewer than 10 events occurred in any treatment group, the incidence density ratio (IDR) was derived as the ratio of the number of subjects with the event per follow-up time between the two treatment arms; exact confidence limits were calculated by using binomial distribution and the ratio of exposures.

Potential confounders included smoking history, race, age, BMI, sex, HbA_1c (at trial baseline and the end of the trial), and lung function (at trial baseline and the end of the trial). Although we expected these covariates to be balanced across the treatment groups at trial baseline, we anticipated imbalances among patients enrolled in the observational, prospective follow-up. The analytic plan specified that if any potential confounder was imbalanced across randomized groups at the prospective follow-up baseline (i.e., a >5% observed difference across groups), and a sufficient number of events occurred, the primary analyses would include adjustment for imbalanced covariates. In the absence of imbalances, covariates would be adjusted for in secondary analyses.

The EMA and FDA reviewed the protocol, and the study design and objectives were discussed in depth with EMA. Final protocol, amendments, and informed consent documentation were reviewed and approved by the institutional review board, the independent ethics committee, or both at each participating investigational center. A protocol summary was posted on ClinicalTrials.gov (clinical trial reg. no. NCT00734591) on 12 August 2008, and summary results were posted 24 September 2012.

Across 25 countries (Argentina, Australia, Austria, Belgium, Brazil, Canada, Croatia, Denmark, Estonia, Finland, France, Germany, Greece, Italy, Mexico, the Netherlands, Norway, Poland, Portugal, Slovakia, South Africa, Spain, Sweden, the U.K., and the U.S.), 327 sites participated in this study. For patients in the prospective follow-up, the first visit occurred 27 August 2008 and the last visit occurred 6 January 2012.

Overall, 7,439 patients contributed 24,409 PY of observation (12,606 PY among EXU patients and 11,803 PY among comparator patients) (includes one additional PY beyond the addition of retrospective and prospective PY due to rounding) (Table 1). Of the total person-time, 15,092 PY (61.8%) accumulated during the original trials, 4,017 PY (16.5%) accumulated between the end of the original trials and the beginning of screening for the prospective follow-up, and 5,299 PY (21.7%) accumulated during the prospective follow-up. The period between the end of a trial and the start of prospective follow-up (for screened patients) ranged from 0 years (for patients who provided consent on the last day of the antecedent trial) to 9 years, with a mean of 2.6 years (median 2.1 years) in EXU patients and 2.4 years (1.7 years) in comparator patients. Patients included in FUSE were followed for a mean of 3.3 years. The mean follow-up time was 1.4 years for group 1, 2.9 years for group 2, and 5.7 years for group 3; follow-up varied little between the EXU- and comparator-treated patients.

Among the 7,439 patients, 3,092 (41.6%) were from sites that did not participate in the prospective follow-up (group 1) and 4,347 (58.4%) were from participating sites (groups 2 and 3) (Fig. 2). More than 60% of patients from participating sites (n = 2,631) participated in the prospective follow-up (group 3). Of these, 1,358 patients (51.6%) had been randomized to EXU in the original trial and 1,273 patients (48.4%) had been randomized to a comparator. Of those who did not participate in the prospective follow-up (n = 1,716; group 2), 691 were unwilling to participate, 608 were lost to follow-up, 95 died before screening (7 of these patients had died during the original trial), and 322 had withdrawn consent during the original trial (Fig. 2). Among reasons for nonparticipation, no meaningful imbalances existed across the EXU and comparator groups, apart from a numeric imbalance among those who died during the original trial (two EXU-treated and seven comparator-treated patients). A total of 93 patients (∼4% of 2,631 participants) did not complete the prospective follow-up: 70 were lost to follow-up, 20 were unwilling to participate, and 3 withdrew or were withdrawn for unspecified reasons. Among the reasons for discontinuation, no meaningful imbalances existed across the EXU and comparator groups.

As expected, all potential confounders were balanced across EXU- and comparator-treated patients at trial baseline. Although we anticipated covariate imbalances across treatment groups among those enrolled in the prospective follow-up, the covariates remained balanced among participants at the prospective follow-up baseline (Table 1). Additionally, potential confounders only measured at the prospective follow-up baseline visit (e.g., asbestos exposure and family history of lung cancer) were balanced across the EXU- and comparator-treated patients. Given this balance and the small number of outcome events, the main analyses were not adjusted for covariates.

Of 3,875 EXU-treated patients, 76 (2.0%) died of any cause; among 3,564 of comparator-treated patients, this number was 87 (2.4%) (Table 2). The all-cause mortality end point was based on reported rather than adjudicated cases. Of the 163 deaths, only 7 were reported as incident primary lung cancer–related deaths. Of the 156 deaths not reported as such, 91 could be adjudicated to determine whether unreported lung cancer was the underlying cause, yielding 1 additional lung cancer death in the EXU group.

There were seven reported incident primary lung cancer–related mortality end points (five in the EXU group and two in the comparator group) (Supplementary Table 2). All seven were classified by adjudicators as “highly likely” or “likely” primary lung cancer mortalities, meeting the primary end point definition. Of the eight patients who died from lung cancer, seven (five in the EXU group and two in the comparator group) were former smokers (Supplementary Table 3). No lung cancer–related deaths occurred during the original trials. Five of six lung cancer–related deaths in the EXU group and one of two such deaths in the comparator group were identified after the end of the original trial but before the start of the FUSE prospective follow-up; these were excluded from the in-study analysis.

There were 18 reported incident primary lung cancer end points (14 in the EXU group and 4 in the comparator group) and 1 lung cancer identified in the EXU group through adjudication, as noted above. Fifteen cases (12 in the EXU group and 3 in the comparator group) were adjudicated as highly likely or likely to be incident primary lung cancer meeting the end point definition, 14 of which occurred in former smokers. When provided, the histological types of lung cancer varied and were clinically unremarkable. Eight of the 12 EXU cases and 1 of the 3 comparator cases meeting the incident primary lung cancer end point definition were diagnosed after the end of the original trial but before the start of the FUSE prospective follow-up and were excluded from the in-study analysis. Two additional cases of lung cancer were reported after the end of FUSE follow-up, both in patients who were exposed to EXU. Although these cases were not included in the results, they were adjudicated by the end point committee for completeness. One case was adjudicated as highly likely and the other as unlikely to be incident primary lung cancer.

The IDR (ratio of the number of subjects with an event per follow-up time between the EXU and comparator treatment arms) for the lung cancer–related mortality primary end point was 2.8 (95% CI 0.5, 28.5) in the continuous analysis and 0.9 (95% CI 0.0, 73.5) in the in-study analysis (Table 2). In the continuous analysis, the absolute rates were 0.5 per 1,000 PY (95% CI 0.2, 1.0) in the EXU group and 0.2 per 1,000 PY (95% CI 0.0, 0.6) in the comparator group; the rate difference (EXU minus comparator) was 0.3 (95% CI −0.3, 0.9). The absolute rates from the in-study analysis were 0.1 per 1,000 PY (95% CI 0.0, 0.5) in the EXU group and 0.1 per 1,000 PY (95% CI 0.0, 0.5) in the comparator group; the rate difference was 0.0 (95% CI −0.5, 0.5).

The IDR for the former-smoker subset of patients with the lung cancer–related mortality end point was 2.3 (95% CI 0.4, 24.0) in the continuous analysis and 0.9 (95% CI 0.0, 71.8) in the in-study analysis. The IDR for the incident primary lung cancer end point was 3.7 (95% CI 1.0, 20.7) in the continuous analysis and 1.9 (95% CI 0.3, 20.7) in the in-study analysis.

The HR (for EXU/comparator) for the all-cause mortality end point was 0.81 (95% CI 0.60, 1.10) in the continuous analysis and 0.83 (95% CI 0.53, 1.28) in the in-study analysis (Table 2). In the continuous analysis, the absolute rates were 6.0 per 1,000 PY (95% CI 4.8, 7.6) in the EXU group and 7.4 per 1,000 PY (95% CI 5.9, 9.1) in the comparator group; the rate difference was −1.3 (95% CI −3.4, 0.7). The absolute rates for the in-study analysis were 3.4 per 1,000 PY (95% CI 2.4, 4.7) in the EXU group and 4.0 per 1,000 PY (95% CI 2.9, 5.4) in the comparator group; the rate difference was −0.6 (95% CI −2.3, 1.0).

FUSE was designed to evaluate whether patients previously treated with EXU in controlled clinical trials experienced incident primary lung cancer–related mortality at a substantially higher rate than patients treated with a comparator. Because of slow sales, the pooled EXU randomized trial study population was by far the largest EXU-exposed group among existing data sources in 2008, including large data sets from electronic health records. While designing the study and in discussions with regulators, it was understood that the interpretation of results would likely have limitations, as the study was not sufficiently powered to draw clear statistical conclusions about lung cancer–related HRs (or IDRs) less than 3.5- or 6-fold, depending on the assumed background comparator group mortality rate. The prospective part of FUSE exceeded its study enrollment goal and had lower-than-expected discontinuation, but some analyses were underpowered because of lower-than-expected comparator incidence rates.

The continuous analysis (including all events that occurred from randomization in the original clinical trial to the end of the prospective FUSE follow-up) yielded a 2.8-fold higher rate in the EXU group than in the comparator group, but the 95% CI (0.5, 28.5) was wide. The in-study analysis (excluding events occurring between the end of the clinical trial and the start of FUSE) did not yield a higher rate of lung cancer–related mortality after randomization to EXU than after randomization to a comparator. Because the 95% CI was wide (IDR 0.9 [95% CI 0.0, 73.5]), an increased rate in the EXU group cannot be excluded, although the rate difference counterpart excludes 0.5 events per 1,000 PY. For primary lung cancer incidence, an higher rate occurred in the EXU group in both the continuous (IDR 3.7 [95% CI 1.0, 20.7]) and in-study (IDR 1.9 [95% CI 0.3, 20.7]) analyses. The rate of all-cause mortality in the EXU group was not higher in either the continuous or in-study analyses; the HR point estimates were below 1, with narrow 95% CIs, likely excluding an increased rate of all-cause mortality in the EXU group (continuous analysis: HR 0.81 [95% CI 0.60, 1.10]; in-study analysis: HR 0.83 [95% CI 0.53, 1.28]). All patients who died from lung cancer were former smokers except one (an incident case), consistent with results from the EXU clinical program as of 2008 and with data from the general population.

During the design phase, we were concerned with potential detection/reporting bias of lung cancer incidence for several reasons: 1) the novel route of EXU administration, 2) mild pulmonary symptoms (e.g., cough) associated with EXU use (2,3), 3) a hypothetical concern that human insulin had potential mitogenic properties via IGF-I receptor binding (4), and 4) the open-label nature of the randomized clinical trials. Because respiratory symptoms occurred more frequently in EXU-treated patients than in comparator-treated patients in clinical trials (5), it was likely that EXU-treated patients would undergo more diagnostic procedures such as chest X-ray and sputum cytology, possibly leading to different rates of diagnosis of latent cases between treatment groups. Thus, detection bias seemed to be a plausible explanation for the imbalance in lung cancer incidence that we identified in the controlled trials that stimulated this study. Further, a 2008 label warning about a possible increased risk of lung cancer among patients taking EXU (6) might have led to more diagnostic procedures during the prospective follow-up among patients who previously used EXU, and consequently to disproportionate discovery and reporting of lung cancer cases.

In light of potential detection and reporting biases, we made several efforts to improve the study’s validity. Each reported lung cancer death and diagnosis was adjudicated by an external expert committee (blinded to exposure group) in order to evaluate whether each reported death was due to primary lung cancer and whether each diagnosed lung cancer was incident (newly diagnosed) and primary. Lung cancer–related mortality was specified as the primary end point, as we considered it to be much less susceptible to detection bias than lung cancer diagnosis, it has been an end point in many epidemiologic studies investigating risk factors for lung cancer (7–9), and lung cancer incidence and mortality rates are highly correlated (10). Finally, for each end point, we specified two analyses a priori: one that included all person-time and events occurring between the start of the original trial and censoring or the end of the FUSE prospective follow-up (continuous), and another that excluded person-time and events occurring after the original trial but before the start of prospective follow-up (in-study).

While five of eight lung cancer cases were fatal during the prospective follow-up, four of these were excluded from the in-study analysis because the diagnosis occurred during the gap between the trial and the prospective follow-up, making interpretation of the lung cancer–related mortality in-study analysis difficult. Despite our concerns about reporting bias for the lung cancer–related mortality end point (specifically lung cancer deaths being underreported in the comparator group), no clear evidence indicated that the lung cancer–related mortality results were affected. Finally, the all-cause mortality end point was not susceptible to detection bias and was the least susceptible to reporting bias; underreports of mortality during the period between the end of the original trial and the start of FUSE prospective follow-up would likely be similar in the EXU and comparator groups.

Overall, these data cannot exclude an increased risk of lung cancer–related mortality associated with EXU use. The higher rate observed in the continuous analysis could be explained by a tumor-promoting effect of EXU among smokers with undetected lung cancer, by reporting bias associated with preferential reporting of EXU cases, or by some combination of these factors. If real, the absolute increased risk of lung cancer–related mortality was small (0.5 per 1,000 PY) and was notably smaller than the estimated lung cancer–related mortality rate in nonsmoking patients with diabetes that we used to calculate our sample size (1.3 per 1,000 PY; data available upon request).

Strengths of FUSE include a geographically and ethnically diverse study population, a high participation rate, little loss to follow-up, and balance among all measured potential confounding variables during the initial trial and at the prospective FUSE baseline. Given that EXU was no longer on the market when the study started, the chosen study population and design were the best opportunities to evaluate the potential effects of EXU on lung cancer incidence and mortality. In addition, this represented a defined source population, with data collected both at the trial baseline and accumulated over the course of the study. This allowed us to compare covariates among the group of patients who agreed to participate in the prospective follow-up and among those who did not or could not; this comparison did not identify important differences. Further, although we expected covariate balance in the pooled data from trial baseline, covariates remained balanced across treatment groups for those who participated in the prospective follow-up, such that adjustment for covariates (conducted in the secondary analyses) did not meaningfully change the effect estimates. Despite a higher risk of lung cancer–related mortality after EXU exposure, EXU users did not experience more all-cause deaths than did users of other diabetes treatments over the study period.

Acknowledgments. The authors thank all of the investigators and coordinators who took part in this study (a complete list of whom can be found in the Supplementary Data). The authors are indebted to the 2,631 participants in the prospective FUSE, to all 7,439 clinical trial participants who contributed retrospective data, to the FUSE Scientific Steering Committee (M.B.B., G.G.K., R.A.W., R.B.C.), to the FUSE Endpoint Committee (M.B.B., chair [not an adjudicator]; R.A.W., chief adjudicator; R.B.C.; Atul Malhotra; and Alan Sandler), to Kevin Sweetland for providing outstanding study management, and to their colleagues Robert Reynolds, Richard Riese, Pamela Schwartz, Susan DeCorte, Susan Gannon, and Vivianne Dillon at Pfizer Medical, Research and Development, and Safety. Study management was provided by Nokwazi Nxumalo, Itumeleng Siweya, and Tebogo Sebata (in South Africa); Rosaria Mangano (in Italy); Andrea Dector, Mario Perez, Gabriela Negrete, and Angel Mario Coll (in Mexico); Kerry Pasculli, Luke Stacey, and Karen Cooley (in Australia); Zofia Ziebolewska and Sylwia Leszczynska-Kosmala (in Poland); Maria Savini, Mariana Rodriguez Batz, and Nydia Reynoso Hunter (in Argentina); Isabel Lizarraga (in Spain); Jenny de Gelder-Bakker and Gaitrie JhinkoeRai (in the Netherlands); Ida Ratih (in the U.K.); Alexander Celedin (in Austria); Vera Weilburg (in Germany); Martin Lucas (in Canada); Merli Laissaar and Moonika Soots (in Estonia); Selim Yasavul (in Turkey); Dieudonne Bwirire (in Belgium); Cissi Akerberg (in Sweden); Gustavo Teixeira, Lenio Alvarenga, and Rosana Fusaro (in Brazil); Pia Korhonen and Pia Eloranta (in Finland); Chrysostomos Mylonas and Theologia Karabatzoglou (in Greece); Eric Heilbron (in Costa Rica); Maja Gacic (in Croatia); Laurence Descamps (in France); Daniela Kollarova (in Slovakia); Jes Hunsballe (in Denmark); Paula Rebelo and Ines Mendes (in Portugal); Mette Hallen (in Norway); and the United Biosource Corporation (in the U.S.) under Bruce Smith. Pfizer funded all study management.

Duality of Interest. This study was sponsored by Pfizer. N.M.G., W.T.D., and J.L. are full-time employees and shareholders of Pfizer. When conducting the study, D.O.K., S.K., and N.C.J. were full-time employees and shareholders of Pfizer. M.B.B. has served as a paid consultant to Pfizer, Forest Laboratories, GlaxoSmithKline, Eli Lilly & Co., Procter & Gamble, and Sanofi. M.B.B., G.G.K., R.A.W., and R.B.C. were paid by Pfizer as consultants for their roles on the FUSE Steering Committee. G.G.K. is the principal investigator of a biostatistical agreement between Pfizer and the University of North Carolina at Chapel Hill, and that agreement provided the structure for his activity on the steering committee of the study reported in this article. G.G.K. is also the principal investigator of many such biostatistical agreements with other biopharmaceutical sponsors, including Arena, AstraZeneca, Eli Lilly & Co., Forest Research Institute (Allergan), GlaxoSmithKline, Merck, Novartis, and Sanofi, although his activities for those sponsors are not related to the content of this article. Information concerning all biostatistical agreements for which G.G.K. is the principal investigator is publicly available through the University of North Carolina at Chapel Hill. R.A.W. has served as a paid consultant to Pfizer, MannKind Corp., and Sanofi. R.B.C. has served as a paid consultant to Pfizer for work related to clinical trials. No other potential conflicts to interest relevant to this article were reported.

Author Contributions. N.M.G. wrote and revised the manuscript. N.M.G. and D.O.K. designed and implemented the study. D.O.K. wrote the statistical analysis plan and reviewed and edited the manuscript. M.B.B. chaired the scientific steering committee, implemented the study, assessed data, and edited the manuscript. W.T.D. designed the study and the statistical analysis plan, produced the final tables, produced post hoc analyses, and reviewed and edited the manuscript. J.L. served as project manager and designed and implemented the protocol, assessed data, and reviewed and edited the manuscript. S.K. designed the protocol, assessed data, and reviewed the manuscript. G.G.K., R.A.W., and R.B.C. were members of the scientific steering committee, implemented the study, assessed data, and edited the manuscript. N.C.J. designed the protocol, implemented and governed the study, assessed data, and reviewed the manuscript. N.M.G. is the guarantor of this work and, as such, 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.

Prior Presentation. The design of FUSE was presented at the 25th Anniversary International Conference on Pharmacoepidemiology & Therapeutic Risk Management, Providence, RI, 16–19 August 2009; the accompanying abstract was published in Pharmacoepidemiol Drug Saf 2009;18(Suppl. 1). The results of FUSE were presented at the 28th International Conference on Pharmacoepidemiology, Barcelona, Spain, 23–26 August 2012; the accompanying abstract was published in Pharmacoepidemiol Drug Saf 2012;21(Suppl. 1).