original article
`
`Diabetes, Obesity and Metabolism 16: 748–756, 2014.
`© 2014 John Wiley & Sons Ltd
`
`Dose-finding results in an adaptive, seamless, randomized
`trial of once-weekly dulaglutide combined with metformin
`in type 2 diabetes patients (AWARD-5)
`Z. Skrivanek1, B. L. Gaydos1, J. Y. Chien1, M. J. Geiger2, M. A. Heathman1, S. Berry3, J. H. Anderson4,
`T. Forst5, Z. Milicevic1 & D. Berry3
`1LillyDiabetes,EliLillyandCompany,Indianapolis,IN,USA
`2Cardiovascular&MetabolismTherapeutics,RegeneronPharmaceuticalsInc,Tarrytown,NY,USA
`3BerryConsultants,Austin,TX,USA
`4DiabetesandCardiometabolicMedicine,Carmel,IN,USA
`5Profil,Hellersbergstrasse,Neuss,Germany
`
`article
`original
`
`Aims: AWARD-5 was an adaptive, seamless, double-blind study comparing dulaglutide, a once-weekly glucagon-like peptide-1 (GLP-1)
`receptor agonist, with placebo at 26 weeks and sitagliptin up to 104 weeks. The study also included a dose-finding portion whose results are
`presented here.
`Methods: Type 2 diabetes (T2D) patients on metformin were randomized 3 : 1 : 1 to seven dulaglutide doses, sitagliptin (100 mg), or placebo.
`A Bayesian algorithm was used for randomization and dose selection. Patients were adaptively randomized to dulaglutide doses using available
`data on the basis of a clinical utility index (CUI) of glycosylated haemoglobin A1c (HbA1c) versus sitagliptin at 52 weeks and weight, pulse
`rate (PR) and diastolic blood pressure (DBP) versus placebo at 26 weeks. The algorithm randomly assigned patients until two doses were
`selected.
`Results: Dulaglutide 1.5 mg was determined to be the optimal dose. Dulaglutide 0.75 mg met criteria for the second dose. Dulaglutide 1.5 mg
`showed the greatest Bayesian mean change from baseline (95% credible interval) in HbA1c versus sitagliptin at 52 weeks −0.63 (−0.98 to
`−0.20)%. Dulaglutide 2.0 mg showed the greatest placebo-adjusted mean change in weight [−1.99 (−2.88 to −1.20) kg] and in PR [0.78
`(-2.10 to 3.80) bpm]. Dulaglutide 1.5 mg showed the greatest placebo-adjusted mean change in DBP [−0.62 (−3.40 to 2.30) mmHg].
`Conclusions: The Bayesian algorithm allowed for an efficient exploration of a large number of doses and selected dulaglutide doses of 1.5
`and 0.75 mg for further investigation in this trial.
`Keywords: AWARD-5, Bayesian adaptive, dose finding, dulaglutide dose, GLP-1, GLP-1 receptor agonist, metformin, type 2 diabetes
`
`Date submitted 24 February 2014; date of first decision 12 March 2014; date of final acceptance 18 April 2014
`
`Introduction
`
`Dulaglutide is a long-acting human GLP-1 receptor agonist in
`development as a once-weekly subcutaneous injection for the
`treatment of type 2 diabetes (T2D) [1–3]. The molecule consists
`of two identical, disulphide-linked chains, each containing an
`N-terminal GLP-1 analogue sequence covalently linked to
`a modified human immunoglobulin G4 heavy chain by a
`small peptide linker [1]. Dulaglutide exhibits GLP-1-mediated
`effects, including glucose-dependent potentiation of insulin
`secretion, inhibition of glucagon secretion, delay of gastric
`emptying and weight loss [1–4].
`Dose selection for the dulaglutide clinical development
`programme utilized an adaptive design within the first confir-
`matory dulaglutide trial (AWARD-5) that enabled exploration
`
`Correspondenceto: Zachary Skrivanek, PhD, Eli Lilly Corporate Center, Lilly Diabetes,
`Indianapolis, IN 46285, USA.
`e-mail: skrivanek_zachary@lilly.com
`
`of seven doses in a dose-finding portion, and possible selection
`of up to two doses. The primary and secondary objectives
`compared the efficacy and safety of selected dulaglutide dose(s)
`with sitagliptin at 52 and 104 weeks, and with placebo at
`26 weeks [5,6]. The results of dose-finding are presented here.
`
`Research Design and Methods
`Eligible patients (18–75 years) had T2D (≥6 months) and
`glycosylated haemoglobin A1c (HbA1c) of >8.0% if on
`diet and exercise alone, or HbA1c of ≥7.0 and ≤9.5% if
`on oral antihyperglycaemic medications (OAM) (any OAM
`monotherapy, or combination of metformin with another
`OAM) and a BMI between 25 and 40 kg/m2. The protocol
`was approved by local ethical review boards, and all patients
`provided written informed consent before trial-related activity.
`The study was conducted in compliance with the Declaration of
`Helsinki and the International Conference on Harmonization
`guideline on good clinical practices [7].
`
`MPI EXHIBIT 1113 PAGE 1
`
`

`

`DIABETES, OBESITY AND METABOLISM
`
`AWARD-5 was a multicentre, randomized, double-blind,
`104-week, parallel-arm trial
`in T2D patients treated with
`metformin, and included an initial dose-finding portion
`(Figure 1). The study design and statistical methodologies
`were previously published [8–10]. Eligible patients entered
`a lead-in period (≤11 weeks). Patients were required to be
`treated with metformin (≥1500 mg/day) for ≥6 weeks prior to
`randomization; other OAMs were discontinued. After lead-in,
`patients were randomized to dulaglutide injection (seven doses
`during dose-finding; only selected dose(s) after dose selection
`occurred), sitagliptin 100 mg once daily or placebo (injectable
`and oral), all in combination with metformin.
`The objective of the dose-finding portion was to identify a
`dulaglutide dose with an optimal efficacy and safety profile,
`and possibly one lower dose to be available in case of an
`observed, unforeseen safety signal with the optimal dose
`during subsequent clinical development. For that purpose, a
`Bayesian adaptive algorithm was used. The algorithm involved
`adaptive randomization of patients to dulaglutide doses and
`evaluation of predefined dose decision rules on a biweekly
`basis. The decisions made by the algorithm were based on the
`posterior probability distributions available for each analysis.
`Posterior probability distributions changed with each analysis
`based on the additional data that had accrued in the interim.
`Patients were adaptively randomized to 1 of seven once-weekly
`dulaglutide doses (0.25, 0.50, 0.75, 1.0, 1.5, 2.0 and 3.0 mg),
`preferentially assigning higher probabilities to doses considered
`to have a more favourable clinical profile, and randomization
`to sitagliptin or placebo at 20% each; 3 : 1 : 1 ratio, respectively
`(Figure 1).
`Four efficacy and safety measures were considered important
`for dose selection based on early phase dulaglutide data: HbA1c,
`weight, pulse rate (PR) and diastolic blood pressure (DBP) [1].
`These measures were used to define criteria for dose selection.
`
`original article
`The selected dulaglutide dose(s) had to have a mean change
`of ≤+5 beats per minute (bpm) for PR and ≤+2 mmHg for
`DBP relative to placebo at 26 weeks. In addition, if a dose was
`weight neutral versus placebo, it had to show HbA1c reduction
`≥1.0% and/or be superior to sitagliptin at 52 weeks. If a dose
`reduced weight relative to placebo≥2.5 kg, then non-inferiority
`to sitagliptin would be acceptable.
`A clinical utility index (CUI) was incorporated in the
`algorithm to facilitate adaptive randomization and dose
`selection [8,9] based on the same parameters used to define
`dose-selection criteria described above. Longitudinal models
`were utilized to estimate HbA1c at 52 weeks and weight, PR and
`DBP at 26 weeks. The CUI was applied to estimate for HbA1c,
`weight, DBP and PR. A posterior probability distribution for
`the CUI and its components were calculated every 2 weeks. The
`CUI was a multiplicative index with a range of possible values
`from 0 to 6, with larger values reflecting a favourable clinical
`profile [9]. Zero values resulted if at least one component
`reached prespecified thresholds for a clinically unacceptable
`outcome (i.e. posterior mean increase in DBP ≥2.5 mmHg).
`Decision rules were based on posterior probability thresholds
`and predictive probabilities for meeting the primary study
`objective.
`During each biweekly interim assessment to support the
`adaptive algorithm,
`the dose with the highest posterior
`probability of having the largest CUI was designated the
`maximum utility dose (MUD). After 200 patients were
`randomized (the minimum sample size required for dose
`selection), if the MUD met predefined selection criteria (CUI
`≥0.6 and predictive probability of non-inferiority versus
`sitagliptin at 52 weeks for HbA1c change from baseline ≥0.85)
`at one of the interim assessments, that dose and possibly a
`lower dose would be selected. The lower dose was the next
`highest dose with an acceptable CUI ≥0.6 and ≤50% the dose
`
`Figure 1. Study design. aMetformin concomitant therapy from lead-in through treatment period (≥1500 mg/day). bLead-in period lasted up to 11 weeks.
`cThe dose finding period (indicated by the blue area) ended at the decision point (29 April 2009) resulting in different exposures within and across
`treatment groups. dAfter 26 weeks, patients in the placebo arm transitioned to sitagliptin in a blinded fashion.
`
`Volume 16 No. 8 August 2014
`
`doi:10.1111/dom.12305 749
`
`
`
`MPI EXHIBIT 1113 PAGE 2
`
`

`

`original article
`
`strength of the MUD to ensure minimal overlap of dulaglutide
`exposure. If
`there was strong evidence that no optimal
`dose existed (i.e. no therapeutic window), the algorithm
`would determine the study should be stopped because of
`‘futility’.
`The potential decision point was also adaptive (could have
`happened any time after 200 patients were randomized and
`up until 400 patients were randomized) (Figure 1). If data at
`the decision point supported selection of 1 or 2 doses, patients
`randomized to those specific doses and the comparator arms
`would continue the study, whereas patients assigned to non-
`selected dulaglutide doses would be discontinued. Additional
`patients would be randomized to selected doses and comparator
`arms using a fixed-allocation scheme to test the primary
`objective. The primary objective was to show non-inferiority
`(margin of 0.25%) of the selected optimal dulaglutide dose
`to sitagliptin in HbA1c change from baseline at 52 weeks
`[11]. Patients followed the same visit schedule and testing
`procedures, irrespective of when they were randomized into
`the trial
`An independent Data Monitoring Committee (DMC)
`external to Lilly provided oversight of the implementation of
`the adaptive algorithm and monitored study safety. The DMC
`fulfilled this role during the dose-finding portion, and contin-
`ued monitoring after dose selection until an interim database
`lock at 52 weeks, at which time the study was unblinded to
`assess the primary objectives. Sites and patients continued to
`be blinded to the treatment allocation until the completion
`of the study. The DMC was not allowed to intervene with the
`design operations. A Lilly Internal Review Committee (IRC),
`independent of the study team, would meet if the DMC recom-
`mended the study to be modified. The role of the IRC was to
`make the final decision regarding the DMC’s recommendation.
`The external Statistical Analysis Center (SAC) performed all
`interim data analyses for the DMC, evaluated the decision
`rules and provided the randomization updates for the adaptive
`algorithm.
`The DMC reviewed additional efficacy measures includ-
`ing fasting serum glucose and fasting plasma insulin; β-cell
`function and insulin sensitivity indices [updated Homeosta-
`sis Model Assessment (HOMA2)]; and lipids. Safety assess-
`ments included vital signs, adverse events, laboratory param-
`eters, hypoglycaemic episodes, electrocardiograms (ECGs)
`and dulaglutide antidrug antibodies. Pancreatic enzyme
`measurements began approximately 7 months after study ini-
`tiation, per regulatory guidance for GLP-1 receptor agonists.
`
`Statistical Analysis
`Interim analyses included all patients randomized and were
`based on all available data in the clinical trial database at
`the time of data transfers. After initiation of the adaptive
`algorithm, updated randomization probabilities were provided
`in a biweekly report by the SAC. The report summarized the
`exposure and the randomization probabilities for the next
`2 weeks. Raw data and Bayesian estimates (posterior means,
`posterior probabilities and posterior predictive probabilities)
`for the four CUI components, the CUI and Bayesian parameters
`related to the algorithm were plotted and tabulated. Posterior
`
`DIABETES, OBESITY AND METABOLISM
`
`probability distributions at an interim analysis represent the
`current knowledge about the parameter of
`interest. This
`Bayesian approach is referred to as ‘active learning’, which
`updates posterior distributions with data from the interim
`to help with decision making [9]. Ninety-five percentage
`credible intervals were calculated by taking the 2.5th and
`97.5th percentiles of the corresponding posterior probability
`distributions. The DMC chair and the lead SAC statistician
`reviewed these reports and were tasked to convene an
`unscheduled DMC meeting if an issue was identified with
`the algorithm or the decision point was triggered.
`Additional
`interim safety reports were generated for
`DMC meetings. These reports included summary tables
`of adverse events, serious adverse events, vital signs, ECG
`parameters, hypoglycaemic events, laboratory parameters and
`antihypertensive medications.
`After final database lock at 104 weeks, data from patients
`randomized in the dose-finding portion were reassessed for
`robustness of
`the dose-finding results (Appendix S1 for
`methods, Supporting information and Table S2 for results)
`Statistical analyses were performed using the sas system®
`version 8.2 or higher and Fortran 77.
`
`Results
`Patient Disposition and Exposure
`Patients were adaptively randomized to the seven dulaglutide
`doses during the dose-finding portion (Table S1). A total of
`230 patients were enrolled prior to the decision point. Of
`these, 199 patients had post-randomization data available to
`contribute to the evaluation of the decision rules. The other
`31 patients enrolled shortly before the decision point, and
`thus, had no available post-randomization data. The number
`of patients randomized to treatment arms and their baseline
`characteristics are provided in Table 1. Sponsor decision was the
`most common reason for early study discontinuation (Figure
`S1); this included all patients from dulaglutide arms who were
`discontinued at decision point because their respective doses
`were not selected.
`(10th biweekly interim analysis),
`At decision point
`mean exposure in all dulaglutide-treated patients was
`11.1 ± 7.5 weeks; placebo- and sitagliptin- treated patients had
`been exposed to study drug for (mean± SD) 9.2 ± 7.2 and
`10.7 ± 7.6 weeks, respectively (Table 2).
`
`Dose Selection
`The results of the first nine interim analyses were used only
`for adjustment of the dulaglutide randomization probabilities
`as specified in the protocol. After the ninth interim, the DMC
`recommended ceasing randomization to dulaglutide 3.0 mg
`because of safety concerns (additional data presented below);
`the IRC endorsed that recommendation. The decision rules
`were applied for the first time at the 10th interim, after 200
`patients had been enrolled. Results of the posterior unadjusted
`mean changes from baseline and mean changes from baseline
`adjusted for comparators for the four components of the CUI at
`this interim are shown in Tables 2 and 3, respectively. Figure 2
`
`750 Skrivanek et al.
`
`Volume 16 No. 8 August 2014
`
`
`
`MPI EXHIBIT 1113 PAGE 3
`
`

`

`DIABETES, OBESITY AND METABOLISM
`
`Table 1. Baseline characteristics, demographics and disposition of randomized patients.
`
`original article
`
`Variable
`
`Sex, n (%)
`Men
`Women
`Age (years)
`Race, n (%)
`Black
`White
`East Asian
`Hispanic
`Others
`BMI (kg/m2)
`Duration of
`diabetes (years)
`HbA1c
`[%, (mmol/mol)]
`SBP (mm Hg)
`DBP (mm Hg)
`
`Dulaglutide
`0.25 mg
`(n = 24)
`
`Dulaglutide
`0.5 mg
`(n = 25)
`
`Dulaglutide
`0.75 mg
`(n = 21)
`
`Dulaglutide
`1.0 mg
`(n= 10)
`
`Dulaglutide
`1.5 mg
`(n = 25)
`
`Dulaglutide
`2.0 mg
`(n = 30)
`
`Dulaglutide
`3.0 mg
`(n = 15)
`
`Sitagliptin
`(n = 42)
`
`Placebo
`(n= 38)
`
`9 (38)
`15 (63)
`57 (7)
`
`0 (0)
`12 (50)
`5 (21)
`6 (25)
`1 (4)
`31 (4)
`7 (4)
`
`7.8 (0.8)
`[62 (9)]
`130 (15)
`78 (9)
`
`13 (52)
`12 (48)
`55 (10)
`
`2 (8)
`9 (36)
`0 (0)
`14 (56)
`0 (0)
`33 (5)
`7 (4)
`
`10 (48)
`11 (52)
`52 (11)
`
`0 (0)
`13 (62)
`1 (5)
`7 (33)
`0 (0)
`33 (5)
`7 (5)
`
`3 (30)
`7 (70)
`55 (9)
`
`0 (0)
`4 (40)
`0 (0)
`6 (60)
`0 (0)
`34 (4)
`7 (5)
`
`10 (40)
`15 (60)
`53 (11)
`
`2 (8)
`10 (40)
`2 (8)
`11 (44)
`0 (0)
`32 (5)
`9 (7)
`
`8 (27)
`22 (73)
`53(11)
`
`3 (10)
`13 (43)
`6 (20)
`8 (27)
`0 (0)
`31 (5)
`7 (5)
`
`5 (33)
`10 (67)
`53 (11)
`
`1 (7)
`7 (47)
`0 (0)
`7 (47)
`0 (0)
`31 (5)
`7 (6)
`
`21 (50)
`21 (50)
`53 (12)
`
`2 (5)
`20 (48)
`4 (10)
`16 (38)
`0 (0)
`32 (4)
`9 (5)
`
`12 (32)
`26 (68)
`53 (10)
`
`2 (5)
`15 (40)
`4 (11)
`17 (45)
`0 (0)
`32 (4)
`7 (6)
`
`8.3 (1.3)
`[67 (14)]
`125 (16)
`76 (7)
`
`8.2 (1.1)
`[66 (12)]
`133 (16)
`82 (8)
`
`7.9 (0.6)
`[63 (7)]
`134 (13)
`79 (10)
`
`8.7 (1.5)
`[72 (16)]
`127 (16)
`77 (8)
`
`8.4 (1.0)
`[68 (11)]
`127 (15)
`79 (8)
`
`8.0 (1.1)
`[64 (12)]
`128 (17)
`76 (9)
`
`8.4 (1.1)
`[68 (12)]
`128 (12)
`77 (6)
`
`8.1 (1.1)
`[65 (12)]
`127 (12)
`78 (8)
`
`Data are means (SD) or n (%) unless otherwise indicated. BMI, body mass index; DBP, diastolic blood pressure; HbA1c, glycosylated haemoglobin A1C;
`SBP, systolic blood pressure; SD, standard deviation.
`
`provides the Bayesian estimates for the CUI and for each
`individual component. All results presented are based on data
`accrued up until the time of dose selection, the decision point.
`Posterior mean (95% credible interval) changes from
`baseline in HbA1c at 52 weeks ranged from 0.82 (−1.13,
`−0.53)% for dulaglutide 0.25 mg to−1.33 (−1.62,−1.00)% for
`dulaglutide 1.5 mg (Table 2 and Figure 2). The posterior mean
`change for sitagliptin was −0.70 (−0.87, −0.56)%. Dulaglutide
`1.5 mg showed the greatest posterior mean change compared
`to sitagliptin at 52 weeks −0.63 (−0.98, −0.20)% (Table 3).
`Posterior mean (95% credible interval) changes from
`baseline in weight at 26 weeks ranged from −0.96 (−1.47,
`−0.33) kg for dulaglutide 0.75 mg to −4.45 (−5.30, −3.70) kg
`for dulaglutide 3.0 mg (Table 2 and Figure 2). The posterior
`mean change from baseline for placebo was −0.52 (−0.99,
`0.01) kg (Table 2 and Figure 2). Across the range of dulaglutide
`doses (excluding discontinued 3.0 mg dose), dulaglutide 2.0 mg
`showed the greatest posterior mean change in weight compared
`to placebo at 26 weeks [−1.99 (−2.88, −1.20) kg] (Table 3).
`For PR, the posterior mean (95% credible interval) changes
`from baseline to 26 weeks for dulaglutide doses ranged from
`1.47 (−1.50, 4.13) bpm for dulaglutide 0.5 mg to 6.26 (3.95,
`10.95) bpm for dulaglutide 3.0 mg (Table 2 and Figure 2). The
`posterior mean change for placebo was 3.27 (1.23, 5.47) bpm
`(Table 2 and Figure 2). Across the range of dulaglutide
`doses (excluding discontinued 3.0 mg dose), dulaglutide 2.0 mg
`showed the greatest posterior mean change in PR compared to
`placebo at 26 weeks 0.78 (−2.10, 3.80) bpm (Table 3).
`The posterior mean (95% credible interval) changes from
`baseline to 26 weeks in DBP for dulaglutide doses ranged
`from −0.02 (−1.74, 1.78) mmHg for dulaglutide 0.25 mg to
`−0.99 (−2.22, 0.20) mmHg for dulaglutide 1.5 mg (Table 2
`and Figure 2). The posterior mean change for placebo was
`−0.37 (−2.88, 1.99) mmHg (Table 2 and Figure 2). Across
`
`the range of dulaglutide doses (excluding discontinued 3.0 mg
`dose), dulaglutide 1.5 mg showed the greatest posterior mean
`change in DBP compared to placebo at 26 weeks [−0.62 (−3.40,
`2.30) mmHg] (Table 3).
`Dulaglutide 1.5 mg was determined to be the MUD [3.1
`(0.7, 4.0)] at the 10th interim assessment (Figure 2). The
`posterior probability that the CUI for this dose of ≥0.6 sample
`size was 0.982 and the posterior predictive probability that
`dulaglutide 1.5 mg would show non-inferiority to sitagliptin
`at 52 weeks, based on a total sample size of 263 in each
`arm, was >0.99 (Figure 2). On the basis of these results, the
`algorithm determined the decision point had been reached, and
`selected two doses: dulaglutide 1.5 mg as the optimal dose and
`dulaglutide 0.75 mg as the second, lower dose (CUI= 1.054).
`It also determined the total sample size for each treatment arm
`in the trial based on the data from the dose-finding portion.
`The population PK/PD results supported the Bayesian-
`based results. A summary of the range of expected responses
`predicted by population PK/PD exposure-response models is
`provided in the Table S2 (Appendix S1 for details in method).
`These model-predicted responses and CUI values supported
`the Bayesian-based results shown in Figure 2. Dulaglutide
`1.5 mg was associated with a similar effect on HbA1c (mean;
`90% confidence interval) −1.27 (−1.72 to −0.84)% and weight
`−3.49 (−5.32 to 2.07) kg to that estimated during the dose-
`finding portion. Dulaglutide 1.5 mg also met the prespecified
`requirements of change in PR ≤5 bpm. No concentration-
`response relationship was identified for DBP, hence, there are
`no results presented for this CUI component.
`Data for all nine treatment arms from the final database up
`to the decision point were summarized for each component
`of the CUI. These reports (data not shown) were consistent
`with the results of assessments based on the datasets used for
`adaptation by the SAC.
`
`Volume 16 No. 8 August 2014
`
`doi:10.1111/dom.12305 751
`
`
`
`MPI EXHIBIT 1113 PAGE 4
`
`

`

`original article
`
`Table 2. Bayesian posterior mean (95% credible interval) changes from baseline and exposure.
`
`DIABETES, OBESITY AND METABOLISM
`
`Dose (mg)
`
`Dulaglutide 0.25
`
`Dulaglutide 0.5
`
`Dulaglutide 0.75
`
`Dulaglutide 1.0
`
`Dulaglutide 1.5
`
`Dulaglutide 2.0
`
`Dulaglutide 3.0
`
`Sitaglitpin
`
`Placebo
`
`HbA1c (%)
`52 weeks
`−0.82
`(−1.13; −0.53)
`n= 13
`−0.95
`(−1.16; −0.75)
`n= 16
`−0.93
`(−1.17; −0.65)
`n= 20
`−1.00
`(−1.25; −0.69)
`n= 8
`−1.33
`(−1.62; −1.00)
`n= 18
`−1.28
`(−1.46; −1.07)
`n= 24
`−1.00
`(−1.30; −0.78)
`n= 10
`−0.70
`(−0.87; −0.56)
`n= 25
`0.01
`(−0.27; 0.27)
`n= 28
`
`Weight (kg)
`26 weeks
`−1.01
`(−1.53; −0.35)
`n = 16
`−1.48
`(−2.04; −0.86)
`n = 22
`−0.96
`(−1.47; −0.33)
`n = 20
`−2.05
`(−2.75; −1.34)
`n = 9
`−2.18
`(−2.68; −1.74)
`n = 22
`−2.50
`(−3.20; −1.78)
`n = 29
`−4.45
`(−5.30; −3.70)
`n = 13
`−0.45
`(−0.89; 0.06)
`n = 35
`−0.52
`(−0.99; 0.01)
`n = 33
`
`Pulse rate (bpm)
`26 weeks
`
`2.12
`(−0.83; 3.84)
`n = 16
`1.95
`(−0.35; 4.15)
`n = 22
`1.47
`(−1.50; 4.13)
`n = 20
`2.56
`(0.19; 4.34)
`n = 9
`3.2
`(1.25; 4.99)
`n = 22
`4.05
`(2.63; 5.70)
`n = 29
`6.26
`(3.95; 10.95)
`n = 13
`−0.90
`(−2.80; 1.09)
`n = 35
`3.27
`1.23; 5.47)
`n = 33
`
`DBP (mm Hg)
`26 weeks
`−0.02
`(−1.74; 1.78)
`n = 16
`−0.26
`(−1.67; 0.81)
`n = 22
`−0.65
`(−1.92; 0.34)
`n = 20
`−0.92
`(−2.36; 0.38)
`n = 9
`−0.99
`(−2.22; 0.20)
`n = 22
`−0.78
`(−2.05; 0.47)
`n = 29
`−0.57
`(−2.70; 1.85)
`n = 13
`−1.22
`(−3.52; 0.75)
`n = 35
`−0.37
`(−2.88; 1.99)
`n = 33
`
`Mean utility
`
`0.733
`(0, 1.639)
`
`1.362
`(0, 2.326)
`
`1.054
`(0.090, 1.923)
`
`2.021
`(0.415, 3.267)
`
`3.052
`(0.721, 4.037)
`
`2.996
`(0, 3.702)
`
`Not applicable
`
`Not applicable
`
`Not applicable
`
`Exposure (weeks)
`mean ± SD
`8.9± 8.5
`
`9.7± 7.1
`
`15.7± 6.2
`
`13.0± 8.8
`
`12.5± 7.4
`
`10.0± 5.2
`
`9.2± .8.8
`
`10.7± 7.6
`
`9.2± 7.2
`
`Statistics presented are posterior means and 95% credible intervals based on data available at the decision point (10th Interim, 29 April 2009). bpm, beats
`per minute; DBP, diastolic blood pressure; HbA1c, glycosylated haemoglobin A1c; SD, standard deviation.
`
`Safety
`The most common adverse events reported during dose-
`finding were gastrointestinal (including nausea, diarrhoea and
`vomiting) and urinary tract infection (Table S3). Nausea and
`vomiting were more common with dulaglutide than sitagliptin
`and placebo. The highest incidence was observed in patients
`with doses ≥1.0 mg. There were a total of three serious adverse
`events (SAEs) (dulaglutide 0.5 mg: pneumonia [1]; dulaglutide
`3.0 mg: gastroenteritis [1] and placebo: cervical dysplasia
`[1]). Approximately 1 year after study discontinuation, a
`patient exposed to dulaglutide 2.0 mg for 6 months was
`diagnosed with an SAE of medullary thyroid carcinoma
`(MTC). The patient had increased calcitonin levels before
`randomization [91.5 pg/ml (reference range 0–11.5 pg/ml),
`evaluated retrospectively] which decreased upon exposure to
`dulaglutide. At the time of dulaglutide discontinuation and
`3 months thereafter, the patient’s calcitonin was 61.7 and
`82.8 pg/ml, respectively. Subsequently, this patient was also
`found to be positive for the RET proto-oncogene mutation,
`indicating a preexisting neoplasm in an individual with high
`risk of MTC. At the time of this patient’s randomization,
`there were no exclusionary criteria for patients at increased
`risk of C-cell neoplasm. Four patients each in the dulaglutide
`2.0 mg, dulaglutide 3.0 mg, and placebo arms, and 1 each in the
`dulaglutide 0.25 mg and dulaglutide 1.0 mg arms, discontinued
`
`the study because of adverse events. Nausea was the most
`commonly reported adverse event that led to discontinuation
`with dulaglutide (2.0 mg: one patient; 3.0 mg: three patients).
`In the placebo group, hyperglycaemia was the most common
`adverse event resulting in discontinuation (two patients).
`The proportion of patients with post-baseline lipase levels
`greater than ×3 upper limit of normal (ULN) was 2.5%
`(dulaglutide 3.0 mg one patient; dulaglutide 2.0 mg two
`patients; sitagliptin one patient). Increase in lipase above
`ULN was the most common treatment-emergent abnormal
`laboratory finding and ranged from 13% for placebo to 46%
`for dulaglutide 2.0 mg. Similar increases in pancreatic amylase
`were reported, but were lower in frequency. There were no
`events of pancreatitis reported during dose-finding.
`Dulaglutide 3.0 mg was discontinued because of increase in
`PR (posterior mean for PR >5 bpm) and higher incidence
`of gastrointestinal adverse events (nausea and vomiting)
`with/without higher increases in pancreatic enzymes compared
`to lower doses.
`
`Discussion
`AWARD-5 was the first Phase 3 trial of dulaglutide. The
`first portion of the trial served as the dose finding and dose
`selection trial for the dulaglutide development programme. An
`algorithm was employed to adaptively randomize patients to
`
`752 Skrivanek et al.
`
`Volume 16 No. 8 August 2014
`
`
`
`MPI EXHIBIT 1113 PAGE 5
`
`

`

`DIABETES, OBESITY AND METABOLISM
`
`Table 3. Bayesian posterior adjusted mean (95% credible interval) changes from baseline.
`
`Dose (mg)
`
`Dulaglutide 0.25
`
`Dulaglutide 0.5
`
`Dulaglutide 0.75
`
`Dulaglutide 1.0
`
`Dulaglutide 1.5
`
`Dulaglutide 2.0
`
`Dulaglutide 3.0
`
`HbA1c (%)
`52-week
`sitagliptin-adjusted
`−0.12
`(−0.38; 0.20)
`n= 13
`−0.24
`(−0.44; 0.02)
`n= 16
`−0.23
`(−0.44; 0.08)
`n= 20
`−0.30
`(−0.58; 0.08)
`n= 8
`−0.63
`(−0.98; −0.20)
`n= 18
`−0.58
`(−0.76; −0.32)
`n= 24
`−0.29
`(−0.62; −0.02)
`n= 10
`
`Weight (kg)
`26-week
`placebo-adjusted
`−0.49
`(−1.36; 0.44)
`n = 16
`−0.97
`(−1.80; −0.12)
`n = 22
`−0.44
`(−1.16; 0.44)
`n = 20
`−1.53
`(−2.44; −0.60)
`n = 9
`−1.67
`(−2.48; −0.92)
`n = 22
`−1.99
`(−2.88; −1.20)
`n = 29
`−3.93
`(−4.84; −2.88)
`n = 13
`
`Pulse rate (bpm)
`26-week
`placebo-adjusted
`−1.16
`(−4.40; 1.90)
`n = 16
`−1.32
`(−4.30; 1.80)
`n = 22
`−1.80
`(−5.50; 2.10)
`n = 20
`−0.71
`(−4.00; 2.50)
`n = 9
`−0.07
`(−3.20; 2.70)
`n = 22
`0.78
`(−2.10; 3.80)
`n = 29
`2.98
`(−0.60; 8.70)
`n = 13
`
`original article
`
`DBP (mm Hg)
`26-week
`placebo-adjusted
`
`0.35
`(−2.60; 3.60)
`n = 16
`0.11
`(−2.40; 3.20)
`n = 22
`−0.27
`(−3.00; 2.50)
`n = 20
`−0.55
`(−3.90; 2.20)
`n = 9
`−0.62
`(−3.40; 2.30)
`n = 22
`−0.41
`(−2.60; 3.20)
`n = 29
`−0.19
`(−2.70; 2.80)
`n = 13
`
`Mean utility
`
`0.733
`(0, 1.639)
`
`1.362
`(0, 2.326)
`
`1.054
`(0.090, 1.923)
`
`2.021
`(0.414, 3.267)
`
`3.052
`(0.721, 4.037)
`
`2.996
`(0, 3.702)
`
`Not applicable
`
`Statistics presented are model predicted means and 95% credible intervals based data available at the decision point. bpm, beats per minute; DBP, diastolic
`blood pressure; HbA1c, glycosylated haemoglobin A1c.
`
`a range of seven dulaglutide doses and to select up to two
`doses. This design enabled simultaneous assessment of efficacy
`and safety measures deemed relevant for dose selection based
`on the results of the completed clinical pharmacology studies,
`optimized patient allocation using data accumulated during
`the dose-finding portion and selected the optimal dose when
`enough data to reach a decision were available. Dulaglutide
`1.5 mg was selected as the optimal dose and dulaglutide 0.75 mg
`was selected as the lower dose for further assessment.
`The data met the protocol prespecified conditions for
`dose selection and the assignment algorithm dropped all but
`two active doses, dulaglutide 1.5 and 0.75 mg, at the 10th
`interim analysis. Dulaglutide 1.5 mg had the highest posterior
`probability of being the MUD. Assessment of the effects of
`dulaglutide 1.5 mg on the posterior probability distributions
`for the four components of the CUI showed that this dose had
`robust glucose and weight lowering effects, and was associated
`with an increase in PR that was smaller than the effect of the
`adjacent 2.0 mg dose. These observations explain the selection
`of the 1.5 mg dose. The 0.75 mg dose met the prespecified
`criteria for a lower dulaglutide dose. The selection of the
`dulaglutide 1.5 and 0.75 mg doses were also supported by
`PK/PD model-based analyses and the results of the efficacy and
`safety assessments up to 104 weeks [5,6,12–16].
`The dose responses for PR and weight are different than for
`HbA1c. It appears, based on this study (Table 2) and previous
`studies [1], that the dose-response attenuates around 1.5 mg
`for HbA1c but continues for weight and PR. This closely linked
`dose effect for weight loss is in line with expectations for the
`GLP-1 receptor agonist class and may have significant value
`
`in combination with other antidiabetes therapies that cause
`weight gain (e.g. insulin).
`Although dulaglutide was commonly associated with GI
`effects in earlier Phase 1 studies, that effect was not considered
`to be dose limiting in the dose range included in the AWARD-5
`study. Phase 1 data indicated that PR and DBP would be dose
`limiting before GI would become a factor in dose selection.
`Because there was no indication of potential dose–response
`effect for SBP, PR and DBP were included in the CUI. In this
`study we did not observe any dose response with DBP, but PR
`was indeed dose limiting. There were no DMC concerns related
`to GI issues with the selected doses.
`the DMC
`After
`the ninth interim data assessment,
`recommended discontinuing the dulaglutide 3.0 mg dose
`because of increases in PR, higher incidence of GI adverse
`events, and higher incidence in pancreatic enzyme values
`above ULN. These changes were not unexpected based on the
`known dose-response effects of the GLP-1 receptor agonist
`class, and observations from completed dulaglutide trials
`[1,12,17,18]. Although the changes in pancreatic enzymes have
`not been completely understood, recent reports have showed
`an increased incidence of asymptomatic increases above upper
`limit of the normal range in lipase and amylase in patients with
`T2D [19–23]. The incidence increases further with exposure
`to GLP-1 receptor agonists; these asymptomatic findings have
`not been associated with any adverse outcome in clinical trials
`[24,25].
`The Bayesian adaptive approach used in AWARD-5 has
`not been used often in drug development, although it offers
`substantial benefits in terms of accuracy and efficiency in
`
`Volume 16 No. 8 August 2014
`
`doi:10.1111/dom.12305 753
`
`
`
`MPI EXHIBIT 1113 PAGE 6
`
`

`

`original article
`
`DIABETES, OBESITY AND METABOLISM
`
`Figure 2. Dose–response model and CUI. CUI and change from baseline in CUI components, posterior means and 95% credible intervals at 6 months
`(DBP, PR and weight) and 12 months (HbA1c) (data available at decision point, the 10th Interim, 29 April 2009). bpm, beats per minute; CUI, clinical
`utility index; DBP, diastolic blood pressure; DU, dulaglutide; HbA1c, glycosylated haemoglobin A1C; PK/PD, pharmacokinetic/pharmacodynamics; PL,
`placebo; PR, pulse rate; SITA, sitagliptin.
`
`comparison to a fixed design [26,27]. Several design aspects [use
`of prior information (information gained from Phase 1 studies,
`the mechanism of action and general knowledge about the
`class), adaptive randomization and dose selection] were utilized
`to increase the efficiency and quality of dose-finding. First, the
`Bayesian approach explicitly used prior information to inform
`the dose response and the longitudinal profiles of the treatment
`arms. This decreased the uncertainty of the CUI, allowing for
`earlier adaptations and decision making. Second, the adaptive
`randomization enabled exploration of more dulaglutide doses
`than would be possible with a traditional fixed-allocation
`strategy, given total sample size constraints. This was achieved
`by allowing data accumulating during the trial to drive
`randomization probabilities, in effect, shifting patients away
`from doses with a less favourable clinical profile to doses more
`probably to meet target clinical profiles (Figure 2). Finally,
`
`the adaptive algorithm selected doses when it determined
`adequate safety and efficacy information was collected; whereas
`a fixed design would stop after a predetermined study length
`irrespective of whether there was sufficient information to make
`a decision. The efficiency of the utilized design is showed by the
`relatively small sample size (230 patients) and short exposures
`(median exposure: 4–16 weeks) needed