F1K-MC-EVDP Statistical Analysis Plan Page 12

F1K-MC-EVDP Statistical Analysis Plan Page 1

1. Title Page Statistical Analysis Plan for Clinical Studies: Efficacy and Safety of Drotrecogin Alfa (Activated) in Adult Patients with Septic Shock

LY203638 [Drotrecogin Alfa (Activated) Septic Shock

A randomized, double-blind, placebo-controlled, multicenter study of drotrecogin alfa (activated) administered as a continuous 96-hour infusion to adult patients with septic shock.

Eli Lilly and Company Protocol F1K-MC-EVDP Phase 3 SAP Version 1

Confidential Information

The information contained in this Statistical Analysis Plan (SAP) is confidential and the information contained within it may not be reproduced or otherwise disseminated without the approval of Eli Lilly and Company or its subsidiaries.

Drotrecogin Alfa (Activated) (LY203638) Draft: 29 April 2018 F1K-MC-EVDP Statistical Analysis Plan Page 2

2. Table of Contents

Section Page

1. Title Page...... 1 2. Table of Contents...... 2 3. Revision History...... 3 4. Abbreviations...... 4 5. Study Objectives...... 5 5.1. Primary Objective...... 5 5.2. Secondary Objectives...... 5 6. A Priori Statistical Methods...... 6 6.1. Analysis Plan...... 6 6.2. General Considerations...... 6 6.3. Adjustments for Prerandomization Covariates...... 8 6.4. Handling of Dropouts or Missing Data...... 8 6.5. Multicenter Studies...... 9 6.6. Multiple Comparisons/Multiplicity...... 9 6.7. Patient Disposition...... 9 6.8. Patient Characteristics...... 9 6.9. Efficacy Analyses...... 11 6.9.1. Primary Efficacy Analysis...... 11 6.9.2. Secondary Efficacy Analyses...... 12 6.9.3. Other Efficacy Analyses...... 13 6.10. Quality-of-Life Analyses...... 16 6.11. Biomarker Analyses...... 17 6.12. Safety Analyses...... 17 6.13. Subgroup Analyses...... 18 6.14. Supplementary Analyses...... 20 6.15. Interim Analyses and Data Monitoring...... 23 6.16. Determination of Sample Size...... 25 6.16.1. Resizing the Study...... 25 7. Unblinding Plan...... 27 8. References...... 29

3.Revision History

SAP Version 1 was approved prior to unblinding.

4.Abbreviations

ANOVA Analysis of variance

APACHE II Acute Physiology and Chronic Health Evaluation II – a scale to assess severity of disease. A higher score indicates more severe disease.

ARDS Acute respiratory distress syndrome

COPD Chronic obstructive pulmonary disease

DMC Data Monitoring Committee

Euro-QoL-5D A health-related quality of life instrument

ITT Intention-to-treat

LOCF Last observation carried forward

MedDRA Medical Dictionary for Regulatory Activities

PAI-1 Plasminogen activator inhibitor 1

PROWESS Recombinant human PROtein C Worldwide Evaluation in Severe Sepsis study.

PT Prothrombin time

SAP Statistical analysis plan

SF-12 Short Form-12. A survey instrument to assess functional health and well-being.

SIRS Systemic inflammatory response syndrome

SOFA Sequential Organ Failure Assessment – a scale to assess organ dysfunction in each of 6 organ systems: cardiovascular, respiratory, renal, hematologic, liver, and central nervous system (central nervous system dysfunction will not be assessed using the SOFA methodology in this study). Scores range from 0-no dysfunction to 4-organ failure.

5.Study Objectives

5.1. Primary Objective The primary objective of this study is to demonstrate that treatment with drotrecogin alfa (activated) 24 mcg/kg/h administered as an intravenous infusion for 96 hours reduces 28-day all-cause mortality in adult patients with septic shock compared with placebo.

5.2. Secondary Objectives The secondary objectives of this study are as follows:  To demonstrate that treatment with drotrecogin alfa (activated) reduces 28-day all-cause mortality in adult patients with septic shock and severe protein C deficiency (baseline protein C level less than or equal to half the lower limit of normal) compared with placebo.  To demonstrate that treatment with drotrecogin alfa (activated) improves cardiovascular, respiratory, and renal organ function compared with placebo.  To demonstrate that treatment with drotrecogin alfa (activated) reduces 90-day and 180-day all-cause mortality in adult patients with septic shock compared with placebo.  To demonstrate that patients treated with drotrecogin alfa (activated) have a similar quality of life compared with patients treated with placebo.  To demonstrate that drotrecogin alfa (activated) has an acceptable safety profile in this patient population.

6. A Priori Statistical Methods

6.1. Analysis Plan The analysis plan for this study will investigate the effect of drotrecogin alfa (activated) treatment on 28-day mortality and the factors associated with mortality. Primary analysis. The primary analysis will investigate the effect of drotrecogin alfa (activated) on mortality. The primary endpoint is 28-day all-cause mortality. Sensitivity analyses will be performed that include patients with unknown 28-day survival status. Secondary analyses. Secondary analyses will further investigate the effect of drotrecogin alfa (activated) treatment on the following: (1) mortality in a population at a high risk of death, patients with severe protein C deficiency (protein C level less than or equal to half the lower limit of normal [≤40%]), (2) organ function, and (3) long-term outcomes of septic shock and quality of life. The safety of drotrecogin alfa (activated) will also be investigated.  Organ function analyses. These analyses will evaluate the change in cardiovascular, respiratory, and renal dysfunction from baseline to postbaseline timepoints.  Long-term outcomes of septic shock and quality of life. All-cause mortality will be assessed at 90 and 180 days postrandomization. The change in quality of life from baseline will be evaluated at 28 days, 90 days, and 180 days postrandomization.  Safety of drotrecogin alfa (activated). The following categories of adverse events will be compared between treatment groups (1) serious adverse events, (2) nonserious bleeding events, (3) serious bleeding events, (4) nonserious thrombotic events, and (5) serious thrombotic events.

6.2. General Considerations The principal investigators for this study, Dr Taylor Thompson and Dr. Marco Ranieri, have identified an academic statistical center (Duke Clinical Research Institute) that, in cooperation with the Academic Steering Committee (Dr. Taylor Thompson, Dr. Marco Ranieri, Dr. Simon Finfer, Dr. Philip Barie, Dr. Jean-Francois Dhainaut, Dr. Ivor Douglas, Dr. Bengt Gardlund, Dr. John Marshall, and Dr. Andrew Rhodes), will be responsible for reviewing and approving this statistical analysis plan (SAP), including all of the prospectively defined analyses of efficacy and safety. The SAP will be filed with the FDA prior to the data being unblinded and, at the same time, will be made freely and publicly available on an academic website. The academic statistical center will be responsible for conducting the primary analysis of the study data, and preparing the main

Drotrecogin Alfa (Activated) (LY203638) Draft: 29 April 2018 F1K-MC-EVDP Statistical Analysis Plan Page 7 report of study results for the Academic Steering Committee to create the primary manuscript and any additional analyses of the study data for subsequent manuscripts that, in the view of the Academic Steering Committee, are material to understanding the efficacy and safety of drotrecogin alfa (activated). The academic statistical center will also be responsible for performing any post hoc analyses requested by the Academic Steering Committee. The academic statistical center will have unfettered access to the full study database and to the randomization codes at the time that the database is locked and the study is unblinded. Lilly or its designee will independently conduct the prospectively defined analyses of efficacy and safety for inclusion in the clinical study report for submission to regulatory authorities. Any discrepancies will be resolved to the satisfaction of both parties. The efficacy analyses will be conducted on an ITT basis unless otherwise specified. The ITT population for this study will consist of all randomly-assigned patients, regardless of whether the patient receives study drug. An ITT analysis is an analysis of data by the treatment groups to which patients are assigned by random allocation even if the patient does not receive the correct treatment or otherwise does not follow the protocol. Efficacy analyses will also be performed for all treated patients, the subset of patients in the ITT population who receive study drug for any length of time. Safety analyses will be performed for all treated patients. The SAP and List of Analyses will include a comprehensive description of the statistical methods and reports to be included in the final study report. No member of the Lilly clinical team or the Academic Steering Committee will be unblinded to the clinical data before all patients have completed the study and data has been locked; therefore, all analyses specified in this SAP are considered prospectively defined. Any change to the data analysis methods described in the protocol will require an amendment ONLY if it changes a principal feature of the protocol. Any other change to the data analysis methods described in the protocol, and the justification for making the change, will be described in the clinical study report. Additional exploratory analyses of the data will be conducted as deemed appropriate. Data for qualitative variables will be presented as proportions (total number of patients, number of events, and percent). For qualitative variables, treatment groups will be compared using chi-square and Cochran-Mantel-Haenszel tests. Relative risks and confidence intervals will be calculated using the Mantel-Haenszel adjusted relative risk method. Data for continuous variables will be summarized using measures of central tendency and dispersion. Statistical tests for continuous variables will be performed using analysis of variance (ANOVA) based on ranked and unranked data (Snedecor and Cochran 1989) unless otherwise specified. Two-sided 5% significance levels and 95% confidence intervals will be used for all efficacy and safety analyses unless otherwise specified. The two-sided alpha level for the primary analysis will be adjusted to account for the planned interim analyses.

6.3. Adjustments for Prerandomization Covariates The primary analysis of 28-day mortality will not be adjusted for any covariate. A supportive analysis of the primary endpoint will be performed in which adjustments are made for important prerandomization covariates. The quoted p-value will be based on the Cochran-Mantel-Haenszel test stratified by baseline APACHE II quartile, baseline protein C activity class (40%, 41% to 60%, 61% to 80%, >80%, and unknown), and age class (<65, 65). The primary quoted relative risk and corresponding confidence interval will be calculated using the Mantel-Haenszel adjusted relative risk method with an adjustment for baseline APACHE II quartile, baseline protein C activity class, and age class. Handling of missing data for adjusted analysis. To conduct the risk-adjusted analysis, baseline APACHE II score, baseline protein C activity, age, and 28-day mortality status are required. If a patient’s baseline protein C activity is not available, the patient will be included in the baseline protein C activity class designated “unknown.” If a patient’s 28- day mortality status is unknown, the patient will be excluded from all mortality analyses. Algorithm for pooling under-represented strata. Forty strata are defined by baseline APACHE II quartile, baseline protein C activity class, and age class. It is expected that there will be relatively few patients in a treatment group within some of the strata. The following algorithm will be used to pool strata prior to conducting the analysis. The algorithm begins by organizing the strata by the following risk adjustment factor hierarchy: (1) baseline APACHE II quartile, (2) baseline protein C activity class, and (3) age class. At the first stage of the algorithm, for a set of 2 strata with the same baseline APACHE II quartile and baseline protein C activity class, but different age classes, the 2 strata will be pooled if one of the 2 treatment groups has fewer than 10 patients in at least 1 of the 2 strata. This pooling algorithm will be assessed individually for each baseline APACHE II quartile/baseline protein C activity class combination. At the second stage of the algorithm, for all remaining strata within the same baseline APACHE II quartile, the original 5-level protein C activity classification will be condensed to a 2-level classification (60% or unknown, >60%) if, for at least 1 of the 5-level defined strata, at least 1 of the 2 treatment groups has fewer than 10 patients in the strata. At the third stage of the algorithm, all remaining strata within the same baseline APACHE II quartile will be combined if at least 1 of the 2 treatment groups has fewer than 10 patients in a strata. No pooling of strata defined by different baseline APACHE II quartiles will occur.

6.4. Handling of Dropouts or Missing Data The last observation carried forward (LOCF) method of imputation will be used to impute missing values in analyses in which imputation is used with the exception of the following analyses:

 Mortality analyses. Patients with unknown survival status will be excluded from landmark analyses of mortality unless otherwise specified (sensitivity analyses that account for patients with unknown survival status will be performed). Patients with unknown survival status will be censored at the last date for which the patient was known to be alive or in the study hospital for all survival analyses.  Quality of life analyses. Patients with missing data for reasons other than death will be excluded from the analyses.

6.5. Multicenter Studies This study is an international, multicenter study with a small expected patient enrollment per site. For this reason, there are no plans to use investigative site as a covariate in the analyses defined in this SAP. For regulatory reporting purposes, reports of significant protocol violations by investigator, patient allocation by investigator, and listings of deaths and other serious/significant adverse events per investigative site will be produced as well as analyses of 28-day all-cause mortality by country, origin, region, and investigative site.

6.6. Multiple Comparisons/Multiplicity The primary efficacy analysis is based on an unstratified chi-square test for the ITT population. Significance levels at the interim analyses and at the final analysis will be adjusted so that the overall error rate is 0.05, as explained in Section 6.15 and the appendix. Multiplicity adjustments for secondary outcomes are not anticipated in the 2- group study.

6.7. Patient Disposition Differences between the treatment groups in the frequencies with which each reason for withdrawal from the study is observed will be summarized and presented. The reasons for discontinuation are completed study, lost to follow-up, investigator decision, sponsor decision, and subject decision. The data will be presented as counts and proportions (number of patients, number of events, and percent). Treatment comparisons will be performed using chi-square tests.

6.8. Patient Characteristics Summary reports of baseline characteristics will be produced for the ITT population and the population of patients with severe protein C deficiency (ITT severe deficiency population). Two-sided p-values from chi-square will be used to summarize the comparability of the treatment groups for categorical variables and ANOVA tests for continuous variables.

The baseline characteristics to be summarized are the following. Demographic characteristics: gender (female, male), age, age classification (<65, ≥65), origin (Caucasian, African, Hispanic, native American, east Asian/Pacific, west Asian, aboriginal and/or Torres Strait Islander), country (Australia, Belgium, Brazil, Canada, Czech Republic, Finland, France, Germany, India, Italy, Mexico, New Zealand, Portugal, Spain, Switzerland, The Netherlands, United Kingdom, United States), region (Europe, United States/Canada, rest of world), patient location prior to hospitalization (home-no support, home-unpaid support, home-paid support, other acute care hospital, other care facility, rehabilitation center, skilled nursing facility, unskilled nursing facility, unknown). Clinical characteristics: number of SIRS criteria met (1, 2, 3, 4), SIRS criteria-body temperature (no, yes), SIRS criteria-heart rate (no, yes), SIRS criteria-respiratory (no, yes), SIRS criteria-white blood cell count (no, yes) evidence of hypoperfusion criteria met (renal only; metabolic acidosis only; hepatic only; renal and metabolic acidosis; renal and hepatic; metabolic acidosis and hepatic; renal, metabolic acidosis, and hepatic), number of baseline organ dysfunctions (1, 2, 3, 4, 5), organ dysfunction class (single, multiple), respiratory organ dysfunction criteria met (no, yes), renal organ dysfunction criteria met (no, yes), hematology organ dysfunction criteria met (no, yes), metabolic acidosis organ dysfunction criteria met (no, yes), time from onset of septic shock to start of infusion, baseline APACHE II score, APACHE II median ( median, > median), APACHE II quartiles, APACHE II Acute Physiology Score, APACHE II Chronic Health Points (0, 2, 5), Glasgow Coma score, APACHE II class (<25, 25), recent surgery (no, yes), steroid use (no, yes), baseline mechanical ventilation status (no, yes), baseline renal replacement therapy status (no, yes), acute respiratory distress syndrome (ARDS) status (no, yes). Infection characteristics: presumed or known site of infection (abdomen, blood, bone, central nervous system, head, heart, lung, other, pleura, reproductive tract, skin/skin structure, urinary tract), source of infection (community, nosocomial), evidence of infection (abdominal infection, bacterial meningitis, bloodstream infections, lower respiratory tract, purpura fulminans, pyelonephritis, skin or soft tissue infection, other), infecting organism identified (no, yes), how infecting organism identified (antigen detection, DNA testing, serology, culture), infecting organism sensitive to antibiotics administered during pretreatment period (no, yes, organism not isolated), bacterial pathogen (no, yes), fungal pathogen (no, yes), viral pathogen (no, yes), parasitic pathogen (no, yes), unknown pathogen (no, yes), time from the start of treatment with initial antibiotic to the start of vasopressor therapy. Patient history: alcohol dependence (no, yes, unknown), chronic liver disease (no, yes, unknown), chronic obstructive pulmonary disease (COPD) (no, yes, unknown), chronic renal disease (no, yes, unknown), congestive heart failure (no, yes, unknown), coronary artery disease (no, yes, unknown), diabetes mellitus (no, yes, unknown), hypertension (no, yes, unknown), immunodeficiency (no, yes, unknown), malignancy (no, yes,

Drotrecogin Alfa (Activated) (LY203638) Draft: 29 April 2018 F1K-MC-EVDP Statistical Analysis Plan Page 11 unknown), pancreatitis (no, yes, unknown), stroke (no, yes, unknown), family history of thrombosis (no, yes, unknown), family history of thrombophilia (no, yes, unknown). Baseline anticoagulant exposure: unfractionated heparin (no, yes), low molecular weight heparin (no, yes), fondaparinux (no, yes). Baseline organ function: lactate level, lactate class (<2.5 mMol/L, 2.5 mMol/L), bicarbonate level, bicarbonate class (<18 mMol/L, 18 mMol/L), pressure-adjusted heart rate, intravenous fluids, central venous oxygenation, vasopressor dosage at the start of the study drug infusion (vasopressors listed on the case report form)1, was the patient still on vasopressor support at or above the inclusion criteria doses (criterion 4B) at the start of the study drug infusion (no, yes), cardiovascular SOFA score, cardiovascular SOFA score as a factor (0, 1, 2, 3, 4), PaO2/FiO2 ratio, SpO2/FiO2 ratio, mechanical ventilation-mean airway pressure, mechanical ventilation-plateau pressure, central venous pressure, respiratory SOFA score, respiratory SOFA score as a factor (0, 1, 2, 3, 4), creatinine level, microalbumin/creatinine level, urine output, renal SOFA score, renal SOFA score as a factor (0, 1, 2, 3, 4), coagulation SOFA score, coagulation SOFA score as a factor (0, 1, 2, 3, 4), liver SOFA score, liver SOFA score as a factor (0, 1, 2, 3, 4). Baseline laboratory data: protein C activity, protein C class (40%, 41% to 60%, 61% to 80%, >80%, unknown), activated partial thromboplastin time, prothrombin time, plasminogen activator inhibitor 1. Baseline microbiological data: microorganisms infecting at least 1 patient in order of most common to least common.

6.9. Efficacy Analyses

6.9.1. Primary Efficacy Analysis The primary efficacy analysis will be the comparison of 28-day all-cause mortality in the drotrecogin alfa (activated) and placebo treatment groups for the ITT population. Patients will be classified as “alive at Study Day 28” or, if dead, “dead at Study Day 28,” regardless of cause of death. The 28-day time point will occur 672 hours from randomization. Patients with unknown survival status will be excluded from this analysis. Treatment comparisons will be performed using a 2-sided Pearson’s chi-square test unadjusted for continuity. Relative risk (risk ratio) and odds ratio estimates with associated 95% confidence intervals will be presented. The primary final analysis 2- sided critical alpha level will be 0.05 adjusted for multiple interim analyses. The 2-sided alpha threshold at the conclusion of the study will be chosen to ensure that the overall type I error for the efficacy hypothesis will be less than 0.05.

1 Postrandomization variable.

Survival analyses will also be performed. Differences between the drotrecogin alfa (activated) and placebo treatment groups in the time to death from randomization through 28 days will be assessed using a log-rank test. Survival curves will be produced using the method of Kaplan and Meier (Kaplan and Meier 1958). Sensitivity analyses of 28-day mortality to account for patients with unknown survival status will also be performed. Two analyses will be performed: (1) all patients with unknown survival status will be assumed to be dead at Study Day 28 and (2) all drotrecogin alfa (activated) patients with unknown survival status will be assumed to be dead at Study Day 28 and all placebo patients with unknown survival status will be assumed to be alive at Study Day 28. Treatment comparisons will be performed using a chi-square test. Relative risk and odds ratio estimates with associated 95% confidence intervals will be presented.

6.9.2. Secondary Efficacy Analyses The following secondary efficacy analyses will be conducted. Twenty-eight-day mortality in patients with severe protein C deficiency. An analysis of 28-day mortality in patients with severe protein C deficiency (protein C activity less than or equal to half the lower limit of normal) will be conducted. Treatment comparisons will be performed using a chi-square test. Relative risk and odds ratio estimates with associated 95% confidence intervals will be presented. Patients with missing survival status will be excluded from the analysis. Long-term mortality. Survival status will be assessed at Study Day 90 and Study Day 180. Patients will be classified as “alive at Study Day 90 (or 180)” or, if dead, “dead at Study Day 90 (or 180).” Because some patients will be lost to follow-up and, therefore censored, mortality rates will be estimated using the method of Kaplan and Meier. Differences between drotrecogin alfa (activated) and placebo in mortality rates at these landmarks will be evaluated using the assumption of asymptotic normality. Standard errors of the mortality estimates within each treatment group will be estimated using Greenwood’s formula. The difference between the drotrecogin alfa (activated) and placebo groups in the time to death from randomization to Study Day 90 (or Study Day 180) will be assessed using a log-rank test. Survival curves will be constructed using the method of Kaplan and Meier. Estimates of relative risks and odds ratios and the corresponding 95% confidence intervals will be presented using the same methods as the analysis of 28-day mortality. Organ function. Organ function will be assessed using Sequential Organ Failure Assessment (SOFA) scores for each organ system. The effect of drotrecogin alfa (activated) treatment on cardiovascular, respiratory, and renal organ function compared with placebo will be evaluated using the following analyses: (1) change from baseline to Study Day 7, (2) change from baseline to Study Day 28, (3) repeated measures analysis, and (4) graphical displays.

For the analysis of change from baseline to Study Day 7, the LOCF algorithm will be used. Descriptive statistics will be presented (mean, median, and standard deviation). The Wilcoxon rank sum test will be used to compare the average change from baseline in the drotrecogin alfa (activated) and placebo groups. A similar analysis will be performed for change from baseline to Study Day 28. In the repeated measures analysis LOCF will not be employed, and the analysis will be based on available data. A linear mixed effects model will be applied using the SAS procedure PROC MIXED. The model will contain baseline SOFA score as a covariate, a random subject effect, and the following fixed effects: continuous time (1, 2, 3, 4, 5, 6, 7, 14, and 28), treatment, and treatment-by- continuous-time interaction and the response variable is the SOFA score at each timepoint. Graphical displays of the repeated measures analysis will consist of a profile plot depicting mean SOFA score over time for the treatment groups and a composite graph of mortality and SOFA scores over time.

6.9.3. Other Efficacy Analyses The following additional efficacy analyses will also be performed: Cause of death. Cause of death will be summarized for those patients who were dead at Study Day 28, at Study Day 90, and at Study Day 180. An overall comparison of the cause of death in the 2 treatment groups will be performed using Pearson’s chi-square test. Patients with missing cause of death will be classified as unknown. Organ Function. Organ function analyses will be performed for the cardiovascular, respiratory, renal, and hematologic systems. Measures of organ function will be assessed at baseline and during the study drug infusion. For each organ system, a hierarchy of analyses will be defined: primary, secondary, and tertiary.  Cardiovascular Primary analysis – Change from baseline to Study Day 7 in cardiovascular SOFA score (see Section 6.9.2 for a description of this analysis). Secondary analysis – A daily status analysis of vasopressor use. For each treatment group and each day from baseline through Study Day 28, the number and percentage of patients within each level of a 3-level categorical response variable: “Yes,” “No,” or “Dead” will be presented graphically. A repeated measures model for categorical response will be employed using a model similar to the model for SOFA scores by the SAS procedure PROC GLIMMIXED. A multinomial distribution with the linked function CUMLOGIT will be used in this analysis. The model will contain a random subject effect and the following fixed effects: continuous time, treatment, and treatment-by-continuous-time interaction.

Tertiary analyses – (1) Lactate clearance: lactate levels will be measured at baseline and every 6 hours for the first 24 hours of the treatment period. Change over time will be compared between treatment groups using descriptive statistics (number of patients, mean, median, and standard deviation). (2) Pressure-adjusted heart rate: pressure-adjusted heart rate will be measured at baseline, every 6 hours for the first 48 hours of the treatment period, and then daily on Study Days 3 through 6. Change over time will be compared between the treatment groups using descriptive statistics (number of patients, mean, median, and standard deviation). Profile plots by treatment will also be presented.  Respiratory Primary analysis – Change from baseline to Study Day 7 in respiratory SOFA score (see Section 6.9.2 for a description of this analysis). Secondary analysis – A daily status analysis of mechanical ventilation use. For each treatment group and each day from baseline through Study Day 28, the percentage of patients within each level of a 3-level categorical response variable: “Yes,” “No,” or “Dead” will be presented graphically. A repeated measures model for categorical response will be employed using a model similar to the model for SOFA scores by the SAS procedure PROC GLIMMIXED. A multinomial distribution with the linked function CUMLOGIT will be used in this analysis. The model will contain a random subject effect and the following fixed effects: continuous time, treatment, and treatment-by-continuous time-interaction.

Tertiary analyses – (1) PaO2/FiO2 ratio: the PaO2/FiO2 ratio will be measured at baseline, every 6 hours for the first 24 hours of the treatment period, and daily on Study Days 2 through 6. Change over time will be compared between the treatment groups using descriptive statistics (number of patients, mean, median, and standard deviation). (2) Oxygenation index: The components of the oxygenation index are measured at baseline and daily on Study Days 1 through 6. Change over time will be compared between the treatment groups using descriptive statistics (number of patients, mean, median, and standard deviation).  Renal Primary analysis – Change from baseline to Study Day 7 in renal SOFA score (see Section 6.9.2 for a description of this analysis).

Secondary analysis – A daily status analysis of renal replacement therapy use. For each treatment group and each day from baseline through Study Day 28, the percentage of patients within each level of a 3-level categorical response variable: “Yes,” “No,” or “Dead” will be presented graphically. A repeated measures model for categorical response will be employed using a model similar to the model for SOFA scores by the SAS procedure PROC GLIMMIXED. A multinomial distribution with the linked function CUMLOGIT will be used in this analysis. The model will contain a random subject effect and the following fixed effects: continuous time, treatment, and treatment-by-continuous-time interaction. Tertiary analyses – (1) Creatinine level: creatinine level will be measured at baseline and every 6 hours for the first 48 hours of the treatment period. Change over time will be compared between the treatment groups using descriptive statistics (number of patients, mean, median, and standard deviation). (2) Urine output: urine output will be measured during the following intervals: from the start of vasopressor therapy to the start of the study drug infusion, the first 24 hours of the treatment period, and the second 24 hours of the treatment period. The average urine output per hour will be calculated for the periods before the start of study drug and after the start of study drug. The mean change in average urine output will be compared between the treatment groups using descriptive statistics (number of patients, mean, median, and standard deviation). (3) Microalbumin/creatinine ratio: this ratio will be measured at baseline and every 6 hours for the first 24 hours. Change over time will be compared between treatment groups using descriptive statistics (number of patients, mean, median, and standard deviation).  Hematologic Primary analysis – Hematologic function will be assess using SOFA scores (0, 1, 2, 3, or 4) at baseline and on Study Days 1 through 7, Study Day 14, and Study Day 28. The effect of drotrecogin alfa (activated) treatment compared with placebo will be evaluated using the following analyses: (1) change from baseline to Study Day 7, (2) change from baseline to Study Day 28, (3) repeated measures analysis, and (4) graphical displays. These analyses are described in Section 6.9.2 under “Organ Function.” Secondary analyses – (1) Plasminogen activator inhibitor 1 (PAI-1): PAI-1 will be measured at baseline and Study Day 7. Change from baseline to Study Day 7 will be compared between treatment groups. The comparison between the treatment groups in change from baseline to Study Day 7 will be performed using descriptive statistics and analyzed as described in Section 6.11.

 Liver Liver function will be assess using SOFA scores (0, 1, 2, 3, or 4) at baseline and on Study Days 1 through 7, Study Day 14, and Study Day 28. The effect of drotrecogin alfa (activated) treatment compared with placebo will be evaluated using the following analyses: (1) change from baseline to Study Day 7, (2) change from baseline to Study Day 28, (3) repeated measures analysis, and (4) graphical displays. These analyses are described in Section 6.9.2 under “Organ Function.” Free-day analyses. Days on vasopressors, mechanical ventilation, and renal replacement therapy will be collected while the patient is alive and in the study hospital. As increased mortality in one treatment group may reduce the number of patients treated with vasopressors, mechanical ventilation, and renal replacement therapy, a free-days analysis will be conducted. For the free-day analyses, a patient will receive a score of 0 to 28 based on the number of days the patient is alive and does not require the specific resource (vasopressors, mechanical ventilation, or renal replacement therapy). In case of death, the patient was not free of the resource for the days remaining in the 28-day study period. For days when the patient was alive, but data were not available, the missing data were imputed using LOCF. Descriptive statistics will be presented. Comparisons between treatment groups will be made using the Wilcoxon rank sum test. Daily status analyses of patients in the ICU or in the study hospital, but not in the ICU. For each day from Study Day 1 through Study Day 28 the number and percentage of patients in each treatment group in the ICU, in the general ward (in the study hospital, but not in the ICU), discharged alive, or dead will be calculated. All ITT patients will be included in these analyses. These data will be presented graphically. A repeated measures model for categorical response will be employed using a model similar to the model for SOFA scores by the SAS procedure PROC GLIMMIXED. Patient location. Patient location will be collected at Study Days 28, 90, and 180. Analyses of patient location will be conducted for patients who survive to the endpoint of interest, that is, 28-day survivors, 90-day survivors, and 180-day survivors. For each time point, the proportion of patients at each location will be compared between the treatment groups using a chi-square test.

6.10. Quality-of-Life Analyses Measures of quality of life and functional status (Euro-QoL-5D and SF-12) will be assessed at baseline and Study Days 28, 90, and 180. Euro-QoL-5D. The primary analysis of quality of life will be the comparison between the 2 treatment groups of the change in total score from baseline to Study Day 28. Analyses of change from baseline to Study Days 90 and 180 will also be performed. Analyses will be performed for the population of patients who survive to the time point of interest (28-day survivors population, 90-day survivors population, and 180-day

Drotrecogin Alfa (Activated) (LY203638) Draft: 29 April 2018 F1K-MC-EVDP Statistical Analysis Plan Page 17 survivors population) and for the ITT population. For the ITT analysis, patients who die will receive a total score of 0 for all time points after the death occurred. For each time point (baseline, Study Day 28, 90, and 180), the percentage of patients who score Level 1, Level 2, and Level 3 for each of the 5 questions will be presented as well as the mean score on the visual analog scale. SF-12. The primary analysis will be the comparison between the two treatment groups of the change in the physical component score and mental component score from baseline to Study Day 28. Analyses of change in the component scores from baseline to Study Days 90 and 180 will also be performed. Analyses will be performed for the population of patients who survive to the time point of interest (28-day survivors population, 90-day survivors population, and 180-day survivors population).

6.11. Biomarker Analyses The following biomarkers will be evaluated at the following timepoints:  Protein C – baseline, Study Days 1 through 7, Study Day 14, and Study Day 28  Activated partial thromboplastin time (APTT) – baseline, Study Days 7, 14, and 28  Prothrombin time (PT) – baseline, Study Days 7, 14, and 28  Plasminogen activator inhibitor 1 (PAI-1) – baseline and Study Day 7  Platelet count – baseline, Study Days 7, 14, and 28 Statistical tests for continuous variables will be performed using ANOVA. Analyses of change from baseline, percent change from baseline, and symmetric percent change from baseline (Berry and Ayers 2006) will be performed for each postbaseline data point. Missing data will be imputed using LOCF. Protein C. In addition to the analyses described above, the following analyses will be performed for the ITT population and the population of patients with severe protein C deficiency: Change from baseline to endpoint. Change from baseline to Study Day 4 and change from baseline to Study Day 7 will be analyzed. The treatment groups will be compared using ranked ANOVA. Change in protein C class. The proportion of patients who are no longer severely protein C deficient at Study Day 4 and Study Day 7 will be compared between treatment groups using a chi-square test. Patients are defined as severely protein C deficient if they have a protein C level less than or equal to half the lower limit of normal; patients are no longer severely deficient when their protein C level rises to above half the lower limit of normal.

6.12. Safety Analyses Safety analyses will be performed on the population of all treated patients unless otherwise specified. Study drug exposure and study drug interruptions/discontinuation. Total exposure to study drug (g/kg) and the duration of infusion will be summarized as measures of central tendency and dispersion. The reasons for study drug interruption and the reasons for study drug discontinuation will be summarized by treatment group. The treatment groups will be compared using a chi-square test. Adverse events. The primary safety objective of this study is to demonstrate that drotrecogin alfa (activated) has an acceptable safety profile in this patient population. To evaluate the safety profile, the following adverse events will be summarized by MedDRA preferred term and system/organ class for Study Days 0 through 6 and Study Days 0 through 28: serious adverse events, study-drug-related serious adverse events, study- drug-related nonserious adverse events, adverse events that led to discontinuation of study drug, serious thrombotic events, nonserious thrombotic events, serious bleeding events, and nonserious bleeding events. The two treatment groups will be compared using Fisher’s exact test. Vital signs. Vital signs were collected as components for measurements of cardiovascular function, such as pressure-adjusted heart rate and oxygenation index. No other analyses of vital sign data will be performed.

6.13. Subgroup Analyses A priori we have identified 9 subgroups of clinical interest that will be reported in the primary manuscript. Each is defined by the presence or absence of a prerandomization variable. The treatment effects and safety profile in cohorts defined by postrandomization events may be examined for both regulatory and safety reporting purposes; these cohorts are described in Section 6.14. Subgroup analyses of 28-day all-cause mortality will be performed for the ITT population. Subgroup analyses of serious adverse events and serious bleeding events will be performed for the population of all treated patients. The primary outcome for planned subgroup analyses will be the same as in the primary analysis, 28-day, all-cause mortality. Reported p-values will not be adjusted for multiple hypothesis testing, but the number of a priori subgroup analyses will be specified in all publications. Analysis. The main analysis for each subgroup will be the Breslow-Day test for homogeneity of odds ratios to determine whether the effect of treatment differs significantly across categories, for example, in patients with an APACHE II score of 25 or more versus patients with a score of less than 25. As a secondary analysis, within each

Drotrecogin Alfa (Activated) (LY203638) Draft: 29 April 2018 F1K-MC-EVDP Statistical Analysis Plan Page 19 level of each subgroup factor, treatments will be compared using chi-square (nonstratified) tests. Since mortality rates may vary across levels of the subgroup factors, adjusted treatment comparisons will be assessed using Cochran-Mantel-Haenszel tests. Adjusted relative risks, odds ratios, and 95% confidence intervals will be calculated using the Mantel-Haenszel adjusted relative risk method. A result for a given subgroup will be defined as consistent with the overall study results if the 95% confidence interval around the relative risk in that subgroup contains the point estimate of the relative risk for the overall study. The following status at baseline will determine the 9 subgroup analyses: 1. Patients with an APACHE II score 25 at baseline versus patients with a score <25. Rationale: In the PROWESS study, the beneficial effect of drotrecogin alfa (activated) was observed in patients with a baseline APACHE II score of 25 or greater. This effect was not observed in the ADDRESS study. This subgroup analysis will clarify whether the treatment effect differs according to baseline severity of illness as indicated by APACHE II score. 2. Patients defined by the number of dysfunctional organs at baseline (1, 2, 3, 4, 5; if the subgroups of 1 organ dysfunction or 5 organ dysfunctions contain less than 200 patients, they will be pooled with the subgroups of 2 organ dysfunctions and 4 organ dysfunctions, respectively). Rationale: In the PROWESS study, the beneficial effect of drotrecogin alfa (activated) was observed in patients with more than 1 dysfunctional organ at baseline. This subgroup analysis will clarify whether the treatment effect differs according to the number of dysfunctional organs at baseline. 3. Patients who have had recent surgery versus patients who have not. Rationale: In the PROWESS study, the observed treatment effect of drotrecogin alfa (activated) was less in patients with severe sepsis who had recently undergone surgery. In the ADDRESS study, mortality was higher in patients who had recently undergone surgery and had only 1 dysfunctional organ at baseline. This subgroup analysis will clarify whether the treatment effect differs in patients eligible for enrollment in the PROWESS-SHOCK study who have recently undergone surgery. 4. Patients defined by the time from onset of septic shock to start of infusion quartile (1st, 2nd, 3rd, 4th). Rationale: Analysis of cohort studies has found that patients treated with drotrecogin alfa (activated) earlier in the disease process are less likely to die. This subgroup analysis will clarify whether the treatment effect differs depending on the time from the onset of septic shock to the start of the study drug infusion.

5. Patients defined by protein C activity at baseline (40%, 41% to 60%, 61% to 80%, >80%). Rationale: Patients with lower protein C activity at baseline are expected to be at greater risk of dying. This subgroup analysis will clarify whether the treatment effect differs according to baseline protein C activity. 6. Patients treated with corticosteroids at baseline versus patients not treated. Rationale: Corticosteroids are a recommended treatment for persistent septic shock and, while their effect on mortality is debated, study data indicate that treatment with corticosteroids has a vasopressor-sparing effect. Patients eligible for enrollment in the study despite treatment with corticosteroids may represent a subgroup of patients who respond differently to treatment with drotrecogin alfa (activated). This subgroup analysis will clarify whether the treatment effect differs in patients who are or are not treated with corticosteroids at baseline. 7. Patients treated with prophylactic heparin at baseline versus patients not treated. Rationale: Previous studies have suggested that the treatment effect and safety profile of drotrecogin alfa (activated) may differ in patients treated with prophylactic heparin. This subgroup analysis will clarify whether the treatment effect and safety profile differs in patients who are or are not treated with prophylactic heparin at baseline. 8. Patients with ARDS at baseline versus patients without. Rationale: Patients with ARDS are a well-defined subset of patients with septic shock and in comparison with previous studies are expected to constitute a larger subgroup in the PROWESS-SHOCK study. This subgroup analysis will clarify whether the treatment effect differs in patients who do or do not have ARDS at baseline. 9. Patients defined by platelet count based on hematology SOFA score (0 to 1 versus 2 to 4, corresponding to a platelet count of 100.000/mm3 versus <100,000/mm3). Rationale: Previous studies suggest that the beneficial treatment effect of drotrecogin alfa (activated) may be increased in patients with sepsis- induced thrombocytopenia although safety concerns may also be increased in this subgroup. This subgroup analysis will clarify whether the treatment effect and safety profile differs in patients stratified according to baseline platelet count.

6.14. Supplementary Analyses The following analyses have been requested from regulatory authorities, involve subgroups identified from prior publications or regulatory inquiries, or are needed for the regulatory study report defined by the International Council on Harmonisation. Concomitant exposure to heparin/anticoagulants. Information on heparin and other selected anticoagulants will be collected from baseline through Study Day 28. The effect of concomitant heparin/anticoagulant exposure administered concomitantly during the study drug infusion on the occurrence of serious and nonserious bleeding events will be examined in the all treated patients population. For these analyses, patients will be classified into one of the following subgroups: concomitant heparin/anticoagulant exposure-yes or -no. Patients who received any anticoagulant (unfractionated heparin, enoxaparin, dalteparin, bemiparin, fondaparinux, ardeparin, certoparin, nadroparin, parnaparin, reviparin, or tinzaparin) for any length of time during Study Days 0 through 6 for either prophylactic or renal replacement indication will be classified as concomitant heparin/anticoagulant exposure-yes. Patients who did not receive an anticoagulant during Study Days 0 through 6 will be classified as concomitant heparin/anticoagulant-no. Analyses will be performed as described for baseline heparin/anticoagulant exposure. Prophylactic heparin discontinuation analyses. Analyses will be performed for the population of patients who discontinue prophylactic heparin during the pretreatment period. This population is expected to be small; therefore, only descriptive statistics will be presented. The following analyses will be performed:  Change in markers of thrombin generation (thrombin-antithrombin complex [TATc] and prothrombin fragment 1.2 [F1.2]). These markers will be evaluated at baseline and every 12 hours for the first 72 hours after the start of infusion. Values at each timepoint and change from baseline to each timepoint will be summarized using measures of central tendency and dispersion for each treatment group. No imputation will be used for these analyses.  Twenty-eight-day mortality. The number and proportion of patients alive and dead at 28 days will be summarized for each treatment group. Patients with missing survival status will be excluded from this analysis.  The number and proportion of patients who experience at least one serious adverse event and serious and nonserious thrombotic event will be summarized for Study Days 0 through 6 and Study Days 0 through 28 for each treatment group. Analyses of SOFA scores. Each organ system (cardiovascular, hematology, liver, renal, and respiratory) will be scored using the SOFA methodology. A patient will receive a score of 0, 1, 2, 3, or 4 on each day (Study Days 1 through 7, Study Day 14, and Study Day 28). Time-averaged SOFA scores will be calculated for each organ system and for total SOFA for the following time periods: (1) Study Days 1 through 4, (2) Study Days 1

Drotrecogin Alfa (Activated) (LY203638) Draft: 29 April 2018 F1K-MC-EVDP Statistical Analysis Plan Page 22 through 7, (3) Study Days 1 through 14, and (4) Study Days 1 through 28. A nonsurviving patient will receive a score of 4 (worst score) for the day of death and every day thereafter. For those days when a patient is alive, but no data are available, LOCF will be used to impute the missing data. For each organ system, all ITT patients with any postbaseline data (including death) on Study Day 0 through Study Day 28 will be included in the analyses. To calculate the time-averaged score, the average of imputed daily SOFA score for each patient will be calculated. The mean time-averaged SOFA score for the 2 treatment groups will be compared using ANOVA based on unranked data. For each organ system, analyses of time to resolution of organ function and the occurrence of new organ function will be performed. Time to resolution is defined as the time for patients with a baseline SOFA score greater than 0 to attain a SOFA score of 0. Time to resolution curves will produced using the method of Kaplan and Meier; differences between treatment groups will be assessed using a log rank test. Occurrence of new organ dysfunction is defined as developing a SOFA score of 2 or greater postbaseline in patients with a SOFA score of 0 at baseline. The percentage of patients in each treatment group who develop a new organ dysfunction will be compared using a chi-square test. Laboratory data. For each chemistry and hematology laboratory analyte, statistical summaries of the change from baseline to postbaseline timepoints (Study Days 7, 14, and 28) will be presented and observed changes compared by treatment group using ANOVA with ranked data. In addition, a summary of the number and percentage of patients in each treatment group who went from a low, normal, or high baseline laboratory value to a low, normal, or high value at postbaseline timepoints based on the central laboratory (chemistry analytes) or local laboratory (hematology analytes) reference ranges. Subgroup analyses. Additional subgroup analyses will be performed using the statistical methods described in Section 6.13. Subgroups to be investigated for 28-day all-cause mortality, serious adverse events, and serious bleeding events include the following:  Gender (female, male)  Age class (<65, 65)  Origin (Caucasian, non-Caucasian)  Investigative site  Country (Australia, Belgium, Brazil, Canada, Czech Republic, Finland, France, Germany, India, Italy, Mexico, New Zealand, Portugal, Spain, Switzerland, The Netherlands, United Kingdom, United States)  Region (Europe, United States/Canada, rest of world)  Evidence of hypoperfusion criteria met (1, 2, 3)

 Respiratory organ dysfunction criteria met (no, yes)  Renal organ dysfunction criteria met (no, yes)  Hematology organ dysfunction criteria met (no, yes)  Metabolic acidosis organ dysfunction criteria met (no, yes)  Baseline mechanical ventilation status (no, yes)  Baseline renal replacement therapy status (no, yes)  Baseline APACHE II score quartiles  Site of infection (present subgroups with 200 or more total patients; combine the rest of into a category of “other”)  Source of infection (community, nosocomial)  Malignancy (no, yes)  Central venous oxygenation (<70%, 70%)

 PaO2/FiO2 ratio (200, >200)  Creatinine clearance (> 50 mL/min, 50 mL/min)  Microorganisms (present subgroups with 200 or more total patients; combine the rest into a category of “other”)  Microbiologically verified infection and bacteremia, fungemia, or positive cerebrospinal fluid culture (no, yes)  Time from onset of first sepsis-induced organ function to start of study drug class quartile Miscellaneous reports. A summary of patient allocation by investigative site will be produced. This report will summarize the number of patients randomized by treatment group at each investigative site. Listings. The following listings will be produced:  Listing of Treatment Assignments (ITT Population)  Listing of Patient Disposition (ITT Population)  Listing of Demographic Characteristics (ITT Population)  Listing of Primary and Secondary Efficacy Endpoints (ITT Population)  Listing of Adverse Events that Led to Study Drug Discontinuation (All Treated Patients)  Listing of Serious Adverse Events (All Treated Patients)  Listing of Serious Bleeding Events (All Treated Patients)

 Listing of Deaths (ITT Population)  Listing of Adverse Events (All Treated Patients)

6.15. Interim Analyses and Data Monitoring Two interim analyses are planned for this study. These interim analyses will be conducted under the auspices of an external and independent DMC to minimize the operational and statistical bias that result from performing interim analyses. The purpose of the DMC is to advise Lilly regarding the continuing safety of study participants and the continuing validity and scientific merit of the study. Only the DMC is authorized to evaluate unblinded interim efficacy and safety analyses. Study sites will receive information about interim results ONLY if they need to know for the safety of their patients. The DMC will be composed of 1 statistician and 5 physicians with relevant critical care, medical, or medical/clinical study ethics experience. All members will be voting members and DMC membership will cover the duration of the study. All members of the DMC are external to Lilly. The first interim analysis will occur after approximately one-third of the planned patient enrollment have completed 28 days of follow-up; the second will occur after approximately two-thirds of patients have completed 28 days of follow-up. If a decision is made to increase the sample size of the study, interim analyses will be inserted in such a way that approximately 500 patients are enrolled between each contiguous pair of interim analyses. Since the first interim analysis is scheduled to take place after 28-day outcomes have been obtained for 500 patients, the additional interim analyses will be conducted at regular intervals between 500 and the new sample size, and the interim analysis that was originally scheduled to take place after 28-day outcomes had been obtained for 1000 patients will be eliminated. The nominal significance level at the final analysis will be chosen to maintain the overall error rate of 0.05 using the conditional rejection probability technique of Muller and Schafer (2004) as outlined in the appendix. The objective of the DMC for the first interim analysis is to recommend whether to terminate enrollment in the study for safety concerns based on prospectively defined guidelines. Because an examination of mortality rate will be a part of the safety review, there will be an alpha spend at the first interim analysis. However, the sponsor has no plans to stop the study for positive efficacy at the first interim analysis. The objective of the second interim analysis is to recommend whether to terminate enrollment in the study for safety concerns, efficacy, or futility based on prospectively defined guidelines. The prescription for alpha spending will follow accepted statistical practices, such as the methodology of O’Brien and Fleming as implemented by Lan and DeMets (Lan and DeMets 1983; DeMets et al. 1995). The alpha spend at the first interim analysis and the guideline for stopping due to efficacy at the second interim analysis will ensure an overall

Drotrecogin Alfa (Activated) (LY203638) Draft: 29 April 2018 F1K-MC-EVDP Statistical Analysis Plan Page 25 false positive (one-sided) error rate less than 2.5%. The final stopping guidelines will be decided in consultation with the DMC committee members and documented in the DMC Charter. The DMC Charter will be submitted to regulatory agencies before the first patient visit. The prospectively defined interim analysis plan will be executed by a group of independent statisticians at a CRO who will be separated by a firewall from the statisticians at the same CRO who will be responsible for the analysis of data at the conclusion of the study. This group of independent statisticians will present interim results to the DMC. Individuals responsible for data management activities and the generation of datasets for interim reporting will not have access to patient treatment assignments. Patient treatment assignments will be provided to the independent statistical group prior to each interim analysis. The independent statistical group will merge the interim datasets and the data set that contains patient treatment assignments to execute the interim analysis plan.

6.16. Determination of Sample Size A sample size of approximately 1500 patients is required to achieve 80% power at an alpha level of 0.05 using a chi-squared test of the null hypothesis that the mortality rates are equal in the 2 treatment groups. The assumptions that underlie the sample size calculations are based upon analyses of results from the PROWESS database. In PROWESS, the placebo mortality rate for patients with septic shock at baseline was approximately 35% and treatment with drotrecogin alfa (activated) was associated with a 20% relative risk reduction. Based on these data, it was assumed that the placebo mortality rate is 35% in patients with septic shock and that drotrecogin alfa (activated) is associated with a 20% relative risk reduction, resulting in a drotrecogin alfa (activated) mortality rate of 28%.

6.16.1. Resizing the Study Due to uncertainty in the mortality rate of patients with septic shock because of changes in the standard of care as well as the possibility that some patients may opt out of their randomly assigned therapy, the sponsor, in conjunction with the steering committee, may choose to increase the sample size during the course of the study if the blinded mortality rate is lower than expected or if there is a greater-than-anticipated number of patients for whom study drug is permanently discontinued due to disease progression/lack of improvement and therapy with commercial drotrecogin alfa (activated) is initiated (opt- out patients). The maximum adjustment to the sample size will be 500 patients for a total maximum sample size of 2000 patients. Mortality rate. After approximately one-half of the patients in the study have been enrolled, the blinded mortality rate will be computed. This calculation will be based upon summary data that have been aggregated across the 2 treatment groups. If the

Drotrecogin Alfa (Activated) (LY203638) Draft: 29 April 2018 F1K-MC-EVDP Statistical Analysis Plan Page 26 estimate of the overall mortality rate is less than 30%, the total sample size may be increased by reconsidering the initial assumptions surrounding the mortality rate in the placebo group and the size of the anticipated treatment benefit. Under the initial assumptions, the overall mortality rate was expected to be 31.5%. Opt-out rate. After approximately one-half of the patients in the study have been enrolled, the blinded opt-out rate will be computed. This calculation will be based upon summary data that have been aggregated across the 2 treatment groups. If the estimate of the opt-out rate is greater than or equal to 3%, the total sample size may be increased. Recalculation of the sample size will be based upon the observed opt-out rate and upon assumptions surrounding the mortality rate and treatment benefit in placebo patients who opt out of their randomly assigned therapy and receive drotrecogin alfa (activated). The purpose behind the augmentation of the sample size is to maintain the nominal statistical power of the study in the event that some placebo patients actually receive commercial drotrecogin alfa (activated).

7. Unblinding Plan

The purpose of this unblinding plan is to maintain the scientific integrity of the study. The following actions/procedures will be put in place prior to any unblinding of the study data:

A. Site Level Unblinding The procedure for site personnel to unblind an individual a patient’s treatment assignment for emergency is described in the protocol. The site monitor is responsible for verifying compliance with the blinding procedures at the investigator site and verifying that access to the patients’ treatment assignments remains restricted from the investigator and site personnel in direct contact with subjects. The investigator and site personnel are instructed to make every attempt to contact Lilly personnel when a patient’s treatment assignment is unblinded at the site. Affiliate personnel document the unblinding records and informs the designated study team member, the lead CDO-CDA associate, who documents the overall unblinding records for the entire study. The documentation is filed in the study files. A final Study Unblinding Summary will be prepared at the end of the study (at the study closeout). When an IVRS (CT-SMS) is used to unblind a patient’s treatment assignment, the computer application will maintain the date, reason for unblinding, and the identification of the person unblinding the treatment assignment.

B. Sponsor/Trial Level Unblinding Interim analysis. The Lilly project statistician is responsible for authorizing members of the statistical analysis center (SAC) access to blinded data to prepare reports for the data monitoring committee (DMC). Members of the SAC will not be involved in the conduct of the study, other than preparation for and attendance at DMC meetings. Only individuals in the SAC within the CRO will have access to unblinded data for the interim analyses. The SAC statistician will keep a running log of individuals given access to unblinded data at each interim analysis. The Lilly project statistician is responsible for authorizing the access of Lilly personnel, such as members of the internal review committee (IRC), to unblinded data. The internal review committee will only be provided access to unblinded data if the DMC recommends modifying the study and the senior management designee decides to convene the IRC. Every attempt should be made to contact the statistician and document the authorization before access is given to unblinded data. A designated systems analyst maintains the records as each individual is given access to the subject treatment assignment or unblinded data. The documentation is filed in the study files.

A designated study team member, the lead CDO-CTM associate, in collaboration with the statistician, will be responsible for keeping a running log of individuals given access to any unblinded study data. This log will include the person’s name, title, date of unblinding, level of unblinding (ie, group or patient), and purpose of unblinding. The following personnel with direct site contact or personnel responsible for data entry and data validation will remain blinded until datalock for the final data analysis is authorized in order to protect the scientific integrity of the study conduct and prevent introduction of bias:  Blinded CRAs/CDAs  Data Management Associate  Statistical analysts  Lilly project statisticians  Systems Analysts  Scientific Communication Associate  CRO Clinical Leads and Project Management The following personnel will be unblinded after the datalock for the interim analysis, in order to perform data analysis, prepare the CSRs, and discuss registration related matters:  SAC Statistician  SAC statistical analysts  DMC

8. References

Berry AD, Ayers GD. 2006. Symmetrized percent change for treatment comparisons. Am Stat 60:27–31.

DeMets DL, Fleming TR, Whitley RJ, Childress JF, Ellenberg SS, Foulkes M, Mayer KH, O’Fallon J, Pollard RB, Rahal JJ, Sande M, Straus S, Walters L, Whitley- Williams P. 1995. The data and safety monitoring board and acquired immune deficiency syndrome (AIDS) clinical trials. Control Clin Trials 16:408–421.

Kaplan EL, Meier P. 1958. Nonparametric estimation from incomplete observations. JASA 53:456–481.

Lan KKG, DeMets DL. 1983. Discrete sequential boundaries for clinical trials. Biometrika 70:659-663.

Muller HH, Schafer H. A general statistical principle for changing a design any time during the course of a trial. Stat Med 2004;23(16):2497–2508.

Snedecor GW, Cochran WG. 1989. Statistical Methods. 8th Edition. Iowa State University Press.

Appendix 1 Algorithm for Adjustment of Significance Level Due to Interim Analyses

This appendix explains the adjustment for multiplicity of testing due to interim analyses. The planned sample size of the study is 1500. There are two planned interim analyses, after 500 and 1000 patients have the primary outcome assessed. After approximately 750 patients have 28 day mortality status determined, an assessment will be made to potentially increase the sample size. The decision to increase the sample size will be made by the steering committee at a sampe size reassesment meeting and will be guided by a power analysis based on the observed blinded aggregate 28-day mortality rate (using data from approximately 750 patients) under the assumption of a 0.8 ratio in the 28-day mortality rate of drotrecogin alfa (activated) to placebo. Although the study will not stop for efficacy at the first interim analysis, a nominal alpha of 0.0001 will be spent at the first interim. At the second interim, the study can be stopped for efficacy if the two-sided p-value for the primary outcome has a nominal p- value less than 0.001. If the sample size is increased at the reassessment meeting and a third interim is added, then, at the third interim analysis, the study can be stopped for efficacy if the two-sided p-value for the primary outcome has a nominal p-value less than 0.001. The nominal significance level at the final analysis will be chosen to maintain the overall error rate of 0.05 in a formal monitoring scheme based on the nominal significance levels and spacings of interim analyses, using the conditional rejection probability technique of Muller and Schafer (2004). The following sections, which provide more detail of the methodology are written in terms of one-sided testing at an overall error rate of 0.025.

1. Study design The study design will include multiple decision points described below.  The first interim analysis (IA1) will take place after 28-day outcomes have been

obtained for n1 = 250 patients in each treatment group. Enrollment may be terminated at IA1 due to safety concerns. Mortality will be examined as a part of

the safety evaluation; therefore, a one-sided level α1 = 0.00005 (corresponding

critical value is c1 = 3.8906) will be spent at IA1.

 Sample size adjustment may be performed after 28-day outcomes have been

obtained for approximately n2 = 375 patients in each treatment group. The sample size will be increased based on a blinded analysis of the 28-day mortality data if the pooled 28-day mortality rate is less than 30%. The maximum sample size will be 1000 patients in each treatment group.

The following two scenarios will be implemented in the study after this decision point:

Scenario 1 (sample size is not adjusted)  The second interim analysis (IA12) will take place after 28-day outcomes have

been obtained for n12 = 500 patients in each treatment group. Enrollment may be terminated at IA12 due to overwhelming efficacy or safety concerns. The study may be stopped due to overwhelming efficacy if the treatment difference is

significant at a one-sided level α12 = 0.0005 (corresponding critical value is c12 = 3.2821).

 The final analysis (FA1) will take place after 28-day outcomes have been obtained for

n13 = 750 patients in each treatment group. The final analysis will be performed at

a one-sided level α13 = 0.0249 (corresponding critical value is c13 = 1.9608).

Scenario 2 (sample size is adjusted)  Two interim analyses (IA22 and IA23) will be included between the decision point and final analyses. These interim analyses will be equally spaced between

n2 = 375 and the adjusted sample size denoted by n24. For example, if n24 = 975 patients per treatment group, the interim analyses will take place after 28-day

outcomes have been obtained for n22 = 575 and n23 = 775 patients in each treatment group, respectively. Enrollment may be terminated at IA22 and IA23 due to overwhelming efficacy, futility or safety concerns. The study may be terminated due to overwhelming efficacy at either interim analysis if the treatment

difference is significant at a one-sided level α22 = 0.0005 and α23 = 0.0005,

respectively. The corresponding critical values are given by c22 = 3.2905 and

c23 = 3.2905. Further, the study may be stopped due to futility at the final interim analysis if the treatment difference is not significant at a one-sided level of 0.6.

 The final analysis (FA2) will take place after 28-day outcomes have been obtained for

n24 patients in each treatment group. The final analysis will be performed at a

one-sided level α24 (corresponding critical value is denoted by c24).

The significance level at the final analysis in Scenario 2 will be determined using the method developed by Muller and Schafer (2004). The computation of this significance level is described in Section 2. The decision rules used in the study are displayed in Figure 1. Figure 1. Interim and final analyses in the SHOCK study (DP, decision point to adjust the sample size).

IA12 FA1 Scenario 1 IA1 DP n12=500 n13=750

n1=250 375 IA22 IA23 FA2 Scenario 2

n22 n23 n24

2. Sample size adjustment algorithm Blinded sample size adjustment at the second interim look will be performed using the method proposed in Friede and Kieser (2004). Let p denote the pooled mortality rate at the second interim look. Further, let r denote the relative risk reduction assumed at the beginning of the study, i.e., r = 0.2. If the pooled mortality rate is less than 30%, i.e., if p < 0.3, the adjusted sample size per treatment group will be set to

2 (z1  z1 ) n24  2 ( p0 (1 p0 )  p1(1 p1)) , ( p0  p1) where z1-x is the (1 - x) percentile of the standard normal distribution, α is the one-sided

Type I error rate, i.e., α = 0.025, β is the Type II error rate, i.e., β = 0.2, and p0 and p1 are the assumed mortality rates in the placebo and Xigris groups computed under the assumption that the relative risk reduction is r, respectively, i.e., 2 p 2(1 r) p p  and p  . 0 2  r 1 2  r

Note that, if the adjusted sample size n24 is less than the original sample size n13 = 750 then n24 is set to 750 patients per treatment group and, if the adjusted sample size n24 is greater than 1000 then n24 is set to 1000 patients per treatment group. The sample sizes per treatment group at the interim analyses in Scenario 2 (IA22 and IA23) are given by n  n 2(n  n ) n  n  24 2 and n  n  24 2 . 22 2 3 23 2 3 To compute the significance level at the final analysis in Scenario 2 (FA2), the conditional rejection probability principle developed by Muller and Schafer (2004) will be utilized. The conditional rejection probability (CRP) is computed at IA1 and is defined as the conditional probability of rejecting the null hypothesis of no treatment effect after IA1 in the original study design given the test statistic at IA1. Let Z1, Z12 and

Z13 denote the test statistics at IA1, IA12 and FA1 in the original study design, respectively. Then the one-sided CRP is computed as follows

CRP   (Z1)  P([Z12  c12 ][(Z12  c12 )  (Z13  c13 )] | Z1) ,

Drotrecogin Alfa (Activated) (LY203638) Draft: 29 April 2018 F1K-MC-EVDP Statistical Analysis Plan Page 34 where the data after IA1 are generated under the null hypothesis of no treatment effect. The CRP can be computed from the multivariate distribution of the three test statistics using methods proposed in Muller and Schafer (2004) and Dmitrienko et al. (2005, Chapter 4). Per the CRP principle, any adaptation can be considered after IA1 as long as the overall

Type I error rate is controlled at level (Z1), i.e., the significance levels at IA22, IA23 and FA2 are chosen in such a way that the probability of incorrectly rejecting the null hypothesis of no treatment effect is (Z1). Let Z22, Z23 and Z24 denote the test statistics at IA22, IA23 and FA2, respectively, based on the data collected after IA1. The critical value c24 is chosen from the equation

1 P((Z22  c22 )  (Z23  c23 )  (Z24  c24 ))   (Z1) .

The probability in the left-hand side of the equation is easy to evaluate since, under the null hypothesis of no treatment effect, the three test statistics follow a standard trivariate normal distribution with the correlation matrix:

 1 1/ 2 1/3     1/ 2 1 2/3 .    1/3 2/3 1 

The associated one-sided significance level is α24 = 1 - (c24), where (x) is the cumulative distribution function of the standard normal distribution.

3. Example To illustrate the blinded sample size adjustment algorithm defined in Section 2, set the overall Type I error rate in the study at a one-sided α = 0.025. Suppose that the 28-day mortality rates in the placebo and Xigris groups at the first interim analysis (IA1) are given by pP = 0.30 and pL = 0.26, respectively. The test statistic at IA1 is

pP  pL Z1   0.9970 pP (1 pP ) / n1  pL (1 pL ) / n1 and the associated one-sided p-value is 0.1594. Given this test statistic, the one-sided conditional rejection probability is (Z2) = 0.0449. This conditional rejection probability is “spent” on the remaining analyses in the study.

Now consider the decision point at n2 = 375 patients per treatment group and assume that the pooled mortality rate is 0.26 and assumed relative risk reduction is r = 0.2. Thus the assumed mortality rates in the placebo and Xigris groups are p0 = 0.2889 and p1 = 0.2311.

Given these mortality rates, the adjusted sample size per group is n24 = 901 patients per group. Note that this sample size is greater than the original sample size n13 = 750 patients per group and Scenario 2 is pursued (total sample size will be increased at the

Drotrecogin Alfa (Activated) (LY203638) Draft: 29 April 2018 F1K-MC-EVDP Statistical Analysis Plan Page 35 decision point). The interim analyses IA22 and IA23 take place after 28-day outcomes are obtained for n22 = 634 and n23 = 768 patients in each treatment group, respectively. The one-sided significance levels to be used at IA22 and IA23 are specified in Section 1

(α22 = 0.0005 and α23 = 0.0005) and the significance level at the final analysis, i.e., α24, is computed from the conditional rejection probability. To compute α24, we first compute the critical value c24 is computed from

P((Z22  c22 )  (Z23  c23 )  (Z24  c24 ))  1  (Z1)  0.9551.

Noting that Z22, Z23 and Z24 follow a standard trivariate normal distribution with the correlation matrix defined in Section 2, the resulting value of c24 is 1.6985 and thus the one-sided significance level is α24 = 1 - (c24) = 0.0447. With this choice of α24, the overall Type I error rate in the study is preserved at the nominal level, i.e., one-sided α = 0.025.

References

Dmitrienko A, Molenberghs G, Chuang-Stein C, Offen W. (2005). Analysis of Clinical Trials Using SAS: A Practical Guide. SAS Press: Cary, NC.

Friede T, Kieser M. (2004). Sample size recalculation for binary data in internal pilot study designs. Pharmaceutical Statistics. 3, 269-279.

Muller HH, Schafer H. (2004). A general statistical principle for changing a design anytime during the course of a trial. Statistics in Medicine. 23, 2497-2508.

Drotrecogin Alfa (Activated) (LY203638) Draft: 29 April 2018