Package: gsDesign 3.11.1.9028

A continuing purpose | Foundations and the early package | Selected milestones | Making the software easier to maintain and use | From statistical information to a trial that can be run | A growing family of packages | New questions, established calculations | A visual record and an open history | References

A design warning: futility, sample size, and overall power | What is being calibrated? | One parameter: CP 0.3 at the first interim | Event-driven survival design with gsSurv() | Targeting predictive power instead | Fixed calendar analyses with gsSurvCalendar() | Two parameters: a risk difference design | Fixed sample size and inflation from futility | What does delta at the bound mean? | Compare all six two-parameter families | Piecewise linear spending: comparing two normal means | A more conservative sequence of CP targets | Practical limits and diagnostics

Specify an effect; retain a spending function | Joint efficacy, futility and harm targets for normal means | Independent reconstruction and verification | All bounds, not only the targeted interim | Compare the fitted spending shapes | Two-parameter futility spending for a risk difference | Hazard ratios require explicit event scale metadata | Diagnostics and interpretation | Built on established computational tools

Introduction | Regulatory context: FDA guidance on OS monitoring in oncology | Design framework overview | Design with non-binding bounds (test.type = 8) | Spending function specification | Summary | Detailed boundary table | Interpreting the boundaries | Boundary crossing probabilities | Harm calibration versus actual stopping | Visualization | Z-value boundaries | Approximate treatment effect at boundaries | Conditional power at boundaries | Spending function plot | B-values at boundaries | Design with binding bounds (test.type = 7) | Comparing binding and non-binding | Efficacy bounds at alternate $\alpha$ levels | Sparse-event mortality monitoring | A harm bound without a separate futility bound | Integer death count boundaries | Sensitivity to excess mortality | Detecting harm versus ruling out harm | Practical considerations | Choice of spending functions | Total harm spending and the number of analyses | Interpreting the harm bound | Harm bound capping | When to use test.type = 7 vs. test.type = 8 | Binding and non-binding harm monitoring | Adjusting the boundaries | References

What is conditioned on? | A Dragalin-related fixed information example | Compare all three methods on the same reference | Same conditional assurance, different sample size and power constraints | Unconditional POS is a different requirement | Verify with established probability calculations | Operating characteristics, not just a target | Sample sizes and approximate effects at all bounds | The final specification is still a spending function | Building on established computational tools with AI | Scope and practical limits | References

Introduction | Summary of method | Binomial endpoint designs | Fixed sample size | 1-sided design | Non-binding design | Time-to-event endpoint designs | Seasonal design with a final zero event rate period | References

Introduction | Assumptions and parameterization | Initial group sequential setup | Example repeated and sequential p-values | Update exact bounds at analysis time | Lightweight runnable simulation | Example with lower-than-planned event rates | Larger offline runs (template) | Notes and extensions | References

Overview | Choose the workflow by the question | Pattern 1: fix study duration and minimum follow-up | Pattern 2: fix rates and minimum follow-up | Pattern 3: fix enrollment and solve follow-up | Calendar time analyses | Power for a fixed operational plan | Integer event and enrollment plans | Stratified enrollment

Motivation | Quick start | What this calculation does—and does not do | How gsSurvPower uses your inputs | Hazard ratio roles | Beta spending in scenario analyses | Analysis timing: calendar time vs. event-driven | Quick decision guide | Relationship to simtrial analysis cuts | Common timing pitfalls | Spending and method | Stratified timing arguments | Power under alternative assumptions | Power under a different hazard ratio | Power over a range of hazard ratios | Multiple timing criteria | Setup | Baseline: design assumptions | Scenario 1: slower enrollment | Scenario 2: higher control failure rate | Controlling effective spending time | User-specified spending time caps with informationRates | Additional and advanced examples | Comparison with gsDesign power plots | Bounds stability | Changing alpha | Binding type planning sensitivity example (test.type = 3) | Event-based timing | Slower enrollment at fixed analysis times | Calendar-based spending | Stratified event targets | Biomarker subgroup to stratified design | Step 1: Design for the biomarker-positive subgroup | Step 2: Power for the overall (stratified) population

Overview | Starting point: SAS PROC SEQDESIGN survival example | Key differences: SAS SEQDESIGN vs. R gsDesign | 1. Event formula | 2. Alpha handling in gsDesign() and gsSurv() | 3. Accrual duration and follow-up time | Reproducing the fractional time design with gsSurv() | gsSurv() with aligned parameters | References

Introduction | Parameters | Example 1: Futility testing only at the first interim | Example 2: Asymmetric lower testing | Example 3: Selective harm monitoring (test.type 7/8) | Harm calibration when futility is skipped | Example 4: Combining selective efficacy and futility | Validation rules | Accessing stored flags | Type I Error Preservation | How it works | Non-binding futility (test.type 4 or 6) | Binding futility (test.type 3 or 5)

Introduction | Parameterization | Exact binomial approach | The time-to-event approach | Generating a design | The time-to-event design | Converting to an exact binomial design | Combined summary table | Checking design properties | $\alpha$-spending | $\beta$-spending | Bound update at time of analysis for example 2 | Analysis of an exact binomial trial | Exact conditional power | Exact confidence intervals | Summary | References

Introduction | The Poisson mixture model | Supporting functions | Scenario assumptions | Examples | Event accumulation | Study design | Design assumptions | Study design and event accumulation | Considerations | References

Introduction | Implementation in gsDesign | Method 1 | Method 2 | Method 3 | Replicating published examples | Summary | References

Overview | Response rate example | Summarizing design properties | Safety monitoring example | Summary | References

Introduction | Example | References

Introduction | Schoenfeld approximation support | Power and sample size with nEvents() | Group sequential design | Information based design | Approximating boundary characteristics | Examples | Lachin and Foulkes design | Model assumptions | Fixed design | Plotting | Event accrual | References

Introduction | The problem considered | Sample size | Examples | Power | Verification with simulation | Group sequential design | References

Introduction | Group sequential design framework | Bounds for testing | One-sided testing | Asymmetric two-sided testing | Spending function design | References

Introduction | Fixed design derivation | Outcome and dropout distributions | Enrollment and trial duration | Deriving design with no interim analyses | Varying enrollment duration to power trial | Group sequential design | Additional parameters | Generating the design | Textual summary | Tabular summaries | Summary plots | Update bounds at time of analysis | Evaluating interim results | References

Overview and notation | Sample size | Testing and confidence intervals | Non-inferiority and super-superiority | Simulation | Power table | Summary | References

Introduction | Design | Update design at time of interim analysis | Testing and conditional power | Predictive power | References

Introduction | Examples | References