agentsclimarketplace

Clinical trial design patterns

Skill choxos/BiostatAgent/plugins/clinical-trial-simulation/skills/clinical-trial-design-patterns

Claude Code plugin marketplace for biostatistics in R — 30 agents, 17 commands, and 45 skills spanning Bayesian modeling (Stan/PyMC/JAGS), indirect treatment comparisons (NMA/MAIC/STC/ML-NMR), tidy R workflows, and clinical trial simulation.

Install
npx -y skills add choxos/BiostatAgent --skill clinical-trial-design-patterns

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 8 stars8 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Common clinical trial design patterns including multi-arm, multi-endpoint, adaptive, and stratified designs. Use when selecting or implementing trial designs.

SKILL.md

9.4 KB, as published. Nobody here has run it

Clinical Trial Design Patterns

When to Use This Skill

  • Selecting appropriate trial design for clinical objectives
  • Implementing multi-arm or multi-endpoint trials
  • Designing stratified trials
  • Planning adaptive designs
  • Understanding design trade-offs

Two-Arm Parallel Design

Standard Design

The most common design: randomize patients to treatment or control.

# simtrial implementation
sim_pw_surv(
  n = 400,
  block = c(rep("control", 1), rep("experimental", 1)),  # 1:1
  enroll_rate = data.frame(rate = 20, duration = 12),
  fail_rate = fail_rate
)

# Mediana implementation
DataModel() +
  OutcomeDist(outcome.dist = "NormalDist") +
  SampleSize(200) +  # Per arm
  Sample(id = "Control", outcome.par = parameters(mean = 0, sd = 1)) +
  Sample(id = "Treatment", outcome.par = parameters(mean = 0.5, sd = 1))

Unequal Randomization

When to Use:

  • Increase exposure to experimental treatment
  • Ethical considerations
  • Resource optimization
# 2:1 randomization (experimental:control)
sim_pw_surv(
  n = 300,
  block = c("control", rep("experimental", 2))
)

# Mediana with unequal allocation
DataModel() +
  Sample(id = "Control", sample.size = 100, ...) +
  Sample(id = "Treatment", sample.size = 200, ...)

Trade-off: Unequal allocation reduces power for same total N.

Multi-Arm Designs

Dose-Finding (Multiple Doses vs Placebo)

# Three doses + placebo
DataModel() +
  OutcomeDist(outcome.dist = "NormalDist") +
  SampleSize(75) +  # Per arm
  Sample(id = "Placebo", outcome.par = parameters(mean = 0, sd = 1)) +
  Sample(id = "Low Dose", outcome.par = parameters(mean = 0.3, sd = 1)) +
  Sample(id = "Mid Dose", outcome.par = parameters(mean = 0.5, sd = 1)) +
  Sample(id = "High Dose", outcome.par = parameters(mean = 0.7, sd = 1))

# Analysis with Dunnett-type comparison
AnalysisModel() +
  Test(id = "Low vs Placebo", samples = samples("Placebo", "Low Dose"), method = "TTest") +
  Test(id = "Mid vs Placebo", samples = samples("Placebo", "Mid Dose"), method = "TTest") +
  Test(id = "High vs Placebo", samples = samples("Placebo", "High Dose"), method = "TTest") +
  MultAdjProc(proc = "HolmAdj")

Active Comparator Design

# Treatment vs Active Control
DataModel() +
  Sample(id = "Active Control", outcome.par = parameters(mean = 0.4, sd = 1)) +
  Sample(id = "New Treatment", outcome.par = parameters(mean = 0.6, sd = 1))

Multi-Endpoint Designs

Co-Primary Endpoints

Both endpoints must be significant for trial success.

# Correlated endpoints
corr.matrix <- matrix(c(1.0, 0.5, 0.5, 1.0), 2, 2)

DataModel() +
  OutcomeDist(outcome.dist = "MVNormalDist") +
  SampleSize(100) +
  Sample(id = list("Control E1", "Control E2"),
         outcome.par = parameters(
           parameters(par = parameters(
             parameters(mean = 0, sd = 1),
             parameters(mean = 0, sd = 1)
           ), corr = corr.matrix))) +
  Sample(id = list("Treatment E1", "Treatment E2"),
         outcome.par = parameters(
           parameters(par = parameters(
             parameters(mean = 0.4, sd = 1),
             parameters(mean = 0.3, sd = 1)
           ), corr = corr.matrix)))

# Evaluation: Conjunctive power (both must be significant)
EvaluationModel() +
  Criterion(id = "Co-primary",
            method = "ConjunctivePower",
            tests = tests("E1 Test", "E2 Test"),
            par = parameters(alpha = 0.025))

Hierarchical Endpoints

Primary must succeed before secondary is tested.

# Primary → Key Secondary → Other Secondary
AnalysisModel() +
  Test(id = "Primary", ...) +
  Test(id = "Key Secondary", ...) +
  Test(id = "Other Secondary", ...) +
  MultAdjProc(proc = "FixedSeqAdj")

Multiple Primary with Gatekeeping

# Two primary, two secondary
MultAdjProc(
  proc = "ParallelGatekeepingAdj",
  par = parameters(
    family = families(family1 = c(1, 2), family2 = c(3, 4)),
    proc = families(family1 = "HolmAdj", family2 = "HolmAdj"),
    gamma = families(family1 = 0.8, family2 = 1)
  )
)

Stratified Designs

Single Stratification Factor

# simtrial stratification
sim_pw_surv(
  n = 400,
  stratum = data.frame(
    stratum = c("Low Risk", "High Risk"),
    p = c(0.4, 0.6)  # Prevalence
  ),
  fail_rate = data.frame(
    stratum = rep(c("Low Risk", "High Risk"), each = 2),
    period = rep(1, 4),
    treatment = rep(c("control", "experimental"), 2),
    duration = rep(100, 4),
    rate = c(0.03, 0.02, 0.06, 0.04)  # Different by stratum
  )
)

Biomarker-Defined Subgroups

# Marker-positive and marker-negative populations
DataModel() +
  OutcomeDist(outcome.dist = "NormalDist") +
  SampleSize(100) +
  Sample(id = "Control M+", outcome.par = parameters(mean = 0, sd = 1)) +
  Sample(id = "Control M-", outcome.par = parameters(mean = 0, sd = 1)) +
  Sample(id = "Treatment M+", outcome.par = parameters(mean = 0.6, sd = 1)) +
  Sample(id = "Treatment M-", outcome.par = parameters(mean = 0.2, sd = 1))

# Pooled analysis (Overall Population)
AnalysisModel() +
  Test(id = "Overall",
       samples = samples(c("Control M+", "Control M-"),
                        c("Treatment M+", "Treatment M-")),
       method = "TTest")

# Subgroup analysis
AnalysisModel() +
  Test(id = "M+ Subgroup",
       samples = samples("Control M+", "Treatment M+"),
       method = "TTest") +
  Test(id = "M- Subgroup",
       samples = samples("Control M-", "Treatment M-"),
       method = "TTest")

Event-Driven Designs

Time-to-Event with Fixed Events

# Mediana event-driven
DataModel() +
  OutcomeDist(outcome.dist = "ExpoDist", outcome.type = "event") +
  Event(n.events = c(300, 350, 400), rando.ratio = c(1, 1)) +
  Design(
    enroll.period = 24,
    study.duration = 48,
    enroll.dist = "UniformDist",
    dropout.dist = "ExpoDist",
    dropout.dist.par = parameters(rate = 0.01)
  ) +
  Sample(id = "Control", outcome.par = parameters(rate = log(2)/12)) +
  Sample(id = "Treatment", outcome.par = parameters(rate = log(2)/18))

PFS/OS Correlated Endpoints

# Correlated survival endpoints
DataModel() +
  OutcomeDist(outcome.dist = "MVExpoPFSOSDist",
              outcome.type = c("event", "event")) +
  Event(n.events = 350, rando.ratio = c(1, 1)) +
  Sample(id = list("Control PFS", "Control OS"),
         outcome.par = parameters(
           parameters(
             par = parameters(
               parameters(rate = log(2)/6),   # PFS
               parameters(rate = log(2)/15)   # OS
             ),
             corr = matrix(c(1, 0.3, 0.3, 1), 2, 2)
           ))) +
  Sample(id = list("Treatment PFS", "Treatment OS"),
         outcome.par = parameters(
           parameters(
             par = parameters(
               parameters(rate = log(2)/9),
               parameters(rate = log(2)/20)
             ),
             corr = matrix(c(1, 0.3, 0.3, 1), 2, 2)
           )))

Adaptive Designs

Sample Size Re-Estimation

Concept: Adjust sample size at interim based on observed effect size.

# Simulation framework for adaptive design
# 1. Generate interim data
# 2. Estimate effect size
# 3. Re-calculate sample size
# 4. Complete enrollment
# 5. Perform final analysis

simulate_adaptive <- function(initial_n, interim_frac, target_power) {
  # Stage 1: Interim
  n_interim <- round(initial_n * interim_frac)
  interim_data <- generate_data(n_interim)
  effect_estimate <- estimate_effect(interim_data)

  # Re-estimate sample size
  new_n <- calculate_sample_size(effect_estimate, target_power)
  new_n <- max(new_n, initial_n)  # Cannot decrease

  # Stage 2: Continue to new_n
  final_data <- generate_data(new_n)

  return(final_data)
}

Response-Adaptive Randomization

Concept: Adjust randomization ratio based on interim results.

Note: More common in Bayesian settings; simtrial/Mediana focus on fixed designs.

Design Selection Flowchart

START
  │
  ├─ How many treatment arms?
  │   ├─ 2 → Two-arm parallel
  │   └─ 3+ → Multi-arm design
  │
  ├─ How many primary endpoints?
  │   ├─ 1 → Single primary
  │   ├─ 2 (both required) → Co-primary
  │   └─ 2+ (any success) → Multiple primary with multiplicity
  │
  ├─ Are there secondary endpoints?
  │   ├─ Yes, hierarchical → Fixed-sequence or gatekeeping
  │   └─ Yes, equal priority → Holm/Hochberg
  │
  ├─ Is stratification needed?
  │   ├─ Yes → Stratified randomization
  │   └─ No → Simple randomization
  │
  ├─ Endpoint type?
  │   ├─ Continuous → Normal-based tests
  │   ├─ Binary → Proportion tests
  │   ├─ Time-to-event → Logrank/survival methods
  │   └─ Count → Poisson/NegBinom tests
  │
  └─ Interim analyses needed?
      ├─ Yes → Group sequential design
      └─ No → Fixed design
END

Best Practices

  1. Match Design to Objectives: Choose design that directly addresses primary question
  2. Consider Multiplicity Early: Plan adjustment strategy during design phase
  3. Stratify When Important: Use stratification for known prognostic factors
  4. Pre-specify Everything: Document design choices before data collection
  5. Simulate Extensively: Validate operating characteristics via simulation
  6. Consider Regulatory Path: Align design with agency expectations

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.