Trial Navigator Assistant
Lesson 5 Decision Assistant
RCTs • Observational Designs • Bias & Confounding — structured, interactive planning
How to use this tool
- RCT Builder: Define arms, ratio, blinding, outcomes → build protocol skeleton. Generate a seeded randomization list and export CSV.
- Observational Designer: Choose design (cohort/case-control/cross-sectional), sampling, exposure/outcome, confounders → get analysis plan.
- Bias & Confounding: Run a bias check, get mitigation strategies, and add them to your consolidated plan.
- Scenarios: Load worked examples to see complete outputs.
- Outputs: Copy/print/export JSON; download randomization CSV when generated.
Note: Educational planning aid — align with ICH-GCP, IRB/IEC, DSMB, and institutional SOPs.
Core concepts
RandomizationBlindingControl groups
ConfoundingIntention-to-treatPer-protocol
CohortCase-controlCross-sectional
Selection/Info/Recall/Observer bias
Lesson 5.1 — RCT Builder
Randomization list (optional)
Preview below (first 20 rows)
No list generated yet.
Protocol Skeleton
Fill inputs and click Build RCT Plan.
Analysis notes
- Primary analysis on ITT set; sensitivity on per-protocol if applicable
- ANCOVA/MMRM with baseline covariates; missing data handling pre-specified
- Multiplicity control for multiple doses/endpoints
Lesson 5.2 — Observational Study Designer
Design & Analysis Plan
Set inputs and click Build Observational Plan.
Method tips
- Cohort: Risk ratios; time-to-event → Cox PH; adjust via multivariable models
- Case-control: Odds ratios via logistic regression; careful control selection
- Cross-sectional: Prevalence ratios/OR; beware temporal ambiguity
- Address confounding: matching, stratification, propensity, regression
Lesson 5.3 — Bias & Confounding Analyzer
Potential biases (check those at risk)
Mitigations planned
Bias Check & Mitigation
Select risks and click Generate Mitigation Plan.
Notes
- Report potential biases & confounders; justify mitigations
- Pre-specify analysis and handling of missing data
Load Example Scenarios
RCT — Antihypertensive
200 pts; 1:1; DB; Δ SBP at Week 12; DSMB; ITT + PP.
Observational — Diet & Heart Disease
Prospective cohort; random sampling; adjust for age/sex/smoking/activity.
Bias — Weight-loss Program
Cross-sectional; risks: selection/info; mitigations: adjust + transparency.
Outputs
Consolidated Plan
Build any section to populate this area.
Summary Tables
RCTs — Participants, Groups, Randomization, Blinding
| Aspect | Description | Example |
|---|---|---|
| Selection | Inclusion/exclusion criteria; random assignment eliminates selection bias | Type 2 diabetes, 30–65 yrs |
| Groups | Intervention vs placebo/SOC/active comparator | New antidepressant vs placebo |
| Randomization | Simple/block/stratified; allocation concealment | Computer-generated 1:1 |
| Blinding | Single/Double/Open; prevents assessor/participant bias | Double-blind supplement trial |
RCTs — Protocol, Outcomes, Analysis
| Aspect | Description | Example |
|---|---|---|
| Protocol | Objectives, design, stats, organization | Cholesterol-lowering drug plan |
| Outcomes | Primary + secondary; timeframe | Asthma attack frequency; QoL |
| Data collection | Clinical assessments, questionnaires, EHRs | HbA1c, adherence, AEs |
| Analysis | ITT; PP sensitivity; pre-specified models | ANCOVA/MMRM; missing data plan |
Observational — Designs & Methods
| Design | Description | Example |
|---|---|---|
| Cohort | Follow exposure→outcome; prospective/retrospective | Diet → heart disease (10y) |
| Case-control | Cases vs controls; past exposures | Pesticide ↔ Parkinson’s |
| Cross-sectional | Single timepoint; prevalence & associations | Obesity × dietary habits |
Bias & Confounding — Types & Strategies
| Type | Definition | Example |
|---|---|---|
| Selection | Non-representative enrollment | Clinic cases vs general controls |
| Information | Measurement error/misclassification | Diet misreporting |
| Recall | Diff. memory of exposures | Cases recall pesticides more |
| Observer | Assessor expectations | Unblinded outcome grading |
| Mitigation | Randomize, blind, match/restrict, adjust | Logistic regression w/ confounders |
