Sensitivity analysis

Sensitivity Analysis

Sensitivity analysis is a statistical technique used in RCTs to test the robustness of study findings by assessing how results change under different assumptions, data handling methods, or analytical strategies. It helps determine whether the trial’s conclusions remain valid when alternative scenarios are considered.

Objectives of Sensitivity Analysis

1. Assess Robustness – Test whether the main findings are stable under varied conditions.
2. Evaluate Missing Data Impact – Explore different ways of handling missing data.
3. Check for Model Dependency – Examine if conclusions depend on the statistical model used.
4. Address Protocol Deviations – Evaluate the impact of excluding or including protocol violators.
5. Support Decision-Making – Provide reassurance to regulatory agencies and clinicians.

Types of Sensitivity Analyses in RCTs

1. Missing Data Handling Sensitivity Analyses

• Complete Case Analysis (CCA): Uses only participants with complete data; may be biased if data are not missing at random.
• Multiple Imputation (MI): Predicts missing values based on observed data and creates multiple datasets.
• Inverse Probability Weighting (IPW): Weighs complete cases to account for the probability of missingness.
• Worst-case/Best-case Scenarios: Assumes extreme outcomes for missing data (e.g., all missing in the treatment group had poor outcomes).
• Pattern Mixture Models: Model different missing data mechanisms (e.g., dropout due to side effects).

2. Alternative Statistical Models

• Per-protocol analysis: Analyzes only adherent participants to estimate efficacy.
• Modified Intention-to-treat analysis (mITT): Includes most, but not all, randomized participants (e.g., excludes those never treated).
• As-Treated (AT) Analysis: Participants analyzed based on the treatment they received, not what they were randomized to.

3. Alternative Outcome Definitions

• Different Time Points: Compare results at different follow-up times.
• Composite Endpoints: Vary which outcomes are combined or how they are defined.
• Alternative Cutoffs: Adjust thresholds for outcome classification (e.g., redefine hypertension based on a different blood pressure level).

4. Subgroup and Covariate Sensitivity

• Stratified Analysis: Reanalyze data within subgroups (e.g., by age, sex, severity).
• Covariate Adjustment: Include or exclude additional covariates in the model to test sensitivity.

5. Excluding Outliers

• Winsorization: Replace extreme values with lower/upper percentile values to reduce outlier impact.
• Trimming: Remove outliers entirely and reanalyze to assess their influence on outcomes.

Example Applications of Sensitivity Analysis

• Diabetes Control and Complications Trial (DCCT): Used multiple imputation to evaluate the impact of missing follow-up data.
• COVID-19 Vaccine Trials: Performed sensitivity analyses by excluding participants with early infections or applying different assumptions about missing data.

Conclusion

Sensitivity analyses are crucial to validate the findings of RCTs. They explore how varying assumptions, missing data strategies, statistical models, and definitions influence the results. Well-conducted sensitivity analyses increase the reliability and generalizability of findings and are especially important in regulatory submissions and clinical decision-making.

See also

• Intention-to-treat analysis
• Per-protocol analysis
• Missing data
• Subgroup analysis

---

Bibliography

1. Piantadosi S. Clinical Trials: A Methodologic Perspective. 3rd ed. Wiley; 2017. Chapter 12: Assessing robustness of trial findings, including sensitivity analysis.
2. National Research Council. The Prevention and Treatment of Missing Data in Clinical Trials. Washington, DC: National Academies Press; 2010. Includes a dedicated section on sensitivity analyses for missing data.
3. White IR, Thompson SG. Adjusting for partially missing baseline measurements in randomized trials. Statistics in Medicine. 2005;24(7):993–1007. Includes sensitivity analysis strategies.
4. Higgins JPT, Thomas J, Chandler J, et al. (editors). Cochrane Handbook for Systematic Reviews of Interventions, version 6.3 (updated February 2022). Cochrane; 2022. Chapter 10: Addressing missing data and conducting sensitivity analyses.
5. Scharfstein DO, Daniels MJ. Going beyond the intention-to-treat analysis: sensitivity analysis for trials with missing data. Statistics in Medicine. 2008;27(13):1307–1326.
6. Thabane L, Mbuagbaw L, Zhang S, Samaan Z, Marcucci M, Ye C, Thabane M, Giangregorio L, Dennis B, Kosa D, Debono VB, Dillenburg R, Fruci V, Bawor M, Lee J, Wells G, Goldsmith CH: A tutorial on sensitivity analyses in clinical trials: the what, why, when and how. BMC Medical Research Methodology 2013, 13:92.

---

Adapted for educational use. Please cite relevant trial methodology sources when using this material in research or teaching.