Intention to treat (ITT)
Analyzes patients based on original assignment regardless if they completed or adhered
Benefits of ITT
Preserves randomization
Minimizes bias
Reflects real-world effectiveness
Avoids overestimating efficacy- gives robust and conservative estimate
Ensures comparability of groups over time- essential for valid causal inference
protects the internal validity of the trial and ensures the results are both scientifically rigorous and clinically relevant
How does ITT preserve randomization
keeps participants in their originally assigned groups, regardless of adherence or dropout.
This maintains the balance of confounding variables achieved through randomization. preventing unknown systematic bias in dropouts
allowing a fair and unbiased comparison of treatment effects across groups regardless of deviations in participant behavior or protocol adherence
How does ITT minimize bias
Prevents attrition bias (i.e., bias introduced when analyzing only those who complete the study).
Avoids cherry-picking favorable results by including all participants
How does randomization reduce bias
by assigning treatment independent of potential confounders
Minimized selection bias
How does PP analysis overestimate efficacy
Per-protocol or as-treated analyses can inflate treatment effect estimates by excluding non-adherent or drop-out participants.
ITT gives a more conservative and robust estimate of the intervention’s impact.
How does ITT preserve statistical power
more conservative estimate of treatment effects, assumes dropouts are random (Could still be biased if really not random e.g. due to treatment AEs)
by maintaining the original, planned sample size throughout the analysis
Why do you need to do sensitivity analysis in ITT
Would need to conduct sensitivity analyses to test how robust results are under assumptions about missing data, dropout etc.
Use LOCF or multiple imputation to account for missing data
Why choose RCT
Random assignment reduces selection bias
Gold standard for cause-effect determination
Control group comparison
justifies use of parametric tests (since any skewedness should be same both groups)
supports generalizability
Enables calculation of risk ratios, NNT, confidence intervals
Minimizes confounding
Standardization/reproducibility of intervention
I.e. most reliable method for evaluating intervention efficacy due to their methodological rigor and ability to control bias
How does RCT reduce selection bias
Balances known and unknown confounding variables between groups
Why choose RCT
Random assignment reduces selection bias
Gold standard for cause-effect determination
Control group comparison
Statistical validity
Minimizes confounding
Standardization/reproducibility of intervention
I.e. most reliable method for evaluating intervention efficacy due to their methodological rigor and ability to control bias
How does RCT assess cause and effect relationships
Controls for temporal ambiguity—intervention clearly precedes the outcome
How does Blinding reduces observer bias
Single- or double-blinding prevents knowledge of group assignment from influencing outcomes or assessment.
How does RCT minimize confounding
Unlike observational studies (e.g., cohort or case-control), RCTs actively control for confounding at the design level, not just in analysis.
How does RCT ensure statistical validity
randomization and concealment to ensure that treatment and control groups are comparable at the start of the study, thereby eliminating selection bias and balancing both known and unknown confounding factors. This methodological rigor ensures that any observed differences in outcomes can be attributed to the intervention rather than pre-existing differences, maximizing internal validity
Randomization justifies use of parametric tests and supports generalizability when well-powered.
Enables calculation of risk ratios, NNT (number needed to treat), and confidence intervals.
Stratified vs. Enriched
Priimary difference lies in patient selection: stratified trials enroll a broad population but balance subgroups, while enriched trials exclude patients unlikely to respond
randomizing within biomarker defined groups (each treatment group has people pos/neg for the biomarker) where enriched is just selecting for it
I.e. different biomarker same treatment all groups vs. same biomarker all groups different treatment
Define stratified
Stratified: ensures characteristic/strata is balanced across treatment groups
Definition: assigning participants to groups to ensure characteristics balanced across
Divide into strata based on characteristic then randomize within stratum
Participants subgrouped into strata based on characteristics, then randomization done AFTER, within each subgroup
Define enriched design
Enriched: selecting participants based on characteristic expected to increase likelihood of success (but not always, idea is to select best chance of observing drug effect-may depend on the biomarker goal)
Definition: intentionally selecting group with a characteristic likely to benefit
Enriched design focuses on one biomarker groups: biomarker positive patients who do/don’t receive the treatment
I.e. same biomarker different treatment
Why choose enriched design
Ensures trial focuses on those most likely to respond, increasing chance of signal detection
Can be biomarker-driven, Clinical characteristics driven, or using predictive modeling
focuses on one biomarker groups: biomarker positive patients who do/don’t receive the treatmentures groups are comparable pre-treatment to confidently interpret post-treatment differences as attributed to intervention not other factors
Define confound
Confounding = third variable associated with both the exposure and outcome, thus potentially distorting the relationship between them
Often demographic, clinical, environmental
How to address confounds
Start by reviewing existing literature to identify possible known confounders
Then control for in statistical tests by adding as covariates
How to account for dropout rates
ITT
Multiple imputation/FIML
Document reasons
Compare groups to ID systemic patterns
