Regulatory Science Virtual Symposium: “Make Informed Decisions: Key Statistical Principles to Clinical Trial Design” Session 5: CTSI Clinical Study Design Types (2022)

1. Objectives
   1. Review of study designs, including clinical trials
   2. Examples of study designs and clinical trials
   3. Alignment of study designs with: research question, data collection, and statistical analysis
2. PICOT Criteria to Develop the Research Question
   1. P = Population
   2. I = Intervention/Exposure
   3. C = Comparison Group
   4. O = Outcome
   5. T = Time
3. Spectrum of Study Designs
4. Descriptive vs. Analytic Study
   1. Descriptive Study: Research questions involving Population and Outcome
   2. Analytic Study: Adds Intervention/Exposure and Comparator Group
5. Observation Study Defined
   1. Are studies where participants already belong certain study groups rather than assigned by the investigators and often exposures or interventions are self-selected
6. Cohort Study
   1. Involve individuals free of outcome, including persons with and without exposure
   2. Follow forward to determine outcome
   3. Statistical methods: Chi-square, logistic regression, propensity scores
7. Case-Control Study
   1. Select persons with (cases) and without (controls) outcome and then go back to determine their past exposure
   2. Statistical methods: Chi-square test, T-test or non-parametric Wilcoxon rank sum, and logistic regression
8. Clinical Trial Defines
   1. Clinical study in which participants are assigned to receive intervention (or no intervention) so that researchers can evaluate effects of intervention on biomedical/health-related outcomes
      1. Not feasible when assigning to an exposure/intervention is not ethical
9. Quasi-Experimental Study
   1. Assign groups without randomization
   2. Considered as “natural” experiments
   3. Design takes advantage of what would already happening or a change in the natural setting; usually utilized in healthcare settings
   4. Comparator is the pre-implementation period
   5. Statistical methods: Regression model for counts
10. Clinical Trial Designs: Parallel Group
    1. Parallel group: each participant assigned to one and only one of the trial interventions
       1. Standard approach for most clinical trials
    2. Statistical methods: Non-parametric Wilcoxon signed rank, non-parametric Wilcoxon rank sum, and intervention effect sizes
11. Crossover
    1. Each participant receives both the experimental and comparator interventions, usually in randomized order, with a wash-out period in between interventions
    2. Each participant acts as their own control, which reduces the variability
    3. Disadvantages: greater likelihood of dropout; must be a stable disease under study; only appropriate for interventions that wash-out and have short-term outcomes
    4. Statistical methods: Paired t-tests, mixed effects regression model
12. Cluster Randomized
    1. Definition: Unit of randomization is a group of persons rather than a single person
    2. Necessary for complex interventions in primary care, health promotion, community/public health settings i.e., schools
    3. Avoids contamination of intervention effects however requires more subjects and blinding is not possible
13. Clinical Trial Designs: Equivalence or Non-Inferiority Trials
    1. Trials that hypothesize that new intervention groups will demonstrate trial outcomes that are the same as (equivalency) or no worse than (non-inferiority) a currently standard intervention
    2. Appropriate if there is a standard and effective intervention avalible
14. Statistical Analysis Plan
    1. Ties directly back to your research questions, aims, and hypotheses
    2. Must address the following: dependent/outcome variables, independent variables, method and frequency of measurement
15. Testing Differences Among Groups
    1. Certain statistical analyses would be more appropriate for particular comparisons i.e., for categorical data, groups are compared with chi-square tests that incorporate the matching (McNemar’s test for proportions)
16. Survival Time Data
    1. Two components: (1) Did the subject have the event (death)? (2) What is the last time the subject was observed?
17. Linear Regression (and Other Regression Models)
    1. Use linear regression to adjust for other variables
18. Other Regression Models
    1. Continuous outcome: linear regression
    2. Dichotomous outcome: logistic regression
    3. Ordinal categorial outcome: ordinal logistic regression
    4. Nominal outcome (not ordered): multinomial logistic regression
    5. Count outcome: Poisson or negative binomial regression
    6. Survival outcome: Cox (proportional hazards) or other “survival” regression
19. Questions?

Accompanying text created by Annie Ly | RKS Project Administrator, SC CTSI  lyannie@usc.edu