1. CHS502 Unit IV-1 Randomized Blinded Trial Ahmed A. Mirza, PhD

2. Unit IV Outline Outline: Designing randomized blinded trial 1.Selecting intervention 2.Selecting control 3.Selecting outcome measurements 4.Selecting participants Measuring baseline variables Randomizing and blinding 1.Random assignments a)Blocked randomization b)Stratified blocked randomization c)Matched pairs randomization 2.Blinding Factorial design 1.The basic 2 × 2 2.Factorial design variations 3.Cluster randomization Nonrandomized between-group design Within-group design Pilot clinical trials Conducting clinical trials 1.Follow-up 2.Adjudicating outcome 3.Monitoring 4.Analyzing the results Designing studies for medical tests 1.General issues a)Gold standard b)Spectrum of severity of diseases c)Other sources of variations d)Blinding e)Costs vs. charges Test reproducibility Test accuracy 1.Sensitivity and specificity 2.Positive and negative predictive values 3.Receiver operating characteristics (ROC) curves

3. This is not an observational study design This is an intervention Probably the best to establish causality – Random: to eliminate confounders – Blinded: to eliminate observer effects be time consuming and laborious Can be expensive Usually is the last step of research after strong evidence and support

4. General design Select a sample from the intended population Measure the dependent and baseline variables Randomly assign subjects and controls BLINDLY administer the intervention (e.g., drug and placebo) Follow-up by measuring the outcome (WHILE BLINDED to group assignment)

5. Intervention It is the critical first step Must consider –Effectiveness and safety Establishing the lowest effective dose with least risks –Dose –Number of interventions Single (keeps it easy) Multiple Using estrogen + progestin therapy increased risk of breast cancer… which one??

6. Controls Control will receive no intervention –Placebo The problem is ethical: How can you withhold treatment? –Give standard care to all + new treatment, this is called “co-intervention” –Reduces causality –If the new treatment is proven effective in premedical trials then offer an equivalence trial (controls take approved intervention while test subjects take the intervention being tested

7. Outcome measurements Clinical outcome: best way to assess the intervention –Quality of life –Better post-intervention Renal function Hepatic function Days in the hospital –Surrogate markers Values that reflect the clinical outcome and is a is immediately affected by the intervention – great predictor –BUN/Creat, AST/ALT,... –Adverse effects

8. Selecting subjects Inclusion criteria –Not too many that will cost too much –Too little not enough power –Select patients that will show the greatest benefit Stratification: Enrolling subgroups according to a predicted outcome

9. Exclusion Criteria –Exposure to treatment/placebo might be harmful Allergies –Co-intervention Existing treatment might interfere with new one –Adherence Unlikely to adhere or continue the study –Moving –Problems with the subject Mental state

10. Measuring baseline variables Collect all possible data about the subjects Useful for later generalizability Comparability Reduction of confounders Risk factors that might effect subgroups Can be used to subgroup subjects to establish “effect modification” or “interaction” Showing that alendronate had effect on women with low bone density but not on women with high bone density.

11. Randomizing Baseline examination Consent Randomly assign by computerized algorithm Must follow rigorous precautions to prevent tampering with subject and control assignment. Observers can be very bias.

12. Best is just a simple randomization to an equal ratio between subjects and control To increase power to smaller studies Blocked Randomization: Stratified blocked randomization: Unequal allocation Randomization of matched pairs (read on page 157)

13. Blinding Blinding: –Subject –Observer Multiple sclerosis study: patients received plasma exchange/medication vs. sham plasma exchange and placebo. Physicians who were blinded show no difference, physicians who were not blinded concluded significant difference in outcome.

14. If you have a problem with… Chance (as discussed before) Bias (as discussed before) Effect-Cause (not possible) Confounding –Prerandomization (Randomization) –Postrandomization (Blinding)

15. What if you CANNOT blind? Control confounders as much as possible Have outcome measured by different observer than ones who worked with the subjects

16. CHS 502 Unit IV-2 Alternative Trial Design Ahmed A. Mirza, PhD

17. Factorial Design Aims to answer more than one question at the same time using a single cohort of participants Treating the sample with both drugs A & B, while maintaining the presence of placebo a & b A total of 4 groups receiving 4 combination of the drugs (AB, Ab, aB, ab) Steps, like the general steps of a randomized blinded trial covered previously, however add: Analysis: compare the two groups receiving drug A with the ones receiving placebo a. Then compare the ones receiving A vs B, and so on.

18. Advantages Can be very efficient Two trials for the price of one Disadvantages Possible interaction between interventions, which reduces power Sample must be appropriate for both interventions

19. Group or cluster randomization Rather than assigning individuals, assign whole groups to receive, or not to receive, the intervention. Naturally occurring groups such as teams, families, classes It reduces subject bias More feasible and cost effective Will reduce power More complex stats and sample size estimation

20. Nonrandomized Between-Group Design Such studies are less effective than randomized trials. Why? Can use “adjustment” to deal with variations amongst base-line variables Nonrondomized studies show greater benefit of the intervention, however confounders might introduce a statistical weakness.

21. Within-Groups Designs Time-Series design: Measurements are made before and after each subject receives the intervention The subject plays both rolls of control and case Lack of concurrent control Learning effects: the subject being exposed to the assessment twice (virtual microscopy) Regression to the mean: due to random variations (blood pressure)

22. Useful to do is starting and stopping the intervention. Provide stronger support that changes are due to the intervention (Statin effects on LDL- Cholesterol)

23. The crossover design Similar to what we discussed before Switching the control group to the test group and test group to the control group Groups taking the drug will take a placebo and the group taking the placebo will take the drug. Problem, “carryover effects”, the time it takes the drug to clear the system and no longer have any effects. Introduce a “washout” period

24. Stages of new Therapy Preclinical: Cell culture and animals Phase I: Unblinded, uncontrolled, check for safety Phase II: small randomized blinded study, check for dose and clinical outcome and Phase III: Large randomized and blinded study, check for clinical outcome Phase IV: After approval. Done by pharm companies to assess side effects and other uses for the intervention.

25. Designing Studies for Medical Tests How reproducible is the test? Intra- and inter-observer How accurate is the test Does it match the gold standard How often does the test effect clinical decision? What are the costs, risks associated with the test? How is going to improve the clinical outcome ?