Intervention Studies Principles of Epidemiology Lecture 10 Dona Schneider, PhD, MPH, FACE

PHCO 0502 Principles of Epidemiology (Schneider) Intervention Studies Subjects are selected from a reference population, the group to which investigators hope to extrapolate their findings Clear, specific definition of subjects prior to selection No subjective decision making!

PHCO 0502 Principles of Epidemiology (Schneider) Intervention Studies (cont.) Individuals are enrolled on the basis of exposure (the investigators control the intervention) Both cases and controls come from the experimental group All subjects (cases and controls) should be at high enough risk for manifesting the outcome so that the study is likely to detect a difference if the intervention works

PHCO 0502 Principles of Epidemiology (Schneider) Potential Problems Selection bias Volunteerism – after screening, the experimental group may no longer be generalizable to the reference population Ethical considerations IRB issues

PHCO 0502 Principles of Epidemiology (Schneider) Informed Consent Describe the overall experience that will be encountered Describe benefits of participation Describe the risks of participation Describe alternatives to participation Describe the extent to which the subject’s information will remain confidential Describe compensation and/or expenses that may be incurred

PHCO 0502 Principles of Epidemiology (Schneider) Informed Consent (cont.) Obtain informed consent prior to a subject’s enrollment in a study A subject may choose not to participate or withdraw at any time without negative consequences Provide the subject with a list of people s/he can contact with further questions regarding the research, his/her rights as a participant, and potential research-related injuries Remember, informed consent is a process, not just a form!

PHCO 0502 Principles of Epidemiology (Schneider) Potential Problems (cont.) Reporting bias Observer bias Watch the experimental group more carefully than the control group

PHCO 0502 Principles of Epidemiology (Schneider) Controlling Bias and Confounding Randomization Distribute known and unknown confounders evenly among treatment groups Occurs after informed consent is provided Sufficient sample size Improves the power to detect a difference Improves the probability of generalizability to the reference population

PHCO 0502 Principles of Epidemiology (Schneider) Controlling Bias and Confounding (cont.) Masking Prevent subjects and study personnel from knowing who is in which treatment group Verify compliance (reduce reporting bias) Pill counting, laboratory studies, interviews of living companions

PHCO 0502 Principles of Epidemiology (Schneider) Controlling Bias and Confounding (cont.) Maintaining compliance with the intervention Home visits Payment at time of visit Telephone and postcard reminders Calendar pill packs Daily logs Pre-study compliance checks Document reasons for noncompliance

PHCO 0502 Principles of Epidemiology (Schneider) Intention to Treat Analysis Once randomized ALWAYS analyzed The analysis must always include subjects who did not comply with the intervention or who did not finish the study If you eliminate those who did not comply, you cannot address the research question – whether offering a treatment program is of benefit Those who comply may be different from the entire experimental group By using only those subjects who comply, you introduce further selection bias and reduce the generalizability of your results

PHCO 0502 Principles of Epidemiology (Schneider) Internal vs. External Validity Large controlled trials usually have a high degree of internal validity Randomization and masking minimizes the risk of confounding and bias, and a large “n” makes it more likely that chance can be ruled out as an explanation of an observed association However, controlled trials also often have poor external validity (i.e., generalizability)

Agree to screening? - YesAgree to screening? - No Meet inclusion criteria? - Yes Wish to continue? - YesWish to continue? - No Agree to randomization? - YesAgree to randomization? - No Experimental Population External Validity in Controlled Trials Respond to letter? - YesRespond to letter? - No Meet inclusion criteria? - No Are they similar? Reference Population

PHCO 0502 Principles of Epidemiology (Schneider) Crossover Studies Subjects begin the study on Treatment A and later switch to Treatment B Patients serve as their own control Variation between individuals remains constant Washout period between treatments reduces residual carryover

Design of a Planned Crossover Trial Randomized Treatment A Group 1 Group 2 Treatment B Group 2 Group 1

PHCO 0502 Principles of Epidemiology (Schneider) Factorial Design Use the same study population to test Drug A & Drug B Assume: The outcomes for each drug are different Modes of action are independent If you need to terminate the study of Drug A, you can continue the study to determine the effects of Drug B instead of beginning an entirely new study.

PHCO 0502 Principles of Epidemiology (Schneider) Factorial Design (cont.) Example: Physician’s Health Study Test aspirin as a means of preventing cardiovascular disease Test beta-carotene as a means of preventing cancer Terminated aspirin arm early due to a significant drop in the risk of first myocardial infarctions Continued beta-carotene arm to completion

PHCO 0502 Principles of Epidemiology (Schneider) Factorial Design for Studying Effects of Two Treatments Both A and B (a) A only (b) B only (c) Neither A nor B (d) Treatment B Treatment A -

PHCO 0502 Principles of Epidemiology (Schneider) To Estimate Sample Size in a Clinical Trial You Need The difference in response rates to be detected An estimate of the response rate in one of the groups Level of statistical significance (  ) The value of the power desired (1 –  ) Whether the test should be one- or two-sided

One-Sided Test Number of Patients Needed in Each Group to Detect Various Differences in Cure Rates;  =.05; Power (1-  ) =.80 (One- Sided Test) Lower of the Two Cure Rates Differences in Cure Rates Between the Two Treatment Groups , , , , , Modified from Gehan E.: Clinical trials in cancer research. Environ Health Perspect 32:31, 1979.

Two-Sided Test Number of Patients Needed in Each Group to Detect Various Differences in Cure Rates;  =.05; Power (1-  ) =.80 (Two- Sided Test) Lower of the Two Cure Rates Differences in Cure Rates Between the Two Treatment Groups , , , , , , , Modified from Gehan E.: Clinical trials in cancer research. Environ Health Perspect 32:31, 1979.