CLA Honors Seminar 3010: Clinical Trials Spring 2007 May 14, 2009 Page 1 of 8 HSEM 3708 - Spring 2009 - Final Exam Name:_________________________________ =============================================================================== 1. Describe and explain what is meant by 'regression towards the mean'. Give an example, different from any that were discussed in class, of where you would expect regression towards the mean to occur. Regression towards the mean occurs when there are two independent sources of variation, between individuals and within individuals, for a random variable, and a sample is taken which includes only individuals whose measured values are in an extreme tail [10] of the distribution. A repeat measurement of these same individuals will tend to have values which are closer to the overall mean value. People who have a high first measurement may be having an unusually "high day", and their true values may be somewhat lower. Their second measurement will tend to be closer to their true mean value. Example: A soybean farmer selects beans from his 2008 crop which are unusually large, and uses these beans as seed for his 2009 crop. He is disappointed to find that, although the 2009 beans tend to be larger than average, they are generally not as large as the beans he selected from his 2008 crop. May 8, 2007 Page 2 of 8 HSEM 3010 - Spring 2007 - Final Exam Name:_________________________________ =============================================================================== 2. List 8 basic principles or ideas in clinical trials with a brief discussion of each. 1. Randomization: randomize treatment assignments to achieve probabilistic balance on factors other than the treatments. 2. Intention-to-treat analysis: an analysis of trial results which includes all people who were randomized, regardless of whether they actually got the treatment. The idea is to give a realistic estimate of the true treatment effect. 3. Survival analysis: A method for analyzing outcomes which are defined as the time to first occurrence of a well-defined event. Survival curves are an essential way to display the results of survival analysis. 4. Double-blind: Masking of the treatment assignment to both patients and doctors in a clinical trial, to minimize the possible biasing effects of expectations if the the treatment assignment were known. 5. Sequential monitoring: A way to provide guidelines for stopping a trial early if there is statistically compelling evidence of a treatment effect, or if there is a very small chance that any significant treatment effect will be observed if the trial goes to completion. 6. Power: Power is the probability that a trial will demonstrate a significant difference between the two groups, assuming a specified magnitude of a treatment effect. Power increases with increasing sample size. 7. Multiple Comparisons: When more than two groups are being compared statistically, the chance that one or more of the comparisons will be statistically significant at a specified significance level will be greater than the chance that an individual comparison will be statistically significant at that same significance level, assuming the null hypothesis is true. This effect is compensated for by using a Bonferroni adjustment of the significance level. 8. Data and Safety Monitoring Committee: A group of experts who are independent of the investigators in a clinical trial, which meets at regular intervals to review data on adverse effects [safety] and effectiveness of treatments and other aspects of the clinical trials operations. The main purposes of the DSMC are to (1) protect patients from harmful treatments, and (2) recommend termination of a trial if the outcome becomes almost certain. May 8, 2007 Page 3 of 8 HSEM 3010 - Spring 2007 - Final Exam Name:_________________________________ =============================================================================== 3. Discuss the relationship between drug companies, the U. S. Patent Office, and the FDA (Food and Drug Administration). How do these various groups affect the conduct of clinical trials? Drug companies invent new drugs which may be promising for treatment or prevention of illnesses. Such drugs can be protected by patents, which will prevent other companies from selling such drugs. Patents are allocated for a period of 17 years. After the patent expires other companies can make and sell the drugs. While the patent is in force, the drug company has a monopoly on the drug and can charge high prices for it and make lots of money because there is no competition. Drug companies must carry out clinical trials to prove that the drugs are SAFE AND EFFECITVE. Such clinical trials must be reported to the FDA. If the drug company is successful in showing safety and effectiveness of the drug, the FDA will grant approval to begin marketing the drug for the conditions it is intended to treat or prevent. The FDA should NOT approve marketing of drugs which are UNSAFE OR INEFFECTIVE. The expense of conducting the clinical trials is borne by the drug companies. They must describe their trial designs and analysis [12] plans in advance to the FDA, and they must adhere to these designs and plans in their final analyses. May 8, 2007 Page 4 of 8 HSEM 3010 - Spring 2007 - Final Exam Name:_________________________________ =============================================================================== 4. Sometimes the results of a clinical trial seem quite definitive, but after the trial is finished, there is not much of a change in clinical practice or the use of medications. Give an example from the studies discussed in class. Speculate on the reasons that this may happen in some cases. Example: Beta-carotene trials. These trials indicated that use of beta carotene is associated with higher incidence of lung cancer in smokers. Yet beta carotene is still sold as a health-food supplement. Why this might happen: the negative effects may be relatively infrequent. Beta carotene, e.g., is a 'natural substance', found in foods which are believed to be healthful, so regulating it would be problematic. More generally, doctors may like a certain drug because it suppresses symptoms, even though its effect on survival or heart attacks is harmful. Doctors do not see enough bad-outcome cases to become convinced that there is a problem. Patients may like the drug because it makes them feel better, even though they know that it increases their risk of a bad outcome. The results of the clinical trial may not become widely enough known. [10] May 8, 2007 Page 5 of 8 HSEM 3010 - Spring 2007 - Final Exam Name:_________________________________ =============================================================================== 5. A clinical trial is conducted in which two drugs, A and B, are being compared. Participants in the clinical trial are people who have migraine headaches. Participants are randomly assigned to take either drug A or drug B for 8 weeks. The outcome for the clinical trial is determined by the participant's answer to the question "In the past 8 weeks, have you had fewer migraine headaches than you usually have?" The choices of answers are yes and no. a) What statistical approach would you take to analyzing the outcome data for this clinical trial ? Chi-square or Fisher Exact Test for a 2 x 2 table. [5] b) Some people quit taking their assigned drug during the trial. This is more likely to happen if they are assigned to drug B than if they are assigned to drug A. How will this affect the analysis of the data? It depends on why they quit taking the drug. If it is for random reasons, and they still show up for the final evaluation, then there may not be much effect on the intent-to-treat analysis. If they stop using the drug because of side effects, this too will emulate what will [5] happen in real life and the intent-to-treat analysis is valid. If they stop taking the drug because they think it is a placebo, then that would not be likely to happen in 'real life' and the likely result is that the treatment effect will be decreased. c) Some people do not come back for the 8-week visit at the end of the trial, so they do not answer the question described above. How will this affect the analysis of the trial? This can have a biasing effect on the results. The people who do not come back cannot be evaluated, and their failure to come back may be related to whether the drug worked or not. A high rate of losses to follow-up can compromise the findings of a clinical trial and damage the plausibility of its findings. Even in the absence of a biasing effect, the sample size will be decreased and this will [5] decrease power. May 8, 2007 Page 6 of 8 HSEM 3010 - Spring 2007 - Final Exam Name:_________________________________ =============================================================================== 6. Describe your understanding of how and why 'monitoring boundaries' are used carrying out clinical trials. Monitoring boundaries are limits on Z-statistics, usually different for different time points in the course of the clinical trial. The purpose is to provide a way of stopping the trial early if there are very strong treatment effects, without affecting much the overall probability of rejecting the null hypothesis if the null hypothesis is actually true. In most cases monitoring boundaries are difficult to exceed early in a clinical trial, but become easier to exceed towards the end. [12] May 8, 2007 Page 7 of 8 HSEM 3010 - Spring 2007 - Final Exam Name:_________________________________ =============================================================================== 7. Suppose you were a reviewer at NIH of grant proposals for clinical trials. List 8 questions that you would want answers to to decide whether or not a given proposal would get funded - brief description (< 15 words) for each. 1. Do you have a clinically significant question which has not been answered already? 2. What evidence do you have that the proposed new treatment will be effective and safe? 3. What is your primary outcome? 4. What is your estimate of the treatment effect under the alternative hypothesis? 5. How large will your sample size have to be? Can you recruit this [12] number of patients? 6. What is your design for the clinical trial? 7. What is the expected duration of the trial? 8. What is your planned statistical analysis? 8'. What are the expected side effects, and what will you do if an excessive number of them occurs? 8''. How will you handle missing data? 8'''. Will there be masking of patients or doctors to treatment assignment or masking of evaluators? May 8, 2007 Page 8 of 8 HSEM 3010 - Spring 2007 - Final Exam Name:_________________________________ =============================================================================== 8. Suppose you were on the Data and Safety Monitoring Board for a clinical trial. There are two treatment groups. Halfway through the trial, the data indicate that there are twice as many deaths in one group as in the other. The official outcome variable for the trial was not death, but rather improvement in pain score. The p-value for comparing the rates of death in the two groups is p = 0.0003. a) Do you think the participants in the trial should be informed of this? Why or why not? Two answers: 1. Yes. Patients entering the trial were assured that there was not [7] evidence favoring one treatment versus the other and that there was equipoise regarding the treatments being tested. This is no longer true. There is fairly strong evidence favoring one of the treatments, there is no longer equipoise, and patients have a right to know it. 2. No. Assuming that a sequential monitoring boundary has not yet been reached, the investigators do not yet have the desired level of proof that one of the treatments is superior. Telling the patients will completely disrupt the trial and the truth may never be known. The trial will end inconclusively. b) Can you think of circumstances where you might recommend continuing the trial in spite of the data on rates of death? Yes. It may be that more deaths early in the trial in one group would be expected, [7] and that this would be outweighed by fewer deaths in that same group later on. Stopping the trial early might prevent investigators from knowing what the long-term effects were and might prevent adoption of a drug which had an overall beneficial effect. Such a pattern of events should have been anticipated before the trial began, and analysis of trial data adjusted as a result.