1.) Type 1 Error: Data indicate that 2 treatments are different, but really are the same. (ex: 10ppl got the vaccine and 3 get the flu (70% protected) and 10 ppl had the placebo and 7 got the flu (30% protected) By chance the vaccine group did not get the flu, and therefore it was not necessarily the vaccine that was better than the placebo.)
-Probability of a type 1 error occuring: =Alpha or p-value (less than or equal to 0.05: generally accept results by interpreting that the treatments really do differ.)
-If data indicate that two treatments are different, how many times would you have to rerun the experiment (using the same protocol, but with new samples.) before the date indicate that the treatments are the same?
ex: Claim: Treatment B works better than A. (p= 0.05) Would have to run this experiment 20 more times to see that A=B, which would be a 5% chance.
2.) Type 2 Error: Data indicate that two treatments are equally effective, but really, they differ. The probability of a type 2 error = 1- Beta. The standard of a solid study is greater than or equal to 80%.
ex: Do Zinc Lozenges Cure Colds?
-Previously perceived placebo rate or cure rate is 15% (cold is better after 48 hours) You want to know how many people to put in the zinc lozenges group and the placebo group to help minimize error. The cold-eeze says that it is 35% more effective than the placebo. On page 151, Table 8-4 shows how to figure this out. The lower of the two cure rates would be 0.15. The answer would be 31 subjects for each treatment group.
-Is the sample size appropriate for the claim that they're trying to make?
3.) 2-Sided Test: Looks at the possibility that treatment A is better than treatment B and the possibility that treatment B works better than treatment A. For this test you need more samples and it is more difficult to show that one is better than the other. This test is used if they want the majority or people to get better.
4.) 1-Sided Test: Possibility that treatment A works better than B. It is easier to show the relationship and statistical significance in this type of a study and you do not have to have as many subjects.
5.) Publication Bias: Bias of publishers to print studies that support the alternative hypothesis rather than the null hypothesis. A null hypothesis shows that two things are equal, which they do not publish, because they are boring and do not show a difference. However, they do show the alternative ones, because they show a difference.
Chapter 8: Case Studies p. 153
The Hypertension Detection and Follow-up Program
-Do blood pressure medicines work effectively in reducing slight increase in blood pressure (diastolic: 90-104)
22,994 subjects -> 10,940 randomized into stepped care or referred care
- Mortality in both groups over a 5 year period were studied.
-The adverse effects of the antihypertensives may be of greater risk then the slight increase in BP
Study of Breast Cancer Prevention Using Tamoxifen
-observed that women using tamoxifen for breast cancer had lower incidence of cancer in the other breast.
13,388 women 35 years of age.
-Endometrial cancer were increased in Tamoxifen group
-smaller studies done in Europe did not show same evidence
*possible reasons for difference = environment, diet, screening tests
