Hypothesis Testing - EBM

Question 1

What is hypothesis testing?

Answer 1

estimates the probability that observed events occurred by chance and a null hypothesis is true

Question 2

what is the null hypothesis?

Answer 2

true difference between two groups is zero

Question 3

rarely have balanced results even if the odds are equal for both groups because of chance

Answer 3

…

Question 4

P value tells us..

Answer 4

how likely is that the observed difference is due to chance alone

conventional boundary (P< 0.05)

Reject the null if its less than that

generally the 5% refers to both tails of the distribution of possible results

Question 5

1/1000 on both sides = 0.02

8:2 on both sides = 0.05

Answer 5

..

Question 6

Why is the measures of effect version important?

Answer 6

There are inherent weaknesses in hypothesis testing… there are other ways to express the results that give us more info from the data

Question 7

Type I error? alpha

Answer 7

False positive

Question 8

Type II error? beta

Answer 8

false negative

Question 9

When we set our threshold P at .05, the likelihood of a type I error when the null hypothesis is true is 5%. We refer to a false negative (treatment truly effective, P > .05, cell c) as a type II or error.

Answer 9

…

Question 10

increasing sample size reduces chance of type II error (greater power)

Answer 10

…

Question 11

positive predictive value (PPV) eqn

Answer 11

A/A+B

Question 12

negative predictive value (NPV) eqn

Answer 12

d/c+d

Question 13

relative risk eqn?

Answer 13

a/(a+b)
///////
c/(c+d)

Question 14

relative risk reduction?

Answer 14

NPV-PPV
//////
NPV

Question 15

risk difference?

Answer 15

NPV-PPV

Question 16

NNT?

Answer 16

100/risk difference

Question 17

odds ratio?

Answer 17

a/b
////
c/d

=
ad/cb

Question 18

what are the challenges to hypothesis testing?

Answer 18

non-inferiority trials
dichotomous vs continous variables
multiple hypotheses

Question 19

what are non inferiority trials?

Answer 19

less expensive, easier to administer, less toxic

sample size is important to reduce risk of a type II error

don’t want to pay for a giant trial but just want to prove a variation of a drug is safe, etc

Question 20

what are dichotomous outcomes?

Answer 20

yes/no
heads to tails
dead to live

Question 21

continous outcomes?

Answer 21

spirometry
cardiac output
exercise capacity

Question 22

what test is appropriate for continous outcomes?

Answer 22

t-test

Question 23

what is the problem with multiple hypotheses?

Answer 23

hypothesis testing breaks down with more than one hypothesis?

secondary results - the chance that these results are correct is less by mathematics

Question 24

confidence intervals become narrower when there are more people in the study

Answer 24

…

Question 25

Using confidence intervals avoids what?

Answer 25

the yes/no dichotomy of hypothesis testing

it also stops the need to argue whether the study should be considered pos or neg

Question 26

NNT can help show you that sometimes confidence intervals, even if narrow, don’t make a result significant enough for clinical tx

Answer 26

..