6 – Clinical Trials II

Question 1

If did a same trial again, would you get EXACTLY the same results?

Answer 1

• NO
• Trails give us an ESTIMATE of treatment difference in the population
  o Larger sample size=better estimate (capture more of the variability)
• *all clinical trials are like this
• *need a way to determine if we can be certain about the estimate

Question 2

Null hypothesis (Ho)

Answer 2

• Assume there is NO difference between treatments
• Trying to reject
• If reject=there is a significant difference
• *always a role of chance

Question 3

Can’t prove a null hypothesis

Answer 3

• Can only NOT reject it

Question 4

Alternative hypothesis

Answer 4

• When we reject the null hypothesis
• There is a DIFFERNENCE between treatments
• *don’t presume or state the direction of difference
• Can’t be tested directly
• TWO-TAILED TEST of they hypothesis

Question 5

2-tailed vs. 1 t-tailed tests

Answer 5

• 2-tailed: do NO know which direcetion the outcome will go if we reject the null hypothesis
• Be suspicious of 1-tailed tests (ex. can use less subjects to get a statistically significant result)

Question 6

Hypothesis testing

Answer 6

• Conduct a test of statistical significance and quantify the degree to which sampling variability may account for the results observed in particular study

Question 7

P-value

Answer 7

• Probability statement
• Describes the CHANCE of getting the observed effect in the outcomes, IF the null hypothesis is TRUE

Question 8

Very small p-value

Answer 8

• Reject null hypothesis
• Unlikely we could have obtained the results if the null hypothesis was true

Question 9

Very large p-value

Answer 9

• Do NOT reject null hypothesis
• Higher probability that we could obtain the observed results if the null hypothesis were true

Question 10

How small of a P-value do we need?

Answer 10

• Decide it before the trail is started
• 5% is the standard level of ‘statistical significance’
• *confidence intervals give more meaningful info

Question 11

‘statistically significance’ and ‘statistically non-significant’ are not necessarily contradictory

Answer 11

• Observed effect is the same
• But one has a low p while the other does not
• *increased confidence interval

Question 12

Statistical significance

Answer 12

• Defines likelihood of achieving this treatment difference by chance alone
• Small trials: very large differences my not be statistically significant
• NECESSARY precondition for consideration of clinical importance
• Indicates NOTHING about the actual magnitude of the effect

Question 13

2 potential errors that can occur

Answer 13

• Type 1 error: false claim
• Type 2 error: missed opportunity

Question 14

Type 1 errors (alpha error)

Answer 14

• FALSE CLAIM
• When find a treatment difference, but there is NO difference
• Need to decided how much type 1 error you are willing to accept (ex. P<0.05)
• Usually set at 5% or 1% (p-value)

Question 15

If alpha at 5%

Answer 15

• 95% certain that a treatment difference is NOT a random result

Question 16

Type 2 errors (betta error)

Answer 16

• More common
• MISSED OPPORTUNITY
• Didn’t show a difference, but there is one

Question 17

Larger the trial and type 2 errors

Answer 17

• less likely to have a type 2 error
• if wont to lower your ‘level’ = NEED MORE SAMPLES

Question 18

Type 2 error ‘level’

Answer 18

• usually set at 20%
• can accept that 20% of the time, I will not be able to detect a treatment difference even though there is one

Question 19

Power and type 2 errors

Answer 19

• power=1-beta (type 2 error)
• if power close to 1=good at detecting a false null hypothesis
  o 80% chance of rejecting a null hypothesis