Statistics and Study Design

Question 1

Hypothesis

Answer 1

Null hypothesis (NH)- No difference between 2 groups
Alternative hypothesis (AH)- There is a difference
There is no difference, unless proven

Question 2

Evidenced Based Medicine

Answer 2

Any opinion can be evidence based
Evidence/facts are notobserved, but interpreted subjectively
Higher level = more likely accurate

Question 3

Replication

Answer 3

Even with high levels of evidence, we need replication

Higher levels do not guarantee certainty

Question 4

Levels of Evidence

Answer 4

I- randomized blind
II- open randomized
III- Large observational >50
A) prospective
B) retrospective
IV- Small observational 10-50
V- Case report/series(<10), expert opinion

Question 5

Observational Studies

Answer 5

Cohort
A) Get baseline info
B) Follow over time
Case control
A) Compare those with condition to those without
Can have confounds
Try to control this by:
A) Stratification=1 variable
B) Regression= multiple

Question 6

Deceptive Results

Answer 6

We look for causal connections
We cannot accept studies at theirface value
Correlation does not equal causation i.e. Caffeine and cancerlink

Question 7

Placebo Effect

Answer 7

Natural healing
Psychological effect
Related to supports of treatment
Can all effect be placebo related to se’s?

Question 8

Confounding

Answer 8

Unseen factorinfluences our results
Randomization prevents
Regression analysis to remove after
Prevention is better

Question 9

Effect Modification

Answer 9

OCPs and smoking
Neither alone has a high risk of blood clots
Combined they modify the effect of smoking

Question 10

Randomization

Answer 10

Equalizes potential confounding factors
More people= less chance of confounding
Flip a coin 2x vs 1000x
Roughly need 50 people
Not always successful-10% difference or more concerning

Question 11

Measurement Bias

Answer 11

If I know the treatment, it may bias my expectations of the results
Blinding prevents this
Partial blinding if strong effects/ side effects

Question 12

Chance

Answer 12

Random chance that a study gives incorrect results

We set 5% as the bar

Question 13

P Values

Answer 13

Chance of seeing results if null hypothesis is true
0.05 generally used “statistically significant”
95% confident was not by chance
Somewhat arbitrary

Question 14

Significance

Answer 14

Statistically means result
Statistical vs clinical
PHQ-9 change of 1/27

Question 15

Generalizability

Answer 15

Who would want to be in a study?
What are exclusion criteria?
Comorbid psych
Comorbid subs
SI
10% would participate

Question 16

FDA Approval Process

Answer 16

Safety assessed
Increasing numbers of efficacy/safety studies
2 positive studies
Typical 10 years
Post-market monitoring