CHAPTER: Stats//Behavioral

Question 1

Type of study that measures odds ratio - whats the equation

Answer 1

Case control 
Retrospective (why can't calc RR) 
Exp grp = cases = diseased
Control = disease FREE
OR = ad / cb

Question 2

Type of study that measures relative risk - equation

Answer 2

Cohort study
Prospective or historical
RR = (a/a+b) / (c/c+d)
Evaluation of the people who got the disease in some way (what put these people at higher risk?!)

Question 3

Name phase 1-4 of a clinical trial

Answer 3

1 = is it safe = healthy vol
2 = does it work = small grp pts w/ disease
3 = is it as good or better = randomly assigned to treatment v placebo 
4 = cant it stay = post-marketing surveillance for long term adverse SE

Question 4

What is an ecological survey?

Answer 4

Study of POPULATIONS

How frequently does x pair with y outcome in pop data

Question 5

What is a cross over study

Answer 5

Rd 1 = a placebo and trt grp
Washout
Rd 2 = grps get the other treatment
Pro = pts serve as own controls

Question 6

In a 2x2 table, what is the x vs y axis

Answer 6

X = disease +/-
Y = test, RF or intervention +/-

Question 7

Equation for incidence + prevalence - which value is affected by disease recurrance

Answer 7

Incidence = # new cases/#people @ risk
Prev = #existing cases / total # ppl
Only prevalence is affected by recurrence (add some back into existing case pool)
Similarly, cures + death only remove from existing cases
If a drug ↑survival but does not cure: ↑prev, no change incidence

Question 8

What is AR + equation

Answer 8

Proportion of disease occurrences attributable to the exposure
AR = (a/a+b) - (c/c+d)
AR is subtraction, RR is division

Question 9

Equation relative risk reduction

Answer 9

Proportion of risk reduction due to intervention

RRR = 1-RR

Question 10

Absolute risk reduction equation

Answer 10

Difference in risk due to intervention
ARR = event rate (control) - ER (trt)
ARR = (c/c+d) - (a/a+b)
ER is one of those two variables, don’t separate a+b values
Switch the AR equation

Question 11

Define + equation NNT

Answer 11

NNT for 1 pt to BENEFIT

NNT = 1/ARR

Question 12

Define + equation for NNH

Answer 12

# pts need to be exposed to a RF for 1 pt to be harmed
NNH = 1/AR

Question 13

Another word for precision is…

How does precision change with random error, SD

Answer 13

Precision = reliability (how close 3 hits are)
↑Random error ↓precision
↑precision - ↓SD (duh)

Question 14

Another word for accuracy is… what type or error changes accuracy?

Answer 14

Accuracy = validity (are you on the target)
Does this test measure what it should
Systemic error ↓accuracy

Question 15

What is procedure bias? How reduce it

Answer 15

Different grps not treated the same
Ex: trt grp spent more time in a specialized hospital
BLIND + PLACEBO

Question 16

What is the Pygmalion and Hawthorne effects? How reduce

Answer 16

Pygmalion = researcher’s belief changes outcome (self fulfilling prophecy)
Hawthorne = subject behavior changes as they know they are being studied (ex: doctors in a quality control study)
Blinding + placebo

Question 17

Best way to control confounding

Answer 17

Matching

Question 18

What type of bias should you keep in mind for screening tests

Answer 18

Lead time bias - did this screening test actually extend someone’s life or just detect the disease earlier

Question 19

What is the mean, med, mode in normal dist vs positive skew vs negative skew

Answer 19

Normal dist: mean = med = mode
+ skew = hump to L, long tail to the R
mOde E > O

Question 20

What is a type 1 error? Alpha?

Answer 20

T1 error = false positive
You said there was an effect but none exists
Alpha = probability making T1 error

Question 21

Why do you compare alpha to the p value?

Answer 21

If p

Question 22

What is a type 2 error? Beta?

Answer 22

False neg
You said my data is cramp, but it really isn’t! There is an effect
B = probability making T2 error

Question 23

What is power? 3 ways to ↑power?

Answer 23

Power = 1-B 
Ability to detect a difference when one really exists (you're subtracting the likelihood that you don't ID a relationship)
↑power if you ↓B
1. ↑SS (pwoer in #s)
2. ↑expected effect size
3. ↑precision of your tests

Question 24

If CI = mean +/- Z(stand error of mean), what is Z for a 95 vs 99% CI?

Answer 24

95% CI, Z = 1.96

99% CI, Z = 2.58

Question 25

If you are given a MEAN and the 95% CI is (-1, 1), is the value stat sig?

Answer 25

NO
Mean 0 has to be out of the CI
You’re average can’t be 0!

Question 26

If you are given OR or RR and the 95% CI is (0-2), is the value stat sig?

Answer 26

NO

You’re base like risk or odds are 1 - can’t be in the CI

Question 27

If you are given 2 groups, and the CI intervals overlap, is there a sig difference between them?

Answer 27

No - cannot overlap

Question 28

What equation tells you the P of at least 1 + sample test result?

Answer 28

1 - P (all events being the same)

1 - P (all 8 tests are neg)^8

Question 29

Stat test for means of 2 grps

Answer 29

T or z test

Question 30

Stat test for means of 3+ grps

Answer 30

ANalysis Of VAriance

Question 31

Stat test for 2+ % or proportions of categorical outcomes

Answer 31

Chi^2

Chi-tegorical

Question 32

Stat test to det linear relationship between 1 dep variable and 2+ indep variables

Answer 32

Multiple linear regression

Question 33

What is the name of the correlation coefficient

Answer 33

r
Varies between -1 -> 0 -> 1
Positive vs negative tells you direction of the line
Closeness of data to that lines tells you 1 vs 0

Question 34

MediCARE vs medicAID

Answer 34

Care for old, disabled, ESRD

Aid the poor

Question 35

Name medicare pts A-D

Answer 35

A = hospital insurance
B = basic medical billing = doc fees, dx tests
C = A+B, delivered by approved private companies 
D = prescription drugs

Question 36

What type of analysis id’s the cause of medical error retrospectively vs prospectively

Answer 36

Root cause anal = look back at ALL problems that lead to an error (fishbone)
Failure mode + effects analysis = inductive reasoning to anticipate how a process might fail