Stats Flashcards

1
Q

Measures of central tendency?

A

Mean
Median
Mode

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Mean?

A

Average value

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

Median?

A

Middle value

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Mode?

A

Frequent value

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Best measure when distribution not skewed?

A

Mean

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

Best measure when distribution skewed?

A

Median

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Standard deviation?

A

Square root of variance

NOT influenced by sample size

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Empirical rule of standard deviation?

A

Need a normal distribution (not skewed).

1 SD = 68% of data (34% on either side of mean)
2 SD = 95% of data
3 SD = 99.7% of data

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Level 1 evidence?

A

Meta-analysis with small CI

At least TWO RCTs with a large sample size

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

RCTs can be in which level of evidence?

A

1, 2, 3 depending on sample size and how many done (need at least 2 for level 1)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

Meta-analysis can be in which level of evidence?

A

1,2 depending on the CI

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Nul hypothesis

A

What we are trying to REJECT

Need p < 0.05 to reject (chance occurrence less than 5%)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

Type 1 error (ALPHA)

A

Nul hypothesis rejected but was TRUE

Error of INTERNAL VALIDITY

Probability of it happening = p value

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

Type 2 error (BETA)

A

Nul hypothesis accepted but was FALSE

Lack of POWER

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

Internal validity

A

Are the results representing what we wanted to measure?

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Reliability

A

Are the results consistent and reproducible?

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
17
Q

Student “t” test

A

Compares means of TWO samples made up of CONTINUOUS VARIABLES

Small samples with N < 30

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
18
Q

ANOVA (“f” test)

A

Compares means of MORE THAN TWO samples made up of continuous variables

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
19
Q

One-tailed test

A

Reject nul hypothesis in ONE direction (active treatment is better than placebo)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
20
Q

Two-tailed test

A

Reject nul hypothesis in TWO directions (active treatment is different than placebo, either better or worst)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
21
Q

“Z” test

A

SAME AS “T” TEST BUT FOR LARGER SAMPLES N > 30

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
22
Q

Chi-square

A

Evaluates ASSOCIATIONS between 2 samples of CATEGORICAL VARIABLES
(percentages, proportions)
Can compare 2 proportions
Can make a table of frequencies

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
23
Q

Pearson test

A

Test of linear correlation between CONTINUOUS VARIABLES
-1 = perfect indirect association
0 = no association
1 = perfect direct association

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
24
Q

Linear regression

A

PREDICTION of results once a correlation is demonstrated between CONTINUOUS VARIABLES

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
25
Multiple logistic regression
PREDICTION of results once a correlation is demonstrated between CATEGORICAL VARIABLES
26
Incidence
Number of new cases / number of people at risk over a certain period of time (REMOVE KNOWN CASES)
27
Prevalence
Number of NEW AND OLD cases / entire population over a certain period of time
28
Establishing causality
- High degree of correlation - Consistency of the correlation - Temporal association - Coherence with contemporary scientific knowledge - Dose-response relation - Reversibility - Biological plausibility - Specificity - Elimination of other explanations
29
Relationship between prevalence and illness duration?
Prevalence is proportional to incidence x illness duration. SO, if illness duration increases (ex: new treatment comes out which prolongs life expectancy), prevalence will ALSO INCREASE.
30
Power
1 - Beta (error) Probability that a difference will be picked up if that difference really exists. Most important factor = SAMPLE SIZE
31
Cohort study
Follow 2 groups (exposed/non-exposed) PROSPECTIVELY to see if they develop an illness. CAUSE to EFFECT Useful when EXPOSURE is rare Longer and more expensive than case-control studies.
32
Case-control
Looking back at 2 groups (ill / not ill) and determine their level of exposure. RETROSPECTIVE EFFECT to CAUSE Useful when ILLNESS is rare
33
Attrition bias
Loss of certain patients for analysis | Fix with LAST OBSERVATION CARRIED FORWARD
34
Association in studies can be due to what?
Chance Bias Reverse causality Confounding
35
Qualities of a good screening test?
``` Inexpensive Easy to administer Little discomfort Reliable Valid Comparable to gold-standard ```
36
PPV and NPV decedent on what?
PREVALENCE
37
Sensitivity
If I HAVE THE DISEASE, will the test pick it up? True positive / all people with disease
38
Specificity
If I DON'T HAVE THE DISEASE, will the test not pick it up? True negative / all people without disease
39
Sensitivity good for?
RULING OUT | Negative result most helpful
40
Specificity good for?
RULING IN | Positive result most helpful
41
Positive predictive value
If I have a POSITIVE TEST, what are chances I have disease? True positive / all positive test
42
Negative predictive value
If I have a NEGATIVE TEST, what are chances I don't have disease? True negative / all negative test
43
High sensitivity means?
Low false negatives.
44
High specificity means?
Low false positives.
45
Ratio vs. proportion vs. odd vs. rate
Ratio is dividing one number by another: 1. Proportion is part/whole so like M : population or F : population 2. Odd is part/non-part like M : F (part and non-part of whole population) Rate is ratio with time as an intrinsic part of the denominator
46
Odds
Part / non-part Whole = 1 Non-part = 1-part Odds = part / (1-part) ***Odds = proportion / 1-proportion***
47
Odds if probability of death is 20%?
``` Odds = 0.2 / (1-0.2) Odds = 0.2/0.8 Odds = 0.25 ``` 25% aka for every person that dies, there are 4 people who live
48
Odds if probability of horse winning is 75%?
``` Odds = 0.75 / (1-0.75) Odds = 0.75 / 0.25 Odds = 3 ``` 300% 3 to 1 odds of your horse winning (3 x 100%)
49
Disease odds ratio
Odd disease among exposed / odd disease among unexposed
50
Exposure odds ratio
Odd exposure among disease / odd exposure among non disease
51
Odds ratio
AD/BC
52
Odds ratio relevant for which type of study?
Case-control (RETROSPECTIVE)
53
Odds ratio interpretation?
OR > 1 means greater odds of association (1.2 = 20% increase in odds of an outcome with a given exposure) OR = 1 means no association between outcome and exposure OR < 1 means lower odds of association (0.2 = 80% decrease in the odds of an outcome with a given exposure)
54
Risk ratio
Proportion so now we look at wholes! Risk of disease among exposed / risk of disease among unexposed Risk of disease among exposed = A / (A + C) Risk of disease among unexposed = B / ( B + D)
55
Risk ratio interpretation
RR > 1 means exposure is associated with a higher risk RR = 1 means no association between exposure and disease RR < 1 means exposure associated with lower risk (may be causally protective)
56
Risk ratios relevant for which type of study?
``` RCTs COHORT studies (can be retrospective or prospective) ```
57
Absolute risk reduction?
Risk of disease among exposure - risk of disease among non-exposure Always positive because it is an absolute value
58
NNT/NNH
1 / ARR (absolute risk reduction)
59
Confidence interval
Mean +/- (1.96 x standard error of the mean) Standard error of the mean = SD / (square root of sample size) 1.96 = t-statistic for 95% CI (most commonly used) If CI crosses 1 = not statistically significant
60
Effect sizes when comparing standardized practices?
Small 0.2 Medium 0.5 Large 0.8
61
Effect sizes when comparing correlation coefficients?
Small 0.1 Medium 0.25 Large 0.4
62
Effect sizes for relative risk?
Small 1.5 Medium 2.5 Large 4.3
63
Pareto principle?
80% of effects are the product of 20% of the causes Pareto diagram highlights the importance of different causes to a phenomenon but does NOT allow for evaluation of the quality of practice of doctors with regards to application of guidelines/recommendations!