Foundations of Scholarship

Question 1

Positive Likelihood Ratio

Answer 1

true positives/false positives

Question 2

Negative Likelihood Ratio

Answer 2

False negative/true negatives;

Question 3

SPIN and SNOUT

Answer 3

a SPecific test is good at ruling IN disease; a SeNsitive test is good at ruling OUT disease; If you get a negative test you can be sure that the person doesn’t have the disease.

Question 4

SPIN and SNOUT

Answer 4

a SPecific test is good at ruling IN disease; a SeNsitive test is good at ruling OUT disease; If you get a negative test you can be sure that the person doesn’t have the disease.

Question 5

Positive Predictive Value

Answer 5

Given you have a positive test, how likely it is that you have the disease. Depends on the pretest probability - the likelihood of having the disease to begin with.

Question 6

Negative Predictive Value

Answer 6

Given you have a negative test, how likely it is that you don’t have the disease.

Question 7

Post-test probability

Answer 7

What is the probability I have the disease given I have tested positive?

Question 8

2x2 table for a diagnostic test

Answer 8

Question 9

Calculation for Sensitivity

Answer 9

Question 10

Calculation for Specificity

Answer 10

Probability of getting a negative test if you don’t have the disease. High specificity = low false positive rate.

Question 11

Likelihood ratios

Answer 11

Likelihood ratios “indicate the extent to which a given diagnostic test result will increase or decrease the pretest probability of the target disorder.”

Can be used to change from pre-test to post-test probability using a nomogram.

Question 12

Purpose of Fagan’s Nomogram

Answer 12

To determine posttest probability with known pretest probability and a likelihood ratio.

Question 13

What is the difference between blinding and allocation concealment?

Answer 13

Not all trielas can be blinded, but in all cases allocation (of the next patient) can be concealed. Allocation concealment refers to the knowledge of the treatment to which the next patient is to be randomized.

Question 14

Why is it the intention-to-treat analysis preferable?

Answer 14

1. We randomize to create balance. As soon as we begin to remove randomized pts from the analysis, we counter that manipulation and open the trial to bias
2. The p-value on the outcome is based on the assumption that patients are randomly assigned to treatment, and failure to include all pts as randomized violates that assumption. Note that we do calculate p-values in non-randomized studies but that is based on further unverifiable assumptions.
3. ITT mimics the real world. For instance, we don’t expect every pt to be 100% compliant and some patients will be 0% compliant.

Question 15

Intention-to-treat (ITT) analysis

Answer 15

We use ITT to prevent bias from occurring in our analysis

more…sarah emailed me notes

Question 16

Downsides to intention-to-treat

Answer 16

Downsides:

• non-compliance in either group throws off the study. When we combine them with the people who were compliant, we water down the actual efficacy of the treatment.
  *

Question 17

Can you ignore missing data?

Answer 17

No. This violates the intent to treat approach. Missing data is not random, certain people are more likely to be lost to follow up than others.

Question 18

Heterogeneity or non-heterogeneity

Answer 18

Heterogeneity refers to variation in methods included in the meta-analysis (or the way the included studies quantified the outcome?). If there is heterogeneity present, you need to use a randomized effects model for analysis. You can use da fixed effect model if there is no heterogeneity and this is preferred if you can use it because the confidence intervals are tighter.

clinical vs statistical heterogeneity?

heterogeneity influences the decision to perform a meta-analysis or not

Question 19

What does the vertical line on a forest plot mean?

Answer 19

It means that RR = 1 and if the confidence intervals cross that line it indicates that the treatment may not yeild any difference in outcome. If your results are to the left, it favours treatment over no treatment.

Question 20

I^2 statistics

Answer 20

Calculates the proportion of variation due to heterogeneity.

0% = no heterogeneity

90% = lots of heterogeneity, so the results of the analysis are almost meaningless

Question 21

Define Systematic Review

Answer 21

“The application of strategies that limit bias in assembly, critical appraisal, and synthesis of all relevant studies on a specific topic. Meta-analysis may be, but is not necessarily, used as part of this process.”

Question 22

Fixed effects models vs random effects models

Answer 22

Random effects models takes into account all of the random things that could have interfered. If you have a very strong significance with a random effects model, it is more likely to be even more significant with a fixed effects model because it maintains that significance even while accounting for random interference.

Question 23

How to assess for heterogeneity

Answer 23

1. Visually
2. Q statistic - has a p value that indicates that heterogeneity is present if it is significant.
3. I^2 statistic tells us the percent heterogeneity that exists

the issue with Q statistics is that they are often underpowered so they can suggest that there is no heterogeneity (insignificant p value) in cases where they are just underpowered and heterogeneity is in fact present.

Question 24

PEO

Answer 24

Mostly for qualitative research designs.

Population

Exposure

Outcome

Question 25

The major difference between descriptive and analytic studies

Answer 25

Analytic studies test a hypothesis.

Question 26

What do the results of an odds ratio tell you?

Answer 26

OR =1 : exposure does not affect odds of outcome
OR >1 : exposure associated with higher odds of outcome
OR <1 : exposure associated with lower odds of outcome

Question 27

What are sequential mixed methods designs

Answer 27

Quantitative → qualitative → interpretation : explanatory

Qualitative → quantitative → interpretation : exploratory

Before intervention (qualitative) → intervention (quantitative) → after intervention (qualitative) → interpretation: sequential embedded design

Question 28

Cross-sectional study design

Answer 28

A snapshot of people with and without the disease and exposure.

Con: Assesses the possible cause and effect at the same time so not actually able to determine which came first - therefore unable to determine cause and effect relationships.

Pro: quick, cheap, easy, and can look at multiple variables

Question 29

Case-control studies

Answer 29

Choose participants based on outcome status and then look back to see what exposures they had by asking directly or examining records. A retrospective design

Pro: food for rare diseases. Good for when exposure may be unethical to allocate in an RCT. Good for outbreak investigations.

Cons: Subject to recall bias. Confounding due to lack of similarity bt cases and controls (selection bias)

Question 30

Cohort studies

Answer 30

Participants are selected based on their exposure status. Can be prospective, when the cohort is followed forward in time, or retrospective, where the cohort is selected from a database at a time in the past and then re-checked in the database at present time to see if they got the outcome.

Pros: Can assess cause and effect. Can study multiple outcomes of a single cause. Better evidence than a cross-sectional or case-controls tudy.

Cons: Prospective is slow and expensive. Not good for rare outomes. Prospective is subject to loss-to-follow-up.

Question 31

Relative Risk

Answer 31

Used to measure association of categorical variables. Requires data on incidence, so can’t be performed for case-control studies.

Equation:

RR = [P(outcome|exposure)]/[P(outcome|no exosure)

RR = [a/(a+b)]/[c/(c+d)]

If RR = s, then the people with the exposure were 3x more at risk of experiencing the outcome compared to those without the exposure.

Question 32

Odds Ratio

Answer 32

Basically the shitty version of RR that we do when we don’t have incidence data (i.e., we did a case-control study). It’s a good estimate of RR when the disease is rare.

OR = (a/b)/(c/d)

Question 33

Attributable risk

Answer 33

Measure of the difference in disease risk between exposed and unexposed groups.

AR = (a/(a+b)) - (c/(c+d))