Epi Bio Review Slides

Question 1

Prevalence

Answer 1

Total # diseased / # total population
- measures disease burden

Also
Prevalence=incidence * duration

Question 2

Incidence

Answer 2

new cases / population at risk during a given time period

Measures rate of appearance of new cases

Question 3

Cumulative Incidence

Answer 3

New cases in a specified time period / Total # at risk at the start of the time period

Question 4

Incidence Density

Answer 4

New cases in a specified time period / # Units of person-time

MOST PRECISE

Because accounts for variable follow-up, including lost-to-follow-up

Question 5

Observational vs. Intervention studies

Answer 5

Observational- investigator just observes and collects data

Interventional- investigator applies some new treatment or intervention and examines the result on a health outcome

^ both can be either descriptive or analytic

Question 6

Descriptive vs. Analytic Studies

Answer 6

Observational or interventional studies can be

Descriptive- measure and report data without addressing a hypothesis

Analytic- address specific hypothesis

Question 7

Observational Descriptive studies

Answer 7

Case report
Case series (*no comparison group)
Cross-sectional
Correlation studies

Question 8

Cross-sectional studies advantages and disadvantages

Answer 8

studies that examine the exposure and outcome at the same point in time (exp. exercise and arthritis)

Advantage – relatively quick and easy
Disadvantage – temporal relationship between exposure & outcome sometimes hard to establish

Question 9

Observational Analytic studies

Answer 9

Cohort studies

Case-control studies

Question 10

Intervention Descriptive studies

Answer 10

Case report
Case series (exposure is chosen by investigator, no comparison group)

Question 11

Intervention Analytic studies

Answer 11

RCT

Question 12

Case series studies

Answer 12

No comparison group. Tracking pts with known exposure & examining records for exposure and outcome.

Question 13

Cohort study

Answer 13

OBSERVATIONAL STUDY
-researcher does not determine exposure status or intervene in any way. Subjects are chosen based on their exposure status.

• multiple outcomes can be assessed
• researchers determine incidence rate of outcome, so can use RR or OR to determine association between exposure and outcome.
• Good for rare exposures, bad for rare outcomes.

Exposure is not randomized –> difficult to match subjects in exposed and unexposed groups, makes cohort studies subject to confounding.

Fewer ethical issues vs. RCT

Question 14

Prospective Cohort Study

Answer 14

subjects assembled based on exposure status at the beginning of the study, then followed up in real time until the study ends at a pre-specified time (ex. ten years).

• Exposure definition should be clear and precise
• Assessment of outcome should be blind to exposure
• Good at establishing true risk, but expensive and need lg #s

Question 15

Retrospective Cohort Study

Answer 15

subjects with historical records of both exposure and outcome are gathered; the follow-up has already occurred in the past.

• Assessment of exposure should be blind to outcome
• Assessment of outcome should be blind to exposure
• MORE BIAS PRONE
• can be relatively quick/inexpensive

Question 16

Case-control study

Answer 16

OBSERVATIONAL STUDY
-All case-control studies start by assembling cases of the disease and controls without the disease. (subjects selected based on their outcome status, cannot be randomized –> more subject to confounding)

• Case-control status should be assessed blind to exposure status. After cases and controls have been identified, exposure should be assessed blind to outcome.
• Multiple exposures can be assessed
• Good for rare outcomes; bad for rare exposures
• Fewer ethical issues vs. RCT
• Relatively quick/inexpensive
• Incidence cannot be calculated (bc started with cases, people who already have disease of interest), so cant calculate RR.
• ONLY odds ratio can be measured

Question 17

Which types of cases are usually preferred for case-control studies?

Answer 17

Incident cases: newly diagnosed, usually preferred

vs prevalent cases: existing at a point in time

Question 18

Disadvantages of cohort studies

Answer 18

• Bad for rare outcomes (selected by exposure status)
• Time consuming (if prospective)
• Several sources of bias (e.g confounding, bc not randomized)
• Loss-to-follow-up

Question 19

Disadvantages of case-control studies

Answer 19

• Bad for rare exposures
• Difficult to choose valid controls
• Many sources of bias (recall & selection)
• Cannot (usually) measure incidence

Question 20

Randomization

Answer 20

Randomization: Allocation to a treatment group based on chance

Helps remove bias

Removes investigator & volunteer treatment preference (e.g. investigator cannot put sickest patients in exciting new treatment arm)

Balances the trial arms = similar risk profiles (esp. large studies)

Question 21

Intention-to-Treat Analyses

Answer 21

participants are analyzed in the group in which they were originally randomized, regardless of whether or not they adhered to their treatment

(exp: If a person allocated to the experimental treatment group doesn’t take the experimental drug, they will still be analyzed as a part of the experimental treatment group)

Question 22

RCT

Answer 22

“Gold standard” –> internal validity (absence of bias) is greater than in other study designs

1. Enroll participants according to strict inclusion and exclusion criteria
2. Allocate participants to treatment groups by randomization
   - Allocation should be blinded to both the participant and the researcher
3. Follow participants for a relevant period of time
4. Ascertain outcome of participants using objective procedures
5. Analyze results (intention to treat analysis)

Question 23

How does placebo reduce potential bias?

Answer 23

By assessing the impact of the treatment over the placebo effect, the investigator can determine the impact of a
new treatment

• Removes investigator influence by blinding
• Removes volunteer influence by blinding.

Question 24

Odds Ratio

Answer 24

• assesses the association between an exposure and an outcome by comparing the odds of the outcome occurring in the exposed group vs. the unexposed group

ad/bc

Question 25

3 criteria for confounding

Answer 25

Confounding occurs when there is an imbalance in risk factors across study arms 1. Risk factor for outcome 2. Associated with the exposure under study (unevenly distributed among exposure groups) 3. Cannot be a mediator in causal pathway between exposure and outcome

Question 26

Avoiding confounding

Answer 26

1. Randomization: helps balance risk factors across study arms (RCT) 2. Restriction (restrict to all women if worried about gender) 3. Matching (Exp. If worried about confounding by smoking - match one exposed (smoker) to one unexposed who smokes) 4. Stratification of results (Adjusting for confounders) - calculate association (RR or OR) within confounding groups 5. Multivariate analysis (adjustment for multiple cofounders at once) - fancy form of stratification; key words are “adjustment” or “control”

Question 27

Selection bias

Answer 27

Measure of association between the exposure and outcome is not representative of the true association in the target population (depression --> stress and aging) *inflated risk ratio To avoid selection bias, study subjects should be assembled in a manner that dissociates the exposure from the outcome; volunteers should not be aware of the study hypothesis primarily a problem for retrospective studies as the outcome has already occurred Randomized trials cannot have selection bias (joined the study before they know their exposure status, and because the outcome has not yet occurred)

Question 28

Random misclassification

Answer 28

- Always towards the null (RR=1) - happens in same groups (exp. subjective measures like in yoga trial - exposure affected) (exp. difficult disease to diagnose accurately - outcome affected), affects both exposure or outcome groups equally - to avoid --> use well-defined and precise measures

Question 29

Non-random misclassification

Answer 29

- Can bias toward OR away from null - effects only ONE of the groups (moms with FAS babies lie about alcohol use --> outcome group affected) (Investigators over-diagnose lung cancer (outcome) in smokers --> exposure group affected) - to avoid--> use blinding of investigators or volunteers

Question 30

Loss-to-Follow-up Bias

Answer 30

- Exp. in RCT, if sickest people leave placebo group to get other tx, results show healthier placebo group --> underestimated difference between experimental tx and placebo - if the loss to follow-up is not related to the exposure or outcome, it occurs at random, and does NOT cause bias. - To avoid --> stay in touch with volunteers

Question 31

Bradford Hill's viewpoints

Answer 31

-NOT needed for RCTs 1. Strong association 2. Consistency across studies 3. Specific result (one cause-->one outcome) 4. Temporality (exposure precedes outcome?) 5. Biological gradient (direct relationship between the risk factor and people’s status on the outcome variable) 6. Biological plausibility 7. Coherence (between epidemiological and laboratory findings) 8. Experimental Evidence 9. Analogy (exp. induced smoking in lab rats --> lung cancer

Question 32

Data types

Answer 32

1. Nominal: unordered (hair color), binary if only 1 of 2 values (HIV --> pos/neg) 2. Ordinal: breast cancer staging (0-IV) 3. Discrete data: restricted to specific values, exp. number live births, number joints with arthritis 4. Continuous: not restricted to any specific value (height, weight, LDL CHL) *use histogram or box plots

Question 33

Positive (right) skew

Answer 33

median < mean | *more common

Question 34

Negative (left) skew

Answer 34

Mean < median

Question 35

Interquartile Range

Answer 35

Compute the difference between the 75th and 25th percentiles.

Question 36

The 68-95-99 % rule for normal distributions

Answer 36

``` In a normal distribution: - 68% of the values lie within 1 standard deviation of the mean - 95% lie within 2 standard deviations of the mean - 99% lie within 3 standard deviations of the mean ``` * The mean, median, and mode are equal * The curve is bilaterally symmetric

Question 37

How to interpret z-scores

Answer 37

z-score: number of ‘Standard deviations’ that ‘the value’ is above or below the ‘Mean’ z = (Value – Mean) / Standard Deviation If a person’s z-score is 0, then you know their data value is average. If it is 1, then you know their data value is slightly higher than average, but not impressively so. If their z-score is 6, then you know their data value is WAY out in the upper tail of the distribution. This could be a good thing (if we are measuring high jump ability, for example) or a bad thing (if we are measuring LDL cholesterol). z-score of 0 --> at the median (50th percentile)

Question 38

Interpretation of the | 95% Confidence Interval

Answer 38

Range of values that we expect includes the true population parameter we are estimating (whereas a sample mean is our best estimate of the population parameter) We are 95% confident that the lower and upper bounds of the interval will contain the true value we are trying to estimate. exp. "95% certain that the true population mean glucose is between 105.7 and 107.9 mg/dl." Bc in repeated samples, 95% of the confidence intervals will contain the unknown population parameter we are trying to estimate. -studies w/ larger sample size --> smaller CI

Question 39

Confidence interval calculation

Answer 39

• If a CI includes the null value, do not reject HO - Example: RR = 1.23 CI = 0.95, 1.43, do not reject Ho • If a CI does not include the null value, reject Ho

Question 40

SE vs. SD

Answer 40

standard error (SE) is the standard deviation of many sample means standard deviation (SD) --> single sample

Question 41

Things that make confidence interval wider (less precise in capturing the true population value)

Answer 41

Increasing the level of confidence, say from 95% to 99%. More variability among the observations (i.e., larger standard error) A smaller sample size

Question 42

Assessing p-value

Answer 42

If P-value < .05 Reject Null Hypothesis If P-value < .05 DO NOT Reject Null Hypothesis "Given the null hypothesis is true, the probability of obtaining a result at least as extreme as the one observed." ``` The p value does NOT tell us: How likely the Ho is true How likely it is that Ha is true The clinical significance of the result Anything about the role of bias or confounding in the study ```

Question 43

Two sample t-test

Answer 43

Used to compare the means of two groups (the outcome is a continuous variable and the exposure is binary) • Example: Comparing the mean blood pressures between men and women Abstract: "compared two means" and outcome is continuous --> two-sample t-test

Question 44

Chi-square test

Answer 44

Appropriate when both exposure and outcome are nominal (often binary) Used to compare two or more proportions (percentages) Allows you to calculate a p-value associated with RRs and ORs • Compare observed values to expected values (expected values if Ho were true) - If observed values are similar to expected values (small χ2), Ho is plausible - If observed values are different than expected values (large χ2), Ho is rejected

Question 45

Type I error rate

Answer 45

Type I error rate: reject HO, when HO is actually true | - α = 0.05 means that there is a 5% chance of a type I error

Question 46

Type II error rate (β)

Answer 46

do not reject HO, when HO should | be rejected in favor of HA

Question 47

Power (1-β)

Answer 47

probability that a statistical test will result in | a correct rejection of HO

Question 48

Pearson Correlation Coefficient (r)

Answer 48

quantifies the strength (magnitude) and direction of linear relationships between two continuous variables ``` Has no units - Ranges from -1 to +1 - 0 = no linear relationship -1 = perfect inverse linear relationship +1 = perfect positive linear relationship ``` Problems: some variables may be related to each other, but not linearly; may distort data if extreme data points exist

Question 49

Linear Regression

Answer 49

y= β0+ β1 X+ ε • Used to determine how the outcome variable (Y) changes as we change the level of exposure (X) • Allows us to quantify the association (slope of the line, β1) • Outcome must be continuous (exp. bp); exposure may be dichotomous, ordinal, discrete, or continuous • The regression line minimizes the space between the data points and the best fit line • Slope (β1) is the change in Y for a one-unit change in X • Intercept (β0) is the value of Y when X is zero

Question 50

Logistic Regression

Answer 50

used when the outcome is binary (yes/no, exp. mortality, MI); produces ADJUSTED odds ratios; can be used with one or multiple exposures

Question 51

Cox Proportional Hazards Regression

Answer 51

outcome is also binary (mortality) but INCLUDES TIME ELEMENT useful for adjusting survival analyses for extraneous variables (i.e. confounders and effect modifiers); yields hazard ratios (like relative risks)

Question 52

Multivariable Regression/Multiple Linear Regression

Answer 52

assesses multiple exposures and potential confounding, used to identify effect modification (same as linear regression, but with at least two exposure variables)