Stats

Question 1

Berksons bias

Answer 1

False observation of negative correlation between two positive traits. Ie only have one. ACTUALLY = UNEQUALLY OBSERVED

Question 2

Hawthorne effect

Answer 2

Observer effect!
Individuals modify behaviour in response to being observed. BUT OBSERVER EFFECT

Also secondary observer effect ie researchers looking at q responses

Question 3

Neyman bias

Answer 3

Selection bias eg v sick or ill excluded from study!!

Ie died excluded - less severe
Ie recovered exluded - more severe

Question 4

Case control

Answer 4

Compare groups of cases and controls re exposure factors often nested eg sample of controls
Clear case definition
ODD RATIO = EST RELATIVE RISK rare disease assumption
Ie cases: exp/non exp divided controls: exp/non exp

Care with matching ie over match/factors affecting exposure as it should be equal in groups

Question 5

Cohort study

Answer 5

Compare exposed and non exposed re incidence of disease
Longitudinal v restrospective/historical

Measure exposure
Match or id confounders for analysis ie counterfactual ideal
Measure outcomes
Analyse
GENERAL PMR
Proportional mortality ratio deaths given cause/total deaths in same time period

SMR - deaths in study pop/expected deaths in general pop. Standardise w age bands/gender

Types of comparison
Internal ie unexposed IN cohort ie exp unexp in cohort of factory workers
External ie similar cohort of factory workers or gen pop. May differ! Gen pop needs to have low rate of exposure!!

Question 6

Cohort study weaknesses

Answer 6

Healthy worker effect ie gen pop sicker if used as controls
Loss to FU ie TIME - bias = difference between groups DIFFERNTIAL
Difficulties w long latency or rare diseases. TIME/RESOURCES!
Limited records/info ie on exp in retrospective

Question 7

Cohort studies strengths

Answer 7

Longitudinal - clarity re sequence CHAIN OF EVENTS
Calculate INCIDENCE and RISK
RARE EXPOSURE
MULTIPLE EFFECTS
Longitudinal avoid enrolment SELECTION BIAS

Question 8

Case control advantages

Answer 8

RARE DISEASES
Long LATENCY - resources!!
Easier to obtain exp data
DYNAMIC POP
MULTIPLE CAUSES/exp

Question 9

Case control disadvantages

Answer 9

SELECTION BIAS
Inefficient RARE EXPOSURE
OBSERVER BIAS
ODDS RATIO not relative risk or CAUSATION

Question 10

Confounder definition

Answer 10

A confounding factor independently determines the risk of disease/outcome studied and is UNEQUALLY distributed between groups ie is correlated with exposure

NOT PART OF CAUSAL CHAIN OF EVENTS
Positive = stronger
Negative = weaker assoc

Question 11

Sources of bias

Answer 11

Unrepresentative of pop/inaccurate info

Selection bias
Information bias
Observer bias
Recall bias

Question 12

Interpretation of associations

Answer 12

Bias
Chance
Confounding
Effect modifier - other causative factor part of causal chain of events

Question 13

Linear regression

Answer 13

X axis = independent variable
Y axis = dependent variable ie effect being studied

Y=mx+c
Correlation coefficient +- 1 = strongest association

Question 14

Distribution

Answer 14

Mode = most freq value ie uni bimodal
Positive skew - long upper tail, pulls mean up
Negative skew - long lower tail, pulls mean down

Mean - weighting or geometric ie log, mean, anti log

Median

Range

Percentiles ie 5th = 5% values below
Quartiles ie 25-75th in box plot

Question 15

Standard deviation

Answer 15

Dispersion of values about mean
NORMAL DISTRIBUTION 95% values within +-2SDs of mean

Coefficient of variation SD/mean x 100

Question 16

Probability ie confidence intervals

Answer 16

Ie of RR or OR
95% confidence that value is located within range
Assume random sample, normal distribution

Question 17

Significance v chance ie P values

Answer 17

Test validity of null hypothesis
P value 0.05 / 5%

Two means - chi2/students T test
Non parametric - distribution free

Question 18

Adjustment re confounders

Answer 18

Stratify
Direct standardisation - eg weight. same weight ie same group size /denominator. Need all age etc specific rates.
Indirect standardisation ie SMR - different WEIGHTS ie age structure of pop. Only need total deaths and pop size. Useful if small numbers

Stats ie mantel henizel
Regression models
Pooled adjustment for residual confounders

Question 19

Measures of occurence

Answer 19

Incidence = new cases/1000/yr
Prevalence ie point/period = cases in given time
Approx incidence x ave duration

Question 20

Sensitivity

Answer 20

Sensitivity = True positives/positive CASES ie true positives + false negatives
PPV = true positives/positive RESULTS ie true positives + false positives

Specificity = true negatives/negative cases ie true negatives + false positives
NPV = true negatives/negative results ie true negatives + false negatives