EBM 4 ; inference

Question 1

when determining the association between E (exposure) 7 D (disease) there are 5 possibilities that describe and explain the relationship

Answer 1

result could be due to;

• CHANCE (p values)
• reversible causality
• causal
• bias
• confounders

Question 2

correlation

linear regression

Answer 2

C; how 2 continuous variables change together
(-1 to +1)
- MEASURES STRENGTH OF RELATIONSHIP

LR; relationship between one variable and another
MEASURES CHANGE IN OUTCOME PER UNIT CHSNGE IN THE EXPOSURE
y = a + bx
y ; outcome
x ; exposure
a ; y intercept
b ; slope

Question 3

confounder

• what?
• 3 features

Answer 3

- another risk factor of a disease 
3 features 
- must be associated with the exposure of interest 
- must be a risk factor of the disease 
- must NOT be in the causal pathway

Question 4

how to reduce the effect of confounders

Answer 4

• RCT ; Randomised ; should mean exposure ; no relationship with cofounders (not v sure WHY THO)

CONTROL IN ANALYSIS
eg STRATIFICATION ;
- estimate OR for different strata
- keep levels of confounder constant so only exposure of interest levels are changing
- average out OR using some sort of equation

Question 5

to control the effect of a no. of confounders

Answer 5

use REGRESSION MODELS

• used in associations where the outcome is binary (presence or absence of disease)
• v similar to line go regression except y value is binary, not continuous
• thus the relationship is ; how x increases and how this affects whether outcome is more likely to be absence or presence of disease

Question 6

BIAS -

• 3types
• within each type you can have ?

Answer 6

1) - selection bias ; occurs in the selection of participants
2) - loss to follow up bias ; participants drop out
3) - measurement bias ; bias in measuring exposure & outcome ;
- performance bias ; occurs when 2 groups in a study are not treated equally e.g. placebo group given extra interventions etc
- detection bias ; when some participants are less likely to report outcomes etc

selection bias ; non differentiated selection bias or differentiated selection bias

NON DIFFERENCIATED SELECTION BIAS
- if any errors in selecting participants occurs equally between the 2 groups then the association between E & D is UNBIASED

DIFFERENCIATED SELECTION BIAS
- if the errors in selecting participants is not equal between the 2 groups then the association is biased and the association will be under or over biased

Measurement bias ; NON DIFFERENCIATED MEASUREMENT BIAS & MISSCLASSIFICATION & DIFFERENCIATED MEASUREMENT BIAS & MISSCLASSIFICATION

N.D.M.B&. MC
- measurement errors which lead to misclassification are equal in all groups then association will be underestimated as usually these errors make the groups more similar.

D.M.B & MC.

• measurement errors lead to misclassification occurs unequally in groups
• association may be over or under estimated - NOT GOOD

Question 7

BIAS IN

• RCT
• cohort studies
• CASE CONTOL

Answer 7

RCT; mentioned above

COHORT;
- differential measurement bias is unlikely in a PROSPECTIVE CS since exposure is measured before outcome

CASE CONTROL STUDIES;
- recall bias ; errors in participants recalling their exposure or non exposure to something e.g. depressed cases are more likely to talk about being bullied than non depressed people i.e. controls

• interviewee bias; some interviewers will be more likely to get info out of participants etc
• also danger of REVERSE CAUSALITY

Question 8

BRADFORD HILL CRITERIA

Answer 8

• temporal sequence ;
  is it def causal and not reverse causality
• strength of association ;
  stronger the RR the more likely the relationship is causal
• consistency of relationship ; if diff ppl did the study with diff participants would they get the same results ?
• biological gradient ie dosage compensation ; with increased exposure does risk of disease increase - more believable
• specifity ; exposure and disease association must be specific - specific outcome not multiple
• coherence ; explanation for the association
• reversibilty ; RCT evidence etc