Critical Appraisal Bias - Blinding/Follow Up

Question 1

Why do we blind Patients

Answer 1

To prevent placebo effect

Question 2

Why do we blind Clinicians

Answer 2

To prevent cointerventions
- Additional procedures to sway the data

Question 3

Why do we blind Data Collectors

Answer 3

To prevent bias in collection

Question 4

Why do we blind Outcome Adjudicators

Answer 4

To prevent bias in decisions regarding outcomes
- When deciding who has what outcomes

Question 5

Why do we blind Data Analysts

Answer 5

To avoid bias in decisions regarding data analysts

Question 6

Open Label Study

Answer 6

No blinding
- Clinicians and Patients no which intervention they have

Question 7

PROBE Design

Answer 7

Patient and Clinician can not be blinded

Can blind Outcome Adjudicators

Question 8

How to control placebo

Answer 8

Blinding

Make outcomes less subjective

Ask patients at the end if they think they received

Question 9

Contamination

Answer 9

Patient receives treatment from the other group

Question 10

Why is loss of follow up important

Answer 10

If the patient is lost to follow up then the study wont be able to record their events and outcomes

Question 11

Loss To Follow Up
- Rule of Thumb

Answer 11

5% Rule
- If 5% or below then risk of bias is low and we can tolerate the

20% Rule
- If 20% or higher than risk of bias is high and we can not trust the data

Question 12

Loss of Follow Up is between 5% and 20%

Answer 12

Assume worst case scenario
- The minimal effect the intervention can have

Even in the worst case there is still benefit

Question 13

Patients lost in follow up vs patients remaining

Answer 13

Both groups have different prognosis
- Usually the patients lost to follow up either are much sicker or much healthier

Question 14

Alternative to 5% and 20% rule

Answer 14

Assume worst case scenario

So for an experimental treatment that reduces a bad event
- Assume the patient’s outcome was the bad effect

Question 15

How to deal with missing data
- Single Imputation

Answer 15

• Carry last observable outcome forward
• Replace with average value of all participants
• Use regression model to predict missing value based on trends

Question 16

How to deal with missing data
- Multiple Imputation

Answer 16

• Use regression methods and perform standard analysis for each imputation cycle
  –> Final analysis takes into consideration variability from multiple imputation cycles

Question 17

Sensitivity Analysis

Answer 17

To assess the effect of missing data
- If the data remains consistient even after slowly changing different variables after multiple imputation then you can trust the data

Question 18

What does Loss to Follow Up depend on

Answer 18

Does NOT depend on total number of LTF

Depends on number of events in each group

Question 19

LTF with low risk of bias

Answer 19

For low risk of bias
- Investigators need to demonstrate that prognostic is similar between group that remains and group LTF
- Even when assuming worst case there is little change in results