drop put

Question 1

2 types of drop out in a study

Answer 1

1. from treatment

2. from study (no questionnaires/ interviews)

Question 2

When is somebody a treatment drop-out?

Answer 2

10 out of 20 sessions? 15? 19?

• you need to define a cut-off (define 2 or more groups), e.g.,
  1. drop-out
  2. part treatment completer
  3. treatment completer
• very subjective
• define beforehand

Question 3

What do you do with people who do not start treatment at all? DO NOT

Answer 3

• problem with replacing or forgetting about them: inflates your treatment effects (because it “messes” with treatment preference)
• worse with more treatment drop-out

Question 4

What do you do with people who do not start treatment at all? DO

Answer 4

• keep them in the study
• they might refuse treatment because they don’t like that treatment
  
  • this will also happen in routine practice
  • it might be associated with certain patient characteristics
• invite for assessments/ questionnaires (even though they haven’t done complete treatment)

Question 5

preferably you do 2 analysis (at the end of your study)

Answer 5

1. what is the effect of receiving treatment (treatment completers only)
2. what is effect of offering treatment (all randomized patients)

Question 6

drop-out from the study, how much is okay?

Answer 6

• even though you might have enough patients, it might be a biased sample
• you always miss data: 10-20% is normal/ quite good
• more than 30-50% is problematic
• too much missing –> your trial failed
  
  • results remain unknown
  • no publications possible

Question 7

what can you do to encourage participants to stay in your study?

Answer 7

• same person every time if they have questions
• personal
• kind, clear
• too much to do
• reminders

Question 8

How do you handle missing data?

Answer 8

• you impute the missing data

Question 9

How do you handle missing data? analyse remaining

Answer 9

• completers only
• advantage: you are sure about these data
• disadvantage: there might be selection, e.g. 10% most severe patients might have dropped out; effect might inflate

Question 10

How do you handle missing data? estimate the scores of the missing data and analyse 100%

Answer 10

• this is called imputation of missing data
• the analyses are called intention-to-treat (ITT)
• advantages: you try to prevent selection bias
• disadvantage: remains estimation

Question 11

How do you impute/ estimate missing data? techniques

Answer 11

• replace by mean score (might be wrong, does not solve bias)
• last observation carried forward (you use the last available measure –> quite conservative, but not always, so you might also overestimate your effect)
• multiple imputation (estimate scores based on what you know from baseline(predict post-test scores), compare with other people with similar scores)

Question 12

multiple imputation

Answer 12

• best option of imputing data
• regression analysis: based on baseline scores + outcomes of observed values predict missing values
• never perfect predictions: repeat e.g. 20 times
• 20 datasets –> 20*analysis
• pool results from 20 analysis

Question 13

intention-to-treat

Answer 13

main analyses= intention to treat= analyze as randomized

Question 14

sensitivity analyses

Answer 14

e. g., analyze 80 treatment completers CBT vs. 100 CAU
- analyze 70 study completers CBT vs. 70 study completers CAU
- -> make that clear before (so you are not accused of phishing)
- are results similar=robust?