Title: Missing Inaction: Why Do So Many People Ignore Missing Data in RCTs
1Missing InactionWhy Do So Many People Ignore
Missing Data in RCTs?
- Temple-Merck Conference 17-Oct-08
- Janet Turk WittesStatistics Collaborative
2Extent of missing primary outcome data
- Cardiovascular outcome trial 1-2
- Cancer progression-free survival 2-20
- Short-term blood pressure trial 5-10
- 12 week pain trial 20-40
- 12 week antipsychotic drug 30-50
- 12 week anti-infective 20-50
- Source informal experience
3What others assume
4What others assume
5What others assume
6What we fear (and assume)
7What we fear (and assume)
- What we have left is different from what was
there at first - We cant characterize what is missing
- What is missing differs by group
8Evidence of inaction hard to ferret out extent
and timing
Time
Missing values imputed with LOCF.
9Rarely apparent in survival curves
10Time to event
11What I am not going to talk about
- MCAR, MAR, not MAR
- Ignorable/non-ignorable
- The effect of missing data on inference
- In sample surveys
- In experiments
- In randomized clinical trials
- Detailed methods of dealing with missing data
12What I will discuss
- Once over lightly of the methods at hand
- Why others dont care about missing values
- Why our protocols encourage missing data
- What we can do to prevent missing data
- Even though prevention is boring
13Underlying principle
- Our method of imputation shouldnt give us better
results than what we would have seen from the
complete cases
14The cards in our deck binary outcomes
- Just ignore the missing observations
- Impute missing on basis of
- Proportion in own group
- Best case all pbo fail all rx succeed
- Worst case all pbo success all rx fail
- Proportion in placebo group (not unreasonable
guess) - Proportion in opposite group (reasonable worst
case) - Multiple imputation
15Problems with the binary cards
- Too many degrees of freedom
- Some methods overstate effect
- Some methods understate effect
- Some methods are unreasonably pessimistic
16Loss of 3 lines of vision
- Two groups treated and control
- 120 eyes per group (one per person)
- 40 in placebo 20 in treated
- Look at relative risk (
- Missing equal in both groups
17Loss of 3 lines of vision impute own group
Imputation leads to increased sample size hence
false precision.
18Binary example loss of 3 lines of vision
1.13
0.31
100 of missing behave the same way not
reasonable.
19Binary example loss of 3 lines of vision
0.6
Not unreasonable to assume that untreated act
like placebo. Best reasonable case
20Binary example loss of 3 lines of vision
0.67
Opposite arm is worst reasonable case.
21What do binary cards do for us
- Bad
- Too many degrees of freedom
- Some methods overstate effect
- Some methods understate effect
- Good
- Sensible cases provide bounds
- Multiple imputation (if we have a good model)
22The cards in our deck continuous outcomes
- Just ignore the missing observations
- Impute missing on basis of mean in
- Own group
- Combined group
- Placebo group
- Opposite group (worst reasonable case)
- Last Observation Carried Forward
- Baseline Observation Carried Forward
- Last rank carried forward
- Multiple imputation
23The cards in our deck longitudinal
- Just ignore the missing observations
- Impute missing by carrying forward
- Last observation
- Baseline observation
- Own group trajectory
- Placebo trajectory
- Opposite group trajectory
- Last rank
- Longitudinal model
- Multiple imputation
- Proschan et al (2001)., J Stat Planning 96 155
- Obrien, Zhang, Bailey (2005). Stat Med 2434
24Longitudinal outcome
- Pain at Day 4
- 325 patients per group
- 250 per group completed
- 7 point scale
- Placebo Treated
- Baseline 5.0 5.0
25Effect size and p-values
26Effect size and p-values
27Continuous outcome
- Placebo Treated
- Baseline 5.0 5.0
- Day 1 3.7 3.2
28Effect size and p-values
29The cards in our deck survival
- Censor when missing
- Assume missing have event
- At same proportion as own, placebo, or opposite
group - Need to decide when the imputed event occurs
- At time of censoring
- At rate in assigned group
30Message
- Analyses produce very different results
- Can affect
- Direction of effect
- Effect size
31Why people dont care about missing data in
outcome trials
- In outcome trials
- we can censor doesnt matter what happens after
people stop drug
32Why people dont care about missing data
- Outcome trials are different from symptom trials
- Who cares about those who dont take drug?
- We know the drug wont work if you dont take
it - I am not interested in what happens after
people stop. - No evidence that the two groups differ in
Prmissing - We are interested in what we observe complete
cases - Too hard/expensive to bring back those who stop
med
33Informed consent documents unclear
- Participation in this study is entirely
voluntary. Your treatment and your doctors
attitude toward you will not be affected should
you decide not to participate in this study - You will be asked to return for follow-up visits
and to provide follow-up information. - If you agree to participate, you may withdraw
from the study at any time without affecting any
benefits to which you would otherwise be
entitled.
34Permissive protocols encourage missing data
- Drop-outs will not be replaced
- Suggests that it would be ok to replace them
- Suggests that analysis will ignore them
- Expect 10 drop out, therefore increase sample
size by 10 - The primary analysis will use the
intent-to-treat pop - The ITT pop is defined as all those randomized
who - The ITT pop is defined as the evaluable group
35Language about withdrawal an outcome trial
- The reason that a subject discontinues from the
study will be recorded in the Case Report Form. - A discontinuation occurs when an enrolled subject
ceases participation in the study, regardless of
the circumstances, prior to completion of the
protocol. -
- The final evaluation required by the protocol
will be performed at the time of study
discontinuation.
36Outcome continuous measure at week 48
- Subjects must be withdrawn from the study (i.e.,
from any further study medication or study
procedure) for the following reasons - At their own or their legally authorized
representatives request - If, in the investigators opinion, continuation
in the study would be detrimental to the
subject's well-being - Occurrence of an intolerable treatment-emergent
adverse event as determined by the investigator
and/or the subject - Failure of the subject to return to the study
site for scheduled visits - Persistent noncompliance
- Pregnancy
37Prevention of missing values
- Education
- What is the effect of various analytic methods
- Why is missing important
- Revise informed consent forms
- Make protocols less permissive
38Education of investigators
- Important to explain to investigators
- need for follow-up
- consequences to the study of failure to follow-up
39Improved informed consent document
- Participation in this study is entirely
voluntary. Your treatment and your doctors
attitude toward you will not be affected should
you decided not to participate in this study - If you agree to participate, you may withdraw
from the study at any time without affecting any
benefits to which you would otherwise be
entitled. - You will be asked to return for follow-up visits
and to provide follow-up information even if you
are not taking study medication.
40Protocols
- Be vigilant about permissive language
- Distinguish between
- Stopping meds
- Stopping active visits
- Withdrawing consent to be followed passively
- Explain to investigators the importance of
follow-up - (even for those who stop study medication)
41Typical language about withdrawal in protocols
- The reason that a subject discontinues from the
study medication will be recorded in the Case
Report Form. - A discontinuation from the study occurs when an
enrolled subject ceases participation a
participant in the study dies, is permanently
lost to follow-up, or withdraws consent,
regardless of the circumstances, prior to
completion of the protocol. -
- An final evaluation required by the protocol will
be performed at the time of study discontinuation
of study medication.
42But, if there are missing data
- Choose analytic methods that
- Dont add false precision
- Are reasonably conservative
- Are interpretable
- Recognize that need for big increase in sample
size - Phil Lavoris rule 1 missing observation needs
three additional - So, if you expect 10 missing, inflate sample
size by 1/3
43Conclusion homework assignment
- Look at all your protocols
- Look at all your model informed consent forms
- Prevent permissive language in the future