Pattern-Mixture Models for Missing Data With Dr Little

So Rod, in the article that you wrote for JAMA, you talk about the use of pattern mixture models for assessing the potential impact of missing data on the validity of the estimates of treatment effects that we get from randomized clinical trials.

So let's start at the beginning and review for our listeners what are missing data and why are they so important.

Yeah.

So missing data are just things that we were trying to record during a clinical trial or a study that didn't get measured.

Perhaps people dropped out before the end of a study, so the values were missing, or sometimes you can have missing values for intermediate variables or the background variables.

So it's quite common to have missing, missing values or gaps in a data set.

The importance i-in a randomized clinical trial is that if you have missing data, particularly on, on outcome variables, then that can mess up the randomization, so it has the potential to lead to biased estimates for treatment effects.

Statistical analysis can handle missing data, but it has to rely on some kinds of assumptions about, uh, the distribution of the missing values, and if those assumptions are not true, then the treatment effects can be biased.

For that reason, although I like to think about methods for analyzing missing data, actually, the best thing is to try and avoid it if it's at all possible.

I think that we all have observed that as much as people try to minimize missing data and missing outcome data, particularly in clinical trials, it's almost unavoidable, certainly in trials that involve large populations or following patients over a long period of time.

You mentioned that many statistical approaches make assumptions about the distribution of the missing values, meaning what would have those values looked like had we been able to observe them.

So let's talk a little bit about those possible assumptions.

So Rod, in the article that you wrote for JAMA, you talk about the use of pattern mixture models for assessing the potential impact of missing data on the validity of the estimates of treatment effects that we get from randomized clinical trials.

So let's start at the beginning and review for our listeners what are missing data and why are they so important.

Yeah.

So missing data are just things that we were trying to record during a clinical trial or a study that didn't get measured.

Perhaps people dropped out before the end of a study, so the values were missing, or sometimes you can have missing values for intermediate variables or the background variables.

So it's quite common to have missing, missing values or gaps in a data set.

The importance i-in a randomized clinical trial is that if you have missing data, particularly on, on outcome variables, then that can mess up the randomization, so it has the potential to lead to biased estimates for treatment effects.

Statistical analysis can handle missing data, but it has to rely on some kinds of assumptions about, uh, the distribution of the missing values, and if those assumptions are not true, then the treatment effects can be biased.

For that reason, although I like to think about methods for analyzing missing data, actually, the best thing is to try and avoid it if it's at all possible.

I think that we all have observed that as much as people try to minimize missing data and missing outcome data, particularly in clinical trials, it's almost unavoidable, certainly in trials that involve large populations or following patients over a long period of time.

You mentioned that many statistical approaches make assumptions about the distribution of the missing values, meaning what would have those values looked like had we been able to observe them.

So let's talk a little bit about those possible assumptions.