Comparing slopes for serial measurements in the presence of missing data

classic Classic list List threaded Threaded
5 messages Options
Reply | Threaded
Open this post in threaded view
|

Comparing slopes for serial measurements in the presence of missing data

AndreasV
We are planning to investigate whether a new treatment alters the progression of a disease over time. We will have two groups of patients randomized to standard of care or the new treatment, and there will be monthly measurements of some continuous outcome over a period of 12 months. According to regulatory guidance, efficacy of a new treatment should preferrably be demonstrated by showing that the decline slopes for the outcome of interest diverge over time.

Since some patients will likely drop out before the scheduled end of the period of observation, and we need to include all randomized patients with any post-baseline data into the analysis, I was originally thinking of using MMRM for alanyzing the data. However, since MMRM treats time as categorical, this may not be the optimal method for comparing slopes.

What would be a more efficient method for comparing slopes while appropriately accounting for missing data - and (how) can this be done in SPSS?

Thanks, Andreas
Reply | Threaded
Open this post in threaded view
|

Re: Comparing slopes for serial measurements in the presence of missing data

Ryan Black
Andreas,

You could employ a linear mixed model where you treat time as a continuous variable in estimating the group-specific slope fixed effects while simultaneously treating time as a categorical variable when accounting for temporal residual correlation.

I assume your data are MAR or MCAR.

Ryan

On Sat, Nov 5, 2016 at 8:52 AM, Andreas Voelp <[hidden email]> wrote:
We are planning to investigate whether a new treatment alters the progression
of a disease over time. We will have two groups of patients randomized to
standard of care or the new treatment, and there will be monthly
measurements of some continuous outcome over a period of 12 months.
According to regulatory guidance, efficacy of a new treatment should
preferrably be demonstrated by showing that the decline slopes for the
outcome of interest diverge over time.

Since some patients will likely drop out before the scheduled end of the
period of observation, and we need to include all randomized patients with
any post-baseline data into the analysis, I was originally thinking of using
MMRM for alanyzing the data. However, since MMRM treats time as categorical,
this may not be the optimal method for comparing slopes.

What would be a more efficient method for comparing slopes while
appropriately accounting for missing data - and (how) can this be done in
SPSS?

Thanks, Andreas



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Comparing-slopes-for-serial-measurements-in-the-presence-of-missing-data-tp5733425.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
Reply | Threaded
Open this post in threaded view
|

Re: Comparing slopes for serial measurements in the presence of missing data

AndreasV
Ryan Black wrote
I assume your data are MAR or MCAR.
... most probably MAR.

Andreas
Reply | Threaded
Open this post in threaded view
|

Re: Comparing slopes for serial measurements in the presence of missing data

Ryan Black
Andreas,

To treat time as continuous and categorical within the same linear MIXED model, you should create two variables with the same data but slightly different names (e.g., "time_cat" and "time_cont" where "time_cat" will be treated as categorical and "time_cont" will be treated as continuous). That said, here's MIXED code:


MIXED y BY group time_cat WITH time_cont
  /FIXED=group time_cont group*time_cont | SSTYPE(3)
  /METHOD=REML
  /PRINT=SOLUTION
  /REPEATED=time_cat | SUBJECT(subject_ID) COVTYPE(<specify error covariance structure>).


HTH.

Ryan

On Sun, Nov 6, 2016 at 11:37 AM, Andreas Voelp <[hidden email]> wrote:
Ryan Black wrote
> I assume your data are MAR or MCAR.

... most probably MAR.

Andreas



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Comparing-slopes-for-serial-measurements-in-the-presence-of-missing-data-tp5733425p5733431.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
Reply | Threaded
Open this post in threaded view
|

Re: Comparing slopes for serial measurements in the presence of missing data

AndreasV
... thanks very much!

Andreas