| Message/Author |
|
|
|
|
| We are working with longitudinal data from at treatment study and want to identify classes of patient with different courses during a 3-year period. Latent class growth analysis with ML for treating missing is the method. Our problem is that in our data there are some cases that only have baseline data and are missing on the remaining 6 time points (early dropouts). Would it be reasonable to include these cases in the analysis or should they be completely removed? |
|
|
|
|
| I would include everyone. Those with only baseline data contribute to the estimation of baseline-related parameters. |
|
|
|
|
Thank you very much for your advice! The thing that worried me is how reliable their group belonging is since it is only based on one value.
Best David |
|
|
|
|
| Right. That depends on what the class formation is mostly driven by - the level or the trend. You will see in the saved cprob values for these subjects how well they classify. |
|
|
|
|
I will take a look at the cprob values! A final question if I may. Would you take any actions on subjects that have many missing values and very low cprob, such as removing them?
Thank you again for taking time to answer my questions. Your help is highly appreciated.
David |
|
|
|
|
| No, but I might report that individuals with many missing data points are non-informative due to equal cprobs. Such as with 3 classes you have a subject with cprob = 1/3, 1/3, 1/3. |
|
|
|
|
Excellent! Thank you so much for taking time to answer my questions. Very helpful! Best
David |
|
|
| Back to top |
