Hello,
I am doing a survival analysis using ~20 or so datasets. I would like to estimate the rate of my outcome at different time periods of my follow-up.
I have a question about the use of strate and stsplit using my data.
I have used stsplit to split up my follow-up time as seen using the code below:
Then I run strate adjusting for the clustering of each cohort as seenn below:
When i do this I get this output:
However, when I do something (at least intuitively to me) seems like it should give me the same result I get vastly different results. For this I run strate and truncate the follow-up time to only include below the first year. This should (I would think) be equal to the first row above but instead I get vastly higher rates (11.65 rate versus 2.13).
Please see below code and output:
Can anyone explain why the rates are so different? Am I doing or assuming something inaccurately about strate?
Thank you for all your help.
Leo
I am doing a survival analysis using ~20 or so datasets. I would like to estimate the rate of my outcome at different time periods of my follow-up.
I have a question about the use of strate and stsplit using my data.
I have used stsplit to split up my follow-up time as seen using the code below:
Code:
stsplit time_split_merge, at(1,2,5,10)
Code:
. strate time_split_mer if age_mer<19 & prosp==1 , per(100) graph cluster(study)
HTML Code:
. strate time_split_mer if age_mer<19 & prosp==1 , per(100) graph cluster(study)
failure _d: inc_dis_merged
analysis time _t: survival_tyrs_merged
id: unique_id
Estimated rates (per 100) and lower/upper bounds of 95% confidence intervals
(30960 records included in the analysis)
The jackknife was used to calculate confidence intervals
41 total clusters identified by study and time_split_merge
+------------------------------------------------------------+
| time_s~e D Y Rate Lower Upper |
|------------------------------------------------------------|
| 0 309 144.5146 2.138193 1.090785 5.851070 |
| 1 100 70.3475 1.421515 0.780483 2.733694 |
| 2 11 49.3407 0.222940 0.049534 0.903587 |
| 5 0 0.2739 0.000000 . . |
+------------------------------------------------------------+
Note: Jackknife confidence intervals are missing because of an insufficient
number of failures in the dataset.
However, when I do something (at least intuitively to me) seems like it should give me the same result I get vastly different results. For this I run strate and truncate the follow-up time to only include below the first year. This should (I would think) be equal to the first row above but instead I get vastly higher rates (11.65 rate versus 2.13).
Please see below code and output:
Code:
. strate if age_mer<19 & prosp==1 & survival_tyrs_mer<=1, per(100) graph cluster(study)
HTML Code:
. strate if age_mer<19 & prosp==1 & survival_tyrs_mer<=1, per(100) graph cluster(study)
failure _d: inc_dis_merged
analysis time _t: survival_tyrs_merged
id: unique_id
Estimated rates (per 100) and lower/upper bounds of 95% confidence intervals
(3519 records included in the analysis)
The jackknife was used to calculate confidence intervals
13 clusters identified by study
Warning: graphical output suppressed - no groups to plot
+--------------------------------------------+
| D Y Rate Lower Upper |
|--------------------------------------------|
| 309 26.5146 11.6540 6.5145 23.7337 |
+--------------------------------------------+
Thank you for all your help.
Leo
