Dear all, I am writing to you about a Survival Analysis problem.
I have addressed this topic in simpler studies but now I would prefer to have your opinion to avoid publishing biased results.
I realize this is a long post. I apologize and thank you in advance.
I have a study where 3 var drugs are given to about 1000 patients. The dataset is in long format as shown below.
The efficacy of the drug is binary (var effective).
The presence of adverse events is binary (var adverse).
Patients are monitored at 6, 12, 24, 36 months (observation ends at 36 months).
The aims of my study are to evaluate retention (time-to-dropout), the efficacy of the drug over time and the presence of adverse events over time.
Concerning retention-time (time-to-dropout), I woluld use a survival-analysis adjusting for drug-efficacy and its adverse events change over time (they are communicated by the patient at each follow-up visit).
I guess you have to use a model with efficacy and adverse events as time-dependent variables. Is it reasonable to specify the model in this way, assuming (for now) proportional risks and modelling time-dependence by interaction terms?
stcox i.drug i.followuptime ## i.effective i.followuptime ## i.adverse
I have found many helph-state examples of time-dependent binary variables but none of them represent mine
I also wonder, if efficacy was non-binary, but ordinal (0, 1, 2) how to specify it.
Finally, I know of the existence of Joint models, in which the longitudinal submodels (for adverse and effective variables) and the one for time-to-dropout would be integrated.
However of these models I have only seen examples (command stjm) where the longitudinal is relative to continuous variables (not binary, as in my case).
I wonder if any of you know how to specify this or not.
Thanks again and again.
Gianfranco
I have addressed this topic in simpler studies but now I would prefer to have your opinion to avoid publishing biased results.
I realize this is a long post. I apologize and thank you in advance.
I have a study where 3 var drugs are given to about 1000 patients. The dataset is in long format as shown below.
| id | followuptime | effective | adverse | failure | observtime | age | drug |
| 1 | 6 | 1 | 0 | 0 | 36 | 42 | 2 |
| 1 | 12 | 1 | 0 | 0 | 36 | 42 | 2 |
| 1 | 24 | 0 | 1 | 0 | 36 | 42 | 2 |
| 1 | 36 | 1 | . | 0 | 36 | 42 | 2 |
| 2 | 6 | 1 | 1 | 1 | 27 | 35 | 1 |
| 2 | 12 | 0 | 0 | 1 | 27 | 35 | 1 |
| 2 | 24 | 1 | 0 | 1 | 27 | 35 | 1 |
| 2 | 36 | . | . | 1 | 27 | 35 | 1 |
| 3 | 6 | 0 | 1 | 1 | 15 | 37 | 2 |
| 3 | 12 | 0 | 1 | 1 | 15 | 37 | 2 |
| 3 | 24 | . | . | 1 | 15 | 37 | 2 |
| 3 | 36 | . | . | 1 | 15 | 37 | 2 |
The presence of adverse events is binary (var adverse).
Patients are monitored at 6, 12, 24, 36 months (observation ends at 36 months).
The aims of my study are to evaluate retention (time-to-dropout), the efficacy of the drug over time and the presence of adverse events over time.
Concerning retention-time (time-to-dropout), I woluld use a survival-analysis adjusting for drug-efficacy and its adverse events change over time (they are communicated by the patient at each follow-up visit).
I guess you have to use a model with efficacy and adverse events as time-dependent variables. Is it reasonable to specify the model in this way, assuming (for now) proportional risks and modelling time-dependence by interaction terms?
stcox i.drug i.followuptime ## i.effective i.followuptime ## i.adverse
I have found many helph-state examples of time-dependent binary variables but none of them represent mine
I also wonder, if efficacy was non-binary, but ordinal (0, 1, 2) how to specify it.
Finally, I know of the existence of Joint models, in which the longitudinal submodels (for adverse and effective variables) and the one for time-to-dropout would be integrated.
However of these models I have only seen examples (command stjm) where the longitudinal is relative to continuous variables (not binary, as in my case).
I wonder if any of you know how to specify this or not.
Thanks again and again.
Gianfranco
