Hi, I would like to get some help using merlin
I'm trying to run a simple model to see if there is an association between biomarker leves assessed during the first year of follow up and the survival
The dataset looks like this
list id mstime mdied dlevels time if inlist(id,1676,1691), sepby(id) noobs
I run the follow model including a random intercept for the time variable and I used EV to see the biomarker effect at the same unit
merlin (dlevels time M1[id]@1, family(gaussian) timevar(time)) (mstime EV[dlevels], family(weibull, failure(mdied)) timevar(mstime)), covariance(unstructured)
Fitting fixed effects model:
Fitting full model:
Iteration 0: log likelihood = -20828.813 (not concave)
Iteration 1: log likelihood = -20762.625
Iteration 2: log likelihood = -20743.905
Iteration 3: log likelihood = -20743.558
Iteration 4: log likelihood = -20743.557
Mixed effects regression model Number of obs = 7,139
Log likelihood = -20743.557
As far as I can see there is a significant association between the biomarker and the time, seeing that biomarker is slightly increasing over the first year of follow-up (0.001, CI 0.000, 0.002), which should be the opposite.
Moreover, there is an association between high level of biomarker with an increased risk of death (0.212, CI 0.059, 0.365)
Is that right? Do you have any suggestion to improve this model?
I would like to introduce even other biomarker with different scale and add others confounder, but I would just make sure first that the simpler model is right
Thank you very much for your help!
I'm trying to run a simple model to see if there is an association between biomarker leves assessed during the first year of follow up and the survival
The dataset looks like this
list id mstime mdied dlevels time if inlist(id,1676,1691), sepby(id) noobs
| id | mstime | mdied | dlevels | time | |
| 1676 | 3359 | 0 | 0.00 | 0 | |
| 1676 | . | . | 5.90 | 13 | |
| 1676 | . | . | 4.90 | 14 | |
| 1676 | . | . | 10.30 | 80 | |
| 1676 | . | . | 7.20 | 108 | |
| 1676 | . | . | 6.70 | 139 | |
| 1676 | . | . | 7.50 | 170 | |
| 1676 | . | . | 7.60 | 211 | |
| 1676 | . | . | 5.40 | 225 | |
| 1676 | . | . | 9.60 | 238 | |
| 1676 | . | . | 13.00 | 255 | |
| 1676 | . | . | 6.50 | 269 | |
| 1676 | . | . | 6.00 | 290 | |
| 1676 | . | . | 7.00 | 325 | |
| 1676 | . | . | 10.20 | 357 | |
| 1676 | . | . | 9.10 | 364 | |
| 1676 | . | . | 5.30 | 400 | |
| 1691 | 3360 | 0 | 0.00 | 0 | |
| 1691 | . | . | 2.00 | 4 | |
| 1691 | . | . | 2.40 | 5 | |
| 1691 | . | . | 9.30 | 8 | |
| 1691 | . | . | 5.30 | 10 | |
| 1691 | . | . | 4.70 | 12 | |
| 1691 | . | . | 4.10 | 13 | |
| 1691 | . | . | 2.80 | 14 | |
| 1691 | . | . | 6.10 | 16 | |
| 1691 | . | . | 9.70 | 17 | |
| 1691 | . | . | 7.70 | 19 | |
| 1691 | . | . | 10.00 | 20 | |
| 1691 | . | . | 8.70 | 22 | |
| 1691 | . | . | 8.70 | 25 | |
| 1691 | . | . | 9.20 | 32 | |
| 1691 | . | . | 15.80 | 39 | |
| 1691 | . | . | 5.80 | 46 | |
| 1691 | . | . | 7.20 | 48 | |
| 1691 | . | . | 6.00 | 52 | |
| 1691 | . | . | 7.50 | 56 | |
| 1691 | . | . | 10.40 | 59 | |
| 1691 | . | . | 16.10 | 80 | |
| 1691 | . | . | 11.50 | 89 | |
| 1691 | . | . | 9.90 | 96 | |
| 1691 | . | . | 8.40 | 108 | |
| 1691 | . | . | 10.70 | 118 | |
| 1691 | . | . | 11.00 | 192 | |
| 1691 | . | . | 13.10 | 223 | |
| 1691 | . | . | 16.10 | 249 | |
| 1691 | . | . | 5.50 | 283 | |
| 1691 | . | . | 12.10 | 347 | |
I run the follow model including a random intercept for the time variable and I used EV to see the biomarker effect at the same unit
merlin (dlevels time M1[id]@1, family(gaussian) timevar(time)) (mstime EV[dlevels], family(weibull, failure(mdied)) timevar(mstime)), covariance(unstructured)
Fitting fixed effects model:
Fitting full model:
Iteration 0: log likelihood = -20828.813 (not concave)
Iteration 1: log likelihood = -20762.625
Iteration 2: log likelihood = -20743.905
Iteration 3: log likelihood = -20743.558
Iteration 4: log likelihood = -20743.557
Mixed effects regression model Number of obs = 7,139
Log likelihood = -20743.557
| Coef. | Std. Err. | z | P>z | [95% Conf. | Interval] | |
| dlevels: | ||||||
| time | .0008842 | .0004317 | 2.05 | 0.041 | .0000381 | .0017302 |
| M1[id] | 1 | . | . | . | . | . |
| _cons | 7.252385 | .122254 | 59.32 | 0.000 | 7.012771 | 7.491998 |
| sd(resid.) | 3.816123 | .0326449 | 3.752674 | 3.880646 | ||
| mstime: | ||||||
| EV[] | .2120668 | .0781024 | 2.72 | 0.007 | .0589888 | .3651447 |
| _cons | -8.003838 | .8442573 | -9.48 | 0.000 | -9.658552 | -6.349124 |
| log(gamma) | -.3646416 | .1064292 | -3.43 | 0.001 | -.5732389 | -.1560442 |
| id: | ||||||
| sd(M1) | 1.704884 | .0925013 | 1.532892 | 1.896174 | ||
As far as I can see there is a significant association between the biomarker and the time, seeing that biomarker is slightly increasing over the first year of follow-up (0.001, CI 0.000, 0.002), which should be the opposite.
Moreover, there is an association between high level of biomarker with an increased risk of death (0.212, CI 0.059, 0.365)
Is that right? Do you have any suggestion to improve this model?
I would like to introduce even other biomarker with different scale and add others confounder, but I would just make sure first that the simpler model is right
Thank you very much for your help!
