Comments (4)
drizopoulos
commented on August 16, 2024
Thanks for your e-mail and interest in my package. Regarding your
question, please check the following:
* If there are any event times that are exactly equal to zero, set those
to a small positive number, e.g., 1e-05.
* Order the longitudinal data according the subject id variable, and
order the time points within each subject with respect to time. You can
use the order() function for this, e.g., my_data <-
my_data[order(my_data$id, my_data$time), ].
* Remove any missing data you have in the variable you are going to use
in the analysis beforehand, e.g., my_data_nomiss <-
my_data[complete.cases(my_data[c("long_outcome", "time", "age", "bmi",
"sex")]), ].
* Assuming the you have the data as described above, and that the event
time variable is also in the longitudinal dataset, construct the dataset
for the survival submodel using my_data_surv <-
my_data[!duplicated(my_data$id), ].
* If you have longitudinal measurements after the event times, these are
not allowed in my package.
* Set both model = TRUE and x = TRUE in the call to coxph().
* If you have any missing data, then first exclude them and then run the
linear mixed and Cox models.
* It is better to use the adaptive Gauss-Hermite rule, i.e., in the
definition of the 'method' argument use, e.g., "piecewise-PH-aGH"
instead of "piecewise-PH-GH".
* It is better to use the piecewise-constant or spline-approximated
baseline hazard instead of the unspecified hazard (i.e., method =
"piecewise-PH-aGH" or "spline-PH-aGH" instead of "Cox-PH-aGH").
* If the knots for the B-spline or the piecewise-constant baseline
hazard are placed in parts where there are no or very few events. You
can override the default placing of the knots by manually setting them
at specific time points based on, e.g., the Kaplan-Meier estimator. In
this case you can use the 'knots' argument.
* The time scale must be the same in both the longitudinal and survival
submodels.
* If in the linear mixed model it appears that some coefficients have
very different magnitudes than others, then try centering or scaling
some of the offending covariates.
…On 5/20/2019 9:11 AM, Chaochen Wang wrote:
Hi, thanks for the amazing package. I am wondering in the two parts of
the joint modeling framework, can they accept patients entering the
study/observation from different time points?
I am dealing with data from a patient database, therefore, in a
real-world setting, patients come to visit the doctor and with their
blood sample drawn at totally different time points/intervals. Some
patients started early some with their initial recording later.
So my question is whether the survival part of the Joint Modeling within
this package can be fitted using the following form?
|coxFit.DF <- coxph(Surv(start, stop, event) ~ sex + age + group, data,
x = TRUE) |
Because when I fitted the above Cox model with the difference between
the times they were followed |time = stop-start|:
|coxFit.DF <- coxph(Surv(time, event) ~ sex + age + group, data, x = TRUE) |
There were errors telling me that some patients had observations after
the events. And when I use the |Surv(start, stop, event)| one, error
message also came up and ask me to refit the Cox model using |x = TRUE|
argument which I had already added.
One of the flexibility of the time-dependent Cox regression model is
that patients are allowed to enter the study from different start
points, If there is also a way to allow for this in the Joint Modeling
framework, I would like to see how and it will be greatly appreciated.
Many thanks.
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fdrizopoulos%2FJM%2Fissues%2F13%3Femail_source%3Dnotifications%26email_token%3DADE7TT6CD3KRU3PO5YZI2XTPWJFL3A5CNFSM4HN7CRR2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4GUUX7EQ&data=02%7C01%7Cd.rizopoulos%40erasmusmc.nl%7C56bd8c5bf49e49eac31f08d6dcf272ed%7C526638ba6af34b0fa532a1a511f4ac80%7C0%7C0%7C636939331201721564&sdata=M4Y%2FsXotdtHzuNQEtmVHN8rEzRlh7BH7RwfnkFoUaUQ%3D&reserved=0>,
or mute the thread
<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FADE7TT6LWGUJ53H2MYMJNVLPWJFL3ANCNFSM4HN7CRRQ&data=02%7C01%7Cd.rizopoulos%40erasmusmc.nl%7C56bd8c5bf49e49eac31f08d6dcf272ed%7C526638ba6af34b0fa532a1a511f4ac80%7C0%7C0%7C636939331201731568&sdata=9RDYt9uk%2F4nmwnuF1cjs1umwcde7m5JoliVaa%2FdUT9I%3D&reserved=0>.
--
Dimitris Rizopoulos
Professor of Biostatistics
Department of Biostatistics
Erasmus University Medical Center
Address: PO Box 2040, 3000 CA Rotterdam, the Netherlands
Tel: +31/(0)10/7043478
Fax: +31/(0)10/7043014
Web (personal): http://www.drizopoulos.com/
Web (work): http://www.erasmusmc.nl/biostatistiek/
Blog: http://iprogn.blogspot.nl/
from jm.
winterwang
commented on August 16, 2024
Hi, I have tried the following with what you have suggested (twice). There is no patient with 0 follow-up time. Missing data were also excluded from the beginning.
Because in your book, when comparing with the time-dependent model, JM should have given us a less underestimated result. However, it seems that it was not the case in my data. The time-dependent model which used a biomarker measured at each visit paid by patients as the main predictor is as follows:
Call:
coxph(formula = Surv(as.numeric(START0), as.numeric(LaboDate),
MACE) ~ logMarker + BASE_AGE + SEX, data = ANA_df,
ties = "breslow")
n= 689548, number of events= 1959
coef exp(coef) se(coef) z Pr(>|z|)
logMarker 0.638440 1.893525 0.138547 4.608 4.06e-06 ***
BASE_AGE 0.024577 1.024881 0.002027 12.125 < 2e-16 ***
SEX -0.535464 0.585398 0.050854 -10.530 < 2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
exp(coef) exp(-coef) lower .95 upper .95
I(log(RESULT_MEAN)) 1.8935 0.5281 1.4432 2.4843
BASE_AGE 1.0249 0.9757 1.0208 1.0290
SEX 0.5854 1.7082 0.5299 0.6468
Concordance= 0.605 (se = 0.006 )
Likelihood ratio test= 252.6 on 3 df, p=<2e-16
Wald test = 238.8 on 3 df, p=<2e-16
Score (logrank) test = 239.5 on 3 df, p=<2e-16
So I was expecting to see a stronger relationship between this biomarker and the hazard of the event.
However, when I fitted the mixed effect linear model and the Cox regression model with your approach, I found that the association is gone:
lmeFit.DF <- lme(logMarker ~ obstime + SEX + BASE_AGE,
random = ~ obstime | PatientId, data = ANA_df)
coxFit.DF <- coxph(Surv(Time, MACE) ~ SEX + BASE_AGE, data = DF_1strow,
x = TRUE, model = TRUE)
jointFit.DF <- jointModel(lmeFit.DF, coxFit.DF, timeVar = "obstime",
method = "piecewise-PH-aGH")
# Call:
# jointModel(lmeObject = lmeFit.DF, survObject = coxFit.DF, timeVar = "obstime",
# method = "piecewise-PH-aGH")
#
# Data Descriptives:
# Longitudinal Process Event Process
# Number of Observations: 689548 Number of Events: 1959 (3.7%)
# Number of Groups: 52511
#
# Joint Model Summary:
# Longitudinal Process: Linear mixed-effects model
# Event Process: Relative risk model with piecewise-constant
# baseline risk function
# Parameterization: Time-dependent
#
# log.Lik AIC BIC
# 589034 -1178032 -1177872
#
# Variance Components:
# StdDev Corr
# (Intercept) 0.1511 (Intr)
# obstime 0.0001 -0.4542
# Residual 0.0853
#
# Coefficients:
# Longitudinal Process
# Value Std.Err z-value p-value
# (Intercept) 2.0248 0.0037 552.4671 <0.0001
# obstime 0.0000 0.0000 -28.0565 <0.0001
# SEX 0.0066 0.0013 4.9324 <0.0001
# BASE_AGE -0.0014 0.0000 -27.9929 <0.0001
#
# Event Process
# Value Std.Err z-value p-value
# SEX -0.5044 0.0507 -9.9453 <0.0001
# BASE_AGE 0.0312 0.0022 14.0130 <0.0001
# Assoct 0.3949 0.4403 0.8969 0.3698
# log(xi.1) -13.8911 0.9143 -15.1927
# log(xi.2) -13.1572 0.9062 -14.5189
# log(xi.3) -12.9483 0.9031 -14.3368
# log(xi.4) -12.6816 0.9021 -14.0575
# log(xi.5) -12.4629 0.9027 -13.8065
# log(xi.6) -12.1712 0.9070 -13.4184
# log(xi.7) -9.6649 0.9121 -10.5962
#
# Integration:
# method: (pseudo) adaptive Gauss-Hermite
# quadrature points: 3
#
# Optimization:
# Convergence: 0
Do you have any suggestion on how to interpret this result? Maybe I can try adding more markers since they were measured at the same longitudinal time points.
from jm.
drizopoulos
commented on August 16, 2024
Indeed most often the bias in Cox is downwards but not always.
From: Chaochen Wang <[email protected]<mailto:[email protected]>>
Date: Friday, 24 May 2019, 9:13 AM
To: drizopoulos/JM <[email protected]<mailto:[email protected]>>
Cc: D. Rizopoulos <[email protected]<mailto:[email protected]>>, Comment <[email protected]<mailto:[email protected]>>
Subject: Re: [drizopoulos/JM] Can the joint modeling frame work accept patients with different start time ? (#13)
Hi, I have tried the following with what you have suggested (twice). There is no patient with 0 follow-up time. Missing data were also excluded from the beginning.
Because in your book, when comparing with the time-dependent model, JM should have given us a less underestimated result. However, it seems that it was not the case in my data. The time-dependent model which used a biomarker measured at each visit paid by patients as the main predictor is as follows:
Call:
coxph(formula = Surv(as.numeric(START0), as.numeric(LaboDate),
MACE) ~ logMarker + BASE_AGE + SEX, data = ANA_df,
ties = "breslow")
n= 689548, number of events= 1959
coef exp(coef) se(coef) z Pr(>|z|)
logMarker 0.638440 1.893525 0.138547 4.608 4.06e-06 ***
BASE_AGE 0.024577 1.024881 0.002027 12.125 < 2e-16 ***
SEX -0.535464 0.585398 0.050854 -10.530 < 2e-16 ***
…---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
exp(coef) exp(-coef) lower .95 upper .95
I(log(RESULT_MEAN)) 1.8935 0.5281 1.4432 2.4843
BASE_AGE 1.0249 0.9757 1.0208 1.0290
SEX 0.5854 1.7082 0.5299 0.6468
Concordance= 0.605 (se = 0.006 )
Likelihood ratio test= 252.6 on 3 df, p=<2e-16
Wald test = 238.8 on 3 df, p=<2e-16
Score (logrank) test = 239.5 on 3 df, p=<2e-16
So I was expecting to see a stronger relationship between this biomarker and the hazard of the event.
However, when I fitted the mixed effect linear model and the Cox regression model with your approach, I found that the association is gone:
lmeFit.DF <- lme(logMarker ~ obstime + SEX + BASE_AGE,
random = ~ obstime | PatientId, data = ANA_df)
coxFit.DF <- coxph(Surv(Time, MACE) ~ SEX + BASE_AGE, data = DF_1strow,
x = TRUE, model = TRUE)
jointFit.DF <- jointModel(lmeFit.DF, coxFit.DF, timeVar = "obstime",
method = "piecewise-PH-aGH")
# Call:
# jointModel(lmeObject = lmeFit.DF, survObject = coxFit.DF, timeVar = "obstime",
# method = "piecewise-PH-aGH")
#
# Data Descriptives:
# Longitudinal Process Event Process
# Number of Observations: 689548 Number of Events: 1959 (3.7%)
# Number of Groups: 52511
#
# Joint Model Summary:
# Longitudinal Process: Linear mixed-effects model
# Event Process: Relative risk model with piecewise-constant
# baseline risk function
# Parameterization: Time-dependent
#
# log.Lik AIC BIC
# 589034 -1178032 -1177872
#
# Variance Components:
# StdDev Corr
# (Intercept) 0.1511 (Intr)
# obstime 0.0001 -0.4542
# Residual 0.0853
#
# Coefficients:
# Longitudinal Process
# Value Std.Err z-value p-value
# (Intercept) 2.0248 0.0037 552.4671 <0.0001
# obstime 0.0000 0.0000 -28.0565 <0.0001
# SEX 0.0066 0.0013 4.9324 <0.0001
# BASE_AGE -0.0014 0.0000 -27.9929 <0.0001
#
# Event Process
# Value Std.Err z-value p-value
# SEX -0.5044 0.0507 -9.9453 <0.0001
# BASE_AGE 0.0312 0.0022 14.0130 <0.0001
# Assoct 0.3949 0.4403 0.8969 0.3698
# log(xi.1) -13.8911 0.9143 -15.1927
# log(xi.2) -13.1572 0.9062 -14.5189
# log(xi.3) -12.9483 0.9031 -14.3368
# log(xi.4) -12.6816 0.9021 -14.0575
# log(xi.5) -12.4629 0.9027 -13.8065
# log(xi.6) -12.1712 0.9070 -13.4184
# log(xi.7) -9.6649 0.9121 -10.5962
#
# Integration:
# method: (pseudo) adaptive Gauss-Hermite
# quadrature points: 3
#
# Optimization:
# Convergence: 0
Do you have any suggestion on how to interpret this result? Maybe I can try adding more markers since they were measured at the same longitudinal time points.
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fdrizopoulos%2FJM%2Fissues%2F13%3Femail_source%3Dnotifications%26email_token%3DADE7TT3CP6GEPRMFGROI773PW6ITBA5CNFSM4HN7CRR2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODWELSDI%23issuecomment-495499533&data=02%7C01%7Cd.rizopoulos%40erasmusmc.nl%7C6463bd1fbd464d5e945b08d6e017612a%7C526638ba6af34b0fa532a1a511f4ac80%7C0%7C0%7C636942788350149318&sdata=UxZpQqdy25tbNWmB8zbkBOkubmmbgyrNJtY39%2BZfhnk%3D&reserved=0>, or mute the thread<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FADE7TT7YAE6XR4244T7VAHLPW6ITBANCNFSM4HN7CRRQ&data=02%7C01%7Cd.rizopoulos%40erasmusmc.nl%7C6463bd1fbd464d5e945b08d6e017612a%7C526638ba6af34b0fa532a1a511f4ac80%7C0%7C0%7C636942788350159311&sdata=tIQyVDzeIMflKNS%2BfgufwDKkr8a%2FDyDFToa9ybqwuFs%3D&reserved=0>.
from jm.
winterwang
commented on August 16, 2024
OK, thanks for your help. So it was not because of any error in my coding?
from jm.
Related Issues (20)
- Query on getting back association parameter from data simulation in weibull-ph-agh
- Error in exp(coefs[1]) : non-numeric argument to mathematical function- fail to run CompRisk HOT 5
- Error in if (w_dif < specs$tol || iter == specs$maxiter) break : missing value where TRUE/FALSE needed
- Error in if (w_dif < specs$tol || iter == specs$maxiter) break : missing value where TRUE/FALSE needed
- Simulating data and GLMMPQL using JM
- issues in running survfitJM
- Error in aeqSurv(Y) : aeqSurv exception, an interval has effective length 0
- Error in d * log.hazard + log.survival : non-conformable arrays In addition: Warning message: In wk * rep(x$P, each = nk) : longer object length is not a multiple of shorter object length HOT 3
- Error in optim(thetas, opt.survPC, gr.survPC, method = "BFGS", control = list(maxit = if (it < : non-finite value supplied by optim HOT 3
- `survfitJM` expects newdata to be ordered by subject, and by time within each subject, else things go wrong
- Error in in optim(thetas, LogLik.splineGH, Score.splineGH, method = "BFGS", HOT 1
- ***caught segfault*** caused by Error in names(fit$coef) <- cname : 'names' attribute [4] must be the same length as the vector [2] HOT 5
- problem running fuction jointModel() in JM package HOT 1
- problem running fuction jointModel() in JM package with method = "Cox-PH-aGH" option HOT 1
- Question about the calculation of standard deviation in the "random effects parameterization"
- Extract fitted lme object from the jointModel HOT 2
- Error in if (t1 || t2) { : missing value where TRUE/FALSE needed
- code graphical time dependent slope
- sample sizes in the longitudinal and event processes differ.
- The interpretation of exp(Assoct) in JM
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from jm.