New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Confusing about the time in MonteCarloGFormula #101
Comments
Hi @Jnnocent
Can you tell me a little more about your data set and study question? It may help me answer your question more clearly. Depending on the data, there may be a better estimator to use. In my experience, I have found |
Thank you! I know more about this article and the code. And I study the MonteCarloGFormula by the following paper Intervening on risk factors for coronary heart disease: an application of the parametric g-formula . On page 1601, it describes the algorithm that it builds the model for each 2-year , which is different form that fitting the same model for all time points (t = {1, 2, 3, ... k}) On page 5 in Keil et al 2014, it is said that
So, I feel confused about how to build the model that fitting the same model for all time points (t = {1, 2, 3, ... k}) or fitting the differsent models for each time points (t = {1, 2, 3, ... k}). |
The Taubman paper and Keil paper both use what is referred to as pooled logistic regression. This is the same approach that The pooled logistic regression process works by fitting a single model to the full data with a term for time (
where
Our outcome model could look something like the following, The pooled logistic models are different for each variable. Following Taubman, the However, |
The As Taubman has it written, I believe there would be no |
I get it! Thanks for you answering in details! ^_^ |
No problem @Jnnocent ! Got a chance to read through Taubman 2009 again. Below are some general thoughts / comments I had, mostly discussion of what is and isn't possible using
Hope this helps |
Based on your risk function for the true data, it doesn't look like that many events occur. Is this correct? It looks like there is only 6 or so events. With so few events, it will be hard to estimate the Monte Carlo g-formula. I wouldn't trust the results from the g-formula. You can try to use year data instead to see if that helps.
Can you explain this a litter further? Based on the graph it looks like original: 0.2 and simulation: 0.45 |
Ahh yeah, so Something you may want to consider is using |
Thank you! I would like to read the paper first! It seems simpler |
The iterative conditional g-formula allows for continuous confounders. I believe they say the covariates are discrete due to how they write it. The last cumulative product is written as Pr(...). For continuous covariates, that term would become f(...). In their application, they have several continuous variables they adjust for (weight, height, risk score)
Not currently. The issue is that confidence intervals with For the interative conditional g-formula, the model needs to output the probabilities, not 0 / 1, for it to estimate correctly. Due to the |
Hi,I am confusing about how to use the time and history data. It seems that we should make a model with data of questionnaires k-1 and k-2, but I couldn't find such code implementation. I found you used data of time k only while fitting you model.
#176 exposure_model
#207 outcome_model
#274 add_covariate_model
The text was updated successfully, but these errors were encountered: