How to understand the causal inference in this method?

[zhiyang0310]

I have a confusion. In my memory, LSTM can only extract correlation. How to understand the causal inference in this method? Quote from the paper: "Building upon well-established causal inference and deep learning methods, our framework emulates randomized clinical trials for drugs present in a large-scale medical claims database. "

[ruoqi-liu]

I have a confusion. In my memory, LSTM can only extract correlation. How to understand the causal inference in this method? Quote from the paper: "Building upon well-established causal inference and deep learning methods, our framework emulates randomized clinical trials for drugs present in a large-scale medical claims database. "

Hello,
As illustrated in our paper, we proposed a deep learning method (based on LSTM) for propensity score estimation (inverse probability of treatment weighting) to adjust for confounding bias. The LSTM is used to model the temporal longitudinal observational patient data and compute the propensity scores, not to understand the causal relationships.
Best,
Ruoqi

[zhiyang0310]

Thanks for your reply. In this paper, where did you establish the causal relationship between drugs and disease? Or there is no causal relationship? On Oct 17, 2023, at 9:54 PM, Ruoqi Liu ***@***.***> wrote:  I have a confusion. In my memory, LSTM can only extract correlation. How to understand the causal inference in this method? Quote from the paper: "Building upon well-established causal inference and deep learning methods, our framework emulates randomized clinical trials for drugs present in a large-scale medical claims database. " Hello, As illustrated in our paper<https://rdcu.be/cc2CP>, we proposed a deep learning method (based on LSTM) for propensity score estimation (inverse probability of treatment weighting) to adjust for confounding bias. The LSTM is used to model the temporal longitudinal observational patient data and compute the propensity scores, not to understand the causal relationships. Best, Ruoqi — Reply to this email directly, view it on GitHub<#6 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AHGPDBVSXC7HG52F7FFP33DX72EX7AVCNFSM6AAAAAA6DN5F3GVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONRWGQ3DSMRRHA>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[ruoqi-liu]

Thanks for your reply. In this paper, where did you establish the causal relationship between drugs and disease? Or there is no causal relationship? On Oct 17, 2023, at 9:54 PM, Ruoqi Liu @.> wrote:  I have a confusion. In my memory, LSTM can only extract correlation. How to understand the causal inference in this method? Quote from the paper: "Building upon well-established causal inference and deep learning methods, our framework emulates randomized clinical trials for drugs present in a large-scale medical claims database. " Hello, As illustrated in our paperhttps://rdcu.be/cc2CP, we proposed a deep learning method (based on LSTM) for propensity score estimation (inverse probability of treatment weighting) to adjust for confounding bias. The LSTM is used to model the temporal longitudinal observational patient data and compute the propensity scores, not to understand the causal relationships. Best, Ruoqi — Reply to this email directly, view it on GitHub<#6 (comment)>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AHGPDBVSXC7HG52F7FFP33DX72EX7AVCNFSM6AAAAAA6DN5F3GVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONRWGQ3DSMRRHA. You are receiving this because you authored the thread.Message ID: @.>

I don't quite understand your question. Are you asking what's the causal graph and how we compute the causal effects?

• In our setting, the causal graph contains three main elements: 1) treatments, which are drug ingredients; 2) potential confounders, which are historical medications and diagnoses before the drug index; 3) outcomes: which are important disease outcomes. (You may also refer to more details of study design in our paper).
• We adjusted for the influence of confounders on treatment assignments through the propensity score method (IPTW) and computed the causal effects as the weighted difference between the treated and control outcomes.

[zhiyang0310]

Yes, I mean causal graph. 👌 On Oct 17, 2023, at 10:25 PM, Ruoqi Liu ***@***.***> wrote:  Thanks for your reply. In this paper, where did you establish the causal relationship between drugs and disease? Or there is no causal relationship? On Oct 17, 2023, at 9:54 PM, Ruoqi Liu @.> wrote:  I have a confusion. In my memory, LSTM can only extract correlation. How to understand the causal inference in this method? Quote from the paper: "Building upon well-established causal inference and deep learning methods, our framework emulates randomized clinical trials for drugs present in a large-scale medical claims database. " Hello, As illustrated in our paperhttps://rdcu.be/cc2CP, we proposed a deep learning method (based on LSTM) for propensity score estimation (inverse probability of treatment weighting) to adjust for confounding bias. The LSTM is used to model the temporal longitudinal observational patient data and compute the propensity scores, not to understand the causal relationships. Best, Ruoqi — Reply to this email directly, view it on GitHub<#6 (comment)<#6 (comment)>>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AHGPDBVSXC7HG52F7FFP33DX72EX7AVCNFSM6AAAAAA6DN5F3GVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONRWGQ3DSMRRHA. You are receiving this because you authored the thread.Message ID: @.> I don't quite understand your question. Are you asking what's the causal graph and how we compute the causal effects? * In our setting, the causal graph contains three main elements: 1) treatments, which are drug ingredients; 2) potential confounders, which are historical medications and diagnoses before the drug index; 3) outcomes: which are important disease outcomes. (You may also refer to more details of study design in our paper). * We adjusted for the influence of confounders on treatment assignments through the propensity score method (IPTW) and computed the causal effects as the weighted difference between the treated and control outcomes. — Reply to this email directly, view it on GitHub<#6 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AHGPDBR7EJUSWQVOIPHS2R3X72IN5AVCNFSM6AAAAAA6DN5F3GVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONRWGUZTAMJUGU>. You are receiving this because you authored the thread.Message ID: ***@***.***>