Real-World Data in Real Life by Axle Training

Response

2025-06-18 17:22:57 UTC

Each aithor if the oublivation are responsibke to understand wheree the limits of causation snd assioatikn aligns in my research i tru to be extreemyl conservedd in stating that the finds are a cause however if with enough statisrocs Prow the the assim can be ande that the fibds are casusal

Madison Far sworth

Response

2025-07-15 23:09:06 UTC

One of the major limitation is selection biases of the participants. To handle this, the authors used propensity score weighting to balance possible cofounding features.

Christina Chance

Response

2025-07-15 23:12:07 UTC

The authors used two different causal inference models to validate the outcomes as well as assessed negative control outcomes to which aligned with expectations and supported the methods efficacy

Christina Chance

Response

2025-07-15 23:13:40 UTC

One of the biggest things we can take away and use is the validation of method efficacy and utilizing other methods and approaches to validate our outcomes (which I think my team did a really good job of).

Christina Chance

Question 1: How do the authors validate the efficacy of their causal inference methods?

2025-07-17 13:08:43 UTC

They validate their pipeline by applying it to “negative control” conditions which are outcomes expected not to be affected by antidepressant use. By demonstrating that neither propensity-score weighting nor their novel embedding-based approach identified spurious effects in these negative-outcome cases, they provide proof that their methods correctly distinguish causal signals from noise.

--Loni Taylor

Question 2: Discuss the limitations of using observational data for causal inference and how the authors addressed these limitations.

2025-07-17 13:11:32 UTC

Observational EHR data pose several challenges:

High‑dimensional confounding
Lack of randomization
Measurement bias

The authors tackled these by:

Propensity-score weighting
Node2Vec embedding o
Negative control outcome experiments

Dual modeling approaches, dimension reduction, and falsification testing improve but cannot fully eliminate inherent biases in observational data.

--Loni Taylor

Question 3: How does this article inform your team project?

2025-07-17 13:15:32 UTC

This article provides a practical blueprint for applying causal inference to EHR or other complex observational datasets by:

Demonstrating the value of using both traditional propensity scores and modern embedding-based methods
Adopting embeddings for medical codes in order to provide an efficient alternative to one-hot encoding when dealing with large vocabularies.
Incorporating falsification tests into our evaluation plan to strengthen our causal claims.
Highlighting the trade-offs between simpler, transparent models and more complex, potentially overfitted ones.

Overall, it guides us to design our project with multiple causal methods, embed high-dimensional features thoughtfully, and validate rigorously via control outcomes, encouraging us to explore hybrid pipelines in our work.

--Loni Taylor

Authors can check for balance and SMD. Amirah.

2025-07-24 15:47:22 UTC

Question 3: How does this article inform your team project?

2025-07-30 23:31:04 UTC

I found these particular techniques to be interesting. I plan to consider them in our team project as well as to bounce ideas with our team's data scientist to see if we have incorporated them already or see if there is a way to incoporate them. ~Briana Lettsome

Question 1: How do the authors validate the efficacy of their causal inference methods?

2025-07-31 00:27:09 UTC

The authors validate the efficacy of their causal inference methods by using negative controls. With this method, they compare outcomes that are not thought to be associated with the exposure with their outcomes of interest. ~Briana Lettsome

Question 2: Discuss the limitations of using observational data for causal inference and how the authors addressed these limitations.

2025-07-31 00:33:39 UTC

One limitation of using observational data for causal inference is addressing unmeasured confounding which can have an effect on model estimates. Another limitation with the use of observational data for causal inference is that, unlike a randomized clinical trial which can capture true causal effects, results from the former estimates a causal effect rather than illustrate a true one. The authors addressed these by using high-dimensional propensity score weighting, using logistic regression and random forest, and embedding medical code representation, which the authors mentioned could address poorly measured confounders. The authors also used negative controls to determine to verify their results. ~ Briana Lettsome