LLM-Extracted Covariates for Clinical Causal Inference: Rethinking Integration Strategies

Target trial emulation (TTE) enables causal questions to be studied with observational data when randomized controlled trials (RCTs) are infeasible. Yet treatment-effect methods often address causal estimation, missingness, and temporal structure separately, limiting their robustness in electronic health records (EHRs), where time-varying confounding and missing-not-at-random (MNAR) biomarkers can reach 50%--80%. We propose a two-stage pipeline for treatment effect estimation from incomplete longitudinal EHRs. First, CausalFlow-T, a directed acyclic graph (DAG)-constrained normalizing flow with long short-term memory (LSTM)-encoded patient history, performs exact invertible counterfactual inference, avoiding approximation errors from variational inference and separating confounding through explicit causal structure. Ablations on four synthetic and one semi-synthetic benchmark with known counterfactuals show that DAG constraints and exact inference address distinct failure modes: neither compensates for the other. Second, because CausalFlow-T requires completed inputs, we introduce an LLM-driven evolutionary imputer that proposes executable imputation operators rather than individual entries, and evaluate it with three large language model (LLM) backends, including two open-source models. Across 30%--80% MNAR missingness, this imputer achieves the best pooled rank over biomarker and causal metrics, leading in point-wise accuracy and temporal extrapolation while preserving average treatment effect (ATE) recovery as statistical baselines degrade. On Swiss primary-care EHRs from adults with type 2 diabetes initiating a GLP-1 receptor agonist or SGLT-2 inhibitor, the pipeline estimates a per-protocol weight-loss difference of -0.98 kg [95% CI -1.01, -0.96] favoring GLP-1 receptor agonists, consistent with randomized evidence and obtained from realistically incomplete real-world EHRs.

The training data of large language models (LLMs) comprises a wide range of biomedical literature, reflecting data from many different patient populations. We investigate how it might be possible to recover information on correlation and causal links between patient characteristics, as a key building block for medical decision making. To avoid the pitfalls of direct elicitation, we propose an approach based on structured comparison questions, specifically patient comparison triplet questions. This is combined with a statistical model for the LLM representation that provides estimates of correlations without access to activations or model internals. Intuitively, we consider how similarity decisions of LLMs based on a first variable are affected by providing information on a second variable for one of the patients being assessed. We then induce prompt-level environment shifts to obtain correlation estimates for different subpopulations, which enables an invariant causal prediction (ICP) approach to obtain conservative candidate parent links. We demonstrate the method in two clinical domains, chronic obstructive pulmonary disease (COPD) and multiple sclerosis (MS). Across prompted environments, the elicited correlations are smooth, stable, and clinically interpretable, yet vary in a statistically significant way that supports downstream invariance testing, such that ICP provides a small set of candidate invariant parent links. These results show that indirect elicitation via triplet comparisons can recover meaningful association structure from LLMs and offer a cautious route from implicit correlations to causal statements that are congruent with LLM answering patterns.

Existing causal-inference benchmarks for LLMs mostly score method descriptions or whether generated code runs, not whether the executed workflow recovers the target causal estimate. CausalVerify studies this verification problem for structured econometric causal-estimation workflows by separating realistic interpretation from verifiable computation. It pairs 259 published economics papers (reconstructed research question, data description, institutional context) with 100 fixed-seed synthetic scenarios that realise CSV datasets for difference-in-differences, event study, instrumental variables, and regression discontinuity designs. Experiment A (real-paper text agreement) scores method-family and direction agreement against four-LLM consensus labels. Experiment B (synthetic execution) runs model-written R code and checks whether the extracted treatment-effect estimate matches a canonical estimator on the same realised dataset; this execution-grounded correctness layer is L2b+, distinct from L2b, which records only whether the code executes. A calibration arm asks whether self-reported confidence separates correct from incorrect workflows. On Experiment B, seven LLMs reach L2b+ pass rates of 10% to 88% at the default 50% tolerance, and 66 of the 426 workflows that execute (15.5%) return a wrong estimate. Execution ranking (L2b) agrees with L2b+ far better than text-direction scoring (L4): Kendall $τ=0.81$ and Spearman $ρ=0.93$, versus Kendall $τ$ between $-0.20$ and $0.10$ for L4. Llama-3.3-70B-Instruct shows the same qualitative gap, and reported confidence does not reliably separate correct from incorrect workflows. The claims are confined to standardized single-shot workflows in these four design families under the evaluated R backend and model panel; the benchmark does not measure general causal-inference ability. Code, data, cached outputs, and a datasheet are released.