Implicit Framing in Obstetric Counseling Notes: A Grounded LLM Pipeline on a VBAC-Eligible Cohort

Baris Karacan, Barbara Di Eugenio, Patrick Thornton, Joanna Tess, Subhash Kumar Kolar · 2026

Read Implicit Framing in Obstetric Counseling Notes: A Grounded LLM Pipeline on a VBAC-Eligible Cohort by Baris Karacan, Barbara Di Eugenio, Patrick Thornton, Joanna Tess, Subhash Kumar Kolar (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-23059.txt.

Introduction

Karacan and four co-authors analyze implicit framing in obstetric counseling notes for patients cleared for both vaginal birth after cesarean and repeat cesarean. A cohort of 2,024 history and physical notes is built from tabular clinical data plus a language-model mining pipeline tied to grounded, verbatim evidence. A zero-shot framework then sorts advice segments into framing groups for how birth options are put. Risk-focused language takes a much larger share in repeat cesarean notes than in vaginal birth notes, with differences backed by statistical testing.

Catalog record

CREATORBaris Karacan, Barbara Di Eugenio, Patrick Thornton, Joanna Tess, Subhash Kumar Kolar
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE48552
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.23059

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Implicit Framing in Obstetric Counseling Notes: A Grounded LLM Pipeline on a VBAC-Eligible Cohort Baris Karacan, Barbara Di Eugenio, Patrick Thornton, Joanna Tess, Subhash Kumar Kolar Abstract Clinical framing—the linguistic manner in which clinical information is presented—can influence patient understanding and decision-making, with important implications for healthcare outcomes. Obstetrics is a high-stakes domain in which physicians counsel patients on delivery mode choices such as vaginal birth after cesarean (VBAC) and repeat cesarean section (RCS), yet counseling language remains underexplored in large-scale clinical text analysis. In this work, we analyze physician counseling language in 2,024 obstetric history and physical narratives for a rigorously defined cohort of patients for whom both VBAC and RCS were clinically viable options. To control for confounding due to medical contraindications, we first construct a VBAC-eligible cohort using structured clinical data supplemented by a large language model (LLM)-based extraction pipeline constrained to grounded, verbatim evidence from free-text narratives. We then apply a zero-shot LLM framework to categorize counseling segments into predefined framing categories capturing how physicians linguistically present delivery options. Our analysis reveals a significant difference in counseling framing distributions between VBAC and RCS notes; risk-focused language accounts for a substantially larger share of counseling segments in RCS documentation than in VBAC, with category-level differences confirmed by statistical testing, highlighting the value of controlled LLM-based framing analysis in obstetric care. I Introduction For women with a prior cesarean, the decision between attempting a vaginal birth after cesarean (VBAC) or undergoing a repeat cesarean section (RCS) is often preference-sensitive, involving trade-offs among maternal and neonatal risks, recovery, and future reproductive plans [ 1 ] . Because both options can be clinically appropriate for many patients, counseling plays a central role in how patients understand these trade-offs. Prior work shows that linguistic framing—how equivalent information is presented in terms of risks, benefits, or recommendations—can shape perceptions and decision-making [ 2 , 3 ] . In obstetric documentation, differences in how clinicians describe VBAC and RCS may therefore reflect systematic framing patterns, motivating large-scale analysis of counseling language when both options are viable. Because not all patients who underwent RCS are medically eligible for VBAC, including ineligible cases would confound the linguistic comparisons by mixing notes in which VBAC counseling was not clinically appropriate. We therefore focus on a VBAC-eligible cohort—patients for whom both delivery modes were clinically viable at the time of counseling. This cohort-controlled design helps attribute observed differences in counseling language between VBAC and RCS documentation to variation in counseling framing, rather than to medical contraindications that would preclude VBAC discussion. Operationalizing this cohort-controlled analysis at scale requires extracting clinically relevant evidence from free-text obstetric history and physical (H&P) narratives. Key eligibility details (e.g., prior incision type or VBAC contraindications) are often missing or incomplete in structured fields but appear in narrative documentation, motivating the use of natural language processing (NLP) methods to recover eligibility cues from the text. At the same time, counseling itself is expressed primarily in unstructured language, where framing is conveyed through subtle lexical and discourse choices that are difficult to quantify with rule-based approaches alone. We therefore use large language models (LLMs) in two complementary roles: (i) to extract grounded evidence of VBAC eligibility from narratives to support cohort construction, and (ii) to categorize counseling segments into predefined framing categories for large-scale measurement of how delivery options are presented in clinical documentation. Motivated by the need for cohort-controlled analysis and scalable measurement of counseling language in free-text obstetric narratives, we make the following contributions: 1. VBAC-Eligible Cohort Construction: We construct a cohort of patients for whom both VBAC and RCS were clinically viable at the time of counseling by combining structured eligibility criteria with LLM-based extraction of grounded, verbatim eligibility evidence from obstetric H&P narratives, reducing confounding from VBAC contraindications. 2. Reliability Analysis of LLM-Based Eligibility Extraction: We evaluate the reliability of LLM-based extraction under different models and prompt configurations by analyzing hallucination behavior, providing practical evidence on how prompt complexity and model scale affect source-grounded information extraction from clinical narratives. 3. Cohort-Controlled Framing Quantification in Counseling Documentation: We apply a zero-shot LLM framework to categorize counseling segments into predefined framing categories and statistically compare framing distributions between VBAC and RCS documentation, complemented by expert review on a subset of segments. The paper is organized as follows. Related work is reviewed in Sec. II . Sec. III describes the data and cohort eligibility criteria. Sec. IV presents the proposed methodology for eligibility evidence extraction and framing analysis. Results are reported in Sec. V . Sec. VI concludes with limitations and future work. II Related Work Implicit framing refers to subtle differences in wording, emphasis, or presentation that can lead audiences to evaluate equivalent information differently. Classic demonstrations show that mathematically equivalent descriptions can produce systematically different judgments and choices [ 4 ] . In healthcare, implicit framing captures the mismatch between an intention to present options neutrally and the often unintentional ways in which emphasis, omission, and evaluative wording can shape patients’ perceptions and decisions [ 5 ] . Empirical studies further suggest that such framing and recommendations have been shown to differ by patients’ socioeconomic status and race/ethnicity [ 6 ] . Obstetric decision-making provides another critical example: the choice between VBAC and RCS is preference-sensitive, yet counseling is often perceived as inadequate or directive. In a study of 68 women, Folsom et al. [ 7 ] reported that although 18 had a calculated chance of more than 70% for successful VBAC, only two met criteria for adequate counseling, and many participants perceived their physicians as recommending RCS. More recently, Kanjanakaew et al. [ 8 ] emphasize that when VBAC is not explicitly presented as an option, patients may be steered toward RCS by default. Together, these findings motivate systematic examination of how VBAC and RCS options are framed in routine counseling documentation. Clinical documentation is intended to monitor a patient’s condition and support communication among members of the healthcare team [ 9 ] . With widespread adoption of electronic health records (EHRs)—now used by more than 80% of U.S. hospitals—these narratives have become a common source for analysis, and prior work notes benefits of the transition from paper to electronic records such as reduced prescribing and documentation errors and improved information flow [ 10 , 11 ] . EHRs include both structured fields and unstructured free-text notes; while structured fields are standardized, free text contains rich clinical detail and discussions not captured in coded fields, but its unstandardized form makes extraction and analysis difficult [ 12 ] . Free-text notes are often information-dense and can include abbreviations and acronyms, typographical errors, inconsistencies, semantic ambiguities, and incomplete information [ 13 , 14 ] . For example, the acronym ”MR” may refer to ”mental retardation” but could also be interpreted as ”mitral regurgitation” without sufficient context [ 15 ] . Clinicians often organize free-text documentation through the SOAP framework (Subjective, Objective, Assessment, Plan), where patient counseling is generally recorded in the ”Plan” section; however, explicit Plan sections are inconsistently labeled and may be merged under combined headings such as ”Assessment and Plan”, making counseling documentation difficult to analyze systematically from raw notes [ 16 , 17 , 18 , 19 ] .

Read the complete text