Registry & Real-World Data Digest — Issue 02

Renee M Cholyway, William C Kethman

Summary: This narrative review in Clinics in Colon and Rectal Surgery surveys how electronic medical records and artificial intelligence are being applied to clinical decision-making, covering prehabilitation, sepsis detection, protocol-driven care, large language models for documentation, and patient triage. The authors highlight both the expanding capabilities of AI-augmented EMRs and acknowledged limitations including data scarcity for rare conditions and opacity in algorithm development. No original data are presented; this is a descriptive synthesis aimed at a surgical subspecialty audience.

Why it matters: For clinicians and healthcare informatics professionals, the review consolidates a wide landscape of EMR-AI applications into a single readable reference. The explicit acknowledgment that data scarcity poses problems for rare diseases and uncommon complications is a meaningful point often underemphasised in enthusiastic AI literature. However, because the piece does not advance methodology, generate new evidence, or rigorously evaluate existing evidence, its contribution is primarily educational rather than scientific.

Expert perspective: Narrative reviews of EMR and AI capabilities are proliferating nearly as fast as the technologies themselves, and this one is competent but not distinctive. The article covers well-trodden territory — sepsis early-warning systems, LLMs reducing documentation burden, patient portal engagement — without critically appraising the underlying evidence base for any of these applications. Readers should note that many AI clinical decision-support tools carry substantial implementation challenges: alert fatigue, workflow disruption, algorithmic bias, and performance degradation when models are deployed outside the populations on which they were trained. These concerns receive only passing acknowledgment.

The mention of data scarcity as a barrier to identifying rare complications is particularly relevant to rare-disease research communities, where small patient populations, heterogeneous genotypes, and complex multi-system disease make AI model training genuinely difficult. Registry data and federated learning approaches are natural responses to this challenge, but they are not discussed.

The claim that AI has "vastly broadened" EMR capabilities warrants scepticism. Many AI-embedded tools show promise in controlled settings but modest or mixed results at scale. The review does not distinguish between tools with robust prospective validation and those that remain largely aspirational. Transparency in algorithm development — cited as a limitation — is a patient safety and regulatory concern that deserves deeper treatment than it receives here.

The absence of a systematic search strategy, quality appraisal, or evidence grading means conclusions should be treated as expert opinion rather than synthesised evidence. Worth reading as a primer; not a definitive reference.

Relevance: Clinical Informatics and Health Data Methods

Source: https://pubmed.ncbi.nlm.nih.gov/42553947/

---

Neha Simha, Hannah Takasuka, Li-Ching Chen, et al.

Summary: This preprint reports a large-scale NLP analysis of 640,345 obstetric clinical notes from 26,178 pregnancies at a single academic medical centre, using a keyword-guided BERT classifier to detect stigmatising language. Stigmatising language was found in 47% of pregnancies, with significantly higher odds among Black patients compared to Asian and White patients, among those with lower educational attainment, and among pregnancies complicated by preterm birth. The study demonstrates that automated NLP methods can systematically quantify documentation bias at scale.

Why it matters: Stigmatising language in clinical notes is not merely a linguistic concern — it can influence downstream clinical decisions, shape trainee attitudes, and perpetuate structural inequities in care. The finding that Black patients and those with lower educational attainment are disproportionately exposed to stigmatising documentation is consistent with well-documented racial disparities in obstetric outcomes. For clinical informatics and health systems, this study illustrates a practical, scalable approach to auditing documentation quality and equity across large EHR datasets — a use case with clear operational value extending well beyond obstetrics.

Expert perspective: This is a methodologically interesting and socially important contribution, though several limitations warrant careful attention before conclusions are treated as definitive.

NLP classifier validity is the central methodological concern. A keyword-guided BERT classifier can achieve reasonable performance, but the abstract does not report precision, recall, F1 scores, or inter-rater agreement against a gold-standard human annotation of a held-out sample. Without this, the reported 47% prevalence figure carries substantial uncertainty. If the classifier has meaningful false-positive rates — plausible when "stigmatising" language overlaps with legitimate clinical descriptors — the prevalence estimate and the between-group comparisons may be systematically distorted. Readers should treat the quantitative estimates as approximate pending external validation.

Single-site academic centre data limits generalisability. Academic medical centres have distinctive patient populations, documentation cultures, and trainee composition. Prevalence and disparity patterns at community hospitals, safety-net institutions, or non-academic settings may differ substantially.

Confounding is a real concern in the disparity findings. Black patients in this cohort may have higher rates of clinical complexity, social risk documentation, or both, which could independently increase exposure to certain language patterns. The association with preterm birth is also worth scrutinising: clinically complex pregnancies generate more notes and more opportunities for any language — stigmatising or otherwise — to appear. Volume of documentation is a confounder that is easy to overlook. Causality cannot be inferred from this cross-sectional, observational design.

Despite these caveats, the study makes a genuine contribution. The scale of the analysis and the application of transformer-based NLP to equity monitoring represent a meaningful advance over prior work relying on manual chart review of small samples. The practical implication — that health systems can and should audit their documentation for bias as a quality-improvement activity — is sound. This is promising work, but preprint status and unvalidated classifier performance mean specific numbers should not be cited as established fact without peer-reviewed validation.

Relevance: Clinical Informatics and Health Data Methods

Source: https://pubmed.ncbi.nlm.nih.gov/42619892/

---

Sarah B Calvert

Summary: This essay, published in Clinical Infectious Diseases, examines the tension between AI-assisted clinical documentation and the preservation of the illness narrative in infectious diseases outpatient practice. The author argues that NLP-based scribing tools, while reducing administrative burden, risk producing flattened, standardised prose that strips away the interpretive, patient-centred language historically embedded in clinical notes. In the era of open notes — where patients routinely read their own records — this loss carries both clinical and therapeutic consequences. The central argument is that clinician authorship of the illness narrative is not merely stylistic but functionally important to patient care.

Why it matters: Clinical documentation has always served dual purposes: medicolegal record-keeping and a site of diagnostic reasoning. In specialties like infectious diseases, where ambiguity is the norm, the narrative structure of a note can itself reflect and communicate the clinician's interpretive process. If AI-generated notes systematically smooth over uncertainty and individualise less, downstream readers — including patients, consultants, and future treating clinicians — may receive a less accurate picture of the clinical situation. The open-notes policy in the United States (mandated since 2021) adds another layer: patients now read these documents, and the language used may influence their understanding of their illness and their relationship with their clinician.

Expert perspective: This is a thoughtful essay rather than an empirical study, and it should be read as such. The argument is intellectually coherent and raises legitimate concerns, but it is supported by reasoning and professional observation rather than data. The claim that AI scribing "flattens narrative nuance" is plausible and consistent with early qualitative reports from clinicians using ambient documentation tools, but it has not yet been rigorously demonstrated through systematic content analysis of AI-generated versus clinician-authored notes. Whether this putative flattening actually affects patient outcomes, diagnostic accuracy, or the therapeutic value patients derive from reading their records remains an open empirical question.

The open-notes dimension deserves particular attention. Research on patient-facing records suggests that language choice — including how diagnoses are framed, how uncertainty is communicated, and whether stigmatising terminology appears — influences patient experience and engagement. NLP tools trained on existing clinical corpora may reproduce or amplify the least patient-centred conventions in that corpus unless specifically designed and audited to avoid doing so.

Less convincing is any implicit suggestion that the status quo ante was uniformly superior. Pre-AI clinical notes varied enormously in quality, completeness, and narrative richness — many were template-driven, copy-forwarded, and filled with generic language long before large language models arrived. The essay would benefit from acknowledging this baseline problem.

The practical challenge — preserving clinician interpretive authorship while gaining efficiency benefits — is real and unsolved. Hybrid workflows, in which AI generates a structured draft that the clinician substantively edits rather than simply approves, may represent a reasonable path, but implementation discipline is required. Institutions deploying ambient scribing tools should be thinking now about documentation quality standards, audit mechanisms, and training that foregrounds narrative responsibility. Overall, this essay makes a valuable contribution by naming a risk easy to overlook in the enthusiasm surrounding AI-assisted documentation; the underlying empirical questions remain open and are worth investigating rigorously.

Relevance: Clinical Informatics and Health Data Methods

Source: https://pubmed.ncbi.nlm.nih.gov/42617064/

---

Glenn K Goodrich, Elizabeth A Bayliss, James Lagrotteria, et al.

Summary: Goodrich and colleagues developed and validated an algorithm combining NLP-style text string searches of clinical documentation with medication order data to identify clinically intended medication discontinuations across five drug classes in adults older than 65. Using a manually reviewed gold-standard cohort of 1,588 individuals, the full algorithm achieved sensitivity ranging from 74–82% and specificity from 75–95% depending on drug class. The combined approach consistently outperformed either text search or discrete order data used alone.

Why it matters: Deprescribing research is hampered by a fundamental measurement problem: administrative claims data cannot reliably distinguish an intentional clinician-directed discontinuation from a lapse in adherence, a coverage gap, or patient non-compliance. If that distinction cannot be made at scale, observational studies of deprescribing effects are severely confounded at the exposure-definition level. This work offers a practical, EHR-based method to partially resolve that ambiguity, which could meaningfully improve the quality of real-world deprescribing evidence generated from large health system datasets.

Expert perspective: The methodological contribution is genuine and the problem being solved is real. Claims data — the workhorse of pharmacoepidemiology — is nearly blind to clinical intent. This paper takes a reasonable engineering approach: combine simple but carefully tailored text strings with structured order data, validate against manual review, and report honest performance metrics.

Several aspects deserve scrutiny. The gold-standard cohort (n=1,588) is modest in size, and the per-drug-class validation samples are small — particularly bladder antimuscarinics (n=80) and antithrombotics (n=160). Performance estimates from samples this small carry wide confidence intervals not reported in the abstract, making it difficult to judge how stable the sensitivity and specificity figures actually are. Small validation samples also increase the risk that algorithm tuning overfits to idiosyncratic documentation patterns in one health system.

The text-string approach — essentially keyword matching rather than large-language-model NLP — is deliberately simple. Simple approaches tend to be transparent, reproducible, and portable. However, drug-class-specific tailoring was required, meaning the approach is not fully generalisable without additional development work for each new medication category or clinical context. Documentation culture, EHR vendor, and clinician habits vary substantially across health systems, and external validation is the obvious next step and a meaningful current limitation.

Sensitivity values in the 74–82% range mean that roughly one in five to one in four intended discontinuations will be missed. For a study of deprescribing effects on mortality, missing 25% of exposures would attenuate effect estimates substantially. Researchers using this algorithm should think carefully about how residual misclassification would bias their particular analysis and whether sensitivity analyses or probabilistic exposure models are warranted.

From a clinical informatics and registry standpoint, the study offers a useful template: define a clinically meaningful construct that cannot be captured with structured data alone, build a gold-standard manually reviewed reference set, develop an algorithm combining structured and unstructured data, and report calibrated performance metrics. This is useful, methodologically careful, incremental progress — a credible foundation, not a solved problem.

Relevance: Clinical Informatics and Health Data Methods

Source: https://pubmed.ncbi.nlm.nih.gov/42593400/

---

Joaquin M Espinosa, Kate Burdekin, Jeanette O Auman, et al.

Summary: This report summarises proceedings from an NIH INCLUDE Project workshop examining clinical data management challenges in Down syndrome research, a condition involving complex, longitudinal, multisystem data across numerous study sites. The workshop produced four main conclusions: shared common data elements and ontologies are foundational for harmonisation; EHR-based data extraction strategies can support scalable research pipelines; AI tools — particularly NLP with human review — can accelerate clinical narrative abstraction; and robust governance frameworks are essential when applying AI to sensitive health data. The authors argue these lessons extend beyond Down syndrome to complex clinical research programmes generally.

Why it matters: Multisite longitudinal research in rare and complex diseases consistently founders on heterogeneous data definitions, inconsistent EHR extraction, and the manual burden of abstraction from clinical narratives — problems that apply directly to rare disease networks and other federated research infrastructures. The explicit pairing of CDE-driven design with AI-enabled workflows, under defined governance structures, represents a maturing operational framework rather than a speculative proposal. The governance emphasis — covering privacy, bias, transparency, and ongoing oversight — is particularly timely given the rapid and sometimes uncritical adoption of AI tools in health data settings.

Expert perspective: This is a workshop proceedings report, not an empirical study, and should be read accordingly. There are no primary data, no outcome measurements, no before-and-after comparisons of data quality, and no formal evaluation of the AI tools described. The "results" are essentially expert consensus statements and workshop outputs — valuable for agenda-setting and community alignment, but not evidence that the proposed approaches actually improve research outcomes in practice. The recommendation that NLP with human review "can accelerate abstraction" is reasonable and consistent with the broader literature, but specific performance characteristics, error rates, and implementation costs in rare disease contexts are not reported.

The CDE harmonisation argument is well-established and uncontroversial in principle; the persistent challenge is adoption and maintenance, neither of which a workshop can fully address. The governance framework discussion is commendably practical — acknowledging bias mitigation and transparency as operational requirements rather than afterthoughts — but governance recommendations from workshop proceedings require institutional will and resources to implement.

For rare disease and registry professionals, the most actionable takeaway is methodological: the combination of standardised data element design upstream with AI-assisted abstraction downstream is a credible pipeline architecture, provided that human review remains integral and governance structures are genuinely implemented rather than nominally endorsed. The risk of overstating AI capabilities in clinical data contexts remains real; this report is appropriately cautious. The broader applicability claim is reasonable but should be tested rather than assumed. Overall, this is a useful, professionally grounded contribution to the clinical informatics literature — informative for those building or modernising rare disease data infrastructure, but it should be treated as a roadmap discussion, not a validated implementation guide.

Relevance: Clinical Informatics and Health Data Methods

Source: https://pubmed.ncbi.nlm.nih.gov/42586352/

---