Chart Abstraction as Shared Infrastructure: From Departmental Cost to Enterprise Capability
September 9, 2026
Modern epidemiology has often been attributed to Dr. John Snow's "famous cholera map." During a large cholera outbreak in the West End of London in 1854, Dr. Snow posited that cholera may be a water borne illness as opposed to the prevailing theory that cholera was transmitted due to "bad air" entering the human body. By geolocating cholera outbreaks alongside sites of water pumps in the neighborhood, he was able to convince public authorities to disable a water pump on Broad Street, thereby stopping the outbreak.
What made the map persuasive was not any single observation, but rather the collation of all the observations. In Dr. Snow's famous infographic (see Figure 1), he took multimodal data inputs, collated the data points onto a map of London, and used this visualization tool to provide meaningful decision making. Every fact he needed was already sitting in the neighborhood. What it lacked was a structure, and until someone built that structure, no one could act on it.
Health Systems are Sitting on the Same Latent Insights
With today's ubiquity of data there is ample opportunity to concatenate information together for myriad use cases across many domains, but the heavy lift of needing to collate data siloes, semantically match data inputs, and provide a unified structure to this data continues to be burdensome. Analogous to how Dr. Snow was able extract meaningful insights from his epidemiologic study once the data was collated properly, a litany of unstructured clinical data (eg: EMR notes, flowsheets, operative reports, radiology reports, pathology reports, PDFs, etc) already holds the answers to the questions service line leaders ask every week. Where is our surgical capacity going? Are we capturing the quality measures we are scored on? Which patients could have been enrolled, staged, or referred sooner?
Yesterday’s Conundrum: Chart Abstraction is a Departmental Cost when it should be a Shared Service
Most health systems fund chart abstraction as a line item inside individual departments. Cardiology hires a registry coordinator, oncology hires a tumor registrar, research hires clinical research coordinators, and quality hires its own abstractors. Each team recruits separately, builds its own workflow, and defends its own budget. The underlying work is close to identical in every case, which means the capability is fragmented precisely where it should be concentrated.
The alternative is to treat abstraction the way health systems already treat imaging or pathology: one capability, centrally operated, that every service line draws on. Structure the data once and the same structured output feeds registry submission, quality and safety reporting, capacity planning, referral analysis, and research enrollment. The marginal cost of the second use case is a fraction of the first, and the third is cheaper still.
The delay and resistance to automate chart abstraction using advanced technology has hindered health systems from recognizing strategic growth opportunities that could be used for service line improvement. Rapid patient and cohort matching, improved near real time quality and safety reporting, and better visibility into surgical throughput are all downstream of the same structuring problem. Solving it once unlocks all of them.
Today’s Solution: An Enterprise Grade Shared Abstraction Service
What is missing is the enterprise level infrastructure to be able to extract this data in a timely manner, structure the data inputs into a platform that can support multiple use cases, and extract insights in a cost effective manner with limited manual oversight.
With that North Star in mind, Brim Analytics was conceived out of the Bioinformatics Department at Vanderbilt University by Professor Daniel Fabbri, PhD in 2024, with the core thesis of activating healthcare data via an enterprise scalable platform enriched with the power of large language models. The approach caught the attention of ARPA-H, which provided grant funding to expand development, and multiple academic institutions including VUMC, UCSF, Johns Hopkins, CHOP, and UNC have been expanding their adoption of the product.
The enterprise scalable platform is designed around the needs of a shared institutional service:
- It is on premise and behind your firewall: PHI never leaves your institution, no data is shared with Brim's servers, and the platform is SOC 2 Type II and HIPAA compliant.
- It requires no coding and no prompt engineering. Coordinators and researchers define variables in plain language, and Brim handles all LLM configuration internally.
- It is human in the loop by design: every AI extracted value is presented alongside the source text that supports it, so abstractors review, confirm, and override rather than blindly accept outputs.
- The documented result is 10x faster abstraction than manual review with an 80 to 90% reduction in chart review time at VUMC.1
Tomorrow’s Impact: Service Line by Service Line
As Brim Analytics has continued to expand across health systems and life science customers across the 2 years, multiple case studies are being published showing the utility and benefit of using this technology to create real-world value across multiple clinical and operational service lines. Below are a few example case studies with additional papers in pre-print awaiting publication.
Surgical capacity and case volume. Assessing a priori surgical risk requires clinical data and notes that no scheduling system reads today. Structuring that input allows ambulatory surgery center scheduling to be built around actual risk rather than conservative defaults, which converts directly into case volume and reduces last minute rescheduling. Brim's work here is documented in a case study on reducing surgical rescheduling from weeks of manual review to hours.
Oncology staging and quality reporting. Historical pathology and imaging reports combined with clinical data can be used to stage patients automatically rather than through retrospective manual review. Accurate and current staging drives appropriate treatment assignment, supports registry compliance deadlines, and feeds the quality measures that determine program reputation and payer position. See the published staging work for methodology and results.
Registry abstraction and benchmarking. Registries are the mechanism by which a service line proves it is good, and until recently they have required a significant amount of manual labor. Increasing the accuracy and throughput of automated chart review lets a program submit complete registries on time, benchmark honestly against peers, and use that position in payer and referral conversations. The economics are laid out in the ROI of patient registries.
Clinical trial enrollment. The same structured output supports patient matching against EMR and genomic data to identify enrollment opportunities before the window closes. This use case carries enough revenue impact to deserve its own treatment, and we will cover the financial mechanics of enrollment delay in a follow up post.
ePROMs and post market surveillance. Structuring patient reported outcomes alongside the clinical record bolsters the utility of patient surveys and supports longitudinal surveillance of pharmaceutical and medical device efficacy and safety, which is the foundation for industry partnership revenue.
These could be five separate projects. But we believe they should be five consumers of one structured data layer. If one structuring layer can serve surgical scheduling, oncology staging, registry submission, trial matching, and outcomes surveillance, then the investment case does not have to be won five separate times inside five separate departments. If the investment decision is made once, at the institutional level, every service line can immediately benefit from this platform and abstraction service.
The Future Is Now: The Question is no Longer Why, but Why Not?
For any health system running active clinical trials or maintaining research registries, the question is no longer whether AI assisted chart abstraction is valid. The question is how quickly and seamlessly it can be deployed, without adding burden to local teams. This is where Brim Analytics excels and is a clear differentiator from other similar platforms. To talk through what a shared abstraction service and platform would look like at your institution, reach out to the Brim Analytics team at enterprise@brimanalytics.com or sign up for a demo.
References
1) Rahman, Ye, Mittendorf, Lenoue-Newton, Micheel, Wolber, Osterman, Fabbri, "Accelerated curation of checkpoint inhibitor-induced colitis cases from electronic health records," JAMIA Open, 2023