Problem
Health Delivery Organizations need information fluidity and local innovation
Healthcare professionals face a growing crisis of information overload. More than three million articles are published annually in biomedical journals, and the data required to assess a single patient has become enormous. Staff at the Duke Institute for Health Innovation (DIHI) found that compiling a new patient’s information before an appointment can take up to two hours. Meanwhile, virtualization has multiplied patient messages and administrative meetings, and many providers still rely on faxes, scanned documents, and paper records alongside their digital systems.
Time and resources are both scarce. As patient volume grows faster than the healthcare workforce, providers must access and share extensive data quickly and reliably. Data fluidity is not a convenience, but a necessity. In high-stakes clinical situations where every minute counts, comprehensive chart review often becomes a casualty of time pressure, forcing providers to make decisions with incomplete information. Academic medical centers are simultaneously under intense pressure to reduce costs and optimize processes, with margins that leave little room for inefficiency. Research and support cuts have further reduced the resources institutions typically use to optimize workflows and address these challenges.
Focused Problem: Vastness of the Electronic Health Record, even for one patient
Electronic Health Records (EHRs) are now ubiquitous in both hospital and ambulatory settings. Their size has become its own problem. “Information overload,” “chart bloat,” and “note bloat” are terms now common in the clinical literature, driven by redundancy, copy-paste habits, automated templates, and regulatory documentation requirements.
This is not merely an administrative inconvenience but a systemic issue impacting patient safety, physician well-being, and the patient-physician relationship. Simulation studies demonstrate that high cognitive load from conventional, data-heavy EHR interfaces is associated with four times as many errors compared to streamlined displays that show only salient information. When critical data is buried in a cluttered chart, a physician’s situational awareness suffers.
The bloat of the EHR has restructured medical work itself. To avoid disrupting the patient relationship during clinic hours, many physicians defer documentation evenings and weekends, a phenomenon often called “pajama time”. In our own Scout study, 400 care providers reported an average of 4.23 hours of pajama time per week (median: 3, SD: 4.21) and an average of 10.43 hours of EHR time within the workday (median: 10, SD: 6.0). Thirteen percent reported spending ten or more hours per week of their personal time with EHRs. High EHR burden is a primary predictor of physician burnout, decreased career satisfaction, and intent to leave practice.
Solution: Scout
Scout is an AI-powered platform for searching and synthesizing patient EHR data. Scout reviews every note, imaging report, pathology report, and other structured data in the patient’s record to deliver exactly the information a care practitioner needs, in the format they want. Scout helps care practitioners save time, reduce cognitive burden, and maintain or exceed care quality standards.
If you have a Duke NetID, you can visit the Scout ‘splash page’ for introductory videos and even a lecture by our principal data scientist, Michael Gao, PhD. Furthermore, you can complete this survey to express interest in becoming an early user – to participate in our IRB-approved research study
Scout functions like an AI assistant, similar in interaction style to ChatGPT, but instead of searching the web, it searches across a patient’s medical record. Scout has been designed with patient safety and data provenance as foundational principles. Every response Scout provides is traceable back to a specific location in the patient’s record, and our system architecture is built to minimize the risk of inaccurate or unsupported output.
We are studying whether Scout outperforms traditional AI tools with features such as
- Full record review with effective EHR integration
- Efficient data processing to enhance speed and minimize unnecessary tokens
- Easy-to-use and granular data provenance
- User-controlled queries
- Simple and loveable design
Early internal testing across thousands of queries has shown:
- A near-zero rate of unsupported or fabricated responses
- High accuracy across a range of query types and clinical domains
- Consistent data provenance, allowing clinicians to verify Scout’s responses directly
Scout will create value across the full care spectrum, supporting clinical, operational, administrative, financial, research, and academic work.
Impact
- Complete tasks that normally take hours in just 1-3 minutes, reducing clinical burden and improving efficiency.
- Significant reduction in mental demand, effort, and temporal demand as measured by the NASA Task Load Index
- A primary conclusion of a study (the prospective, randomized, evaluator-blinded, two-period crossover trial publication is forthcoming) was that Scout maintained quality, meeting “non-inferiority” criteria in accuracy, completeness, and relevance compared to manual EHR workflows.
