Shawn Bierman

MedBeacon

A sepsis alert clinicians can trust

In testing, it caught as many sepsis cases as standard scoring rules, with about a third fewer false alarms.

My role
Cofounder, Product and Technology
Years
2021–2026
Built for
Hospitals on Oracle Health (Cerner) Millennium
Platforms
iOS, PowerChart, web
Company
MedBeacon, Inc. (formerly Health Apogee), a team of fewer than five
Built with
AWS (EKS, Bedrock) and MongoDB Atlas
MedBeacon Desktop census board: a card for each Emergency Department patient with a sepsis risk score, active clinical events, and SEP-1 bundle timers.

The problem

About one in three patients who die in a US hospital has sepsis,1 and each hour of delayed treatment is linked to a higher risk of death.2 Alerts built on standard scoring rules typically use only vital signs and lab values, so they leave out clinical notes, which have been shown to improve early sepsis prediction.3

What we built

An AI sepsis detection and alerting platform for hospitals running Oracle Health (Cerner) Millennium. Its core idea was “Silent Sepsis” detection: generative AI read nurse notes to catch patients whose vital signs still looked normal but whose charting suggested they were declining. Threshold-based rules missed those cases, so MedBeacon paired the AI with the clinical rules behind standard sepsis alerts. Behind it, a HIPAA-eligible AWS architecture on EKS, MongoDB Atlas, and Amazon Bedrock, with business associate agreements (BAAs) in place for the cloud and the database.

MedBeacon Mobile, a native iOS app for clinicians on the move

MedBeacon Mobile patient screen: sepsis suspected, hypotension and hypoxia flags, a Suggested Interventions button, and a blood pressure trend.
A patient’s clinical events by body system, recent lab results, and the latest clinical notes.
Sepsis bundle orders: laboratory, treatment, and diagnostic orders, each marked done or with an estimated time.

MedBeacon Integrated, inside PowerChart, where clinicians already work

The MedBeacon watchlist inside PowerChart: patients ranked by sepsis risk score, each with a trend, the signals behind it, and an Acknowledge button.
A patient’s MedBeacon summary inside PowerChart: a sepsis risk score of 92 and rising, score drivers, vital signs, labs, and recommended actions.

MedBeacon Desktop, a census board with a unit-wide view of risk

A patient’s detail page: a sepsis risk score of 88 and rising, suggested interventions, and the SEP-1 bundle checklist.
The census board as a list, sorted by acuity, with each patient’s clinical events, bundle timers, and risk score.

All screens show demo data with made-up patients.

My part

What I did

  • Shaped what the product should be, with our Chief Product Officer.
  • Designed the workflows and user experience.
  • Prototyped every demo on all three surfaces.
  • Directed three outsourced engineering teams of up to five, delivering all Phase 1 deliverables on a $150K fixed-fee build and scoping a $250K follow-on.
  • Governed the AWS architecture: approved the stack and executed the cloud and database BAAs.
  • Built HIPAA security and access governance, including least-privilege contractor access, required security checks on code, and quarterly access reviews.
  • Led the Cerner Millennium integration partnerships, from vendor evaluation and selection through technical issue resolution.
  • Built a SEP-1 Prospect Finder that ranks 4,600+ U.S. hospitals by public CMS sepsis bundle compliance, to target sales outreach.
  • Reported to the CEO and board.

What others did

  • Our Chief Product Officer shaped the product with me.
  • Outsourced engineering teams developed and maintained the platform.

Results

Tested on 150 ICU patients from a public research database (eICU-CRD), using each patient’s data as it stood 6 hours before diagnosis.4

ApproachSepsis cases caughtFalse-alarm rate
Standard scoring rules74%57%
AI alone82%70%
AI plus clinical rules82%37%

Notes

  1. Rhee C, et al. “Incidence and Trends of Sepsis in US Hospitals Using Clinical vs Claims Data, 2009–2014.” JAMA. 2017;318(13):1241–1249. Sepsis was present in 35% of hospital stays that ended in death. doi.org/10.1001/jama.2017.13836
  2. Seymour CW, et al. “Time to Treatment and Mortality during Mandated Emergency Care for Sepsis.” N Engl J Med. 2017;376:2235–2244. Each hour of delay before antibiotics were given or the 3-hour treatment bundle was completed was associated with higher in-hospital mortality (odds ratio 1.04 per hour). doi.org/10.1056/NEJMoa1703058
  3. Goh KH, et al. “Artificial intelligence in sepsis early prediction and diagnosis using unstructured data in healthcare.” Nat Commun. 2021;12:711. Adding clinical notes improved accuracy over clinical measures alone for warnings 12 to 48 hours before sepsis onset. doi.org/10.1038/s41467-021-20910-4
  4. Patients from the eICU Collaborative Research Database Demo v2.0.1 (doi.org/10.13026/4mxk-na84), described in Pollard TJ, et al., Sci Data. 2018;5:180178 (doi.org/10.1038/sdata.2018.178). The results come from MedBeacon’s internal validation in May 2026 and have not been published. Of the 150 patients, 34 had sepsis. The 95% confidence intervals are roughly ±10–15 points on the share of cases caught.