Admission Intelligence

The Challenge

Preventable hospital admissions cost US healthcare more than $33B annually and it is estimated that more than 15 percentof all adult inpatient stays with a primary expected payer of Medicare were potentially preventable (HCUP and NIH). Studies indicate that the preventable admission problem disproportionately impacts disadvantaged communities and individuals, especially individuals at an economic disadvantage.

A one size fits all approach won't work. The scope of the challenge means improving preventable admission rates requires a data driven architecture that can operate at scale and bring the best decision making and automation tools available to life.

Under the following guiding requirements:

Targeted Care
- Enhance and empower the experts. Enable clinicians and subject matter experts to target care and attention to the patients most at risk of admission
- Personalized automation. Automated communication to patients must meet them where they are in life and connectivity. The system must be adaptable to a wide range of circumstances.
Modern Architecture and Infrastructure
- Utilize modern software infrastructure to provide automated communication, feedback, and outcome tracking across a broad population
- Build for scale. The problem is huge and requires massive amounts of data and decision making
Self Improving Systems
- Close the loop. The architecture must capture feedback and outcomes to improve models, rules, and alerts
- Observability is key. Utilize the available data to track outcomes in as close to real time as possible.
Data and Models at the Core
- Bring models to life. The architecture must enable new models and machine learning techniques must be brought
- Adapt to the data. Decoupled systems and an event driven architecture allow models to evolve and get to production faster and with bigger impact

Hackathon Submission

The submission consists of two parts.

The first part of the submission is a reference architecture that describes a set of services and technology that could support a real world deployment with the scale, automation and software based decision making required to make a meaningful impact on preventable admissions in US health care.

The second part of the submission is a proof of concept implementation of the decision services and dashboard systems of the architecture. The code for the submission is available on the admission-intelligence github page and the dashboard is deployed on Azure and can be viewed here

The decision table and rules implementation for the submission are available on the github page here.