About the challenge

Acute deterioration in hospitalized patients can occur rapidly and unpredictably. Hospitals already collect large amounts of information—from manually recorded observations and electronic health records to continuous bedside monitoring and emerging ambient sensing technologies. The challenge is not simply whether AI can predict deterioration. It is determining which data and monitoring approaches can detect meaningful deterioration early enough to improve clinical action.

The 2026 Scope-AI-Thon challenges teams to scope an AI-enabled system for early detection of acute patient deterioration in a hospital setting. Teams will investigate the evidence, identify a clinically meaningful problem, determine the right data and sensing approach, and design how an automated signal should lead to an appropriate clinical response.

The Starting Point: Four Monitoring Approaches

Teams should begin by examining four broad categories of automated deterioration monitoring systems:

· Manually entered data systems: Information deliberately entered by clinical staff, including intermittently recorded vital signs, patient-reported symptoms, and nursing observations.

· Electronic records–based systems: Algorithms using existing EHR data such as laboratory results, medications, clinical notes, demographics, imaging findings, and trends in recorded vital signs.

· Integrated physiological monitoring systems: Continuous or frequent streams from bedside monitors, pulse oximeters, cardiac monitors, ventilators, wearable biosensors, and related devices.

· Novel sensing and ambient technologies: Emerging approaches such as contactless sensing, computer vision, acoustic monitoring, pressure-sensitive beds, and other AI-enabled ambient technologies.

Teams are not expected to assume that the newest or most technically sophisticated approach is best. They should compare what each approach can observe, how quickly it can detect change, what it misses, how difficult it is to integrate into hospital workflow, and what evidence exists that it can improve meaningful patient outcomes.

The Core Scope Question

Which category—or combination of categories—has the greatest potential to reduce adverse outcomes from acute patient deterioration, and what would a credible AI-enabled detection and response system look like?

Requirements

Your Mission

Each team should turn this broad question into a specific, defensible system proposal. The work should proceed from clinical problem to evidence, data, analytics, and response—not from a preferred model or technology.

1. Define a meaningful deterioration event or target patient population where earlier detection could plausibly improve care.

2. Review evidence on relevant monitoring approaches and explain why the selected category or combination is appropriate for the problem.

3. Identify the signals and data needed for detection and distinguish what is already available from what would need to be newly collected or integrated.

4. Specify an appropriate AI or analytics approach and the decision it is intended to support.

5. Design the response workflow that connects detection to clinical action.

6. Identify the major failure modes, implementation barriers, and evidence gaps that would need to be addressed before deployment.What to Build

The Data Challenge

A central part of the Scope-AI-Thon is determining what data are actually needed. Teams may work with representative healthcare datasets, including MIMIC ICU and emergency-room data, but they should not allow the available dataset to define the clinical problem.

Teams should examine which variables are available, their timing and frequency, missingness, whether they are collected continuously or episodically, and whether they represent the signals needed for the proposed system. If the concept requires bedside monitors, wearable devices, ambient sensors, or other information not present in the provided datasets, teams should explicitly identify those requirements and explain how the sources would be integrated.

From Detection to Response

A strong proposal must go beyond prediction accuracy. Teams should explain how an automated signal becomes a clinical decision and action:

Patient → Data / Sensing → Detection → Decision → Notification → Clinical Response

This means addressing practical questions: Who receives the alert? At what threshold and with what urgency? What information accompanies it? What action is expected? What happens if the alert is ignored or contradicted by clinical judgment? How will the system manage false positives, false negatives, alert fatigue, missing data, sensor failures, and model errors?

Evidence Matters

Teams should use peer-reviewed evidence to understand what is already known and where important gaps remain. The provided appendix offers representative starting examples across the four monitoring categories, including early-warning scores based on manually entered observations, machine-learning systems using EHR data, wearable physiological monitoring, and contactless ambient monitoring. Teams should expand beyond these examples as needed and critically evaluate whether reported improvements concern prediction performance alone or meaningful clinical outcomes.

What Teams Will Deliver

· Clinical problem and target patient population

· Comparison of relevant monitoring approaches and supporting evidence

· Proposed data and sensing architecture, including important gaps

· AI/analytics approach and the decision it supports

· End-to-end clinical response workflow

· Risks, failure modes, implementation barriers, and evidence gaps

· A concise recommendation explaining why the proposed system is worth testing

How Proposals Will Be Evaluated

Criterion

What Judges Will Look For

Problem importance

Is the deterioration problem clinically meaningful and clearly scoped?

Evidence

Does the team understand what is already known and distinguish evidence from assumptions?

Data and AI scoping

Are the required signals, data sources, limitations, and analytical approach credible?

End-to-end design

Does the proposal connect sensing and detection to a realistic clinical response?

Feasibility

Could the proposed system plausibly operate in a hospital environment?

Risk awareness

Does the team address false alarms, missed events, workflow burden, technical failures, and other failure modes?

Potential impact

Is there a credible path from earlier detection to better patient outcomes?

Scope-AI-Thon 2026 - AI for Detection of Acute Patient Deterioration

Prize breakdown

Timeline

About