AI Inpatient Risk Monitoring Market Map: Improving patient outcomes with data
This is part of Elion’s weekly market map series where we break down critical vendor categories and the key players in them. For more, become a member and sign up for our email here.
Early hospitals were among the riskiest places to seek care—sepsis, post-surgical gangrene, and other infections ran rampant, drastically increasing morbidity. While today’s hospitals are much safer, patients still face risks such as infections, strokes, falls, and other forms of acute deterioration. Effective interventions exist, but their success depends on identifying issues early, before they escalate.
Turning Data Into Insights
Modern healthcare generates a wealth of data from bedside monitors, lab results, imaging reports, and clinical notes in the EHR. While the data is abundant, the real challenge lies in continuously monitoring these inputs to detect signs of deterioration before they manifest into crises. AI Inpatient Risk Monitoring tools address this challenge by leveraging machine learning and statistical models to track patient data in real-time and alert clinical staff to potential risks.
Integration with EHR systems and other data sources is essential, but the core difficulty lies in model precision. Models must reliably identify real risks (true positives) without overburdening staff with false alarms (false positives). Overly sensitive models trigger frequent alerts for non-issues, leading to alert fatigue—where clinicians, inundated with notifications, begin to ignore them. Conversely, if a model isn’t sensitive enough, critical signs of deterioration may go unnoticed, jeopardizing patient safety.
Current Models’ Results
In early iterations, predictive models struggled to perform effectively. For example, a 2021 JAMA study found that Epic’s Sepsis Model correctly identified only 33% of relevant cases. However, more recent developments have significantly improved accuracy. A follow-up study reported that an improved model now identifies 86% of true positives. Still, the model’s 33.8% positive predictive value means that three alerts are typically needed to confirm a single sepsis case—highlighting the trade-off between sensitivity and precision. As demonstrated in this paper comparing different early warning models, this is clearly an area of active development.
Differentiating Risk Monitoring Vendors
Generally vendors differentiate based on what type of risk they are monitoring for.
Sepsis remains one of the most challenging conditions to detect early due to its varied manifestations and frequent comorbidities, and in addition to Epic’s Sepsis Model, vendors such as Luminare and AITRICS remain focused on refining sepsis detection, while others tackle different aspects of patient deterioration.
Bayesian Health, Biobeat, CLEW, and Etiometry develop tools to monitor the condition of critically ill patients, often in the ICU.
Kinometrix offers a unique approach by predicting fall risk from EHR data, allowing staff to allocate resources more efficiently. Similarly, Healthplus.ai’s PERISCOPE focuses on preventing post-surgical infections by identifying at-risk patients and optimizing care plans.
Meanwhile, Aidoc and Viz.ai integrate EHR and imaging data to respond swiftly to time-sensitive events, such as strokes and cardiac emergencies.
These technologies are unique in that their accuracy and precision not only save substantial time for clinical staff, but they have the potential to save patient lives during their most vulnerable times. The progress of these tools moves forward not only with quality of the models, but also with the increasingly available input and monitoring data. With progress against patient deterioration like this, it’s hard not to be optimistic about healthcare in another hundred years.