Briefs · Frontier Science
The Problem with Medical Alerts Is Not Too Few, but Rather Too Many
The problem with medical alerts is not too few, but rather too many — and most of them are wrong.
AMC Health has announced an AI-driven clinical alerting framework for virtual care. It replaces fixed thresholds with longitudinal, patient-specific baselines and time-series pattern detection. Six years of data and more than 15 million physiometric readings — weight, blood pressure, heart rate — sit behind it.
Under old threshold alerting, only 24% of red alerts were associated with actual clinical deterioration. Three in four alarms cried wolf. The new framework is projected to raise alert precision by nearly 20%, cut total alert volume by 45%, and cut clinician burden by 45%, while preserving outcomes.
The tricks are individual. Weight alerts follow each patient’s own percentage change and variability; blood pressure and heart rate follow longitudinal variability; behavioural signals, such as a patient suddenly measuring themselves more often, count too.
AMC’s programmes already average a 23% reduction in all-cause hospital admissions versus controls, with savings above $300 per enrolled member per month, up to $670 in some populations. Rigorous evaluation against control groups is planned, with oxygen saturation and blood glucose to follow.
Conclusion: good alerting is not louder alarms, but rather quieter ones that mean something.
Read the original at ACCESS Newswire
Summary in our own words. The original belongs to its publisher.