Why alert fatigue is a design problem, not a content problem
The instinct when a safety event happens is to add an alert. It's an understandable response, and almost always the wrong one in isolation. Every clinical decision support system that has been running for more than a few years accumulates alerts the same way: one at a time, each individually justified, until clinicians start clicking through all of them without reading any of them.
The content is rarely the problem
Most alert content is accurate. The failure mode isn't that alerts say the wrong thing. It's that there are too many of them, presented with the same visual weight, interrupting the same workflow, regardless of whether they represent a genuine safety risk or a minor administrative reminder. Attention is a finite resource, and a system that spends it indiscriminately eventually gets none at all.
What tiering actually looks like
- Severity-based visual treatment, so a critical interaction doesn't look like a routine reminder.
- Configurable thresholds per service, because what counts as noteworthy varies by specialty and patient population.
- A documented override path with a required reason, so alerts inform a decision rather than block one outright.
- Ongoing review of which alerts actually get overridden without documented justification, which is a signal the alert itself may be miscalibrated.
Governance has to own this, not engineering
The uncomfortable truth is that fixing alert fatigue usually means someone with clinical authority has to say no to a new alert, or retire an old one, even when a specific incident originally justified it. That's a governance decision, not a technical one. The platform's job is to make tiering, configuration, and override tracking possible, not to decide where the line sits for your service.

