Alert fatigue: what actually works to reduce it?

Our clinical decision support committee just finished reviewing our alert override rates and... they're not great. We're seeing 85%+ override rates on drug-drug interaction alerts, which basically means clinicians have been trained to click through everything. We've started tiering alerts and suppressing low-severity ones, but I'm curious what other organizations have done that actually moved the needle. Specifically interested in: - How you categorized alert severity (what framework?) - Whether you involved frontline clinicians in the review process - Any measurable outcomes after changes (override rates, near-miss events, etc.) The literature is helpful but I'd love to hear real-world implementation stories.