Patients as Time-Series: High-Stakes Clinical Alerting With Ope...
summary
A concrete blueprint for longitudinal clinical alerting using OpenSearch 2.13 HCAD with per-entity models and one-document-per-measurement indexing, a patient simulator that emits realistic vitals with Gaussian noise, a FastAPI alerting service polling every 5 seconds and maintaining a 20-reading sliding window, and a dashboard plus Slack outputs. It uses shingle size 8 (8-minute context), anomaly scoring per patient, and Bayesian sequential updating to map probability bands to clinical actions, reducing false alarms while catching gradual decline.
AI-written summary · about
chapters (15)
- 0:00Patients as time series
- 1:03Introduction
- 2:49Snapshot queries fail
- 5:22Longitudinal data queries
- 6:04Neurology case study
- 9:39Clinical alerting pillars
- 14:19OpenSearch for anomaly detection
- 14:40Longitudinal indexing design
- 16:25Snapshot vs longitudinal queries
- 18:12Trajectory drift detection
- 23:14Cost of false alarms
- 25:19Bayesian alerting
- 28:11Live demo: patient dashboard
- 30:49Beyond healthcare applications
- 32:11Key takeaways
related talks
23mFinding Needles in the Observability Haystack
49mThinking in time series, the new solution to your pro...
50mApplied Alerting Philosophy
27mAlert Fatigue to Action: Observability as the Backbone of Practical AIOps with OpenSearch
32mVital Signs: OpenSearch and Bioobservability
40mNext-generation Alerting and Fault Detection
25mSymptom-based Alerting for Machine Learning - What I Learned...
22mFrom Telemetry to Insight: Building AI-Powered Observability Pipelines with OpenSearch