Research map / Large language models / AI and society
Data Monitoring for Large Scale Public Health Data
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
A doctoral thesis summary describing a human-in-the-loop data monitoring framework (FlaSH outlier detection plus extreme-value ranking and a reviewer interface) deployed at Delphi to surface outbreaks and data issues in large-scale public health data.
Problem
Public health data reviewers cannot manually inspect the millions of daily data points from modern nonstationary, noisy, revised, and seasonal public health streams, and existing outlier/outbreak detection methods produce too many false alerts.
Approach
Simple, scalable outlier detection (FlaSH) that explicitly models public health properties such as weekday effects and reporting lag, combined with extreme value theory and the hierarchical data structure to rank the most outlying points, and a reviewer interface for analyzing ranked outliers.
Key results
- Deployed FlaSH-based outlier detection let data reviewers identify notable outliers 9.1x faster than the prior approach.
- A 3-month longitudinal evaluation of the ranking interface showed a total average 53x efficiency increase over manual review.
- Reviewers identified approximately 200 significant outbreaks, data issues, or changes in disease dynamics from 15 million new data points weekly.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use contextual outlier ranking with extreme value theory and the hierarchical structure of public health data | Public health data monitoring · Delphi public health data streams | time to identify notable outliers: 9.1x faster | Helped |
| Add an interface for reviewers to analyze ranked outliers | Public health data monitoring · Public health data review | efficiency: 53x efficiency increase | Helped |
| Use the proposed univariate outlier detection method instead of previous outlier detection methods | FlaSH · Public health data streams | outlier detection performance | Helped |
Limitations
- Existing methods produce many false positives and negatives and require constant parameter tuning.
- Foundation time series and language models are untested on epidemiological data and may be incompatible with high nonstationarity.
Experimental setup
Datasets: Delphi public health data streams COVID-19 case data reviewer-ranked outlier survey ground truth
Metrics: alerts/outliers reviewed speed (9.1x faster) efficiency increase over manual approach (53x) number of significant outbreaks/data issues detected (approx
Baselines: existing outlier and outbreak detection methods deep learning outlier baselines prior ranking approach manual review
Builds on: FlaSH
Closest papers by meaning
- Outlier Ranking for Large-Scale Public Health Data (AAAI 2024 · similarity 0.84)
- Aligning Time-series by Local Trends: Applications in Public Health (AAAI 2025 · similarity 0.80)
- BAND: Biomedical Alert News Dataset (AAAI 2024 · similarity 0.80)
- SPATIALEPIBENCH: Benchmarking Spatial Information and Epidemic Priors in Forecasting (NeurIPS 2026 · similarity 0.78)
- The State of Data Curation at NeurIPS: An Assessment of Dataset Development Practices in the Datasets and Benchmarks Track (NeurIPS 2024 · similarity 0.78)
- Context Clues: Evaluating Long Context Models for Clinical Prediction Tasks on EHR Data (ICLR 2025 · similarity 0.77)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.