Electronic health record giant Epic debuted a new feature that monitors county-level health trends and issues alerts when elevated rates of illness are detected.
The health alerts, developed by Epic Research, use statistical models applied to real-world medical records to detect when the rate of a health condition in a county is higher than expected. Each alert is reviewed by the Epic Research team, which includes clinicians and data scientists, before being published, Epic said in a blog post. This manual review step assesses whether the alert is clinically meaningful and appropriate for public reporting, the company said.
Active Health Alerts can be viewed through this Epic dashboard and users can subscribe to receive Health Alerts by email.
The Health Alerts tool has so far flagged elevated rates of acute bronchiolitis, acute tonsillitis, measles, strep throat and viral gastroenteritis in some areas of Illinois, Missouri, Tennessee, Arkansas and South Carolina.
The alerts are based on data in Epic's Cosmos platform, which contains patient records for 300 million patients from 2,067 hospitals and 47,000 clinics.
The company notes that Health Alerts are based on data from Epic Cosmos' participating organizations, and the data does not reflect complete case counts for a condition but rather evaluates population-level trends in disease activity.
The Epic Research team's detection process monitors county-level diagnosis rates using ICD-10-CM diagnosis codes. The goal of the Health Alerts tool is to surface conditions that are both increased from historical rates and accelerating, Epic said in its description of the Health Alerts feature.
Alerts are most likely to appear for acute conditions, communicable diseases and rare or unexpected spikes rather than for chronic conditions or predictable seasonal patterns.
A condition is flagged for review when it meets a number of criteria, including a year-over-year increase in the diagnosis rate, which accounts for normal seasonal variation, and accelerating growth in diagnosis rates. Conditions that meet the first two criteria are then evaluated using a Farrington improved algorithm, a well-established method in public health surveillance, Epic Research said.
"This model uses three years of historical data to establish expected baseline rates, further adjusting for seasonality and long-term trends, and confirms that the observed increase is statistically significant and not likely due to random variation," Epic Research said in an explanation of its methodology.