The FDA Sentinel Initiative uses real-world data from over 20 data partners covering 300+ million patient records (90% of U.S. population with health coverage). Unlike FAERS, Sentinel knows exactly how many people were exposed to a drug, allowing it to calculate the true risk of adverse events.
Comparing Sentinel vs. FAERS detection capabilities
Actual adverse events: 0
Reported cases: 0
Detected rate: 0%
Actual adverse events: 0
Detected cases: 0
Detected rate: 0%
Important Note: FAERS only detects about 1-10% of adverse events, while Sentinel can detect the true incidence rate by using denominator data from real-world patient records. This tool illustrates the magnitude of underreporting with traditional systems.
The U.S. Food and Drug Administration doesn’t wait for patients to get hurt before acting. Instead, it uses big data to catch problems before they spread. This is the heart of the FDA Sentinel Initiative - a national system that monitors the safety of drugs, vaccines, and medical devices after they’ve been approved and are being used by millions of people.
Before Sentinel, the FDA relied mostly on voluntary reports from doctors and patients through the FDA Adverse Event Reporting System (FAERS). But this system had big gaps. Only about 1% to 10% of serious side effects get reported. Many people don’t know they should report a reaction. Others don’t connect a new symptom to a medication they took months ago. And even when reports come in, the FDA rarely knew how many people were using the drug in the first place - making it hard to tell if a side effect was rare or common.
Sentinel changed all that. Launched in 2016 after years of testing, it doesn’t collect data in one place. Instead, it asks trusted partners - like insurance companies, hospital systems, and clinics - to run queries on their own data. The data never leaves their servers. The FDA sends a question: "Did more people taking Drug X have a stroke in the last 90 days compared to those taking Drug Y?" Then, each partner runs the same analysis on their own records and sends back only the results. No names. No personal details. Just numbers.
The system runs on a network of over 20 Data Partners, covering more than 300 million patient records - about 90% of the U.S. population with health coverage. These partners include giants like Kaiser Permanente, Humana, and Optum, as well as smaller regional health systems. Each one uses electronic health records (EHRs) and insurance claims data. Claims data tells you what was billed - a prescription filled, a hospital visit, a lab test. EHRs go deeper: they include doctor’s notes, lab results, vital signs, and even diagnoses written in free text.
The magic happens because every partner uses the same software tools. This means the same question gets answered the same way everywhere. If one hospital finds 15 cases of liver injury linked to a diabetes drug, and another finds 12, the FDA can combine those numbers with confidence. That’s something you can’t do with FAERS, where reports come in messy, inconsistent, and unverified.
When a red flag pops up - maybe a spike in heart attacks after a new blood thinner hits the market - Sentinel can respond in weeks, not years. Traditional studies take time to design, fund, recruit patients, and wait for outcomes. Sentinel already has the data. It just needs the right question.
Early on, Sentinel mostly used insurance claims. That gave it a broad view but a shallow one. It knew someone got a prescription, but not why. Was the drug prescribed for high blood pressure? Or for an off-label use? Did the patient have kidney disease? Those details matter.
Starting in 2019, the FDA pushed hard to add EHR data. Today, over 89% of U.S. hospitals use certified electronic health record systems. Sentinel’s Innovation Center now works with partners to extract meaning from unstructured notes - like a doctor writing, "Patient reports muscle pain after starting statin." That kind of detail used to be invisible. Now, natural language processing (NLP) tools scan thousands of notes to find patterns. Is muscle pain happening more often with one statin than another? Are older patients more at risk? These insights help the FDA make smarter decisions.
The Innovation Center has four main goals: better data infrastructure, smarter ways to turn raw data into usable signals, stronger methods to prove cause-and-effect, and faster ways to detect hidden risks. They’re even using machine learning to predict which drugs might be risky before large-scale problems emerge.
This isn’t theory. Sentinel has directly influenced FDA actions:
One of the most powerful tools built on Sentinel is PRISM - the Postmarket Rapid Immunization Safety Monitoring system. It tracks vaccine safety in near real-time. During the COVID-19 pandemic, PRISM helped detect rare cases of myocarditis in young men after mRNA vaccines. The FDA used that data to update recommendations within weeks - saving lives by balancing risk and benefit.
Compare Sentinel to FAERS:
| Feature | FDA Sentinel | FAERS (Traditional Reporting) |
|---|---|---|
| Data Source | Millions of real-world patient records | Voluntary reports from doctors and patients |
| Denominator Data | Yes - knows how many people used the drug | No - only counts reports, not total users |
| Speed | Weeks to detect signals | Months to years |
| Detail | Includes lab results, diagnoses, prescriptions | Basic symptom descriptions |
| Verification | Data quality checked before analysis | Reports often unverified |
| Population Coverage | Includes elderly, children, pregnant women | Underrepresents vulnerable groups |
Sentinel doesn’t just find problems - it finds them faster, in more people, with more context. It’s especially good at spotting risks in older adults, pregnant women, and people with multiple chronic conditions - groups often left out of clinical trials.
It’s not perfect. One big issue: EHR data is messy. One doctor writes "hypertension," another writes "high BP." One system codes a heart attack as I21, another as I22. The Innovation Center spends a lot of time standardizing these terms.
Also, Sentinel only sees what happens inside the healthcare system. If someone has a side effect but never goes to a doctor - or if they go to an out-of-network clinic that doesn’t share data - it won’t show up. Mental health events, substance use, or symptoms dismissed as "normal aging" often go unnoticed.
And while Sentinel can find patterns, proving a drug caused a problem still takes careful analysis. Just because two things happen together doesn’t mean one caused the other. That’s why the Innovation Center runs "emulated trials" - using real-world data to mimic the design of a clinical trial and test whether the signal holds up.
The system is evolving fast. In 2023, the FDA announced a major funding boost - $304 million - to upgrade Sentinel into what some call "Sentinel 3.0." This version will:
Experts call Sentinel a "learning health system" - one that gets smarter as it uses more data. Dr. Robert Ball, former FDA director, said it could become the blueprint for how the world monitors drug safety. Other countries - like the UK, Canada, and Australia - are now building their own versions.
The bottom line? Sentinel isn’t just a tool. It’s a shift in how medicine stays safe. Instead of waiting for harm, it watches for it - in real time, across millions of lives. And that’s how we protect public health in the age of big data.
FAERS relies on voluntary reports from doctors and patients, which often miss cases, lack context, and don’t include how many people used the drug. Sentinel uses real-world data from millions of patient records, knows exactly how many people were exposed, and can detect patterns faster and more accurately. It’s proactive, not reactive.
No. Sentinel never moves personal data to a central location. Data stays with each partner - like a hospital or insurance company. The FDA sends a coded query, and the partner runs it locally. Only aggregated results - like "15 more cases of liver injury among 500,000 users" - are shared. No names, Social Security numbers, or addresses are ever transferred.
Patients don’t directly opt in or out. Sentinel uses existing healthcare data that partners already collect for billing and care. But each partner follows federal privacy rules (HIPAA) and may offer patients the right to opt out of research use of their data - just like any other health study. The FDA doesn’t have access to individual records.
It can take as little as 2 to 6 weeks to analyze a signal once a question is asked. That’s much faster than traditional studies, which can take years. For example, during the pandemic, Sentinel identified rare heart inflammation after mRNA vaccines in under a month, helping the FDA update safety guidance quickly.
The FDA oversees Sentinel, but it’s run by a network of partners. The Sentinel Operations Center handles day-to-day queries. The Innovation Center develops new tools. The Community Building Center connects with researchers and data partners. Harvard Pilgrim Health Care helped launch the system, but today, multiple academic and private organizations manage parts of it under FDA oversight.
