Efforts to slow the spread of Covid-19 in the U.S have been stymied by the lack of an effective national surveillance system that can track the emergence of suspected and confirmed new cases in real-time. This is in part because important patient data is trapped in siloed electronic health record systems that don’t communicate well with each other; a new case of suspected Covid-19 in one health system could be invisible to another in the same community, blinding the systems and public health officials to Covid-19 hotspots that may be developing right in front of them.
If we could see such hotspots as they appear we could direct resources where they’re needed most to treat new patients and contain transmission. Leveraging an existing platform that already accesses the electronic health records of 200,000 physicians and health care providers across more than 400 hospitals nationally, our firm has built an automated, real-time surveillance app that integrates with existing electronic health records (EHRs) from different companies. This solution could provide early warning capability, forecast surges, and help providers plan coordinated responses. The EHR-agnostic app will be offered for free this year to the dozens of health systems that use our clinical platform beginning with Geisinger Health System in July. Atrium Health, Community Health Network, Vidant Health, AdventHealth and other health systems have confirmed that they plan to rapidly follow.
We initially assumed that the CDC or other government agencies already had this type of capability. But we discovered that existing “syndromic surveillance,” which tracks emerging infectious disease and other public-health issues, is not automated and does not consistently provide the complete information needed to support robust containment and mitigation strategies. As Kaiser Health News reported, clinicians and public health officials sometimes must print out data from EHRs and manually fill in and fax reporting forms. Some CDC syndromic surveillance forms can take up to 30 minutes to complete, and the lag between a patient receiving a positive Covid-19 test and the reporting of that data can be as long as seven days. CDC syndromic surveillance systems, in fact, may have provided false reassurance by reporting an artificially low rate of infection through the ILI-NET surveillance program in mid-April 2020 in California, Florida, and Michigan, even though the virus was running rampant in those states at the time. This potential blind spot may have occurred because surveillance primarily tracked emergency department patients who, fearing infection, had started to avoid ED visits.
Without an effective national or state system for disease surveillance and monitoring, the U.S. response to Covid-19 and future pandemics will continue to face handicaps and have critical knowledge gaps. For instance, today we continue to lack clear, real-time, nationwide data detailing:
- Where and in which settings patients are presenting with Covid-19-like symptoms;
- The local hospitals that are likely to admit positive cases after they are detected;
- Risk- and severity-adjusted information that allows provider systems to predict the supplies they will need to care for their specific patient populations;
- Which patients are likely to become sicker over time;
- How physicians are caring for positive patients, including therapies and prescription medications used in treatment;
- Whether care provided to patients with Covid-19 is consistent with the latest scientific evidence; and,
- Where the disease is surging in real time.
Our application can’t yet do all of this, but it’s a start and it demonstrates how a surveillance app integrated onto disparate existing EHRs could. To the best of our knowledge ours is the first such app to show this capability. More than a dozen existing commercial products that leverage similar functionality and EHR access could be repurposed to do this.
The ability of our app to access cloud-based patient records in real-time across EHR systems depends on a little-known federal law called PAMA (The Protecting Access to Medicare Act of 2014). PAMA requires that most U.S. physicians use digital tools that enable the review of patient information in EHRs as the patient is being treated. Our app integrates with a decision-support tool that is used by more than 200,000 physicians and other clinicians as they order medical imaging procedures across nearly 35 health systems. When imaging is ordered, the app — with appropriate permission — can access the patient record. The app uses natural language processing and machine learning to scan clinicians’ free-text patient records and orders for terms including “trouble breathing” and “loss of sense of taste” amongst many other Covid-19-related signs, symptoms, and other indications of infection.
In tests, we have been able to rapidly identify patients who are presenting with signs and symptoms associated with Covid-19 syndrome. By subsequently associating these flagged patients with their Covid-19 test results we believe that the app can be trained to be highly sensitive and specific, identifying infected patients with a relatively low rate of false positives and negatives. Early testing has shown our symptom false-positive rate to be approximately 5%, which we expect to improve as the software continues to learn. We are now working to expand the app’s access beyond the narrow pool of patients receiving imaging orders to all patients receiving Covid-19 tests, and, from there, we will work to expand to a broader group of patients that have not yet been tested. Data will be collected when patients are evaluated by telehealth technology, in physicians’ offices, in urgent care and in emergency departments. This information is accessible through major EHRs and includes patients’ symptoms, signs, medications, laboratory tests, and other data.
The version that will survey all patients receiving Covid-19 tests, with or without an imaging study, will be rolled out in July 2020, before a potential second wave forecasted for fall or winter of 2020. Ultimately, when the app’s surveillance is expanded to the broader population, it will predict patients’ risk status and, with the ability to detect upticks in suspected cases from a 7-day moving average, the need for hospital beds. This forecasting ability will also allow providers to schedule elective surgeries and procedures to avoid Covid-patient surges.
Because central coordination of Covid-19 surveillance data is largely occurring at a state level, we are in discussions with state and federal officials to send data feeds from health systems to state health departments to automate what is mostly an antiquated and manual process. Patient privacy will continue to be carefully protected as manual processes become automated.
The passage of the PAMA law in 2014 has serendipitously enabled the rapid development of a potentially effective, real-time Covid-19 surveillance system. The required EHR platforms have already been installed in virtually every physician’s office and, as we have shown, straightforward modification of existing apps running on those systems can be readily rolled out. This approach has enormous potential for tracking the current pandemic, detecting subsequent waves of Covid-19, and modernizing population-level surveillance of emerging infectious disease.