The Penn Medicine Center for Health Care Innovation (Philadelphia, PA) was selected as the winner of ECRI Institute's 11th Health Devices Achievement Award for its development of a technology platform for creating customized apps that prompt care providers when actions are required for targeted clinical populations.
Apps built on the "Agent" platform, as it is called, can pull real-time data from multiple electronic health record (EHR) and clinical data sources to address a clinical need. Apps can be designed to identify patients who require a particular intervention (as defined by the use case), to alert care providers when actions are required, and to enable care teams to collaboratively manage tasks on a shared dashboard. By bringing to clinical teams relevant, timely data that previously was not easily accessible, the Agent platform helps care providers make better use of this data to improve patient care.
The Health Devices Achievement Award recognizes innovative and effective initiatives undertaken by member healthcare institutions to improve patient safety, reduce costs, or otherwise facilitate better strategic management of health technology. ECRI Institute announced in May 2017 that the Penn Medicine Center for Health Care Innovation received the top honors this year. For details about the other submissions that achieved recognition, see
The Health Devices Achievement Award: Recognizing Exceptional Health Technology Management.
ECRI Institute congratulates the project team members: Katherine Choi, MD, Yevgeniy Gitelman, MD, Damien Leri, MSEd, MPH, Roy Rosin, MBA, David Asch, MD, MBA, and Shivan Mehta, MD, MBA, MSHP, from the Penn Medicine Center for Health Care Innovation, along with the supporting team from the Penn Medicine Information Services department.
David T. Jamison, Executive Director, Health Devices Group (left), presents the 11th
Health Devices Achievement Award to project leaders from Penn Medicine. |
The Challenge
To proactively identify patients who require a particular intervention—using real-time data from multiple EHR and clinical data sources—and to communicate the need for such actions to the providers caring for those patients.
The Landscape
1. The delivery of optimal patient care requires that care providers have access to relevant information in a timely manner.
a) Providers need to be apprised of:
(1) Events that can affect a patient's care—for example: an expiring order for a critical medication
(2) Actions that may be required to keep those patients safe—for example: conducting an extubation risk screen for intubated patients
b) Providers may also wish to identify cohorts of patients who will require similar interventions, at times extending across service lines or care teams.
2. Although valuable clinical information is available to care providers, barriers exist to obtaining relevant information in a timely manner.
a) Much of the information that care providers need can be accessed through EHRs and other data storage systems.
b) However, traditional EHRs are built as information stores, not information delivery systems. They rely on providers actively seeking out the data they need. As a result:
(1) Providers incur the burden of screening data one patient at a time, as dictated by the EHR.
(2) A provider's situational awareness is only as good as the timing and quality of his or her last search.
c) Additionally, data may be stored in disparate information systems. At the University of Pennsylvania Health System (Penn Medicine), for example, the needed data might reside in the EHR or in a data warehouse system.
3. The Penn Medicine Center for Health Care Innovation was tasked with developing a system that could inform the care providers who are responsible for certain populations of patients about the actions needed during the care of those patients.
a) The Center for Health Care Innovation
describes its purpose as facilitating "the rapid, disciplined development, testing, and implementation of new strategies to reimagine health care delivery for dramatically better value and patient outcomes."
b) The Center's team takes an iterative and agile approach to designing, testing, and implementing solutions. The organization specifies that it strives to experiment quickly at low cost, scaling only when high-impact solutions are identified.
The Process
1. The Penn Medicine Center for Health Care Innovation team established the goal of creating a technology platform that could:
a) Extract actionable data about specific cohorts of patients from various data storage systems
b) Deliver that data to the appropriate care providers—as push notifications or on a shared dashboard
c) Be customized for specific use cases
2. The result was an innovative dashboard and alert platform called "Agent"—an application that can be used by Penn Medicine providers to address specific clinical information needs, as outlined below.
3. Features of the Agent platform include the following:
a) Data extraction—Custom-built APIs (application program interfaces) enabled by the Penn Medicine Information Services department allow data to be pulled from the University of Pennsylvania Health System's EHRs and other data warehouses.
b) Data delivery—The platform can deliver information in two ways:
(1) A secure push notification channel (e.g., text, e-mail)
(2) A "pull" view using a dashboard with a live-updating patient list
c) Application customization—The team works closely with clinicians to identify potential use cases (i.e., problems that can be solved) and then uses the Agent platform to build an application to address the problem.
4. Use cases that have been implemented to date include the following:
a) Preventing missed doses of critical medications due to an expiring order
b) Improving compliance with conducting extubation risk screens for intubated patients
c) Managing, in real time, patients with a history of high emergency department utilization
d) Coordinating transitions of care for patients on hemodialysis from inpatient care to an outpatient hemodialysis unit at the time of their discharge
The Results
Penn Medicine reports that the Agent app-development platform has enabled teams across multiple entities within the University of Pennsylvania Health System to improve patient safety and reduce harm, as illustrated by the following use cases:
1. Preventing missed doses of critical medications due to an expiring order
a) Challenge to be addressed:
(1) In an effort to limit unnecessary antibiotic usage, Penn Medicine implemented automatic stop dates for antibiotics at 7 days.
(2) One unintended consequence of this effort, however, is that it led to orders for medications expiring without the awareness of the primary team. If the patient needed to remain on the medication beyond 7 days, missed doses would sometimes result.
b) The goal for the Agent team was to design an intervention that could decrease missed doses of critical medications that result from the delayed response to an expiring medication order.
c) The Agent solution involved a combination of the following:
(1) Sending an automatic reminder to front-line clinicians about expiring medication orders (limited to only critical medications not reordered at the time of expiration)
(2) Providing a list of expired medication orders to the inpatient pharmacist so the pharmacist could verify, reorder, or engage the front-line clinician to address impending expirations before doses were missed
d) The effectiveness of the Agent app was assessed using a before-and-after comparison that looked at one antibiotic as a marker for missed doses.
(1) In the two months after the start of the intervention, the number of missed doses was decreased by 45% compared with before the intervention.
(2) Over a six-month period prior to the intervention, 9 patients missed more than 5 consecutive days of the antibiotic. In the first two months after the intervention, only 1 patient had missed 5 consecutive days.
(3) On average, pharmacists were catching one case every other day in which an antibiotic expired inadvertently at two hospital sites.
(4) During the first year, more than 350 interventions were made by pharmacists at five hospital sites using Agent.
2. Improving compliance with conducting extubation risk screens for intubated patients
a) Challenge to be addressed:
(1) Penn Medicine's medical intensive care unit (MICU) had developed a high-risk extubation pathway in response to several sentinel events related to surgical airway emergencies following extubation. Compliance was poor initially, and extubation risk screens for intubated patients were often not conducted.
(2) One barrier to improving compliance was that information about whether a risk screen had been conducted had to be pulled manually and could only be accessed retrospectively. This prevented the MICU director from being able to provide real-time education to front-line clinicians, who rotated through biweekly, about the screening protocol.
b) The Agent app helped improve compliance by:
(1) Allowing the MICU director to follow the cohort of ventilated patients
(2) Notifying the MICU director if a screen had not been performed
c) With this information, the MICU director could, within the hour, remind the relevant provider to complete the screen.
d) Compliance in completing extubation risk screens for intubated patients in the MICU increased from 43% before the intervention using Agent to 93%.
e) Additionally, the app is used to inform anesthesiologists every morning about the high-risk patients.
f) Penn Medicine reports that:
(1) Improving the awareness of and preparation for high-risk extubations has been valuable from a patient safety, clinical effectiveness, and quality improvement standpoint.
(2) Use of the Agent app in the MICU eliminated the types of safety events that continued to occur in other ICUs, where Agent had not been deployed. This has led to broader adoption of Agent in additional ICUs within the health system.
3. Identifying, in real time, patients with a history of high emergency department utilization
a) Challenge to be addressed:
(1) The family medicine department established a high-touch Priority Access Program for patients with a history of high emergency department (ED) utilization patterns. This program was intended to help staff intervene to address the patients' underlying needs so that care could be delivered in a lower-cost setting, when appropriate.
(2) However, enrollment in this program was hindered by an inability to quickly identify patients who were superutilizers while those patients were in the ED.
b) The Agent platform provided staff with a dynamic patient list and dashboard view, allowing the family medicine team to identify and track these patients so that interventions could target the underlying causes of high utilization—in particular, transportation needs, medication non-adherence, and lack of mental health access.
c) With Agent:
(1) The family medicine team has been able to manage more than 300 high-utilization patients outside the ED in the past year. In one case, a patient who returned to the ED repeatedly in July because of hot weather was supplied with an air conditioner (to replace his broken one); this addressed his underlying need and reduced his visits to the ED.
(2) Over the course of a year, ED utilization by patients identified as superutilizers was decreased by more than 36%, and utilization of primary care services by this group increased by over 50%.
(3) The cost of care for this group decreased by $270,000 last year.
4. Identifying patients who require coordination with an outpatient hemodialysis unit at the time of their discharge
a) Challenge to be addressed:
(1) Dialysis patients require special communication between the inpatient renal team and outpatient nephrologist.
(2) The inpatient teams struggled to find out when patients would be discharged.
b) The Agent platform was used to identify such patients and to notify renal fellows.
(1) In less than a year's time, hundreds of notifications have been sent to renal fellows to help coordinate discharges for specific, eligible patients.
(2) Preliminary surveys showed that 20% of the time, the renal team would have otherwise not been aware of the patient's discharge. In additional cases, the message provided by the Agent app prompted them to contact the outpatient unit hours or days earlier.
5. Other applications: Examples of additional applications for which the Agent platform has been used to date include the following:
a) Pushing evidence-based pathways to providers treating patients suspected of having a gastrointestinal bleed, delivering this information at a time when it can inform the care provider's decision making.
b) Enrolling patients who are discharged with IV antibiotics in a monitoring and management program that has helped decrease readmission rates.
c) Sending push notifications to PICC (peripherally inserted central catheter) team members about incoming orders for ultrasound-guided IVs, informing them of these orders in real time, while they are still in the care areas.
d) Enrolling eligible patients for clinical trials—while they are still in the hospital—using a real-time patient list populated based on eligibility and exclusion criteria.
Key Takeaways
1. The Center for Health Care Innovation team notes that forming partnerships with passionate clinical champions was absolutely critical to the success of the examples listed above. To help assure that each app will successfully meet a clinical need, the team:
a) Solicits ideas for potential use cases from front-line clinicians and staff
b) Works closely with those individuals throughout the design and testing phases
2. This collaboration facilitates another key to the team's success: rapid iteration. App designs are adjusted based on feedback from the end users, leading to designs that effectively meet the users' evolving workflow and complex needs.
3. A key technological factor is the need to access real-time data from multiple EHR and clinical data sources.
a) The Information Services department custom-built APIs for this purpose.
b) This allowed the app-development team to build, use, and experiment with custom views and to adapt to the workflows of the clinical teams who wanted to use that data dynamically.