Apply by Jan. 31 for Penn PORTAL career development funding to advance Learning Health System science. Learn more.

MEND

Three clinicians standing together, one holding a tablet showing the MEND logo

MEND

Enhancing mental health engagement, navigation, and delivery

Project status

Implementation
Scale

Collaborators

Cecilia Livesey, MD

Eleanor Anderson, MD

Innovation leads

Awards

UPHS Quality and Patient Safety Award, 2019

Funding

Innovation Accelerator Program

Opportunity

An estimated 50 percent of all Americans experience a mental health condition at some point in their lifetime, and approximately one in three individuals are simultaneously coping with physical and mental illness.

Delays and difficulties in identifying, engaging, and coordinating care for hospitalized patients with underlying mental health challenges can adversely affect patient outcomes, increase acute care utilization, and negatively affect patient and provider satisfaction. 

Data from the Hospital of the University of Pennsylvania (HUP) showed that in 2018, the presence of psychiatric comorbidity increased length of stay (LOS) by 22 percent, 30-day readmissions by 30 percent, and emergency department (ED) visits by 97 percent. In 2019, a survey of Penn Medicine general medicine providers revealed that 98 percent felt psychiatric comorbidities compromised their quality of care delivery.

The traditional inpatient psychiatric consult model is reactive, relying on a care team's perception and vocalization of need amidst the acute complexities that led to an individual's admission. It uses high-cost providers and is disconnected from longer-term efforts to address patients' psychosocial needs.

Intervention 

Mental Health Engagement, Navigation & Delivery (MEND) flips the mental health paradigm by integrating proactive psychiatry into existing inpatient care.

The program focuses on three key touchpoints.

  • Engagement: A machine-learning algorithm scans admitted patient records for over 2,000 buzzwords correlated with mental illness. The resulting MEND score identifies patients likely to benefit from mental health care on the first day of their hospitalization.
  • Delivery: MEND clinical teams, comprised of a part-time psychiatrist, full-time nurse practitioner, and full-time social worker, round with medical teams to discuss the psychiatric needs of patients on MEND floors. They offer advice on medication management, interpersonal dynamics, and crisis de-escalation and deliver mental health interventions throughout a patient's stay in the hospital. With all team members continuously and collaboratively working at the top of their license, only one in five cases requires the expertise of a MEND psychiatrist.
  • Navigation: Still in development, the MEND dashboard supports transitions to ongoing mental health support, whether that be an appointment with a mental health specialist embedded in Penn Medicine's primary care practices or specialty services in the community.

Impact 

By leveraging data, enabling proactive care, and shifting to a less physician-dependent staffing model, MEND offers a scalable and sustainable inpatient psychiatric care model that leads to improved care delivery and better patient outcomes.

Before the launch of MEND on two floors at HUP, about six percent of patients received psychiatric consultations, and these took place closer to discharge than admission. Since implementation, 44 percent of patients have received a proactive consultation, an eightfold increase, with the average connection occurring within 24 hours of arrival to a MEND floor.

MEND's proactive, comprehensive, and collaborative approach has driven a culture change and vast improvement to provider satisfaction, with one trainee labeling the service "life-changing." By shifting the work of mental health support onto the MEND team, the model enables medical teams to concentrate on acute medical concerns, leading to a reduction in length of stay across MEND floors of up to one day.

MEND has scaled to seven floors across HUP and Pennsylvania Hospital. By refining the algorithm, embedding the MEND workflow into the electronic health record, and leveraging technology to connect patients to ongoing care, the team is working diligently to implement MEND's proactive and integrated model as the standard of behavioral health care across all hospital settings at Penn Medicine.

Innovation Methods

Personas

A persona is a detailed description of a fictional character who might use a product or service. The characteristics of personas are derived from actual data and insights gathered about a target demographic. ...

Personas

MEND seeks to provide proactive support for the full spectrum of mental health conditions. With such a diverse pool of end users, one can quickly lose sight of individual stories and needs. Early in the project, we developed personas for individuals with an undercurrent of mental health concerns that were not solely defined by their...

Personas

A persona is a detailed description of a fictional character who might use a product or service. The characteristics of personas are derived from actual data and insights gathered about a target demographic.
 
Creating personas can help you understand users' needs, experiences, behaviors, and goals. Utilizing personas can help ensure that the product being developed caters to the preferences of the target user, not the team designing it.
 

Personas

MEND seeks to provide proactive support for the full spectrum of mental health conditions. With such a diverse pool of end users, one can quickly lose sight of individual stories and needs.

Early in the project, we developed personas for individuals with an undercurrent of mental health concerns that were not solely defined by their symptoms.

Reviewing and iterating in-depth personas for end users helped our team keep top of mind the neurodiversity of our patients and underscored the fact that a one-size-fits-all approach was not feasible for this patient population.

Assumptions matrix

An assumption is a statement about something that must be true for your solution to work. When you have defined a solution you'd like to test, ask yourself, "What must be true for this to work?" Once you have a...

Assumptions matrix

Using the assumptions matrix, we identified three "uncertain" and "devastating" assumptions that needed to be tested. First, we tested the assumption that we would have access to data. This was uncertain, given heightened privacy concerns around behavioral health data. Second, we tested the assumption that we could generate signal out of this...

Assumptions matrix

An assumption is a statement about something that must be true for your solution to work.

When you have defined a solution you'd like to test, ask yourself, "What must be true for this to work?" Once you have a complete list, plot your assumptions on a 2x2 matrix where one axis is how certain you are that your assumption is accurate and the other is how detrimental it will be if it is not.

Mapping assumptions will help you determine what you need to test to de-risk a potential solution. Assumptions that you are uncertain about and that are crucial for your solution to work are your riskiest assumptions.

Download template

 

Assumptions matrix

Using the assumptions matrix, we identified three "uncertain" and "devastating" assumptions that needed to be tested.

First, we tested the assumption that we would have access to data. This was uncertain, given heightened privacy concerns around behavioral health data.

Second, we tested the assumption that we could generate signal out of this data. Through a few chart reviews and quickly generated word clouds, we saw that the free note text was a promising insight source.

Third, we tested whether this signal would be meaningful to clinicians. We deployed a very early version of the algorithm to understand how the MEND team would use this data and how accurate it needed to be to garner trust.

Through these experiments, we justified investment in developing a natural language processing algorithm, which continues to be iterated on today.