University of Pittsburgh

Project Summary

Most young adults between the ages 18-26 in the United States are not prepared for the transition from pediatric to adult care, evidenced by myriad poor health outcomes prevalent in this population such as sexually transmitted infections, unintended pregnancy, depression, anxiety, suicidality, obesity, and suboptimal management of chronic medical conditions, at staggering societal costs. The majority of these health problems are preventable. Young adults tend to forego seeking routine primary care (PC), only presenting to emergency department (ED) or urgent care centers with acute problems. ED providers have neither time nor training to address most health issues identified during treatment and typically recommend primary care follow-up. Unfortunately, per local and national statistics, approximately 70% of these patients never follow-up.

eMCARE, is a two-way, adaptive, automated text messaging (TM) and app-based (planned) system to promote participation of young adults in routine primary care after an ED or urgent care visit. eMCARE conducts brief automated dialogue with participants for up to three months post- ED discharge and tailors messaging and feedback to each individual based on his/her risk factors, presenting problem and motivation level in order to optimally promote PC engagement. Connecting young adults to PC via a system like eMCARE that can be easily integrated into healthcare delivery will address the need and reduce costs associated with annual, preventable healthcare.

Researchers

Dana Rofey, PhD
Associate Professor, Psychiatry
University of Pittsburgh

Mayank Goel, PhD
Assistant Professor, Institute of Software Research and Human-Computer Interaction Institute
School of Computer Science
Carnegie Mellon University

Elizabeth Miller, MD, PhD
Professor in Pediatric and Public Health, University of Pittsburgh
Director of the Division of Adolescent and Young Adult Medicine
Children’s Hospital of Pittsburgh

Clifton Callaway, MD, PhD
Executive Vice-Chair of Emergency Medicine
Professor (tenured) od Emergency Medicine
Ronald D. Stewart Endowed Chair of Emergency Medicine Research
University of Pittsburgh

Sami Shaaban, MBA
Co-Founder/CEO of NuRelm

Project Summary

5.7 million patients are admitted to Intensive Care Units (ICUs) in the United States each year. Optimal evidence-based care for ICU patients is delivered by a multidisciplinary team, who round on patients with an accompanying computer workstation-on-wheels (WOW). However, the team must process many tasks and often fail to provide consistent evidence-based care, leading to preventable morbidity and mortality. An alternative and highly innovative approach is to leverage the WOW as a ‘smart’ listener and oral prompter, supporting checklist completion through dynamic adaption and interaction with the interprofessional care team at the point of care, offering an efficient approach to increase evidence-uptake, which has been shown to improve ICU outcomes.

The team’s solution, Aviva, is a voice-interactive virtual assistant that listens to the team rounding discussion, identifies gaps in the use of evidence-based practices, and selectively prompts the team to consider evidence-based practices when indicated – thus increasing evidence-uptake and improving ICU outcomes. Aviva uses automatic speech recognition (ASR) and natural language processing (NLP) technologies. It is applicable to other team-based care situations in healthcare and across the industry. The project team consists of critical care intensivists, computer scientists, and medical informaticists, and is uniquely poised to develop and deploy this technology to the clinic.

Project Status

The research team is continuing to carry out commercially orientated research on this topic.

Researchers

Andrew J. King, PhD
Postdoctoral Scholar, Biomedical Informatics
University of Pittsburgh

Derek Angus, MD, MPH, FRCP
Professor and Chair, Department of Critical Care Medicine
University of Pittsburgh

Jeremy Kahn, MD, MS
Professor, Critical Care Medicine
University of Pittsburgh

Christopher Seymour, MD
Associate Professor, Critical Care Medicine
University of Pittsburgh

Gregory F. Cooper, MD, PhD
Professor and Vice-Chair, Department, Biomedical Informatics
University of Pittsburgh

Project Summary

One of the major disadvantages of magnetic resonance imaging (MRI) is the time per exam.  A typical clinical exam slot is one hour in length but can extend to be as long as 90 minutes. This limits patient throughput and increases the per-exam cost, which typically is around $1,500 or even higher, due to the high infrastructure costs including equipment maintenance. In addition, extended time spent in the equipment increases stress for patients and their families, in particular in the pediatric population. Long MRI exam time requires sedation for very young patients or other patients who could not otherwise tolerate lying still in the magnet bore for such a long period of time. Current research has shown the harmful effects of sedation (especially repeated sedation) on neurodevelopment in children. Shorter MRI exam times should substantially reduce the number of patients who need sedation.

Reducing the exam time of an MRI scan would, therefore, increase patient comfort and compliance, throughput, and decrease the per-exam cost significantly. This could ultimately improve patient outcome and safety via extending the use of MRI to areas not typically used today, such as replacing X-ray CT scans in patients admitted to the ER with suspected head trauma.

fastMRI (fMRI) is a new type of MRI scanning method that would significantly reduce the time and cost of the traditional MRI scan. The technology involves software incorporating multi-contrast sequences that speed up the exam time, “intelligent” data under-sampling schemes, and machine learning for image reconstruction. fMRI can be applicable to a wide variety of MRI applications, including neurological, musculoskeletal, cardiac, and abdominal imaging.

Researchers

Vanessa Schmithorst, PhD
Associate Professor, Radiology
University of Pittsburgh

Ashok Panigrahy, MD
Professor and Radiologist-in-Chief, Radiology
University of Pittsburgh

Rafael Ceschin, PhD
Assistant Professor, Radiology
University of Pittsburgh

Project Summary

Perioperative stroke, defined as stroke that occurs during surgery or within 30 days after surgery, causes significant mortality and disability. Perioperative strokes are severely debilitating, affecting around 50,000 people every year in the US and are expected to rise as operations are increasingly performed on older patients with more complex conditions. The associated direct and indirect costs are approaching $16 billion.

Inadequate identification and treatment of perioperative stroke reduces the clinical effectiveness of the surgical procedure and can lead to disabilities or death. A perioperative stroke occurs from brain ischemia secondary to decreased blood flow to the brain, usually due to a blood clot traveling to the brain. Early identification of large strokes can result in life-saving mechanical clot removal.

Realtime Evaluation for Adverse Events using Intraoperative Neurophysiological Monitoring (READE-IONM) uses a proprietary machine learning algorithm to automatically mark or identify changes in intraoperative monitoring of brain function with an electroencephalogram (EEG), indicating ischemia in all regions of the brain. On detection of ischemia using READE-IONM, the neurologist can recommend interventions to the surgical team, including pausing the surgery or performing lifesaving mechanical clot removal, which can substantially decrease death and disability associated with perioperative stroke.

Project Status

The research team is continuing to carry out commercially orientated research on this topic.

Researchers

Parthasarathy Thirumala, MD
Associate Professor, Neurological Surgery and Neurology
University of Pittsburgh

Shyam Visweswaran, MD, PhD
Associate Professor, Biomedical Informatics
University of Pittsburgh

Amir Mina, MD, PhD
Medical Student – Biomedical Informatics
University of Pittsburgh

Project Summary

Abdominal aortic aneurysm (AAA) is the 13th leading cause of death
in westernized countries. If left untreated, aneurysm growth may lead
to high-morbidity aortic rupture. Early endovascular repair is the goal
for most AAA treatment tracking regimes, as it reduces overall costs of
AAA intervention when compared to direct costs associated with AAA
monitoring over a 4-year period. Currently, the only variables clinicians
have available to quantify small AAA complication risk are maximum
diameter and rate of AAA growth. There are only a small number
of software packages that increase rupture risk estimation beyond
simple diameter measurements which solely focus on biomechanical
properties, do not include patient history (age, gender, BMI, etc.), AAA
shape irregularities, or AAA progression over time, and have yet to show
predictability in rupture.

Aneurysm Prognosis Classifier (APC) uses the latest advancements in
machine learning to predict the complication risk of a small AAA. APC
algorithm provides clinicians with an objective, predictive tool to guide
surgical intervention decisions before symptoms appear. There are
currently companies performing computational simulations reporting
traditional wall stress metrics, but they have not demonstrated the ability
to predict AAA outcomes consistently.

Researchers

David Vorp, PhD
Professor, Bioengineering
University of Pittsburgh

Michel Makaroun, MD
Chief, Division of Vascular Surgery,
Department of Surgery
University of Pittsburgh

Nathan Liang, MD, MS
Assistant Professor, Division of Vascular
Surgery, Department of Surgery
University of Pittsburgh

Timothy Chung, PhD
Research Specialist V, Department of
Bioengineering
University of Pittsburgh

Project Summary

Chronic pain impacts more than one third of the U.S. population,
generating total annual healthcare costs exceeding $600 billion. A key
contributor to the growing financial burden of chronic pain is the fact that
patient response to different pain treatments varies dramatically between
individuals. As a result, providers making critical therapeutic decisions
for managing pain have little idea whether their chosen treatment will be
of any benefit to a given patient, leading to repeated treatment failure,
reduced quality of life, and increased risk of opioid addiction.

Personalized Pain Treatment (PPT) is a shared decision-making tool that
utilizes machine learning to predict an individual patient’s response to
various pain treatments. The PPT algorithm applies Bayesian analyses
to an extensive repository of medical records and patient reported
outcomes to generate a report detailing the individualized probability of
treatment success. Unlike other personalized medicine prediction tools,
which are limited to forecasting drug responses, PPT can also analyze
the potential efficacy of physical therapy, psychology, and other biopharmaceutical treatments, or combinations thereof. Furthermore, PPT
is unconstrained by any requirement for the collection and processing of
biological samples, allowing the generation of treatment reports in real
time at the point of care.

Researchers

Ajay Wasan, MD
Professor, Anesthesiology
University of Pittsburgh

Jong Jeong, PhD
Professor, Biostatistics
University of Pittsburgh

Gregory Cooper, MD, PhD
Professor, Biomedical Informatics
University of Pittsburgh

Project Summary

Each year over 246,660 women are diagnosed with breast cancer in the
U.S., with 80% classified as estrogen-receptor-positive (ER+). Currently,
~50% of ER+ breast cancer patients receive adjuvant chemotherapy,
which has substantial side effects and toxicity, in addition to the primary
treatment (most often surgery or radiation). Only 4% of the patients benefit
from these therapies with no recurrence of breast cancer in the next 10
years. Currently, there are only limited risk prediction tools available in the
clinic for diagnostic and prognostic testing, and to guide decision-making
about whether to offer adjuvant chemotherapy to a patient.

PI-predictor is a computational model that combines standard pathology
parameters (a manual version is already available through Magee
Womens Hospital) and radiological imaging features obtained via
magnetic resonance imaging and digital mammography to replace the
Oncotype DX assay. PI-predictor is a fast (<5 mins) and cost-effective
method with no additional cost of using clinically readily available
pathology and imaging data.

Researchers

Shandong Wu, PhD
Associate Professor, Radiology
University of Pittsburgh

David Dabbs, MD
Professor, Pathology
University of Pittsburgh

Rohit Bhargava, MD
Professor, Pathology
University of Pittsburgh

Project Summary

Cancer is a heterogeneous disease composed of various cancer cell,
clonal sub populations and other types of cells that comprise the tumor
microenvironment (TME). The heterogeneity within the TME is a major
challenge for accurate diagnostic and prognostic tests, and the spatial
context of the cancer cells and stromal cells, including the migratory
immune cells within the TME, must be determined to properly diagnose
the specific disease subtype and optimal treatment options.

Spatial Pathology Powers Cancer Diagnostics (SPDx) is a digital pathology
software analytics tool that enhances the practice of pathology
through the development of new machine learning software tools to
computationally guide pathologists’ decisions. Unlike competing digital
pathology tools that only analyze digital whole slide images…

in the absence of spatial context and intra-tumor heterogeneity,
SPDx provides objective and measurable spatial guidance for tissue
structures and biomarker relationships that include measures of spatial
heterogeneity within the patient’s tissue slides. The incorporation of
spatial heterogeneity measurement into pathological workflows enables
precision medicine approaches to be incorporated into diagnostic and
prognostic activities, including prediction of tumor metastases and
optimal therapeutic treatment planning. SPDx provides value through
faster, quantitative, objective, and more accurate decisions for both
current and next-generation digital pathology workflows.

Project Status

SPDx formed a startup called SpIntellx.

The SpIntellx research team has raised funding from private investors and
has also received a Phase I Small Business Innovation Research (SBIR)
award from the National Science Foundation.

For more information, visit spintellx.com or reach out to the PHDA at
healthdataalliance.com/contact.

Researchers

S. Chakra Chennubhotla, PhD
Associate Professor, Department of
Computational and Systems Biology
University of Pittsburgh

D. Lansing Taylor, PhD
Professor, Department of Computational
and Systems Biology
University of Pittsburgh

Project Summary

Word sense disambiguation is a fundamental problem, particularly in
clinical natural language processing (NLP). High accuracy acronym
and abbreviation disambiguation is important for all clinical NLP tasks,
as 71% of identified abbreviations in clinical text could be ambiguous
in their meanings. Clinical NLP can unlock critical patient case details
from unstructured clinical texts, such as patients’ health records. The
downstream decisions relying on clinical NLP can be incorrectly applied,
in both clinical and research settings, if word, abbreviation, and acronym
ambiguity is incorrectly interpreted.

Clinical Abbreviation Resolution Engine (CARE) with Deep Sequential
Learning is a deep learning method that addresses word sense
disambiguation to significantly improve text information extraction from
electronic medical record sources of high value, such as admission notes,
consults, and discharge summaries. CARE has the potential to improve
clinical NLP from an 80% accuracy rate (where it has been stalled for
years) to over 95% by addressing these key additional word tokens.

Project Status

The research team is pursuing commercial paths for the innovative solutions that they have built in this project.

For more information, reach out to the PHDA here.

Researchers

Daqing He, PhD
Professor, Informatics
University of Pittsburgh

Project Summary

Precision oncology aims to treat patients based on the genomic makeup
of their individual tumor. However, current methods are limited in their
approach and scope. For example, immunotherapy drugs are currently
used as first line treatment for many solid tumors, including melanoma,
without prior genetic testing. While around 30% of patients benefit from
this approach, no solution exists to select better treatments for the large
group of non-responders, leading to suboptimal outcomes and higher
costs. To help address this problem, DioneX has developed algorithms
that can estimate disease mechanisms of individual tumors and predict
the most effective, personalized treatment for each patient.

The core technology of DioneX is TDI, Tumor Driver Identification,
an engine that employs causal inference modeling and data mining
integration of genomic and transcriptome characterizations of individual
tumors. The algorithm estimates the causal relationships between gene
alterations (M) and molecular phenotypes (P) within each individual tumor.
Using Bayesian causality analysis theory, a patient-specific predictive
model of the individual disease mechanism is constructed. This novel
approach is designed to provide oncologists with data-driven decision
support that improves diagnosis and selection of individualized, targeted
cancer treatments.

Project Status

The TDI team received follow-on funding after the completion of their PHDA project and continues research toward creating a commercial solution.

For more information, reach out to the PHDA here.

Researchers

Xinghua Lu, MD, PhD
Professor, Biomedical Informatics
University of Pittsburgh

Gregory Cooper, MD, PhD
Professor, Biomedical Informatics
University of Pittsburgh

Project Summary

Complex human diseases and most cancers often remain undetected
until they become incurable or challenging to treat. Early diagnosis is
frequently impossible or dangerously invasive by current methods.
Treating advanced chronic diseases and cancer consumes the majority
of health expenditure. To solve these problems, AuguryDx provides a
platform technology that can be used to develop non-invasive, disease
specific, early tests for screening or diagnosis of a variety of diseases. The
AuguryDx platform is currently being validated through proof-of-concept
studies in women’s health and infant diseases, with plans for developing
oncology applications in the near future.

Project Status

The AugruryDX research team is exploring paths to market for their technology.

For more information, reach out to the PHDA here.

Researchers

David Peters, PhD
Associate Professor, Department of
Obstetrics, Gynecology and Reproductive
Sciences
University of Pittsburgh

David Finegold, MD
Professor, Genetics
University of Pittsburgh

Tianjiao Chu, PhD
Statistician, Department of Obstetrics,
Gynecology and Reproductive Sciences
University of Pittsburgh

Project Summary

Preventable medication errors cost an unsustainable ~21 billion dollars
annually. Consequently, healthcare spending has shifted from a feebased
model to one focused on value and cost savings. Payors rate
institutions, pharmacies, and providers use quality metrics to justify
payment, some of which are based on patient medication experiences.
These groups are spending millions of dollars annually to prevent
medication use problems through medication reconciliation tasks and
education in order to improve their ratings, retain patients, and build
efficiency. Patient engagement is a key aspect to achieving high quality,
affordable care. The uptake of health-focused personal technology is
exploding, but few products target medication outcomes, an estimated
more than $161 million market.

MEDIvate is a simple-to-use, patient-centered smartphone application
that empowers patients and providers to achieve great medication
outcomes. Current market alternatives/barriers are costly, cumbersome,
and often still paper-based. MEDIvate’s approach will be successful
because it makes patient medication lists up-to-date, portable, and easy
to share. Current medications are added directly to the app from EHRs
or by the patient, and are always accessible. Patients trigger easy sharing
of their personal medication history with their healthcare providers at the
point of care. This ensures accuracy to reduce medication errors and
saves time to improve transitions of care. MEDIvate is also a personal
medication coach. It reminds patients to take their meds and intelligently
links key facts/educational videos on-demand from pharmacist experts

Project Status

The MEDIvate team is continuing research and exploring use cases in multiple patient populations.

For more information, reach out to the PHDA here.

Researchers

Philip Empey, PharmD, PhD
Associate Professor, Pharmacy
and Therapeutics
University of Pittsburgh

James Coons, PharmD
Associate Professor, Pharmacy
and Therapeutics
University of Pittsburgh

Philip Empey, PharmD, PhD
Associate Professor, Pharmacy
and Therapeutics
University of Pittsburgh

James Coons, PharmD
Associate Professor, Pharmacy
and Therapeutics
University of Pittsburgh

Project Summary

Despite the use of targeted therapies, lung cancer remains a significant
problem with a 5-year survival rate of 18%, and 25% of the annual cancer
deaths. Inflammation is a critical component of lung tumor progression,
and maintaining immune homeostasis in lung cancer patients is critical to:

1. Decrease inflammation and enhance the therapeutic effects
of chemotherapy.

2. Reduce anti-inflammatory co-therapy that causes severe side
effects.

3. Reduce tumor-promoting myeloid cells and macrophages that
promote tumor growth and drug resistance.

OncoBioelectrx is a drug-free, implantable immunotherapy
neuromodulation device designed to stimulate anti-inflammatory
pathways to achieve inflammatory homeostasis that can simultaneously
repress inflammation and boost antitumor immunity.

Project Status

The research team is optimizing their technology on all fronts and continues to take a multidisciplinary approach.

For more information, reach out to the PHDA here.

Researchers

Charles Horn, PhD
Associate Professor, Medicine
University of Pittsburgh

Gutian Xiao, PhD
Professor, Microbiology and Molecular
Genetics
University of Pittsburgh

Lee Fisher, PhD
Assistant Professor, Physical Medicine and
Rehabilitation
University of Pittsburgh

Gary Fedder, PhD
Professor, Electrical and Computer
Engineering
Carnegie Mellon University

Christopher Bettinger, PhD
Associate Professor, Biomedical Engineering
Carnegie Mellon University

Project Summary

Cardiovascular disease (CVD) is the leading cause of morbidity and
mortality and causes 1 in 3 deaths in the U.S. (a total of 800,000 annually).
CVD has taken a disproportionate toll on many racial and ethnic groups
that have higher rates of CVD and its risk factors, and CVD accounts
for about one-third of the disparity in potential life-years lost between
blacks and whites. A prominent example of a CVD risk factor that varies
based on race and ethnicity is vulnerable atherosclerotic plaques, a major
culprit for CVD events. However, patient and subtype-specific CVD risk
assessment is currently lacking from clinical practice. Thus, there is a
great need for tools to detect, diagnose, and stratify patients that would
allow doctors to provide tailored interventions to high-risk groups.

CADidME is a risk prediction tool that characterizes CVD events
according to specific subpopulation characteristics, including
metabolomic data. CADidME tools will enable precision CVD diagnosis
and management to reduce adverse CVD events in patients. CADidME
uses comprehensive metabolomics profiles to standardize and replace
multiple biomarker screening tests to benefit consumers, clinicians, and
healthcare insurance through simplification and cost-effectiveness.

Project Status

The CADidME research team has established novel techniques for biomarker discovery and is exploring high impact applications.

For more information, reach out to the PHDA here.

Researchers

Vanathi Gopalakrishnan, PhD
Associate Professor, Biomedical Informatics
University of Pittsburgh

Steven Reis, MD
Professor, Medicine
University of Pittsburgh

Project Summary

Hospital acquired pressure ulcers (HAPUs) are areas of localized skin
and soft tissue destruction caused by prolonged pressure in debilitated
patients. The estimated PU prevalence is 3 million patients at a cost of
$3.6 billion per year. Several interventions and preventive measures are
recommended to avoid hospital acquired pressure ulcers, including:
patient repositioning, proper nutrition, pressure-relieving support
surfaces, pneumatic mattresses, and skin care. Repositioning patients in
bed is particularly a key preventative measure and a target of opportunity
for low-cost innovative technology-based solutions.

Pressure Ulcer Monitoring Platform (PUMP) provides solutions for
improving compliance with patient repositioning through nursing
intervention solutions. It combines (1) a low-cost, but sophisticated
wearable sensor that automatically detects when patients are
repositioned and wirelessly records the event in the medical record,
which is more suitable to patients with shorter lengths of stay, (2) a
second sophisticated sensor device placed under the wheels of each
hospital bed, which is more suitable for patients with longer lengths of stay or for those patients that are not suitable for a wearable
device, and (3) an electronic alert system via mobile phone SMS to
change nursing behavior and increase compliance.

PUMP aims to minimize the operational and maintenance efforts by doctors and nurses, reduce obtrusiveness for patients, and to achieve the highest system reliability and minimal system cost.

Project Status

The PUMP research team completed their PHDA/CCA project and a team member went on to found Pitt spinout company eWear Technologies, which is developing wearable sensor technologies.

For more information, reach out to the PHDA here.

Researchers

Peter Rubin, MD
Professor, Department of Plastic Surgery
University of Pittsburgh

Mingui Sun, PhD
Professor, Department of Electrical &
Computer Engineering
University of Pittsburgh

Project Summary

Falls are the leading cause of fatal and nonfatal injuries among adults
aged 65 years or more. This finding, and the aging of our nation, suggests
that national attention on the problem of falls will continue to increase.
Tools that can help reduce falls are badly needed in the skilled nursing
facility setting, where 45-64% of the patients experience a fall each year.
The mean incidence of falls is 1.7 falls per bed per year, 10-25% of which
result in fracture or laceration. The market for skilled nursing facility fall
prevention tools includes approximately 15,700 facilities that provide care
for roughly 1.4 million residents. Treatment of falls in the nursing home
is estimated to cost about $5 billion per year and can result in further
litigation risks.

Fall Sentinel is an automated risk monitoring system that applies a
validated patient level prediction model to determine patient fall risk
based on medication administration and Minimum Dataset data.
Fall Sentinel technology functions by processing a stream of electronic
clinical data to provide highly patient-specific and clinically actionable
alerts to clinicians when residents transition to a state of unacceptably
high risk for experiencing a fall while exposed to a potential drug
interaction or other fall-associated medication weak spot.

Project Status

The research team completed their PHDA/CCA project and is now pursuing multiple lines of research to develop data-enabled solutions with commercial potential.

For more information, reach out to the PHDA here.

Researchers

Richard Boyce, PhD
Associate Professor, Biomedical Informatics
University of Pittsburgh

Steven Handler, MD, PhD
Associate Professor, Geriatric Medicine
University of Pittsburgh