Digital Health in the Clinical Trenches : A Reality Check

We often hear that healthcare lies on a precipice of transformative change. Each day brings news of an innovation poised to alter the practice of medicine. Technophiles announce that digital health advances such as electronic health records (EHR), mobile health (mHealth) applications, wearable biosensors and telehealth permit inexpensive and seamless data collection and processing, allowing previously unimaginable delivery of meaningful data from patients to healthcare providers, administrators and analysts. Put simply : We’re told by medical futurists, like Eric Topol, that modern technologies are dramatically transforming healthcare for the better.

We need a reality check. This grand vision is still barely perceptible in the clinical trenches of everyday healthcare delivery. For those of us who take care of one patient at a time, looking them in the eye, hearing their stories and making countless decisions every hour, the lofty talk of digital health innovations has almost no relevance. To date, the most prominent sign of healthcare’s digital transformation is the often-maligned electronic health record (EHR) that physically and emotionally separates patients and their providers.

Despite pervasive advertisements for Apple HealthKit and Google Fit, we see very few current examples of wearable biosensors driving point-of-care decisions in clinical practice. There are more than 165,000 mHealth applications in the Apple iTunes store, but only a fraction of these communicate with EHRs to create actionable insights at population scale. And although telehealth has made inroads, it is still a distant reality for the vast majority of clinicians. Most doctors still care for patients the way they always have: face-to-face in a clinic with minimal technology. Moreover, although the tech community is constantly abuzz with possibilities of digital health, the evidence supporting these technologies is woefully inadequate, and advocates frequently argue beyond what current data support.

However, my own hospital, Cedars-Sinai Medical Center, began offering direct connectivity of Apple HealthKit, FitBit, and other wearable biosensors to our EPIC EHR system. Within days, several hundred patients signed up for the service and are now actively transmitting their biosensor data into the health record. This was an incredible leap forward and a digital triumph for Cedars-Sinai, which is now among the largest health systems in the world to offer this service to its patients. But this also raises questions: Who is in charge of the data? Should we monitor these data in real time and make clinical decisions? How will these data improve the value of care that we provide to our patients? All of these are research questions, and our team and others are working to provide answers. In the rest of this essay, I will discuss a model for how digital health can improve value from the perspective of the digital trenches, written from the perspective of a health services researcher, and provide examples of how we are addressing these pressing questions throughout our health system at Cedars-Sinai.

Healthcare Delivery as a “Biological System”: Implications for Integrating Digital Health Into Clinical Care

Despite under-penetration of digital technologies and still inadequate evidence to support their use in practice, these innovations may still play a vital role in delivering healthcare value. To build a case for connecting patients and providers with digital health, we need to first perform another reality check about a related topic: coordination of care. Once we consider theory versus reality around this sacrosanct principle of healthcare reform, we will better understand how digital health can deliver value on a large scale.

In a traditional fee-for-service healthcare system, a sick patient is worth more than a healthy patient and high volume services drive profit. But in a capitated population health model, such as an accountable care organization (ACO) or a Medicare Advantage arrangement in which providers are incentivized to reduce the total cost of care within fixed boundaries or risk financial penalties, care coordination becomes paramount. When performed correctly, care coordination can improve quality, reduce hospital admissions, avoid service duplications, de-fragment care delivery and reduce unnecessary expenditures. This all makes good sense.

Still, there is a distinction between the theory of coordinating care and the structural realities of healthcare delivery. Although the integrated delivery model treats covered lives as a single population to be managed through coordination, the ultimate unit of care delivery – the most “cellular” structure of the organism – is the patient-provider relationship itself. Population health is ultimately delivered one interaction at a time, in one room at a time, generally between one doctor and one patient at a time. These interactions, when multiplied by tens of thousands every day, form the cellular foundation of the healthcare delivery organism. Therefore, to understand population health, we need to understand the inherently messy, nonlinear, unstructured and deeply human interaction that occurs between patients and their providers.

The biological analogy is apt. Bear with me for a second as I extend this analogy, because it pays dividends to understand the role of digital health in complex health systems. Consider the tightly coordinated interplay between micro and macro structures in a living system. Cells multiply to form organs; organs serve functions; functions prompt behaviors; and behaviors accompany cognitions and emotions. The organism assumes higher-order features at each level. The cellular structures give rise to thoughts, feelings and even quality of life – phenomena that bear no obvious resemblance to their underlying cellular origin. Although the organism assumes emergent characteristics when considered as a whole, those characteristics are epiphenomena rooted in the most granular, underlying structural unit. In the case of population health, that’s the patient-provider relationship.

There’s more to this biological analogy. The organism collects signals from innumerable stimuli, integrates and analyzes the signals through its central nervous system, and then feeds back to regulate its microstructure. Stress and anxiety, for example, trigger a neuro-hormonal cascade that alters the function of organs and their constituents. Physical exertion triggers urgent shifts in physiologic pathways to support the dynamic needs of the whole organism. In case after case, the biological system remains highly efficient by coordinating its micro- and macroscopic levels. In most cases there are redundant or back-up systems that maintain vital internal communications in case of failure. The key to success lies in the “three Vs” of any big data system: volume, variety and velocity. By coordinating a huge volume of data from a large variety of sources, and then processing the data with unparalleled velocity, biologic systems can adapt instantaneously to circumnavigate threats.

The same principles should apply to an integrated health delivery organization. Patients interact with providers on the most granular level; decisions are made; decisions lead to actions; actions consume resources and ultimately drive patient and economic outcomes. The organization’s “nervous system” collects incoming data and monitors its performance with population-level vital signs. Administrators track total cost of care; pharmacy managers scan for prescriber outliers; utilization managers watch for physicians with unusual practice patterns; case managers monitor high-risk patients for missteps, decrements or maladaptive behaviors. Co-located care teams integrate incoming data, run prediction algorithms, and spin data into information and knowledge.

Equipped with this knowledge, the integrated nervous system fires signals back to the periphery: Outlier physicians receive audit results and counseling. Case managers pay high-risk patients timely visits. Utilization reviewers remind clinicians about care pathways. Pharmacy benefit managers deny inappropriate prescription requests. And so forth.

This analogy yields key insights to maximize efficiency; it also creates a blueprint for how to intervene with digital health technologies. First, the insights:

1. Start at the ground floor by understanding and optimizing the patient-provider interaction.
2. Recognize that providers working in the trenches are not in a position – nor should they be – to understand the organization as a singular unit. Interventions targeted at the clinical environment must be designed within the constraints and expectations of people working at that level.
3. The health delivery organization is an inter-dependent system – not just a collection of parts. This interdependence mandates a network of finely tuned, highly reliable connections that ensure accurate and timely data exchange.
4. Higher-order coordinating centers must collect and process a large volume and variety of data with_velocity. _
5. The coordinating centers must convert data into information and knowledge. This requires both “big data” algorithms to find signals in the noise, and “thick data”analysis layers to understand the rich context of the data. That means skilled and experienced humans need to interpret the zeros and ones streaming into the coordinating center.

6. The coordinating center must complete the loop by feeding back to the ground floor. The feedback should be specific, actionable and understandable to those on the receiving end in the clinical trenches.

   Digital Technologies: Serving Micro and Macro Units of the Health Delivery Organism

What does this all mean for digital health technologies? The biological analogy creates a blueprint for how to implement digital health in an integrated care model. Our Health Services Research Department is using this blueprint to start from the ground floor and work our way up to the top of the healthcare organization.

For example, surgeons and nurses in the clinical trenches told us that post-operative length of stay is largely driven by confusion about when to feed patients. We developed a biosensor that adheres to the abdominal wall, monitors digestion sounds via computer, and displays the results on a bedside screen. The clinicians thought this was good, but not good enough. They asked for a “stoplight” with three colors – red for “no feeding,” yellow for “start liquids,” and green for “start solids.” So, we created new stoplight algorithm and found the solution was well received. Moreover, we conducted analytics with this sensor and can now predict who will develop bowel paralysis after surgery with around 80% accuracy. This is an example of research that requires cross-discplinary work from physicians, surgeons, statisticians, graphic designers, computer scientists, and electrical engineers.

Another inpatient problem of great concern is the use of narcotics and related complications. We are working with our hospitalist program in partnership with our psychiatric department to explore the role of virtual reality goggles to transport our inpatients to far away worlds, peaceful landscapes, and immersive, positive topographies in an effort to distract them from their pain and liven-up their hospital stay. We will see if this approach can reduce pain medication usage, increase satisfaction, and reduce length of stay.

On the outpatient side, we learned that short visits and heavy documentation requirements prompt doctors to stare at the EHR and not the patient. Doctors suggested that a computer report generated by a patient at home could provide a head start by obtaining a good patient history prior to the clinic visit. We worked with expert physicians to create an mHealth app, called “MyGiHealth," that takes a home-based history and found that it performed as well as physician-derived histories. The app “translates” the patient report into a full narrative history and makes it available through the EHR to streamline the encounter.

Moving higher up in the organism’s anatomy, we learned from procedure unit managers that patient “no shows” often undermine efficiency and throughput. We built an algorithm that scours the EHR, looks for patterns, and predicts patient absenteeism for gastrointestinal procedures with 85 percent accuracy. Now we can overbook the GI unit with a “fast track” scheduling procedure to fill the expended vacancies in the schedule. Similarly, we are working with artificial intelligence experts to examine how a back pain algorithm can determine when magnetic resonance imaging (MRI) is needed, and when it isn’t.

At still a higher level, we are examining how to manipulate the EHR “nervous system” to provide timely feedback by delivering evidence-based decision support at the point of care. Using “Choosing Wisely” treatment recommendations as a guideline, we can inform physicians when their decisions are at odds with best practices. We are exploring how to parlay this capability to improve our performance as a health system.

At the highest level of the biomedical “organism,” we are examining models in which mid-level providers employ telehealth, mHealth, wearable biosensors, EHR data, natural language processing (NLP) and even social media at scale to proactively monitor high-risk patients in real time. By combining the volume, variety, and velocity of big data with judicious consideration of thick data, the surveillance hub may ultimately provide value by coordinating the micro and macro units of the health delivery system while respecting the realities in the clinical trenches.