Care Redesign

Embracing Technology to Save Primary Care

Article · August 30, 2018

Inviting Disruption

American health care has experienced the most turbulent decade in history, and, if the past year is any indication, there is much more tumult ahead. With the enactment, and then threat of repeal, of the Affordable Care Act (ACA); mergers and acquisitions among and between hospital systems, insurance companies, physician groups, and retail pharmacies; the emergence of retail and virtual care; and the entrance of companies such as Amazon, Alphabet, Apple, and others, the roller coaster ride has only just begun. Moreover, the system is extremely expensive (with Americans spending in excess of $3.5 trillion on health care per year), nearly impossible to navigate, and hugely inefficient, while at the same time producing truly average outcomes at best — and that’s being generous.

In most industries, when companies become bloated, inefficient, and costly while the products that they produce remain largely unchanged, companies cease to exist and the entire industry is primed to be disrupted beyond recognition. Somehow, until now, health care seems to have gotten a pass. Nonetheless, as W. Edwards Deming famously said, “every system is perfectly designed to get the results it gets,” and we deserve to be disrupted.

What Is the Role of Primary Care?

As problematic and complicated as health care is, the proposed solutions, directed at lowering costs while improving access and outcomes, have often simply added more complexity, bureaucracy, controversy, and cost. At the center of many proposed solutions to the American health care crisis has been the push to increase access to primary care. The concept is that improved access to basic and preventative medical care will result in better overall health, which means fewer hospitalizations, less reliance on emergent and urgent care, lower overall morbidity and mortality, and, ideally, decreased system costs.

In addition, pragmatically speaking, when a patient interacts with a single provider, there is less reliance on specialists and testing. In fact, many insurance companies and health systems require primary care providers to control access to higher-cost specialists, who by the nature of their work perform expensive procedures, prescribe expensive drugs, and order expensive tests.

By having the primary care provider serve as the master care coordinator, or gatekeeper, the assumption is that we can avoid unnecessary health care and save money. After all, if we can prevent illness and decrease acute exacerbation of chronic illness, then we can decrease the need for acute care and the complexity of such care when it is inevitably required. This approach is intuitive and makes perfect sense: Get a flu shot, and avoid getting the flu. Get a colonoscopy, have a polyp removed, and avoid developing cancer. And so on.

This approach is especially suitable for the elderly, with novel companies like Oak Street Health creating highly robust programs for Medicare beneficiaries that not only include preventative, chronic, and acute care, but that also feature sophisticated care coordination services that address many of the social determinants of health (e.g., loneliness, transportation, and activity) in an integrated way while relying less on providers. The results have been impressive: Such companies and programs have significantly improved the lives of Medicare beneficiaries while decreasing health care spending and, by being smart and taking risk, making a nice profit in the process. Is this experience generalizable to the non-Medicare population?

New Math or Old Math, the Numbers for Primary Care Just Don’t Work

It may seem intuitive that adding more primary care providers is the solution, but there is a simple barrier: The numbers just don’t add up, and math, as it turns out, is important. Pragmatically, we do not have nearly enough providers to deliver traditional primary care. Even if there were enough providers, it would be far too expensive to have a health care provider, physician, or advanced practice provider — or maybe any human — do the work. Finances aside, being a primary care provider is a difficult, if not impossible, job.

Let’s consider a primary care provider who is expected to deliver all acute, chronic, and preventative care to a panel of 2,000 patients (most have far more) while following every evidence-based guideline and recommendation. It would take this provider 21.7 hours a day, 7 days a week, to adequately care for her panel, meaning that, in reality, patients would not be getting comprehensive, evidence-based, guideline-directed care.

Even with the explosion in the number of advanced practice providers, which has most assuredly improved access and alleviated some of the burden on primary care physicians, there is still a long way to go before there would be sufficient numbers of providers to deliver care as intended or envisioned. And that still does not address the reality that if a human dedicated close to 24 hours a day, 7 days a week to staying abreast of every patient and keeping up to date on new treatments and discoveries, it still would not be enough time.

Where does that leave us? Given these constraints of time, money, and resources, is it simply a fool’s errand to continue on this path? Before we answer this question, let’s first consider the history and evolution of primary care.

It’s Been a Long Time Since the Turn of the Century

At the turn of the century, the majority of care delivered by primary care physicians was symptomatic; that is, patients typically sought care because of an injury (often industrial), acute illness (mostly infectious disease), or pregnancy. In that setting, preventative care amounted to sage “parental-esque” advice and a handful of vaccines. Today, nongeriatric primary care is largely focused on preventative and asymptomatic care.

Although preventative care is unequivocally crucial to caring for communities, it has been difficult to convince asymptomatic adults who feel healthy to see their doctor until they are symptomatic. Conditions such as diabetes and hypertension don’t hurt — until they do, by which time it’s often too late. This challenge is particularly difficult when the patient must pay out of pocket, has high-deductible insurance, must take time off from work, or must wait endlessly for an appointment.

Access: More than a Buzz Word

In the current system, providers need to be as fully booked as possible to meet demand and generate income. Therefore, when a patient has an acute, episodic, or unscheduled need for care, it can be nearly impossible for the patient to see his or her actual primary care provider. The most common scenario is that a patient with a same-day need will be referred to an urgent care center or emergency department. In the unlikely event that the patient can be seen by a provider in the practice, the provider will likely be someone whom he or she has never met. In this setting, the idea that a primary care provider can avoid unnecessary testing and treatment because of an ongoing relationship with the patient and a comprehensive understanding of his or her condition is simply a specious concept.

How can we design a system that can deliver high-quality preventative medical care, quarterback and monitor specialty referrals, keep track of test results, and meet the need for acute, episodic, and unscheduled care while being reliable and cost effective? The answer must focus on deploying technology rather than relying on humans. Health care can no longer be the last bastion of industry where adding technology increases rather than decreases cost and complexity.

Your BOT Will See You Now

The use of machines to deliver care may seem like a controversial or even provocative concept, but let’s consider routine preventative care. Most, if not all, preventative care is specifically described in robust evidence-based guidelines, which are applied based on patient and environmental factors and, eventually, “omics” data (i.e., genomics, proteomics, metabolomics), and data from passive (i.e., wearable) physiologic and biochemical monitors. If we were to adopt the lessons of industry, most, and maybe all, preventative care (i.e., maintenance) would be algorithmic; that is, care would be based on key data points from myriad sources and would be monitored by a learning algorithm, and the patient would be prompted to complete certain tasks and answer questions as directed by machine learning.

Essentially, we would be building a primary care BOT (e.g., Google Duplex). UnitedHealthCare’s Motion program, for example, deploys a wearable device (e.g., FitBit, Apple Watch, etc.) to subscribers, who are refunded a portion of their premium if they meet certain exercise goals as directed and monitored by the device. The preliminary results have been impressive: Patients move more and are sick less, UnitedHealthCare spends less money, patients get money back, and the entire program is driven by an inexpensive piece of technology, is backed by robust and complex data science, and is driven by machine learning.

Why do we need a human to confirm that a patient should lose weight, not smoke, exercise, wear a seat belt, and get a flu shot; that a middle-aged woman should have mammogram; or that an elderly hypertensive smoker should have an ultrasound to screen for an abdominal aortic aneurysm? The answer is that we don’t.

An apt analogy is a modern airplane, in which the pilot inputs standard variables (e.g., destination and time of departure) and the computer culls a multitude of additional variables from sensors, analyzes data too voluminous for a human brain to decipher, makes adjustments, and only brings in the pilot when conflicting data or urgent conditions require human judgment. Similarly, machine learning, data science, remote monitoring, and prescriptive analytics should be harnessed to manage an individual’s primary care needs (and, eventually, most nonprocedural care needs, including specialty care).

Take the ultrasound example mentioned above. The patient in this scenario, a 65-year-old man, has hypertension and a history of smoking, both of which are basic historical factors that would be available in any electronic medical record. The patient would simply receive an automated text message, email, or call instructing him to arrange an abdominal ultrasound, or even better, asking the patient to simply confirm the ultrasound has been arranged for him. The ultrasound demonstrates a 7-cm abdominal aortic aneurysm. The images are then sent to be reviewed by a vascular surgeon who has access to all the patient’s data. The surgeon and the patient can then have a conversation about the risks and benefits of repair or replacement.

In this example, no human is required until the surgeon gets involved. To take it a step further, if BOT-initiated behavioral activation had been engaged earlier in the process (for example, with automated prompts reminding the patient to maintain a better diet, exercise, and stop smoking), perhaps no aneurysm would have formed in the first place.

What about a more complicated primary care scenario? Let’s consider a 50-year-old patient with obesity and diabetes. In such cases, treatment is typically directed at decreasing hemoglobin A1C as an indirect measure of glucose control; however, we know that what really controls glucose is diet, exercise, and medications. For those of us with more than a few pounds to lose, we know that losing weight is one of the single greatest challenges a human can undertake, and convincing patients to lose weight is seldom successful.

Dexcom, a company that produces continuous wearable glucose monitors, has developed a unique app that works on any smart phone. The app tracks glucose levels, sends an alert when the patient has eaten something that has precipitously increased the glucose level, and prompts the patient to get up and walk at the appropriate time after a meal. By continuously measuring glucose with use of this app, the patient can also learn the ideal time to take his or her oral hypoglycemic medication.

As it turns out, not all diabetic patients are identical. Although a prescribed medicine may be taken one, two, three, or four times per day, these recommendations are based on what works for the general population as a whole, and the ideal timing for an individual patient may be different by a number of hours. This simple app, backed up by not-so-simple math, has had remarkable results: Hemoglobin A1C levels have plummeted (sometimes reducing medication) and patients have lost weight (helping them to feel better).

Ouch, That Hurts . . .

For acute, episodic, or unscheduled needs, one can easily imagine a time in the near future when a patient, from the comfort of home, will be able to digitally request an assessment and enter key factors that will be married to already established patient-specific data pulled from the electronic universe. Information may even include the patient’s previous “omics” information and physiologic data from a wearable device.

A machine-learning algorithm will then be applied following evidence-based guidelines, with the machine (BOT) considering the hundreds or thousands of possible permutations in the context of acute symptoms and virtually consulting a human when needed. The machine will be able to easily discern between an acute health care issue (e.g., the patient has chest pain, headache, or runny nose) and a routine need (e.g., the patient has run out of metoprolol). This scenario is not so different from a traveler who calls an airline and is asked by a machine if he or she is calling about an existing reservation, a new flight, or a flight change.

The provider (when needed) would then be presented with the key components of the patient’s history (i.e., the algorithm assumptions) as well as the pertinent diagnostic or treatment recommendations based on prescriptive analytics. Next, the provider would confirm both the assumptions and the treatment recommendations and would assess the need for resources such as laboratory testing, imaging, or specialty care.

Finally, the provider would make the determination as to whether the patient could be treated virtually, would need to be evaluated face-to-face in an acute care environment (e.g., an urgent care center or emergency department), or could wait for a scheduled appointment with a generalist or specialist. Perhaps a specially equipped ambulance with testing equipment and an examination room would be dispatched to the patient’s location, where a paramedic would evaluate the patient while a physician provides guidance virtually. In cases in which medication is needed, perhaps a drone would be deployed deliver the medication within the hour.

Using Humans for the Hard Stuff and Leaving the Basics to Machines

When we consider the current system, in which robust evidence and clinical decision support have been embedded into provider workflows, we can see that guideline compliance has increased dramatically, resource utilization and costs have decreased substantially, and, most importantly, patient outcomes have improved significantly.

If we can take it a few steps further by (1) employing wearables and other devices (e.g., FitBit, Apple Watch, Amazon Echo) to passively collect waveform vital sign data; (2) allowing data science, machine learning, and prescriptive intelligence to monitor patients for physiologic distress well before it is symptomatic and even manage a portion of primary care; and (3) using providers sparingly and respectfully when human judgment or specialty or procedural care is needed, we will have done what every other industry has — we will have used technology to make our product better, faster, and cheaper. And we will have saved health care in the process.

New call for submissions ­to NEJM Catalyst

Now inviting longform articles

Connect

A weekly email newsletter featuring the latest actionable ideas and practical innovations from NEJM Catalyst.

Learn More »

More From Care Redesign
Impact of PCSP on Patient Satisfaction at Providence Heart Clinic

Transforming Specialty Practice in Pursuit of Value-Based Care: Results from an Integrated Cardiology Practice

Despite significant primary care reform around patient-centered medical home models, specialty care remains fragmented, with poor communication between primary care and specialists. How should specialty practices be reformed to deliver more coordinated, patient-centered care?

Michael Bennick Yale New Haven Hospital Medical Director of the Patient Experience - Yale Living History Project

The Living History Project: Open-Ended Patient Interviews Create a Therapeutic Bridge

A program at Yale has students conduct open-ended interviews with patients about their lives, their hopes, their values, and what they most want their medical team to know — creating the opportunity for human connection and a better care experience.

Fisher02_pullquote hypertension guidelines

Hypertension Guidelines: Achieving 90% Success

Focused and innovative health systems are managing to control blood pressure for 9 in 10 patients, which is well above the national average of 50% to 60%.

Health Care Organizations Are Moderately Effective in Using Data

Survey Snapshot: Using Data for Change

NEJM Catalyst Insights Council members discuss how data and analytics are being used at their organizations, both now and with the future in mind.

Percentage of U.S. Adult Hemodialysis Patients Achieving Dialysis Adequacy, 2013-2016. Data will be released in early 2019.

Innovation in Dialysis: Continuous Improvement and Implementation

The U.S. dialysis sector has been criticized for its lack of innovation, but this criticism disregards the kidney community’s success in creating — and continuously improving on — dialysis as a safe, globally scaled, quality-oriented outpatient therapy.

Cleveland Clinic Time-to-Treatment Cancer Programming Overall Scorecard 2015-2017 Sample

Reducing Time-to-Treatment for Newly Diagnosed Cancer Patients

How Cleveland Clinic initiated a multidisciplinary program to reduce time-to-treatment and accomplish a 33% reduction.

Treatment Authorization Increases and Rapid Boost in New Mexico Medicaid Members Treated for Chronic HCV

A Collaborative Model to Expand Medicaid Treatment Coverage for Chronic Hepatitis C Virus

How managing the benefit coverage expansion for the treatment of HCV in New Mexico was successfully achieved after less than 2 years.

Data Analytics Improves Clinical Care

Care Redesign Survey: How Data and Analytics Improve Clinical Care

Data and analytics are a key means for clinicians, clinical leaders, and executives to transform health care delivery. Yet health care organizations have work to do in getting measures right and much to learn about effective use of data, according to our most recent Insights Council survey.

Nobody Wants a Waiting Room sketch

Nobody Wants a Waiting Room

A study in system change.

Orszag02_pullquote - In Defense of the Hospital Readmissions Reduction Program HRRP

In Defense of the Federal Hospital Readmissions Reduction Program

In the current debate about HRRP, the evidence tilts toward no effect or a beneficial one on mortality, says the former Director of the U.S. Office of Management and Budget.

Connect

A weekly email newsletter featuring the latest actionable ideas and practical innovations from NEJM Catalyst.

Learn More »

Topics

Hypertension Guidelines: Achieving 90% Success

Focused and innovative health systems are managing to control blood pressure for 9 in 10…

Primary Care

182 Articles

“Breaking Bread” to Combat Burnout

Can a simple dinner create community among health care providers?

Sustainable Financing for Complex Care Management…

Care management should be payer-agnostic at its core.

Insights Council

Have a voice. Join other health care leaders effecting change, shaping tomorrow.

Apply Now