Conversation

Informatics Innovations: How to Achieve “Medicine-Based Evidence”

Innovations in biomedical informatics and computational health informatics are creating opportunities to move from generic evidence-based medicine at the point of care to patient-specific medicine-based evidence. In addition, technology is opening patient engagement opportunities with smartphone apps, but there are some undesirable consequences to consider.

  • Namita Seth Mohta, MD &
  • Kenneth D. Mandl, MD, MPH
January 27, 2020
Summary

In a discussion with NEJM Catalyst, Kenneth Mandl, MD, offers insights on innovations in biomedical informatics and computational health informatics that are creating opportunities to move from generic evidence-based medicine at the point of care to personalized medicine-based evidence. In addition, the growth of health care smartphone apps that can be integrated with electronic medical records is enhancing patient engagement and improving the information available to clinicians; but the introduction of data that straddles the HIPAA line creates challenges that stakeholders will need to address.

Headshots of Kenneth Mandl and Namita Seth Mohta.

Namita Seth Mohta, MD, interviews Kenneth Mandl, MD, MPH, Director, Computational Health Informatics Program, Boston Children’s Hospital.

Namita Seth Mohta:

This is Namita Seth Mohta for NEJM Catalyst. I am speaking today with Dr. Ken Mandl, Director of the Computational Health Informatics Program at Boston Children’s Hospital and the Donald A.B. Lindberg Professor of Pediatrics and Professor of Biomedical Informatics at Harvard Medical School. Ken’s work has helped define the field of biomedical informatics. He has long advocated for patient participation in producing and accessing data. Dr. Mandl, we are delighted to have you joining us today.

Ken Mandl:

Thank you.

Mohta:

Let’s start with some basics. What are biomedical informatics and computational health informatics? Educate us a little bit about how these fields are relevant to the NEJM Catalyst audience of health care executives, clinical leaders, and clinicians who are listening to our conversation.

Mandl:

Let’s start with an example. The rounding team in a hospital is at the foot of the bed of a patient with new-onset epilepsy, and the medical student is questioned by the senior resident as to what medication to prescribe. The way she answers this now is to recite from her understanding of a published report of a clinical trial. Traditionally, this has been called evidence-based medicine. Now, that study may have been performed on a population not resembling the patient in the bed. Trial participants may be healthier, with less comorbidities, a different racial and ethnic background, and taking fewer, if any, other medications at all.

The alternative approach has been somewhat humorously referred to as medicine-based evidence.1-3 If you were that patient, wouldn’t you want to know, for each possible medication choice, what the outcomes were for all patients like you? The health care system collects a lot of these data, but not in a way that lets us readily answer that question for that patient. Under President Obama, the U.S. invested $48 billion to promote the purchase and use of electronic medical records. We still don’t have this sort of medical intelligence.

Biomedical informatics and computational health are disciplines focused on evolving effective uses of biomedical data, information, knowledge for scientific inquiry, problem-solving, and decision-making all toward improving human health. In a nutshell, biomedical informatics is about transforming medicine into a data-driven discipline engaging patients, doctors, researchers, public health practitioners, and technologists in producing and using data for insight, diagnosis, management, discovery, and care.

Using that example of the patient with new-onset epilepsy seen on rounds, how do we capture those health states, those medications, and those outcomes? How do we represent the data? How do we eliminate bias in the data? For example, if the sickest patients tend to be prescribed drug A, if we’re not careful about bias it’ll appear that drug A produces the worst outcomes. How do we add genomics and RNA expression data to the decision-making process? How do we develop accurate and helpful machine learning and artificial intelligence to guide physicians to make the right decisions? How do we guide physicians about what are the right decisions? Hint: it’s not by texting them or filling up their email inbox. How do we do that in the workflow in a way that works for doctors?

Mohta:

Talk to us about the relationship between machine learning and artificial intelligence and this notion of biomedical informatics and computational informatics.

Mandl:

Let’s look at it from the point of view of the physician, and let’s look at it from the point of view of getting something like a machine learning algorithm into the physician’s workflow in a way that can help them make a decision. In order to do that you need to understand what the physician’s workflow is, and right now it is very much focused on the use of the electronic medical record to do a lot of documentation and input a lot of orders for medications and other things. We know that these systems are quite specific in the way they’re designed. They’re old pieces of software from the pre-Internet era that are generally sold to chief financial officers to manage billing and the revenue cycle, and they weren’t initially designed to perform clinical functions or to support clinicians in their workflow.

The point of care is somewhat of a walled garden, and if we want to even introduce a single algorithm that pushes a patient toward the decision, for example, about a particular medication, it’s hard to add that to an electronic health record workflow. We developed an approach to this; it’s called SMART or SMART on FHIR4 [Substitutable Medical Applications and Reusable Technologies on Fast Health Interoperability Resources], and it’s about getting the interoperability so that different pieces of software can talk to each other in a way that makes for a seamless experience for the physician. At the beginning of the Obama administration when they were spending $48 billion to promote the use of electronic medical records, it included payments to physicians under the Meaningful Use program. It became clear to us that this could produce a walled garden around the point of care. And in a perspective article I wrote with a colleague in the New England Journal of Medicine5-8 we recommended that electronic medical records begin to look like iPhones instead.

The iPhone features an API or an application programming interface so that a software developer could create a third-party app and connect it to the iPhone easily. The iPhone user would choose that app from an app store and the app was substitutable. It could be added or deleted from the iPhone easily. We were just learning about the power of this model. It turns out that when we wrote the piece at the end of 2008, Obama had been elected and there were 10,000 apps for the iPhone, when it was published in March of 2009 in the New England Journal of Medicine, after the inauguration, there were 50,000 apps for the iPhone. One year later as we were beginning our project called SMART, there was more software code written for the iPhone platform than any software project in history.

This idea that we caught, just as the iPhone apps model was hockey-sticking, is one we thought should be applied to electronic medical records. You should be able to add an app or delete an app from it. That app should run all over the health care system and it should be, therefore, part of an ecosystem where apps compete with each other on functionality, design, value, usability.

When a health system purchases an electronic medical record system it does not have much choice in terms of what functionality it receives next, but there are periodic upgrades to those systems. In an apps model there could be thousands of apps that clinicians could choose from, and this would produce some key desirable properties in the system. One, an innovator creating an app could distribute it through an app store instead of having to sell it one hospital at a time to a CIO. Second, a rheumatologist could have a completely different view of the electronic medical record through this app than an obstetrician, for example — and they need different views. Third, these apps would compete with each other and the best apps could win. It would drive down the price of apps and increase choice for the end user.

Mohta:

What are some examples of these kinds of functions or apps that could be integrated with my iPhone/EHR?

Mandl:

One might be an app that uses machine learning or artificial intelligence to, based on your history, recommend a medication for you. Another would be — let’s use a real-world example — there’s an app called Meducation that won an apps contest9 we had about 7 or 8 years ago that now is commercially available, and it does a really interesting thing. You can connect it to the electronic health record, and it takes a medication list for that patient, and it joins that medication list to an incredible resource they have of dynamically created medication instructions in 24 different languages. These are super user-friendly, sixth-grade reading level, illustrated medication instructions completely in contrast to the package inserts that come with these medications that are virtually unreadable and unusable by patients. That app works across multiple versions of electronic medical records, different vendor products, and it also has patient- and physician-facing versions, and it’s also available through app store models. This is an example of a substitutable and reusable app that becomes turnkey to add to the electronic medical record.

Mohta:

It all sounds great. But there are always potential unintended consequences to an approach like this. Two come to mind for me — and there may be more: One is how patient privacy is respected if there’s all this patient data flowing through these third-party vendors or organizations. And then the second one is this notion of standardization; if we are allowing doctors or other care providers to download whatever app they want to, how do we ensure that there is consistent quality of care provided to a patient no matter what provider they see?

Mandl:

Those are two great questions. The privacy aspect is well managed within this ecosystem within the context of the health care organization. Think of these apps as an extension to the electronic medical record. Those apps are completely covered under HIPAA. [Regarding standardization,] they probably won’t be selectable by just any doc within the organization, but — more like other software that makes its way into a hospital environment or a practice environment — will be chosen carefully by a chief information officer or a counterpart to the chief information officer. There will be security and privacy reviews. It does create a set of challenges around managing those security and privacy reviews across potentially a large number of apps. But it’s a similar set of issues to what any hospital system deals with, with any piece of software that they install.

However, there is another set of issues that arises when the patient connects an app. And this is an important moment as we look at what ecosystem we want out there in patient-facing health information technology to support patients in their diagnostic odysseys, or in managing their conditions or the conditions of their children, for pediatric patients. An interesting thing happened around the SMART project. I had successfully lobbied for a few sentences in a law called the 21st Century Cures Act that requires that this Steve Jobs–like programming interface be in place for all electronic health record systems and that it provide access to all the data in the patient’s record.

Now there’s a proposed rule that’s about to go into effect, it’s being reviewed by the Office of Management and Budget, that makes the SMART interface essentially the law of the land, so there will be a way to connect apps to electronic health records.10 And further, the rule enables the patient to connect an app of her choice to the electronic medical record. An interesting tension has arisen in the public conversation around this rule, and you have to understand a small amount about HIPAA to understand where the dilemma arises. When the data are in the electronic health record they are part of what’s called a covered entity under HIPAA — all health care organizations are covered entities — and the data are regulated by the U.S. Department of Health & Human Services’ Office of Civil Rights under HIPAA. Once the patient has requested a copy of her data to come across this programming interface and into an app of her choice, the phone uses the SMART interface to let you download a copy of your data right onto your iPhone. And that’s available at hundreds and hundreds of health care systems.

At that moment — when the data jumps from the hospital electronic medical record onto the iPhone — it’s no longer regulated by HIPAA. Instead, it’s regulated by the Federal Trade Commission, and at that point the FTC only has the authority to regulate those data in such a way as was promised to the patient in the terms and conditions that they agree to. These are those little terms and conditions that we all agree to every time we download an app. And you know in the history of the world no one has ever washed a rental car, and no one has ever read the terms and conditions associated with an app, but we all assert that we do.

So there’s this important opportunity to provide to the patient a clear set of terms and conditions in a large font, in human-readable language at a fourth- or a sixth-grade reading level, where promises are made and then enforced on the other side by a regulatory agency. At this moment there is no such standard for that. There’s some concern that when we open up these APIs under the law — the 21st Century Cures Act and under the rule that puts that law into specific effect — that we will have predatory apps that patients may sign up for.

We’re at this interesting moment where there’s a tension involving patient autonomy. Patients deserve a copy of their data to be able to bring to a researcher or a specialist who can help them with their condition, or to an app that can help them manage their condition with artificial intelligence, genomics, machine learning — things that we want to bring into the health care process to help with discovery and better care management. But how do we protect patients as they exercise their individual right of access to those data? At the same time, how do we avoid being paternalistic and risk not allowing patients to exercise their own choice in how they will best be benefited by the use of their data?

Mohta:

You raise a critical tension that exists not only in the sphere that we’ve been talking about today, but I would put forth is a very real one that providers and patients are dealt with in terms of all aspects of the care journey. Thank you for speaking with NEJM Catalyst today.

Mandl:

Thank you so much. It’s been my pleasure.

See allMost Recent in Health Technology