Physician burnout is greatly impacting healthcare professionals and their ability to effectively care for patients.
The long-term stress reaction has many causes — from staffing shortages in specific areas to the very nature of clinical practice — but one in particular stands out: the burdens of additional documentation, specifically when it involves electronic health records.
Administrative tasks such as data entry can often be demoralizing for clinicians, says Esteban Rubens, principal for the healthcare AI practice with NetApp. “It’s not what they signed up to do,” he notes, “and they just don’t necessarily see a way out.”
Fortunately, AI and other forms of machine learning are now helping to address this issue, Rubens adds, making clinicians more productive and content in their work and allowing them to spend more time doing what they do best: caring for patients.
He recently spoke with HealthTech about physician burnout and some of the ways in which AI can help in the delivery of higher quality care.
HEALTHTECH: How do you see administrative tasks affecting caregivers and their ability to administer professional care to patients?
Rubens: We have some anecdotal evidence of people leaving the profession because they don’t want to have to deal with repetitive tasks and data entry.
And then there’s this phenomenon of concierge doctors. The doctors take payment from the patients directly, and they don’t really have to deal with any of the software that is required for providers who work in a standard organization.
HEALTHTECH: What role do you see for technology, such as AI and machine learning, in helping to alleviate physician burnout?
Rubens: In cases in which there’s not a lot of customization in technology, it can be frustrating for the physicians and caregivers in general. The technology at this point is just this piece of software that is being used for record keeping and to make sure that you're entering everything in every field.
Whereas, when you talk about technology in terms of artificial intelligence — and really more properly of deep learning as a subset of AI — that’s something that, even though we’re talking about software, is really meant to alleviate this.
It’s not meant to increase the burden of record keeping and adding fields to an application that a provider has to fill out or click through. It’s the idea that you can get a deep learning algorithm, or model, trained properly with the right kind of data, which is a big deal because coming across this data is certainly not easy. When you take something like that and you integrate it into the workflow, that can help alleviate this burden.
HEALTHTECH: Are there any specific examples of this type of technology in use that you’re able to share?
Rubens: I like this example from Geisinger in Pennsylvania. It’s doing a project in early intervention in cardiovascular disease, where it’s training models to help do these early interventions so that the patients don’t get that sick to begin with — which is clearly better for the patient. And it’s better for the caregivers too, because the patients get less sick.
The way they do this is they have access to a lot of data from their EHR, from their radiology and cardiovascular PACS [picture archiving and communication system], and some insurance data that gives them some social determinants of health. And they have used that data to train models and then flag some patients for early intervention offers.
So that’s kind of the perfect example of something that is very impactful for patients and caregivers. But that can really only be done if you have the right infrastructure given the scope of the data that they’re dealing with.
HEALTHTECH: For healthcare organizations hoping to introduce this type of technology, where do you suggest they start?
Rubens: The most important thing for people who are considering this technology is to realize that everybody should be doing this. That everybody, in healthcare in particular, has access to different patient populations, and when you train a deep learning model with certain patient data, that model is going to do very well with those patients, and similar patients. So everybody has a lot to contribute to this sort of overall effort for deep learning in healthcare.
Then the other key thing to think about is that you can start very small. It doesn’t require a massive investment, it doesn’t require hiring a bunch of data scientists, and you also don’t need to start from scratch — you don’t need to reinvent the wheel.
The model’s already coded, and the recommendation is you don’t really need to recreate it because it’s just a mathematical algorithm. The thing that really matters is how you train that model — what data you use, how you curate that data, and so on. The barrier to entry is really very low.