How AI supports women's health

News
Video

Kirstin E. Leitner, MD, assistant professor of clinical obstetrics and gynecology at Penn Medicine, discusses how the artificial intelligence program "Penny" can be used to improve care in obstetrics and gynecology.

Contemporary OBGYN:

Hi, I'm Celeste Krewson with Contemporary OB/GYN and I'm here with Dr. Leitner to discuss artificial intelligence and women's health. Do you want to introduce yourself?

Kirstin E. Leitner, MD:

Sure. My name is Kirstin Leitner. I'm an assistant professor of clinical obstetrics and gynecology at Penn Medicine, and I provide clinical care at the hospital at the University of Pennsylvania here in Philadelphia.

Contemporary OBGYN:

So, to get started, can you provide a brief overview of how artificial intelligence can help in women's health?

Leitner:

Sure, I think there's a huge amount of potential with artificial intelligence to improve the care that we provide to our patients and women's health. Specifically, I've been involved in an automatic text messaging program, to provide ongoing support to patients after their birth, so, for postpartum specifically. But I think outside of my own clinical program, which we'll talk about a little bit going forward, the ability for artificial or augmented intelligence to help providers provide care more equitably, and more effectively, and allow them to focus on the you know, more problematic patient care at hand is really sort of untapped, and in many ways unknown. To this point, I think there's only good things that can come out of our ability to work with AI.

Contemporary OBGYN:

That all sounds really interesting. Can you provide a brief overview of Penn Medicine’s Healing at Home program?

Leitner:

Sure, so I alluded to it a little bit. So, the Healing at Home program is a program that we actually started back in 2019, with the goal of using an innovative approach to support patients during their postpartum journey. Understanding that a lot of patients felt after their heavy touch prenatal care time that the postpartum time can feel like a low touch time when they're doing a lot of their own, you know, recovery and medical care at home without their providers necessarily guiding them as directly. And so, we tried to find a way that we could use technology to help support patients, which would not only support patients equitably and consistently and provide sort of a basis for that support after delivery, but also hopefully decrease the burden on providers who maybe felt that a lot of the anticipatory guidance and things like that could be sort of automated in a customized way to patients.

And so, what our program actually does is it's a text messaging program, which uses natural language processing. We've partnered with a company called Memora Health to provide this service to our patients. And patients receive text messages at certain intervals after they're discharged from the hospital for the first 6 weeks postpartum, so they get anticipatory guidance about what to expect about themselves, their baby, feeding their baby, whether it's with formula or breast milk, as well as the opportunity to ask questions of that augmented intelligence robot. So, they can ask us any questions in their own language about things they might be wondering about. “Is it normal for my baby's umbilical cord to fall off at this point?” Or “feeding is painful, can you help me with that?” So, we've sort of automated responses to all these types of questions that patients have, and additionally, we've automated the ability to screen patients for things like new onset postpartum depression, through the Edinburg Postnatal Depression Screening Questionnaire, as well as screenings that we've designed to sort of encourage patients to remain engaged in feeding programs, and especially if they're breastfeeding, to sort of ensure that they're meeting the goals of breastfeeding during this early time.

Contemporary OBGYN:

Can you talk a bit about the AI machine “Penny” and how it helps women and moms?

Leitner:

Yeah, so that's sort of really the concept of this natural language processing robots, which we've taught, again, through our collaboration with Memora Health, to be able to respond to patients’ questions without them having to necessarily go through a yes or no answer question triage. So, they can literally just ask a question, and ideally get the appropriate response. Of course, sometimes that might not work because it is a robot, and I think all of us have interacted with these chatbots in the past, like, that's not really the question I was asking. Right? So, we've built a couple of ways that we can ensure that patients are more likely to get the answer that they expect. The first is that when they're enrolled in the program, we educate them on the ability to say “text” me if they feel like this robot is not answering my question, I need a real person to interven here. It's kind of like that red flag, yellow flag, I need a little extra help. If they text “text me,” our clinical team gets an alert, and we go into the dashboard about once a day. So, we can look at those alerts and respond back because usually, once I see the question, I can kind of understand what the patient's asking. And we also educate the patients on best practices and how to interact with this robot. So, the robot does best with single questions one at a time and simple ways to ask that question. So occasionally, patients will give us too much background as to what question they have. So, like, all these things are going on, I have this question. So, it may be 2 or 3 sentences. And the robot has a difficult time interpreting that question. So, we encourage patients to use simple, 1 sentence questions and to try to separate their questions that they have more than 1 question by a little bit of time to give the robot a chance to answer.

The other important thing about the development of this robot is there were a couple of times where it felt like we really needed to make sure that we weren't giving people a false sense of reassurance, right? If they're texting us in with a question about swelling, for instance, and their legs, the vast majority of the time, that's going to be a normal change that happens postpartum, related to all of the fluid shifts and all the IV fluids they may have received in the hospital. But occasionally, it can be a sign of a more serious complication like a deep venous thrombosis, or DVT. And so, one of the things we purposefully designed into this robot is sort of, the robot gets alerted to a question that requires additional triage before we can really provide that guidance to them. So, if somebody says, “Hey, I'm having swelling in my legs,” the robot says, “Oh, that could be normal. But let me ask you a few other questions.” And that's where we get into this more algorithmic approach, where instead of a natural language processing answer, we got a yes no decision tree that the robot goes through. And then in the end, if there's a warning sign, I, as a clinician receive an alert that somebody has abnormal swelling or swelling, that's concerning for a DVT. But if it seems like all of the sort of red flags are negative when the robot asks the questions, they receive anticipatory guidance along the lines of, you know, swelling, it's very common, if it persists, call the office, etc. So, it's kind of a multi-pronged approach, again, mostly natural language processing, which we've tried to educate our users as to how best to interact with the technology with kind of a fallback of this algorithmic approach for the more concerning clinical signs that could develop during this period.

Contemporary OBGYN:

That all sounds great, thank you. Are there any developments you want to make to Penny in the future? And is there any sort of impact you're aiming for?

Leitner:

Yeah, I mean, it's a great question. We are honestly constantly developing the program, and the program is always changing is sort of like the best. And that's sort of how it's designed to work, right? These robots learn as you use them, they become better over time. But of course, there are some ways that we're looking to potentially redesign the program going forward, and a few things that we've been able to accomplish already, and a few things that I would say we want to do in the future. So, we've been able to build the program in a way that we have a basis program for everybody, and then we can customize responses on top of that based on the patient's clinical characteristics. So, for instance, somebody who has a vaginal delivery may need very different anticipatory guidance to somebody who's had a C section. They may have different questions, but they also need different guidance around a return to activity, lifting restrictions, and things like that need to be customized to the clinical type. And so, as we think of clinical types, one of the things we're looking to add in the near future is a clinical type for gestational diabetes, for instance, or preexisting diabetes, so that those patients may get additional educational information around the need to do their follow up glucose tolerance testing, for instance, or to follow up with their primary care doctor for management of their diabetes.

We have started doing blood pressure screening through the program, which even for low risk patients, so for patients without gestational hypertension, which we found to be a really valuable component of our program. And the other patient type that we're really looking to develop even further is a patient type where a patient may have their infant in the ICN or in the NICU. So obviously, those patients have a very different experience in terms of their postpartum, and they are still often going back to the hospital after they've been discharged. And their infant's needs are very different than the infant's needs for wellborn infant. And so, designing a program that can still allow patients to feel supported and have infant specific information to either a preemie or baby in the ICN are some of the goals that we're looking forward to doing over the next few years.

Contemporary OBGYN:

That all sounds great. We're just about ready to wrap up, but is there anything you want to add first?

Leitner:

No, I mean, I think if anyone's interested in using text messaging, there's so much out there and I'd say there's probably even more out there now than there was in 2019, when we started on this journey. I think the biggest thing that you need to do as you start to think about how AI can help you in your patient care space, is this an acceptable method that your patients are going to be engaging in, because you can design the coolest thing in the world, but if your patient doesn't use their phone to text you back, it's not going to work, right? And the same thing, we can design all these really cool tools with extra apps or extra reminders to the patient portal, but if our patient’s really not engaging in that, it's not the way to get through to them. So, I think the key thing before you even start on this journey, is making sure that whatever technology you're looking to use is the one that your patients are going to engage with.

Contemporary OBGYN:

Thank you for speaking with me today.

Leitner:

Thank you so much.

Related Videos
Addressing maternal health inequities: Insights from CDC's Wanda Barfield | Image Credit: cdc.gov
Addressing racial and ethnic disparities in brachial plexus birth Injury | Image Credit: shrinerschildrens.org
Innovations in prenatal care: Insights from ACOG 2024 | Image Credit:  uofmhealth.org.
The impact of smoking cessation on pregnancy outcomes | Image Credit: rwjmg.rwjms.rutgers.edu
Maximizing maternal health: The impact of exercise during pregnancy | Image Credit: cedars-sinai.org
The importance of nipocalimab’s FTD against FNAIT | Image Credit:  linkedin.com
Fertility treatment challenges for Muslim women during fasting holidays | Image Credit: rmanetwork.com
CDC estimates of maternal mortality found overestimated | Image Credit: rwjms.rutgers.edu.
Related Content
© 2024 MJH Life Sciences

All rights reserved.