OR WAIT null SECS
Results of a study show the efficacy of an artificial intelligence chatbot for patients with asthma and pregnant women.
Yolande Pengetnze, MD, MS
Artificial intelligence (AI) chatbots have an opportunity to individualize care to specific patients and improve outcomes. Findings of a recent study showed that such a program proved advantageous for both patients with asthma and pregnant women in reducing emergency department visits and preterm delivery rates.
Yolande Pengetnze, MD, MS, senior medical director at Parkland Center for Clinical Innovation (PCCI) spoke to HCPLive® about a chatbot program that her colleagues developed to improve outcomes for patients with asthma and pregnant women, and how the technology can be used at other health systems and for other chronic conditions.
Editor's note: The following interview has been lightly edited for style and clarity.
HCPLive: Can you break down the study and its key findings for patients with asthma?
Pengetnze: I’ll start by framing it so you have an understanding of where we were coming from. The way we take care of problems at PCCI is by leveraging 3 key things. One is our machine-learning driven predictive analytics ability, so that we can accurately predict who is going to be at higher risk of having an issue within a specific timeframe and the information is communicated to the right people in the continuum of care to ensure interventions.
On the other side, when we design interventions and we discuss with clinicians, we leverage our clinical insights and evidence-based knowledge, just so that the programs that we're developing are sound, innovative enough, and reasonable and impactful.
The third thing we do, specifically for our patient engagement activities, is that we leverage our experience and published literature in terms of cost-to-consumer and patient behavior. It’s important when we want to engage patients to make sure we perform the specific activities that would allow them to not only gain information but be motivated to change their behavior and hopefully improve their health outcomes and empower them to take care of their health.
When we combine all these expertise and assets, it allows us to develop educational and motivational tools that we tailor to patients’ risk profiles and we distill to them in a way that enhances their engagement with the platform.
So, the study I was I was going to present and discuss at HIMSS in Orlando was going to describe this approach and how it impacted pediatric asthma patients and pregnant women from the Parkland Community Health Plan, which is a large North Texas Medicaid HMO.
The first step was to develop a machine-learning risk prediction algorithm for both asthma and preterm birth. In asthma, we predicted which patients were at higher risk for ending up with asthma within the subsequent 3 months and in pregnant women, we predicted the women who ending up having a preterm delivery during an ongoing pregnancy.
We also leveraged some evidence-based education tools that we used to develop a bank of text messages. The goal was to make these messages short and focused. One message communicated exactly 1 message or piece of information. We didn't want to overwhelm patients with information or transfer them to go and read paragraphs on a website. We actually gave them the information they needed in a simple manner through a short, motivational or educational text message.
Then we broke the patients based on the risk category that they landed in, whether they were high-, medium-, or low-risk based on their risk profile, and we would send them messages that tailored to the specific conditions. For asthma patients, they would receive between 3-5 messages every week. Some of the messages were just education indicating them about asthma to motivate them to change some of the behaviors to reduce exposure to asthma triggers, and some just reminders to take their medication or see their doctors. We would send them a symptom survey that would evaluate how their asthma symptoms were doing.
Typically, we would send 1 message per week for each patient and based on the response they'd send us, if the responses said, “Hey, my asthma is doing well,” they would just wait for 7 days and see when the symptoms were probing. If they did say the asthma was poorly controlled, we would resend that information within 48 hours just to see if the symptoms were improving or worsening and that would trigger us to refer them to a case management group.
A third survey we sent to patients was a 3-item satisfaction survey taken once every month or every quarter, depending on the group of patients, just to assess how they liked the program.
What we found was that within 1 year of entering the program, compared with patients who did not participate in the program, the program participants had a 24% drop in asthma-related emergency department visits, which was about six-times the observed rate among nonparticipants. There was a 23% increase in outpatient visit attendance, which were most likely preventive visits, and a 14% increase in their use of preventive asthma mitigation medications—we call them controller medications.
HCPLive: How did the program benefit pregnant women?
Pengetnze: Among pregnant women, we saw similar trends. For the very high- and medium-risk groups, patients received 5 messages every week tailored to educate them about risk behaviors associated with preterm birth, such as smoking cessation. The messages also encouraged them to seek care if they had any symptoms of that usually precede preterm deliveries such as rising bleeding, pelvic pain, or vaginal discharge. And we did encourage them to attend prenatal visits.
Women at low risk received standard education. We also implemented a three-item satisfaction survey with that subgroup, and again, compared with women with a similar risk profile who did not participate, those who did had a 24% increase in their prenatal visit attendance and a 27% drop in the preterm delivery rate.
We defined preterm birth as any birth under 35 weeks of gestational age. So, we could see the measurable impact.
HCPLive: Were patients satisfied with the program?
Pengetnze: Overall, what we found was that, whether in the pediatric asthma group or the pregnant women group, there was a high satisfaction and retention for the program. For instance, for pediatric patients, 85% of folks who had been enrolled in the program remained in the program over the 12 months, which is a truly high retention rate compared to what's published in the literature.
So, the impacts that we found and truly what we thought were the lessons to be learned from this implementation was the fact that we targeted high-risk patients and tailored messages to the patient's risk profile, and it might have contributed to program success. Also the fact that we used short, focused motivational messages, and we implemented a continuous integration such that, instead of sending people punctual information once every month or once in a lifetime to go and read through a paragraph, we actually took a paragraph, broke it down in specific messages and just sent them to these patients over time. That kept them maintained and engaged, but at the same time optimized the touch points so that they didn't feel overwhelmed by the number of messages they received. We hypothesized that those program characteristics were what drove success for this specific intervention.
HCPLive: How did you develop the messages that were sent?
Pengetnze: For both asthma and pregnancy, we pulled known evidence-based education tools that are available out there. The Asthma National has an education program plenty of educational tools out there, which are basically descriptive paragraphs that tell patients what is good for them, what is bad for them, what is asthma, what medications they should be using and how they should be using it. And we supplemented it, of course, with clinical knowledge from our subject matter experts.
With the pregnancy cohort, we pulled evidence-based educational materials that are very much available to pregnant women and providers that we typically share with pregnant women, some of which came from publicly available messages. For instance, from the text baby program, and we took those, and a couple of our interns sit down and break them into short messages with focused information, of course, under the supervision of clinical subject matter experts.
We were able to take those tools and break them down into specific messages that constituted the message banks that we used in the program.
HCPLive: How can a primary care facility or a large health system implement a chatbot program like this?
Pengetnze: The way we've designed it, this is truly a turnkey program. The message bank is uploaded on a text messaging platform that's HIPAA-compliant. So, if a primary care provider is interested in this program, truly, the only thing that we require right now is for patients to consent to participating, because it's truly not part of usual care yet. As long as it’s not part of the usual care, they'll need patients to consent. If a patient gives consent, we need them to provide us with their phone number, and once we upload their phone number on the platform and we have their risk stratification identified, then we can engage them in a text messaging program tailored to the risk profile. But again, it's a turnkey program that is ready to be implemented as long as we have patient consent and a mobile phone number.
HCPLive: Aside from these instances, what are some implications for the use of chatbots for conditions like diabetes or hypertension?
Pengetnze: We are actually working on all those as well. So, the same principle can be applied for hypertension and diabetes. Pull, very well-known and evidence-based educational tools for hypertension and diabetes, but then work on it. The brunt of the work here is to make sure that we create motivational and educational messages that are focused—that's what we need to do. So, take those education tools that existed and break them into short messages that educate folks on specific aspects of their disease, whether it's hypertension or diabetes, and then upload that on the platform. The platform is 1 that has a lot of capacity because it’s cloud-based. We can pull the information on the platform, and then enroll patients onto the education program.