Conversational AI improves ‘fourth trimester’ maternal care at Penn Medicine – Healthcare IT News

18 minutes, 49 seconds Read

The U.S. is considered one of the most dangerous high-income countries in the world in which to give birth. Meanwhile, there’s a shortage of OB-GYNs, leaving more than 2.2 million women of childbearing age living in so-called maternity care deserts with limited obstetric care.


World’s Leading High-rise Marketplace


The “fourth trimester,” or postpartum time-period, is often a forgotten “trimester” of pregnancy, yet plays a critical role in parental, newborn and overall population wellbeing, explained Dr. Kristin Leitner, an assistant professor of clinical obstetrics and gynecology at Penn Medicine.

“While undergoing many physical and emotional changes after birth, patients may also suffer complications such as infection, thrombosis and hypertensive disorders, as well as the new onset or exacerbation of mental health disorders and other chronic diseases,” she noted. 

“The potential for medical complications postpartum is particularly concerning as more than half of pregnancy-related deaths occur after birth; however, in traditional care models, 90% of visits during the perinatal time-period are during pregnancy alone.

“Since 2018, the American College of Obstetricians and Gynecologists has recommended that care during the postpartum period should be an ‘ongoing process’ rather than the traditional one-time postpartum visit at 6-12 weeks postpartum,” she continued. 

“Even this traditional one-time visit is only attended by as few as 50% of patients in the United States, causing many patients to miss the opportunity for early identification of postpartum complications, routine health maintenance, chronic disease management, and other postpartum goals such as breastfeeding and contraception.”

This gap between postpartum patient needs, clinical recommendations and reality of healthcare access presents a significant challenge to patients and practicing providers. Innovative methods of identifying needs and providing ongoing care for the postpartum patient are needed without added burden to already over-extended providers, she added.

“Tragically, non-Hispanic Black women suffer maternal morbidity and mortality at rates far above those of the overall population,” Miller said. “Therefore, any intervention designed to decrease maternal morbidity and mortality must aim to intentionally target this population with acceptable, feasible and effective solutions targeted specifically to this underserved population.

“When we piloted a postpartum support program at our hospital – the Hospital of the University of Pennsylvania in Philadelphia – we found new parents have a lot of questions about their recovery, feeding and baby care after birth despite being given significant numbers of resources in the hospital,” she continued.

These questions were wide-ranging, such as: Is the swelling in my feet normal? Why isn’t my baby latching? Is my bleeding normal? How soon should I feed my baby after their last meal? When will the umbilical cord fall off? The list goes on.

“At the time, in 2018, mobile health applications had been used for problem-based postpartum support, focused on specific individual conditions regarding postpartum recovery such as breastfeeding, blood pressure monitoring and weight loss,” Miller noted.

“While many of these interventions had shown great promise in improving compliance with care and reducing healthcare disparities, there were little to no comprehensive technological interventions to support patients holistically during the fourth trimester that provide support throughout these realms,” she added.


In 2018, Penn Medicine started its Healing at Home program with the goal of using an innovative approach to support patients during their postpartum journey.

“As part of Healing at Home, we optimized patient workflows on the postpartum unit with the goal to decrease length of stay while in the hospital after birth,” Miller explained. “We recognized, however, that if we decreased the length of stay, we wanted to ensure patients felt supported and had a connection with their care team after discharge.

“This is where we started to conceptualize the solution of a mobile, text message-based solution,” she continued. “Our hospital had great success with text messages to support postpartum patients after discharge in the setting of hypertension of pregnancy (a program called Heart Safe Motherhood); however, we concluded from a small pilot of approximately 90 postpartum patients that to provide holistic support would provide significant complexity to our solution.”

Patients needed highly complex and individual responses to inquiries; for instance, about physical recovery specific to delivery mode (vaginal versus cesarean), to care of newborns (diapering and umbilical cord care), and infant feeding difficulties (pain with breastfeeding, difficulty pumping and uncertainty with preparing formula).

“This complexity led us to conclude that a ‘simple’ algorithmic approach was unlikely to be successful in providing this population with the holistic support required,” Miller said.

So, Penn Medicine began working to create a postpartum conversational agent (chatbot) with artificial intelligence vendor Memora Health, through which the health system was able to leverage natural language processing to provide a conversational experience for patients and layer complex patient needs (by patient type) onto the experience.

“First, a frequently asked question bank was used to generate accurate mapping of questions to the appropriate responses,” Miller explained. “Second, surveys (standardized conversation templates designed to collect patient data) were created by patients’ clinical characteristics (for example, breast milk versus formula fed).

“Third, creation of anticipatory guidance specific to patient clinical characteristics was planned,” she continued. “Finally, algorithms for potentially acute clinical concerns were designed and layered onto the program. Throughout this process we incorporated personal touches into responses, such as patients’ or infants’ names and worked to develop a consistent and empathetic tone.”

By providing ongoing contact for patients as well as the ability to alert clinicians to patient concerns, the team hoped to improve patient experience and outcomes and reduce burden to the care team, highlighting for them which patients really needed direct attention to a clinical need while answering “easy” questions automatically.


The program offers text messaging that uses natural language processing to guide postpartum patients through their care journey for the first six weeks after they are discharged from the hospital. By using automated and conversational text messaging to communicate with patients around routine postpartum care, clinicians can focus on the cases that are more pressing and require more complex medical attention.

“We realized many of the questions patients followed up with after leaving the hospital were common ones that could be efficiently answered,” Miller noted. “We just had to find that technology and ensure that it was comprehensive enough to provide our patients with the same personalized care we deliver as providers.

“Once patients leave the hospital, they receive text messages at certain intervals, so they get guidance about what to expect for themselves and their baby and feeding challenges, as well as the opportunity to ask any questions that might come up – no matter what time of day or night,” she continued. “These new parents often have questions about the more typical postpartum activities like when they can return to exercise, how to care for common symptoms such as hemorrhoids, how to store breastmilk, and the baby’s sleep patterns.”

The AI-enabled technology allows new mothers to ask these questions and receive intelligent, personalized responses that Penn Medicine has helped to inform as the clinical care team.

“In other words, the technology is able to respond to patient questions without them having to wait on hold or send a portal message,” Miller said. “In most use cases, they can ask a question via SMS and get the appropriate response immediately.

“While we of course recognize that automated processes sometimes have kinks, we’ve made sure to plan for these,” she added. “Our team has built ways to ensure that responses are accurately reflective of what patients expect to receive from their doctor.”

One way the team has done this is through education with those enrolled in the program, letting them know they’re able to prompt Penny (the name of the chatbot) to have a real person intervene if they so choose.

“If they text ‘TEXT ME,’ our clinical team gets an alert, and we go into the dashboard to respond back to that person manually,” she said. “We also created a number of clinical algorithms to assess someone’s potential level of need. For instance, while lower-leg swelling can be a benign symptom following childbirth, it can also be a warning sign of serious complications such as, in this case, postpartum preeclampsia or deep venous thrombosis.

“Through clinician-developed automated decision tree algorithms, patients answer additional questions, which allow clinicians to determine which patients with this symptom require additional evaluation (such as an ultrasound for a blood clot) and which can be reassured about this common postpartum symptom,” she continued.

Not only does the patient receive messaging their constellation of symptoms is concerning, but the clinical team also is alerted by the Memora Health platform.

“Overall, we think this multi-pronged approach, enhanced through AI technology, is able to efficiently solve a longstanding problem we’ve experienced in caring for new mothers,” Miller said. “In implementing this tool, we’ve made sure to include patient feedback so they feel supported throughout their entire postpartum care journey.”


Through ongoing training of Penny and the underlying technology, Penn Medicine has seen more than 70% of patient questions correctly answered by conversational AI.

“This percentage gave us confidence that patients were receiving timely, evidence-based answers to questions about their care while reducing the number of routine questions clinicians need to answer so they can focus on more complex patient concerns,” Miller reported. “In some situations, Penny was unable to answer questions because we did not have clinician-curated content for those specific patient questions, so we were able to work with the Memora Health team to develop appropriate responses and optimize the program accordingly.”

Miller said high engagement and patient satisfaction are major results.

“For patients, the program received very high engagement rates – more than 98% of patients interacted with Penny at least once, 93% completing a survey and more than 52% asking at least one question – and earned a 92% satisfaction rate, showing this program is easy to use and patients are getting that personalized care they need and deserve,” she said.

“This number represents that not only are we helping inform the clinical care they need, but patients appreciate and are approving of the tools we are using to keep them healthy and safe,” she continued. “Importantly, we found Black patients were statistically more likely to promote the program compared to white patients. As we look to solutions for the maternal health crisis, we must find technologies that specifically target and support disproportionately impacted populations.”

Miller also said identification of new onset postpartum hypertensive disorders has been a big plus coming out of the use of this technology.

“We started screening our patients who had no previous diagnosis of hypertensive disorder of pregnancy with our program,” she said. “We are fortunate that in our health system all parents already have a blood pressure cuff to check their BP during pregnancy.

“While screening postpartum patients, we found 3% of patients have new onset hypertensive disorder of pregnancy, of whom approximately 45% have no symptoms of elevated blood pressure,” she continued. “In addition to blood pressure screening, making it easy for patients to ask questions about symptoms has also allowed us to detect this potentially serious condition.”

Many postpartum patients suffer headaches due to fatigue, dehydration and sleep deprivation, but a severe headache can be a sign of preeclampsia.

“We have found when a patient identifies a headache as particularly severe, they often also have a concurrent hypertensive disorder,” she said. “A particular patient comes to mind, someone with a severe headache who messaged our program. The clinical team that received this alert was able to assess the patient through the platform and detected a severely elevated blood pressure.

“She was ultimately diagnosed with and treated for preeclampsia with severe features,” she noted. “After she recovered, she shared this feedback with us: ‘It helped me, I’m a single mom and can’t always come in to see about myself. Without this, I would have probably died,'” she added.

Screening for postpartum depression has been another aspect of the new technology that has led to significant results.

“More than 40% of patients completed screening for depression using our platform via completing the 10 question Edinburgh Postnatal Depression Screening questionnaire,” Miller reported. “Of patients who completed this screening, 25% screened as at risk for depression. Detecting abnormal screens allows our team to connect patients back to their clinical team sooner for management, counseling and potential medication therapy.”


When used correctly, AI can have a profound impact on patient care, Miller said.

“As we all know, the healthcare workforce shortage combined with burnout that so many of my colleagues experience poses a danger to patient care,” she said. “If there is a way to incorporate intelligently designed tools like what we are using at Penn Medicine, I encourage my peers at other healthcare provider organizations to do so.”

As long as one has the buy-in of patients, in Miller’s opinion, there are only good things that can come from the ability to use AI technology. She offers a few pieces of advice for providers considering using conversational or mobile health applications in patient care:

  1. “Before you even start on this journey, make sure whatever technology you’re looking to use is the one your patients are going to engage with, because if you don’t have the support of your patients, the technology will be useless.”

  2. “Engage the stakeholders in this process, from the very beginning. Using partners or departmental leadership in your testing phase (ask them to ‘think like a patient’) can garner excitement and buy-in from this very important part of your team.”

  3. “If you are working with a vendor outside of your EHR, start budgeting for and thinking about requesting integration early in the process. Failure to do so will make implementation and scale much more difficult.”

“By automating certain processes, we can provide more comprehensive, equitable and effective care experiences,” said Miller. “No one wants to leave the hospital feeling like they do not have a lifeline through communication with someone from their clinical team, especially in the days and weeks after having a baby. Tools like this give patients the peace of mind, knowing they can ask any question, at any time of day or night, and get a response.”

Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki
Email him: [email protected]
Healthcare IT News is a HIMSS Media publication.

This post was originally published on 3rd party site mentioned in the title of this site

Similar Posts