The US opioid crisis has had a significant effect on women of reproductive age both nationally and regionally [1,2]. Increasing numbers of women report opioid misuse during pregnancy; overall, of women who were surveyed as part of the Pregnancy Risk Assessment Monitoring Study [3], 21% of those using prescription opioids during pregnancy reported misuse, and 36% of those reported wanting or needing to cut back their use [4].

The postpartum period can heighten the risks of adverse outcomes and overdose among women struggling with opioid use disorder (OUD) [5]. Reasons for this include exacerbation of psychiatric comorbidities, increased experiences of partner violence, infant hospitalizations for Neonatal Opioid Withdrawal Syndrome (NOWS), as well as possible engagement with child welfare systems related to the substance use [5-8]. Mounting evidence has shown that women are more likely than men to face barriers to treatment entry and access, have a faster-growing rate of hospitalizations due to substance use, and a faster increase in the rate of overdoses, all of which point to the need for interventions specific to this population [9-11].

Loneliness and social isolation are compounded during the perinatal period and contribute to OUD and treatment outcomes [12]. Specifically, both loneliness and social isolation have been linked to increased rates of use, amount, relapse, and overdose, even when controlling for medications for opioid use disorder (MOUD) treatment engagement, employment, relationship status, and mental health diagnoses [13-15]. Compellingly, the model of social homeostasis suggests that opioid drugs in particular target neurobiological circuits implicated in social bonding, which have been posited to increase vulnerability to opioid misuse or addiction in the absence of meaningful social connection [16]. Feeling connected to others is especially critical during postpartum periods as loneliness is associated with postpartum depression [17,18] and poor social support can impact treatment outcomes [19].

Recent research highlighting the effect of maternal loneliness on worsening mental health outcomes has suggested that the use of text messages for support can be a more useful intervention than engagement with social media or online forums [20]. Beyond providing patients with easy-to-access information and a sense of connectedness, mobile health (mHealth) can also reduce provider burden in caring for a high-need population [21]. While access concerns to the internet and smart devices have been raised, this is generally not a problem among racially diverse and lower-income populations [22].

Educational content delivered asynchronously and “in real words” may be a way to overcome provider burden and simultaneously address patients’ concerns and questions. Indeed, mHealth technologies, such as apps, chatbots, and SMS messaging, allowing texting and feedback from digital support mechanisms to provide clinical education and support in a relational context, have been shown to be acceptable to perinatal patients with a range of clinical needs [23,24].

The limited mHealth technologies that exist for the support of perinatal OUD indicate high acceptability and usability. Digital delivery of education on NOWS, infant care, postpartum care, and maternal addiction recovery has the potential to empower a historically stigmatized patient population and support treatment adherence by increasing access to clinical support [25-27]. While these studies demonstrate that mHealth can be a valuable communication and education tool for perinatal patients with OUD, there are no studies that explore the acceptability of using an interactive text-based chatbot for these purposes.

Current technology models for perinatal patients with OUD cite the importance of patient-centered design [25,27]. Rooted in user-centered design, an approach to product development by focusing on target users [28,29], participatory design engages the intended users as stakeholders in a collaborative research process to ensure patient needs are effectively met [30]. Qualitative feedback from patients with OUD and health care providers can be used to facilitate the development of equitable, high-quality mHealth technologies to sensitively support postpartum individuals, promote addiction care, and address mental health disparities [31,32]. There is high utility in conducting a needs assessment to identify barriers, facilitators, and the specific needs of the target population to develop the most useful and acceptable intervention [33]. Therefore, qualitative feedback is critical to the development of text message content and question-and-response information for a postpartum OUD support chatbot that reflects the topics of importance, resources of interest, and preferred wording, tone, and language of its intended population.

To this end, we conducted an in-depth needs assessment with postpartum individuals with OUD and health care providers engaged in the treatment of perinatal patients with substance use concerns to adapt our team’s previously developed “Healing at Home” for postpartum support [34], a 24/7 2-way messaging chatbot, to the needs of a perinatal population with OUD. We will call this adapted intervention the “Penny Chatbot.” Penny will be a rule-based chatbot that sends anticipatory guidance, responds to keywords, and follows “if/then” decision tree logic.

We conducted a needs assessment using semistructured interviews to gather data on the needs and preferences of perinatal patients with OUD and the clinicians who care for them. We shared this information with a stakeholder advisory board composed of health care providers engaged in OUD care (including clinicians, social workers, and certified recovery specialists) and parent participants with lived experience of OUD to develop educational and supportive content relevant to OUD and recovery in the postpartum period. This work was guided by the COREQ (Consolidated Criteria for Reporting Qualitative Research) guidelines [35].

Two stakeholder groups were recruited to take part in 30‐60 minute semistructured interviews as part of the assessment. The mother sample included patients from the Perinatal Resources for Opioid Use Disorder (PROUD) clinic who were biological mothers who had given birth to a live infant within the last 2 years. Potential patients were referred by clinicians and approached by the study team via phone call or text message. The provider sample included certified peer recovery specialists, social workers, and clinicians from the departments of Obstetrics and Gynecology, Family Medicine, Addiction Psychiatry, Emergency Medicine, and Psychiatry. Provider participants all had extensive experience treating or working with perinatal patients with OUD, which made their expertise valuable for chatbot development. Providers are aware of the most commonly asked questions, important education relevant to postpartum OUD, and the facilitators and barriers to mHealth and treatment delivery for the target patient population. We determined that approximately 12 patients per group would be sufficient to reach thematic saturation and identify issues that would be useful for content development [36]. Semistructured interviews were conducted with 12 perinatal participants with OUD and 11 providers.

Patient interviews were conducted in-person and clinician interviews were conducted via Zoom (Zoom Communications, Inc). All interviews were conducted individually, audio-recorded with participants’ permission, deidentified, and then transcribed by a professional transcription service. Interviews were conducted by a female senior research coordinator who has extensive experience collaborating with perinatal patients with OUD, both in her role as a researcher and due to her cross-training as a birth and postpartum doula. She acknowledged that patients’ willingness to openly disclose their opinions and experiences with perinatal OUD may be hindered by a power imbalance, fear of stigma, or social desirability. Prior to initiating the interview, she took time to build rapport with the patient participants. She assured patients that their responses were anonymous and there were no right or wrong answers. She added that their expertise was important and the research team wants to better understand how to support perinatal individuals with OUD.

Separate interview guides were developed and used for each study group to collect both perspectives and reflect a holistic understanding of stakeholders’ perspectives and values to inform chatbot content. The interview guides were developed using input from stakeholders with lived experiences (ie, certified recovery specialists [CRSs]) and providers who routinely work with the patient population. Patient interviews included questions to gather data on patient-oriented content, such as “What questions do you have about managing your OUD care during your pregnancy?” We also posed questions related to delivery method and types of messages sent, including “How often would you like information tailored to your gestational age or baby’s age ‘pushed’ to you as a text message?” and “How do you see yourself using a texting option to get information about your healthcare, your baby’s development, or your recovery?”

Questions for clinicians and CRSs focused on commonly asked patient questions, for example, “What are some of the most frequently asked questions from patients about managing OUD in the postpartum?” The interview also probed for ways in which clinicians have successfully or unsuccessfully engaged patients in other telehealth platforms and anticipated patient reactions to a text-support platform, including “Do you use messaging systems to stay in touch with your patients between visits?” “How do you think your postpartum patients with OUD would respond to a texting platform to support them and help answer questions for them?”

Codebooks were developed using an integrated approach [37]. This process was informed by modified grounded theory [38], combining inductive coding of emerging themes within the data along with deductive codes informed by a priori determined attributes of interest [39] (ie, specific perinatal and OUD treatment information, and messaging information).

Qualitative analysis was conducted by staff at the Mixed Methods Research Lab (MMRL) at the University of Pennsylvania in a manner consistent with institutional review boards (IRBs) and federal Health Insurance Portability and Accountability Act (HIPAA) guidelines and regulations. One of the MMRL coders is a master’s level statistician with a background in public health, maternal mental health, and psychology who possesses more than 4 years of experience. The other coder is a BA-level trainee under the supervision of the master’s level statistician and has a background in psychology and statistical analysis. They both undertook extensive training in qualitative research methodologies and software packages when they were onboarded. MMRL faculty and staff are highly trained and expert in managing, conducting, and analyzing protocol designs involving grounded theory and other methodologies. Scientific oversight for this study was provided by MMRL director and coauthor PFC, MD, MSCE, who has more than 20 years of qualitative and mixed methods experience, and other MMRL faculty and senior researchers.

Using NVivo software (version 15; Lumivero), 2 coders from MMRL independently conducted open coding of 2 to 3 transcripts per study group, identifying initial themes and patterns in the data. Preliminary codebooks were developed through meetings in which coders compared coding and discussed interpretations. The codebooks were iteratively refined during coding and interrater reliability (IRR) tests to ensure clarity and consistency. For the IRR tests, codes were applied to approximately 25% of the dataset per group (patient n=3; provider n=3), double-coded independently by both coders. Coders then met through multiple sessions to review and discuss discrepancies and reach consensus until strong agreement was reached for all codes (Cohen κ>0.75). Following this process, the remaining data were coded independently by the primary coder (MM) using the finalized codebooks, cross-referencing identified thematic categories against the raw transcripts to ensure consistency between the data and our findings. Analyses progressed as codes were then grouped into broader themes and constructs for each group. Themes were also analyzed across participant groups to identify patterns, noting any overlap or variation in similar topics.

This study was reviewed by the University of Pennsylvania IRB (protocol number 853729). According to 45 CFR 46.104, category 3, 2, the proposal met eligibility criteria for IRB review exemption. Pregnant patients were recruited to participate in the study. To protect a potentially vulnerable group from undue influence or coercion, all patient participants were required to be aged 18 years or older, able to consent on their own, and have no contraindications for participation. If there was any reason during the informed consent process that led to suspicion that the participant’s mental state was impaired enough to cause doubt about the validity of the consent statement, they were excluded. Participants were informed that participation is completely voluntary and they may opt out at any time. Those who chose to participate provided consent electronically through REDCap (Vanderbilt University) in accordance with the IRB policies on collecting electronic consent. Data were anonymized and deidentified. Participants were compensated US $20 on a Greenphire ClinCard for their participation in the one-time interview.

Table 1 reports the demographics of the sample. Patients were 50% White, the majority were single, and most had low incomes. Half the sample were in current treatment for OUD at the time of participation and half were currently pregnant. All patients had given birth to a child within the last 2 years, although child age data were not available.

This study reports on a qualitative needs assessment among perinatal women with OUD and the providers that care for them to determine what content would be most useful to include in an educational and supportive chatbot for those with perinatal OUD. Study participants universally reported that the Penny Chatbot may be a beneficial addition to care. Considering the unique and complex needs of a perinatal patient population with substance use disorders, providers suggest that Penny could promote recovery maintenance and engagement in treatment. Addiction care requires frequent check-ins. For a patient population that faces high treatment dropout and relapse rates along with barriers to in-person care, providers indicated that digital technology may be useful in increasing access to clinical support, bolstering provider-patient trust, and reducing after-hours call burden. Our findings are consistent with prior literature indicating that patients and providers alike agreed on the utility of digital means to connect patients to care [41,42]. Patients and clinicians agree that a balance of interactive chatbot content on substance use recovery, safe baby feeding practices, general postpartum and infant care, and connection to vetted resources would be helpful for postpartum mothers. Patients indicated that their greatest area of concern is understanding the impact of MOUD on their infant while breastfeeding. Patients also expressed strong interest in text-based support because of the availability, convenience, and instant responsiveness of a chatbot. Even so, they highlighted the importance of messaging that feels personal and sensitive to their needs. Given that the chatbot addresses key areas of concern and feels individualized to users, patients and clinicians believe that the Penny Chatbot may be acceptable and useful, with enthusiasm for the adapted intervention. As a result, our work will inform the development of supplemental material for the existing chatbot platform, specific to this patient population.

This study contributes to the existing literature highlighting the acceptability of mHealth to enhance clinical care for perinatal patients and those with substance use disorders. In samples of postpartum patients who do not use substances, chatbots have been found to be acceptable and usable for addressing postpartum mental health needs [23,24]. This mirrors our findings in that the majority of patients in this study reported that they would like and/or use a postpartum support chatbot.

Currently, the majority of digital health applications in the substance use field are devoted to substance use screening, telehealth visits, and virtual peer support [43]. Nonetheless, SMS messaging has been found to be feasible and acceptable as part of SUD treatment in addressing multiple dimensions of engagement, including motivation, self-efficacy, social support, and appointment attendance [44]. This study expands on perinatal OUD interventions, seeking to harness interactive, 2-way messaging to provide text-based clinical support to postpartum patients that may promote recovery, bolster patient-clinician trust, and help address loneliness.

This study also adds to the existing knowledge base on the value of participatory design. Through in-depth interviews with patients and healthcare providers who engage with perinatal OUD, we have obtained comprehensive and applicable qualitative feedback to inform the adaptation of a chatbot for postpartum OUD support.

To address the expressed interest in postpartum recovery information, we will maintain the existing content on vaginal delivery and cesarean section recovery (with planned individualization, as requested by patients) and modify the pain management recommendations to mirror the advice of perinatal addiction specialists. Content will include education on nonopioid pain management options, alternative therapies, monitoring for overuse or misuse, adjusting MOUD, and learning how to have open communication about pain management with providers.

There was ample support for education on infant withdrawal. Patients indicated they frequently use the internet to ask about what to expect and how to help their infant. Parenting messaging for all users with custody of their infant will include information on the “Eat, Sleep, Console” (ESC) protocol [45], Intensive Care Nursery (ICN) visiting information, signs of withdrawal, and tips for reducing or addressing withdrawal in newborns once discharged (swaddling, pacifiers, and minimizing smoke exposure).

The most common patient concern was safe infant feeding practices. Previous studies have pointed to discrepancies in breastfeeding among mothers using MOUD or abstaining and have highlighted the need for more support in this area for postpartum mothers with OUD [46]. Our findings suggest that patients want to know how substances affect their child’s development and want education on lactation and formula feeding while on MOUD. The chatbot will emphasize the safety of MOUD while breastfeeding [47]. We will also develop content on formula adjustment during infant withdrawal, what to do if a patient returns to use while breastfeeding, and the impact on infants of using different psychoactive substances while breastfeeding, including marijuana, nonprescribed benzodiazepines, and stimulants.

While mentioned by patients infrequently, providers indicated that mental health screening and support are critical for the patient population. To monitor risk for worsening perinatal anxiety and depression, the Edinburgh Postnatal Depression Scale (EPDS) [48] will be administered through the chatbot. Providers also noted high rates of IPV among perinatal individuals with OUD, so we will use the Extended-Hurt, Insulted, Threaten, Scream (E-HITS) [49] screening tool to monitor for IPV through the chatbot. Concerning scores will be escalated to the clinical team for direct outreach and patients will be texted resources.

Patients universally reported interest in increased postpartum OUD support. Information on how to secure a lactation consultation, free baby food and diapers, family support, infant soothing, and postpartum mental health resources provided by local public health and social organizations will be available through the chatbot. The chatbot will also have 24/7 functionality. Users will be able to ask questions at any time of day. By texting the word “Question,” patients can prompt a sequence of “if, then” responses to get the desired answers they seek. Questions that cannot be answered or require clinical attention, such as a request to be connected to a mental health therapist, will be streamlined to their care team.

Patients wanted messaging to feel individualized and personal. Chatbot messaging will be developed to sound warm and encouraging, as if the patient is texting a friend. Patients will receive a warm greeting with the first message of each day, as well as a thoughtful and brief introduction before the delivery of sensitive content related to substance use. Providers warned that too many messages may overwhelm patients. Messaging will be limited to no more than 3 times a day to minimize the burden of message fatigue, and messages will decrease in frequency as the 6-week protocol goes on. To enhance personalization, the chatbot will use the patient’s name, infant’s name, and provide their provider’s phone number. We will also opt users in and out of text message content depending on what information corresponds with their birth and postpartum experience. Patients will receive either cesarean or vaginal delivery recovery content and lactation or formula feeding content. Patients who do not go home with their infant due to the child’s extended neonatal intensive care unit (NICU) stay or family services involvement will not receive information related to daily pediatric care.

The Penny Chatbot was described as a potentially beneficial supplemental support to postpartum individuals with OUD. As with other digital interventions for people with OUD, our data suggest that an engaging, sensitive, and individualized chatbot has value in increasing access to education, recovery maintenance, and vetted resources [50]. Providers warned that the chatbot’s effectiveness may be impacted by patient distrust, escalation protocol, and text fatigue. However, we will address these concerns by introducing the chatbot through a warm hand-off by a trusted clinician and giving a detailed explanation of aspects of the platform, such as its privacy and confidentiality. Patients will be made aware that a member of the care team will be notified and will follow up with them in the event of certain concerning responses, such as with mental health or intimate partner violence screening measures. Finally, we will ensure that patients are only receiving relevant and desired messaging. Users will be able to unenroll and stop messaging at any time.

This needs assessment demonstrated the application and usefulness of participatory design. Involving perinatal patients with OUD and clinicians who engage in addiction treatment, perinatal care, or a combination of both as stakeholders in the research process ensures that the adaptation of the Penny Chatbot will effectively address user needs, preferences, and concerns [51]. This needs assessment directly guides text-based content development, resource selection, screening measures, and overall communication style: frequency, wording, and tone.

We acknowledge that there are limitations in this study. The patient sample consisted of patients from a single perinatal OUD treatment center at the University of Pennsylvania, limiting the generalizability of the data to all perinatal patients with OUD. Further, all clinicians who were interviewed were employees of the same health system, limiting the generalizability of study findings.

Postpartum chatbots offer an interactive experience, providing 24/7 clinical answers, maternal mental health resources, and embedded intervention [34]. Text message–based delivery is ubiquitous and adaptive to the needs of an immediate postpartum patient population. mHealth technology access is not limited to higher-income populations; thus, chatbots can address barriers associated with stigma, affordability, and limited access to in-person medical care [52,53]. Chatbots have the potential to increase access to evidence-based mental health and substance abuse care for underserved groups [21]. As barriers to OUD care, especially during the perinatal period, include clinic access factors and stigma, we believe that the development of a postpartum OUD chatbot program using the information described herein addresses a gap in the existing knowledge base on mHealth for perinatal SUD as a way to provide a cost-effective and easily disseminated tool to improve OUD treatment outcomes [54].

Considering this, our team will use these data to further refine and test the Penny Chatbot. Future research on postpartum OUD chatbots should further explore and expand on content areas most useful and desired by this population, as well as ways clinical teams can manage ongoing monitoring to address urgent needs in a responsive manner. An in-depth participatory-driven needs assessment allows for the adaptation of the existing chatbot to meet the needs of a high-risk, difficult-to-reach population. As we move forward with testing of the chatbot for acceptability and feasibility, data from the needs assessment allow us to iteratively refine its content, delivery, and safety.