We present a digital phenotyping protocol specifically designed for pregnant populations using smartphones and Garmin smartwatches to continuously measure behavioral, physiological, and contextual data. Our design addresses the inherent challenges posed by pregnancy, including a nonfixed study duration and unpredictable delivery-triggered phase transition, by spanning from the third trimester through 6 weeks post partum. This protocol also takes into account the engagement-burden tradeoff with data collection and increased physiological variability during pregnancy through various data streams and development specifically for digital phenotyping, which is the in situ quantification of an individual’s phenotype using data from their digital devices, such as smartphones and wearable sensors [1]. This approach allows for the continuous objective measurement of behavior, physiology, and environmental context in someone’s everyday life, offering a detailed understanding at the individual level [1]. Data in this study are collected both passively via automatic sensing from ubiquitous devices and actively through ecological momentary assessments (EMAs), which are brief surveys completed by participants in their everyday environments. By overcoming the limitations associated with retrospective self-report methods, this approach allows for a more comprehensive understanding of pregnancy-related behaviors, physiological processes, and maternal and infant outcomes.

Conventional analytical techniques that rely on retrospective self-reporting can be affected by recall bias and thus fail to capture critical temporal changes [1]. In contrast, digital phenotyping uses high-frequency real-time data [1]. Previous perinatal digital phenotyping studies, such as the Mom2B study [2] and the Postpartum Mothers Mobile Study [3], have shown the feasibility of smartphone-based prediction of outcomes, EMA for post partum and breastfeeding, and mobile tracking of maternal health. Although the Mom2B study relied primarily on smartphone data, our protocol extends this work by integrating synchronized wearable data and multiple app-based platforms. We also adapted EMAs from the study by Demirci and Bogen [4] on primiparous women and the combination of EMA and smartphone data from the Postpartum Mothers Mobile Study [3]. By continuously and objectively capturing behavioral and physiological data, our protocol provides robust insights into the complex interactions influencing maternal and infant health outcomes. This comprehensive approach can significantly enhance our understanding of the pregnancy and postpartum experience. In addition, by collecting a multitude of data that adapts after participants deliver, we can understand participants’ behavior in a more comprehensive way than digital phenotyping studies that stay the same during different stages of pregnancy or the postpartum period.

The first challenge that this protocol must address is the nonfixed study duration.

The duration of an individual’s pregnancy is highly variable, and in clinical practice, the estimated due date (EDD) is 40 weeks of gestational age [5]. In the Early Pregnancy Study [6], even after excluding pregnancies affected by complications or preterm births, the variation in human gestational length was 37 days. Only approximately 5% of people deliver on their assigned 40-week gestational age EDD, and 66% deliver within 7 days of their EDD [5]. The variability in delivery timing means that although the duration of the third trimester is defined as 12 weeks, beginning at 28 weeks of gestation and culminating at 40 weeks of gestation [7], many people will have a third trimester that is shorter or longer. The duration of the third trimester can range from less than 1 week, if the baby is delivered at 29 weeks, to approximately 14 weeks. For example, a participant delivering at 30 weeks 0 days gestation would have 2 weeks of data collected in the third trimester and 6 weeks of data collected in the postpartum period for a total of 8 weeks. Conversely, a participant delivering at 41 weeks and 6 days of gestation would have 14 weeks of data collected in the third trimester and 6 weeks of data collected in the postpartum period for a total of 20 weeks. The nonfixed third-trimester study duration creates a challenge, as compensation in a study is typically for a set amount of time. In addition, data storage amounts cannot be pre-estimated, and research apps typically need to know the study duration before commencement. In addition, to best understand pregnancy, EMA questions need to transition from prenatal-focused questions to postpartum-focused questions. Given these inherent biometric challenges, we developed a protocol that adapts to both longer and shorter study durations.

The variance in the third-trimester duration for participants creates the challenge of an unpredictable delivery-triggered phase transition, which is the change in EMA questions that occurs because of the participant delivering their baby. We collect data in 2 phases, antepartum third trimester and postpartum period, which are segmented by delivery. There is no way to know for certain exactly when the baby will be delivered, and the delivery date will be different for each participant. There will be a discrepancy between the EDD (40 weeks) and the actual delivery date, and between the actual delivery date and the date on which the research team becomes aware that the participant’s pregnancy has ended. This may lead to a delay when the postpartum EMA questions are triggered. The unpredictable delivery-triggered phase transition will also lead to some participants providing a different amount of data before and after delivery, with participants who have a longer third trimester providing a greater amount of data before delivery. This variability introduces complexity when developing a longitudinal data collection protocol aimed at capturing critical behavioral and physiological changes during pregnancy, the puerperium, and the postpartum condition. The protocol must account for transitioning from prenatal data collection streams to the postpartum data collection streams in a way that can be implemented variably after delivery without leading to delays in postpartum data collection.

This protocol aims to investigate the engagement-burden tradeoff. Research data suggest that pregnancy is a health care state associated with high rates of engagement with online searches. One study reported that 90% of pregnant individuals obtained supplemental health care information online [8]. In addition, 50% of pregnant individuals installed a health-tracking app on their phones during pregnancy [9]. This provides a unique opportunity to leverage this increased engagement to collect more data than is typically collected from other populations. However, we must balance this opportunity with minimizing the burden on the participant while still emphasizing increased engagement. Previous studies have prioritized minimizing the burden on participants by including only 1 detection device, collecting fewer data streams, or implementing either passive or active data collection, and not both simultaneously. This protocol must leverage the opportunity for increased data collection while also incorporating multiple touchpoints with clinicians and the use of apps and devices that participants are familiar with.

The pregnant population is vulnerable and experiences increased physiological, psychological, and social variability [10,11]. Several physiological and potentially pathophysiological changes occur during pregnancy [12,13]. Subjective data regarding these changes are collected with surveys at multiple points throughout the pregnancy and after delivery. To address the vulnerability of pregnant participants, this protocol must have questions tailored to and validated for the pregnant population and not make assumptions about the participants. Objective physiological data are obtained via passive sensing with a wearable device.

Many previous digital phenotyping studies investigating pregnancy have focused exclusively on the postpartum or perinatal period or have used the same protocol for both phases [2-4]. By focusing exclusively on 1 period, key changes that occur between the pregnancy and postpartum periods are missed. In addition, these protocols did not need to account for the uncertainty in delivery timing or adjust their protocol during the postpartum period due to having 1 consistent protocol or focusing on 1 period [2-4]. In contrast, our protocol uniquely spans the duration of each participant’s third trimester and resumes immediately during the postpartum period for 6 weeks of new data collection. Our protocol supports digital phenotyping by combining passive data collection from Garmin smartwatches and participant smartphones with active labeling through EMAs. Over an approximate 18-week period, from the third trimester through the 6-week postpartum period, we continuously track behaviors and physiological data relevant to pregnancy outcomes, such as delivery, newborn feeding practices, and mental health and well-being.

This protocol collects data passively through a smartphone (Beiwe app [14]) and smartwatch (Labfront application [15]). It collects data prospectively and actively through participant survey completion, EMAs, and participant data logging in the Huckleberry app [16]. The use of the Beiwe app addresses the nonfixed study duration for EMAs and smartphone data by allowing flexibility in study length. The use of the Labfront application for smartwatch collection also allows for flexibility in the length of the study. The unpredictable delivery-triggered phase transition is also managed by the Beiwe app, as the platform allows us to trigger a change in EMAs after delivery. The engagement-burden tradeoff is addressed by adding more active data collection through EMAs and having participants log data in the Huckleberry app. We allow participants to log their data into the Huckleberry app whenever it is convenient for them and balance the burden of EMAs by implementing this only once a day. Finally, to ensure that the protocol is pertinent for this population with increased physiological variability, we have reviewed and adapted all 350 survey questions and 14 unique EMA questions. In addition, we incorporate all study visits with in-person medical appointments and have clinicians and doulas on staff to help with any situation that may arise (Figure 1).

An initial cohort of approximately 30 nulliparous (females who have never given birth to a living child) participants will have data collected from the start of their third trimester through the 6-week postpartum period. A starting point of 30 participants was selected to ensure fair compensation. The inclusion criteria are that the participants should (1) be pregnant, (2) be between 14 and 28 weeks of gestation, (3) be nulliparous, (4) be aged ≥18 years, (5) own a smartphone, and (6) speak English. The participants will be enrolled before 28 weeks of gestation. Participants must own a smartphone to be able to answer the daily surveys and collect data. Participants must speak and read English, given that the daily survey questions are currently provided in English only. Participants will be excluded if they (1) are unable to provide consent, (2) have a pre-existing diagnosis of anxiety or depression, (3) are unable to wear a smartwatch, or (4) have a medical contraindication to breastfeeding. Participants who are unable to provide consent will be excluded. We will exclude participants with a formal diagnosis of anxiety or depression to reduce the effect of confounding variables, as pre-existing anxiety or depression can impact pregnancy outcomes [17]. Participants must also be able to breastfeed, as we will track breastfeeding rates and attitudes during the postpartum period as one of our main outcomes. Finally, someone who is unable to wear a smartwatch for the entire period is excluded, as they would not be able to collect physiological data such as heart rate, sleep, and activity.

Participants will be recruited in 4 ways to allow for maximum reach. This population has increased contact with physicians, allowing for increased opportunities for recruitment. First, informational posters will be affixed to walls in the waiting room at the Tufts Medical Center’s obstetric clinical areas. Second, the Tufts Medical Center’s obstetrics and gynecology prenatal practice patients will be messaged with a notification regarding the study at around 20 weeks of gestation through a health care provider-patient online portal. Third, participants can be recruited through phone calls from 20 to 24 weeks of gestation. Finally, participants can be approached at a midtrimester (14-28 weeks of gestation) appointment in person. Eligible participants will be identified by the Tufts Medical Center research team using information from their medical charts. This screening log will be stored in a password-protected folder and deleted after the trial completion. To reduce the time spent in the obstetrics and gynecology office and for patient convenience, participants will be able to walk through the informed consent process over the phone if they elect and sign consent forms electronically. Once individuals choose to participate, they will be guided by a research coordinator regarding the enrollment process, which starts with scanning a QR code that brings the users to a brief enrollment survey. This will be conducted at an in-person prenatal care appointment (at approximately 24-28 weeks of gestation). This enrollment survey provides the participants’ email and due dates to the Northeastern University research team, allowing the instant transfer of information from people who are only interacting in person at the Tufts Medical Center with the Tufts staff to the Northeastern University research team. Participants will install all the necessary applications and set up their smartwatch with guidance from the research coordinator and a study-specific installation physically provided to them.

To ensure consistency across the Beiwe and Huckleberry apps and the Labfront application, all data will be aligned using standardized timestamps. Before study initiation, each participant’s smartwatch will be synchronized to the designated study time zone. During data processing, all streams will be harmonized based on these timestamps, allowing for the precise alignment of physiological, behavioral, and EMA data. A unique study ID, whose key is stored separately from the data, will link participant data across platforms, ensuring the proper integration of multiple data sources. This approach not only enables reliable data merging but also enhances robustness by allowing the research team to account for missing information in 1 platform with corresponding data from another. All data are stored in password-protected folders on the Northeastern University server. Only approved members of the research team will have access to these data. Data will be deidentified as much as possible.

Data quality checks will be performed every morning between 8 AM and 10 AM by the research team to assess completion rates and wear time and to ensure that data values are within the expected range. Low wear time, missing data, and missing surveys will result in an email being sent from the research team to the participants to remind them of how to properly wear the watch, open the Garmin application to synchronize their daily watch data, and answer the surveys. If missing data are due to a technical difficulty, the research team will find a solution or give the participant a new device. If participants no longer meet the requirements, for example, can no longer wear the watch or are leaving the country, they will be removed from the study. If participants withdraw or are removed from the study, their data will be deleted and not included in the analysis. The Northeastern University and Tufts Medical research teams will have regular meetings to ensure proper trial conduct and to communicate any changes that need to be made. The research team will also, nonsystemically, keep track of any reported harm that occurs to participants; however, we predict that harm will be unlikely to occur. Potential risks for participants include discomfort or rash from the smartwatch, emotional discomfort when answering EMA questions or survey questions, and a potential loss of confidentiality and data security.

This protocol has been approved by the Tufts Health Sciences Institutional Review Board (STUDY00005835 protocol version 5.14.2025). This study is being conducted in collaboration with the Tufts Medical Center, which is Health Insurance Portability and Accountability Act–compliant. All members of the research team have undergone training on research involving human participants. We will obtain informed consent from all participants. The privacy and confidentiality of participants will be protected. Participants are compensated for their participation in this study.

Certain aspects of this protocol require manual and timely tasks to be done. An example of this is that available data collection apps supporting digital phenotyping are not set up to account for a nonfixed study length. The exact data of childbirth is unpredictable, and a delay from self-reporting and implementing the change in the EMAs should be anticipated. Handling this transition phase manually is increasingly tedious as the study becomes larger, implying that scalability is constrained by the time of the research coordinator. Another example of a manual task is transferring the data collected in the Huckleberry app to the research team. In the future, an app that provides the participant with their health information should also provide research protocols for a more seamless experience. An additional limitation of the protocol is that it requires participants to wear the wearable device provided by the study team. It would be better for participants if the data collection tools were device-agnostic, allowing them to use their own devices. Finally, this protocol may have a higher than usual dropout rate because of not meeting the criteria of a viable pregnancy, due to the many complications that can occur during pregnancy. We mitigate this partially by starting data collection as late as the third trimester, which reduces the opportunity for complications compared to data collection starting earlier in the pregnancy.

We intend to use participant feedback and completion rates from this study to refine the protocol for a future large-scale study of pregnancy with varying conditions. We will also use the results of longitudinal data analysis from this study to determine the effect size of factors, such as activity, feelings of support, mental health, mood, and breastfeeding intentions during the third trimester, on pregnancy, breastfeeding, and mental health outcomes. These estimated effect sizes will inform power analyses in future research endeavors.