Does anyone fit the average? Describing the heterogeneity of pregnancy symptoms using wearables and mobile apps

Introduction

There is an unacceptable lack of evidence-based knowledge of and support for the pregnancy experience, internationally. Accordingly, there is unnecessary maternal morbidity and mortality associated with ineffective tools to identify early enough when a pregnant individual is at risk for a serious complication[1,2]. We have not until recently had the technology to adequately measure the needed scope and resolution of the dynamic pathophysiological processes that occur during pregnancy in near real time[3]. The paucity of existing research on pregnancy experiences, symptoms, related complications and conditions stems largely from un-representative study samples of pregnant individuals and uses aggregate approaches to characterize study data that often comprises subjective recollections of symptoms. The ‘aggregate approach’ combines individual level data and finds average trends within the group. This approach conveys how the average individual experiences pregnancy, in the context of the specific study sample used. There are a number of challenges with the aggregate approach that apply widely to the study of all diseases [4], but are uniquely relevant to pregnancy experiences that result from complex interactions between pre-pregnancy individual characteristics and proximal perinatal environments[5–8].

The expansion of the digital health technology (DHT) space over the past decade has given rise to novel, remote tools that can be worn on the person including wearables and smartphone apps, herein referred to as ‘personal DHTs’. Personal DHTs can be used to capture rich, semi-continuous multimodal (of different types) health-related information[9] that is uniquely relevant to pregnancy[10]. Consumer-grade wearables are capable of tracking semi-continuous measures of physiology (e.g. heart rate (HR), heart rate variability (HRV), body temperature, blood pressure, and oxygen saturation), behavior (e.g. activity, social activity, relative location, and sleep duration and quality) and metabolic information (e.g. weight, body fluid volumes) that include many in-clinic objective measures of health and additional measures not routinely captured that could provide objective measures of risky pregnancy-related symptoms[9,11]. Research approaches involving the collection of such data and in the use of machine learning and artificial intelligence (AI) techniques are showing promise for the personalized objective detection of symptoms such as gait[12], cognition[13], and mood[14] in other disease populations. The value in such data capture is in its objective nature, limiting reliance on subjective recollections of symptoms, ability to be captured outside healthcare settings, and in that it could encompass broad domains (pathophysiological, environmental, and psychosocial) that affect pregnancy experiences. Further, personal DHTs could enable the personalized monitoring (at the n-of-1 level) of pregnancy experiences and symptoms and be returned back to the user in near real time.

With the ubiquity of the smartphone[15], and the increasing popularity of wearable smart devices[16] such as the smartwatch and the smart ring, the use of such tools for monitoring pregnancy-related signs and symptoms and greatly transforming current knowledge of this important metamorphosis is a potential reality. While the body of research involving personal DHTs in pregnancy is small, a number of small preliminary studies have shown potential utility in using wearables for tracking activity, sleep, and physiologic metrics such as HR and HRV as predictors of pregnancy-related outcomes such as preterm birth and delivery readiness[17–23].

The aim of this paper is to elucidate the variability of the pregnancy experience at the n-of-1 level using multimodal personal DHT data from the Better Understanding the Metamorphosis of Pregnancy (BUMP) study - a US-based, participant centric digital health research study that follows individuals from preconception to the fourth trimester (three-months postpartum). Specifically, the study objectives were to: 1) characterize individual level pregnancy self-reported symptoms and objective wearable features compared to the aggregate; 2) determine whether pregnancy self-reported symptoms and objective wearable features can differentiate pregnancy-related conditions; and; and 3) describe associations between self-reported symptoms and objective wearable features.

Results

The BUMP study enrolled 524 participants. This analysis included 406 participants who had at least one data point across the features examined and completed the study through birth. Characteristics of the sample can be found in Supplemental (Supp) Table 1. Participants were mostly white, 87% had a bachelor’s degree or higher, 82% were currently employed and working, and 34% were of advanced maternal age (35-40 years). In this study sample, there were 65 participants that experienced gestational hypertension, 61 that experienced postpartum depression, 43 that experienced preterm birth, 41 that experienced gestational diabetes, 39 that experienced postpartum hemorrhage, and 40 that experienced preeclampsia (3 of which experienced eclampsia). The distribution of body mass index across participants can be found in Supp. Figure 1.

Objective 1 - Characterize individual level pregnancy self-reported symptoms and objective DHT features compared to the aggregate using data from personal DHTs

Figure 1 shows the proportion of participants reporting moderate to severe self-reported symptoms. In the aggregate, participants reported the following trends in self-reported pregnancy symptoms: increasing swelling, shortness of breath, gait changes and cognitive disturbances from the first to third trimesters, higher rates of fatigue during the first and third trimesters compared to the second trimester, higher rates of nausea and vomiting during the first trimester compared to all other trimesters, and increasing mood disturbances, peaking in the fourth trimester. Supp Figures 2a-g includes individual level raw data (coloured lines) and cohort averages (black line) across self-reported symptoms. On average, there were unique contours in self-reported symptoms over the four trimesters; however, there was high variability in self-reported symptoms at the n-of-1 level that show dynamic changes within individuals over pregnancy and into the postpartum period (Supp Figures 2a-g).

• Download figure
• Open in new tab

Figure 1. Proportion of participants reporting moderate to severe symptoms (nausea, shortness of breath, gait, swelling, fatigue, cognition and mood) over the four trimesters of pregnancy

Similarly, cohort averages of objective features show unique contours over pregnancy (Figure 2-4). For example, on average, HR increases during the first two trimesters, and starts to decrease during the third trimester leading up to birth, while peripheral fluid volume steadily increases into the third trimester. HRV lowers with gestational age, but starts to increase leading up to delivery. The paradoxical inflection of HRV leading up to delivery is observed in the aggregate; however, is highly variable at the n-of-1 level, even among healthy pregnant individuals (participants who were not overweight, or obese, are under 35 years of age, and had no pre-existing medical conditions prior to pregnancy, or during pregnancy) (Figure 5). As observed in the self-reported symptom data, there is also high variability in objective feature data at the n-of-1 level both at baseline and over the course of pregnancy (Figures 2-4 and Supp Figure 3).

• Download figure
• Open in new tab

Figure 2. Individual level data (coloured lines) and cohort averages (black lines) for Bodyport peripheral fluid and weight over the four trimesters of pregnancy

• Download figure
• Open in new tab

Figure 3. Individual level data (coloured lines) and cohort averages (black lines) for Garmin heart rate, respiratory rate, and activity, over the four trimesters of pregnancy

• Download figure
• Open in new tab

Figure 4. Individual level data (coloured lines) and cohort averages (black lines) for Oura night time heart rate, heart rate variability, and respiratory rate over the four trimesters of pregnancy

• Download figure
• Open in new tab

Figure 5. Average Oura nighttime HRV and n-of-1 level trends for eight unique healthy participants over pregnancy

Objective 3 - Describe associations between objective DHT features and subjective self-reported symptoms

Mean Spearman correlations across the examined self-reported symptom and objective feature pairs were close to 0 for all of the examined subjective/objective pairs with the exception of self-reported swelling and peripheral fluid measured from the Bodyport scale (Supp Table 2). Figure 6 shows individual level Spearman correlation strengths across all unique participants for two symptoms (swelling and shortness of breath). Self-reported swelling was correlated with peripheral fluid status in most participants (Figure 6). However, across all of the other self-reported symptoms - objective feature pairs, there was high variability in correlation strength across unique participants. In particular, there was a similar proportion of participants showing both positive and negative correlations between all of the examined feature pairs (Figure 6, Supp Figures 8a-f).

• Download figure
• Open in new tab

Figure 6. Individual level correlations between daily self-reported swelling and Bodyport Peripheral Fluid (top) and daily self-reported shortness of breath (dyspnea) and Garmin respiratory rate (bottom)

Discussion

Self-reported symptoms and objective features reflective of physiologic, metabolic and behavioral changes over pregnancy, are markedly heterogeneous across individuals. While on average, pregnant individuals show unique contours of certain subjective self-reported symptoms and objective features from wearable devices, these patterns are misleading for many unique individuals. In particular, aggregate nighttime HR and RR were higher among individuals in pregnancy related-condition groups compared to healthy individuals - a difference that was observed early in pregnancy through birth, while HRV steadily decreased with gestational age, but showed an inflection at approximately nine weeks leading up to delivery. This phenomenon, that could reflect an individual’s physiological readiness for delivery, has been observed in the aggregate among other studies[18,23,26]. However, at the n-of-1 level, this aggregate trend is highly heterogeneous, even among healthy pregnant individuals.

Personal DHTs offer exciting opportunities to capture individual level high-resolution health-related information across multiple domains of health that could be used to glean a deeper insight into pregnancy pathophysiological changes and potential targets for early intervention. One of many values in leveraging the full richness of n-of-1 multimodal DHT data is in finding enough individuals who follow similar trajectories (e.g., the digital doppelganger) that might form new archetypes. These archetypes can then be used as the foundation for the development of early interventions. This is in contrast to the classical aggregate approach of finding groups of individuals who look similar to each other on average. These preliminary data demonstrate the potential challenge in identifying such archetypes owing to the observed high heterogeneity.

Subjective experiences of symptoms are limited by information biases. Objective measures of health enable an unbiased measure of an underlying biological or psychological process that could be used to identify when certain symptoms are becoming risky before the subjective experience arises. The degree of utility of objective measures of health to strengthen the subjective symptom relies on the nature of the symptom itself. For example, the subjective experience of swelling could be argued as an easier symptom to notice due to clear physical changes in the body as opposed to the subjective experience of cognition or mood that relies on individual level perceptions. These examples can be seen as differing examples of ground truths. Indeed, we demonstrated here that there was a more consistent concordance between self-reported swelling and the objective feature of body peripheral fluid volume among pregnant individuals compared to the other objective and subjective pairings. This finding may also highlight that certain objective features are more confounded than others. In line with this, across most of the selected self-reported subjective and objective features, there was high heterogeneity in correlation strength across participants with an almost equal number of participants showing positive and negative correlations. This finding highlights the challenges and complexity in leveraging single objective feature data to enhance subjective symptoms.

Materials and Methods

Supporting Information

Supp. Table 1. Study sample characteristics

Supp. Table 2. Mean Spearman correlation across participants for select subjective self-reported symptoms and objective features pairs

Supp. Figure 1. Body Mass Index (BMI) distribution across participants

Supp. Figure 2a. Individual level (coloured lines) and cohort averages (black lines) of subjective self-reported swelling over pregnancy

Supp. Figure 2b. Individual level (coloured lines) and cohort averages (black lines) of subjective self-reported shortness of breath over pregnancy

Supp. Figure 2c. Individual level (coloured lines) and cohort averages (black lines) of subjective self-reported gait changes over pregnancy

Supp. Figure 2d. Individual level (coloured lines) and cohort averages (black lines) of subjective self-reported low mood over pregnancy

Supp. Figure 2e. Individual level (coloured lines) and cohort averages (black lines) of subjective self-reported nausea over pregnancy

Supp. Figure 2f. Individual level (coloured lines) and cohort averages (black lines) of subjective self-reported fatigue over pregnancy

Supp. Figure 2g. Individual level (coloured lines) and cohort averages (black lines) of subjective self-reported changes in cognitive function over pregnancy

Supp. Figure 3 - Individual level data (coloured lines) and cohort averages (black lines) for Oura ring sleep efficiency and temperature deviation over the four trimesters of pregnancy

Supp Figure 4a - Box plots for weight and peripheral fluid measured from the Bodyport Cardiac Scale across clinical groups, healthy participants and all enrolled participants in the BUMP cohort

Supp Figure 4b - Box plots for heart rate and steps measured from the Garmin smartwatch across clinical groups, healthy participants and all enrolled participants in the BUMP cohort

Supp Figure 4c - Box plots for heart rate, heart rate variability, respiratory rate, sleep efficiency and body temperature deviation measured from the Oura smartring across clinical groups, healthy participants and all enrolled participants in the BUMP cohort

Supp Figure 5 - Cohort averages and 95% Confidence Intervals (CIs) for weight and peripheral fluid content measured from the Bodyport Cardiac Scale across pregnancy condition and healthy groups over the four trimesters

Supp Figure 6a - Cohort averages and 95% Confidence Intervals (CIs) for night time heart rate measured from the Oura ring across pregnancy condition and healthy groups over the four trimesters

Supp Figure 6b - Cohort averages and 95% Confidence Intervals (CIs) for night time respiratory rate measured from the Oura ring across pregnancy condition and healthy groups over the four trimesters

Supp Figure 6c -Cohort averages and 95% Confidence Intervals (CIs) for night time heart rate variability measured from the Oura ring across pregnancy condition and healthy groups over the four trimesters

Supp Figure 6d - Cohort averages and 95% Confidence Intervals (CIs) for sleep efficiency (proportion of time spent asleep) measured from the Oura ring across pregnancy condition and healthy groups over the four trimesters

Supp Figure 7a - Cohort averages and 95% Confidence Intervals (CIs) for heart rate measured from the Garmin watch across pregnancy condition and healthy groups over the four trimesters

Supp Figure 7b - Cohort averages and 95% Confidence Intervals (CIs) for daily steps measured from the Garmin watch across pregnancy condition and healthy groups over the four trimesters

Figure 8a. Individual level correlations between daily subjective cognitive function and Nback score

Figure 8b. Individual level correlations between daily subjective gait and Bodyport Sway Velocity

Figure 8c. Individual level correlations between daily subjective mood and Active Task Emotional bias point

Figure 8e. Individual level correlations between daily subjective mood and Garmin Active Time

Figure 8f. Individual level correlations between daily subjective nausea and vomiting and Bodyport peripheral fluid volume