This study is a prospective analysis of real-world user data from a digital BP-coach (Manoa app) in the period from April 2020 (release of the Manoa app) to November 2021 in Germany. The app was freely available in Germany during this period in all major app stores (iOS, Android). The Department of Nephrology and Hypertension, Hanover Medical School initiated this analysis. It was approved by the local ethics committee (9033_B0_K_2020). Users gave their informed consent that their data may be evaluated for scientific purposes.

Manoa is a conversational smartphone app which collects user information and provides coaching and support to promote hypertension self-management. It also covers topics like healthy lifestyle (e.g., healthy diet, physical activity) and medication adherence. The app was developed in collaboration with Hannover Medical School (MHH).

Core element is the blood pressure diary. The digital coach provides individual feedback on the user's blood pressure level, based on the results from a structured HBPM protocol. Encouragement by the digital Coach “Manoa” (chatbot), reminders, and tracking are used to promote home monitoring. The app motivates users to measure their blood pressure twice a day (every morning and evening) for 6 days (HBPM period). After completion of the first HBPM period, the user is asked and reminded periodically (every 4 weeks) to monitor his or her blood pressure. The app also allows the user to reschedule the HBPM period. Users can record their blood pressure manually, scan the result via the smartphone camera or import the blood pressure values directly from a compatible measurement device. For the analysis, the first HBPM period (baseline) was considered as successfully completed if the user had taken 2 measurements per day (in the morning and in the evening) on at least 6 days within a period of 6–14 days. For the following HBPM periods, 3 complete measurement days (each with one measurement in the morning and in the evening) within 7 days were sufficient.

During the first HBPM period, the user receives relevant information (e.g., about the correct blood pressure measurement at home) through puzzles, articles or videos, for example. After completing an HBPM measurement period, the user receives evaluative feedback and further recommendations based on his blood pressure values. In addition, the app offers a visualisation of the blood pressure values in a diary. This diary can be exported and shared with the doctor.

The participants downloaded the app voluntarily and self-motivated. Persons under 18 and pregnant women were excluded from app usage. The analysis is based on all users with self-reported diagnosis of hypertension who started a first measurement week (n = 3175). Depending on the analysis, subgroups were defined as indicated.

Hypertension was diagnosed according to current guideline recommendations (5). Users were classified in normal BP, high normal BP, grade I hypertension and grade II hypertension according to outcome-driven thresholds, as previously published (7, 16, 17). Home BP was analysed at baseline and 8–16 weeks after successful completion of the first HBPM period (follow up).

The analyses were performed with the open source statistical software R or with GraphPad Prism 9.0. Characteristics of users were summarised as numbers and percentage for categorical variables and mean and SD for continuous variables. The change in the blood pressure level (mean) was analysed with a paired t-test. Differences in categorical variables were compared through a χ2 test. A generalized additive mixed model (GAMM) that considers the repeated measures and a potential non-linear relationship between BP values and day of measurement was applied in the analysis using the R package mgcv (v1.8.40) (18, 19). One result of the applied GAMM models is the reported effective degrees of freedom (EDF) which is a measure of the extent of observed non-linearity, i.e., the curvature of the smoothing spline function. A value of EDF = 1 indicates a linear relationship between BP and time, whereas EDF close to 2 indicates a parabolic curve shape. The identified relationships along with the 95% confidence intervals are plotted and the discussion of non-linearities is based on the plots.

Pearson correlation test was performed to evaluate the relationship between change in systolic BP (follow-up—baseline) and systolic BP at baseline. Piecewise linear regression was used to test if baseline systolic BP predicted change (follow up—baseline) in systolic BP. Cohen's kappa was computed to assess the agreement between different HBPM protocols in classifying hypertension.

From April 2020 to November 2021, datasets from 3,175 app-users with self-reported arterial hypertension were analysed, while the app was freely available in all major app stores in Germany. The baseline characteristics are summarised in Table 1. The mean age (SD) of the participants was 61.6 (10.4) years. 55.6% of the users were older than 60 years and 1,527 (48.1%) of them were female. In the subgroup of users who reported their last OBP (n = 2666), the average reported OBP (SD) was 138.7 (14.4) /83.9 (9.7) mmHg, which is in the controlled range for systolic BP (SBP) and diastolic BP (DBP) (<140/90 mm Hg). However, 54.1% of users had at least one OBP value (systolic or diastolic) which classified them as uncontrolled. We also queried previous HBPM experience of users at the beginning of the coaching program (n = 3983). 51.0% of users had performed HBPM regularly, while 47.8% of users had performed HBPM sporadically and 1.3% of users had never performed HBPM (Table 1).

Current guidelines recommend the completion of a structured HBPM protocol to evaluate ambulatory blood pressure values (5). Therefore, we first analysed the rate of self-reported hypertensive users who successfully completed the first HBPM period with the digital coach (completion of at least 6 BP measurement days). Of 3,175 users with self-reported hypertension who enrolled in HBPM, 81.3% (2582) completed the first day. During the ensuing HBPM interval, there was a steady decline in active users by 1%–2% per day, resulting in an overall successful completion rate of 74.1% (2,352 out of 3175) in this strictly self-motivated and app-coached setting (Figure 1A). Thus, the majority of users was lost during the HBPM initiation process (18.7%), while a smaller fraction of users (7.2%) was lost during the actual measurement period. 91.1% of users (2,352 out of 2582) who started the first measurement day completed the entire HBPM period (Figure 1A). In the subgroup of users who had uncontrolled hypertension on the first day of the HBPM period, the same relative change of user numbers was observed in the subsequent measurement days (Figure 1B). The different age groups did not differ in their usage behaviour (Supplementary Figure S1).

The digital coach supports users in implementing and documenting a structured HBPM-protocol. However, despite the widespread use of HBPM, the exact number of days required for an adequate classification is still matter of debate and accuracy in relation to BP categories is unclear. A recent meta-analysis concluded that a three-day protocol is sufficient in most cases, but should be extended to seven days in case of doubt (20). Furthermore, it is currently unclear how baseline or final (definitive) BP category influences BP course and HBPM protocol requirements. We therefore analysed temporal BP patterns and BP course stratified by BP categories of participants of the first HBPM period.

Across all participants, mean SBP (SD) was highest after the first measurement day [132.6 (13.0)]. During the following measurement days, systolic and diastolic blood pressure decreased significantly over time. Using generalized additive mixed modelling (GAMM), systolic BP decreased in a linear fashing with a slope of -0.25(3) mmHg/day (EDF = 1, p = 2.14 × 10⁻¹⁴), whereas diastolic BP decreased non-linearly (EDF = 1.71, p = 6.9 × 10⁻⁸) with changes being greatest between day 1 and day 2 (Figure 2A). A linear approximation over the whole time period resulted in a slope of −0.13(2) mmHg/day (p = 2.8 × 10⁻¹⁰).

To understand BP volatility in more detail, we next stratified users according to their initial hypertension category (based on day 1 BP). In this analysis the significant changes (details see caption of Figure 2) in blood pressure over time were most pronounced in the categories at the respective ends of the diagnostic spectrum, with the normotensive user group even showing an increase in blood pressure over time. In users categorised as high normal and grade I hypertension only minor changes were detected over time (Figure 2B). Although the sign of the temporal BP patterns across the BP categories differs (increasing vs. decreasing BP), a stable BP level was reached at around day 4 of measurement.

Various studies have shown that an increase in measurement days is associated with better reproducibility and accuracy (17, 20, 21). Since correct BP-classification (e.g., normotension, grade I HTN, grade II HTN) informs prognosis and therapeutic decisions, we examined how the increase in measurement days affects user BP-classification. Overall, the proportion of patients to be reclassified compared to the previous day (based on the cumulative mean) decreased with an increasing number of consecutive measurement days. On day 2, agreement with the previous day was 74.3%, while on day 5 > 90% of users maintained their classification (Table 2). This finding is based on a stabilisation of the mean BP-value over the course of the HBPM period, which can be attributed to the increase in the measured values. However, BP reclassification over time and time to reach definitive BP class was not uniform. At the extremes of the diagnostic spectrum (normotension, grade II HTN), reclassification over consecutive measurement days occurred less frequently and a high proportion of users matched their final category early on (defined by the mean of a complete HBPM period, Figures 3A,B). Of note, 89.8% of users with BP readings in the grade II range were correctly classified at day 2. In contrast, users in BP categories high normal and grade 1 hypertension required more days to decrease BP class volatility and reach their definitive BP class, with 90.4% and 89.8% correctly classified, respectively, at or after day 5 (Figure 3B). Thus, a higher number of measurement days is required for adequate classification when BP values are close to the therapeutic threshold. However, accuracy was improved in all categories when measurements until day 6 were integrated.

Analysis by BP classification tracking revealed an intensive exchange of users between directly neighbouring BP categories during the early to middle phase of the HBPM period, which continued at low level until the end of the HBPM period (Figure 3C). In contrast, more distant class switching occurred less often and early on (Figure 3C). Of note, a BP classified as normal after day 2 was never derived from a previous hypertensive reading and never switched to a hypertensive class, while a BP classified as grade I received significant contributions from the high normal class until the end of the measurement period (Figure 3C).

Since the greatest BP variability occurred in the early phase of HBPM, we examined the impact of excluding measurement day 1 on diagnosis and classification of hypertension. If the readings of the first day were omitted, the mean SBP (SD) decreased significantly from 131.7 (11.2) mmHg to 131.5 (11.3) mmHg (p < 0.0001 C.I. = 0.21–0.10). To investigate whether this change is clinically relevant, we analysed the relative change in the distribution of BD categories after exclusion of day 1. When the first day was omitted from HBPM, BP categories changed in 8.2% of cases (189 out of 2307). Different BP categories were affected to varying extent: while there was a 95.7% agreement in normal BP, accordance with full 6-day protocol in grade I hypertension was only 87.7% (Figure 3D). However, a more detailed analysis showed that excluding the first day was therapeutically relevant for a small proportion of patients. 8.5% of the patients initially classified as grade I HT were downgraded to high normal BP after omitting the first day (56 out of 659). In contrast, 4.1% of the users classified as high normal by a day 1–6 protocol were reclassified as grade I hypertension (32 out of 788). Overall, 3.8% of all users were affected by a therapeutically relevant reclassification after exclusion of day 1 (88 out of 2307, Figure 3D).

To study the association of app usage and blood pressure change, we analysed BP values at baseline and 8–16 weeks after successful completion of the first HBPM period. Previous studies suggest that the BP lowering effect of app-based BP management tools can be detected after more than 8 weeks (22, 23). If a user performed more than one HBPM period within 8–16 weeks after completion of the first measurement week, the most recent HBPM period was defined as the follow-up period. Baseline characteristics of this subgroup (n = 864) are described in Table 3.

Baseline mean (SD) SBP was 131.2 (10.8) mmHg, which was in the controlled range. After 8–16 weeks, the mean (SD) systolic blood pressure decreased significantly by 2.6 mmHg to 128.6 (8.9) mmHg (p < 0.0001, 95% C.I. = −3.09−2.02; Table 4, Figure 4A). Mean (SD) DBP decreased from 80.7 (7.8) at baseline to 79.1 (7.3) at follow up (p < 0.0001, 95% C.I. = −1.94−1.28; Figure 4B). In the subgroup of users with uncontrolled systolic BP (n = 384), mean (SD) SBP at baseline was 138.8 (9.9) mmHg and decreased significantly to 132.7(8.9) at follow up (p < 0.0001, 95% C.I. = −6.99−5.24; Table 4). Lower BP levels also resulted in changes in hypertension control. While at baseline, 55.6% of hypertensive users were controlled for systolic and diastolic BP, this proportion increased significantly to 68.1% at the follow-up HBPM period ([X² (1, N = 864) = 10.92, p = .0009, Table 4]. Since the antihypertensive effects were most pronounced in users with high blood pressure, Pearson correlation coefficient was computed to assess the relationship between change in SBP and baseline SBP. Above a SBP of 125mmHg at baseline, there was a strong correlation with change in follow-up BP [r(624) = −.55, p < 0.0001], while below 125mmHg the degree of correlation was lower [r(236) =−.18, p = .004]. Participants with a very low baseline SBP showed a modest increase of blood pressure (Figure 4D).