Hypertension is the leading attributable risk factor for cardiovascular disease and death, globally [1]. The 2 blood pressure (BP) monitoring methods commonly used in primary care settings are home BP monitoring (HBPM) and office-based monitoring. HBPM is superior to office-based measurements as a predictor of cardiovascular disease and all-cause mortality [2]. Furthermore, the prognostic value of HBPM improves with the number of home BP measurements that patients are able to provide to their health care providers [3], thus emphasizing the importance of having a reliable means of collecting and reporting home BP data at each office visit. A systematic review and meta-analysis of 37 randomized controlled trials (RCTs) comparing HBPM with office BP measurements for effectiveness in reducing BP showed HBPM to be more effective in achieving BP control [4]. Nonetheless, at least one-third of known hypertensive patients in Singapore do not have adequate BP control per conventionally recommended levels [5]. This is especially problematic among the elderly in whom the prevalence of hypertension (73.9%) is higher, as is the rate of uncontrolled BP (75.9%) [6]. The lack of reliability in the conventional HBPM method could be an important contributing factor in the failure to achieve effective BP control and cardiovascular risk reduction in these patients. The shortcomings of conventional HBPM using handwritten logbooks are well known and include inaccuracies, underreporting of data [7], and failure to bring logbooks to clinic visits [8]. The purpose of HBPM is undermined and the value of reported measurements is diminished without an effective means of making accurate home BP readings available to clinicians. It is these considerations that motivate and necessitate the exploration of more reliable methods of communicating home BP values to health care providers.

Mobile health (mHealth) technology has been increasingly evaluated for chronic disease management. mHealth is broadly defined as any health care practice supported by mobile devices and their functionalities [9]. With the widespread use of smartphones in recent years, mobile apps have gained attention as an mHealth modality [10]. A content analysis of the top 107 apps for hypertension management showed that 72% include a tracking function for BP values [11]. This simple feature, when coupled with the wireless data transfer capabilities of Bluetooth-enabled BP monitors, would allow a mobile app to function as an electronic logbook that is more convenient to use and more readily accessible to clinicians than a handwritten log. Singapore, with a 91% smartphone penetration rate, holds favorable conditions to utilize a wireless platform in a clinical setting [12]. However, there are relatively few studies in the literature that present a quantitative comparison between a smartphone app and manual logbook in terms of their respective reliability as a recording tool for HBPM by patients. In addition, because operating technological devices is highly user-dependent, specific patient factors associated with the effectiveness of smartphone app–assisted HBPM need to be explored.

Our pilot study aimed to begin addressing the abovementioned knowledge gaps by assessing whether there is any benefit in using mHealth technology (smartphone app) to store home BP values compared with using handwritten logbooks in terms of making these records available at clinic visits. The primary aim of our RCT was to compare the home BP recording fidelity over a 3-week period using a smartphone app versus a handwritten logbook in the Singaporean hypertensive patient population. Fidelity was defined as the percentage of scheduled home BP readings that are compliant with the HBPM regimen and are successfully reported at the follow-up visit. The null hypothesis postulated no difference in BP recording fidelity between a smartphone app and handwritten logbook. As there was no a priori basis for postulating greater fidelity with the app, the null hypothesis was tested against a 2-sided alternative, although the desired outcome was higher recording fidelity with the smartphone app. Secondary aims were (1) to explore associations between participant characteristics and the recording fidelity within each study arm and (2) to explore the effect of time on the weekly recording fidelity in each study arm. With detailed participant demographic data as well as HBPM records obtained via a logbook and the app, we compared the fidelity of the 2 home BP recording modalities, identified patient characteristics associated with higher fidelity for each monitoring method, described the attenuation pattern of fidelity in each study arm, and compared the fidelity of the 2 recording modalities on a weekly interval.

This study was an open-label, parallel group RCT of 2 study arms with a 1:1 allocation ratio.

The study protocol was approved by the SingHealth Centralised Institutional Review Board (reference #2017/2014) and registered under ClinicalTrials.gov (NCT03209024).

Patients were enrolled in the study via convenience sampling. Patients in the polyclinic waiting area were approached, a brief explanation of the study was provided, and prescreening questions were administered. Interested patients were subsequently screened for eligibility based on the study inclusion/exclusion criteria, and written informed consent was obtained from eligible patients. The informed consent procedure included an explanation of the purpose of the study, the study procedures and visit schedule, participants’ responsibilities and rights, and confidentiality of medical records.

A computer-generated treatment allocation sequence accommodating 80 patients was generated by the study statistician (JCA) using permuted block randomization with blocks of size 6 and one block of size 8. Sequentially numbered, opaque, sealed envelopes (SNOSEs) were prepared, with each envelope containing a treatment group assignment. The primary investigator enrolled all patients, whereas allocation concealment and sequential dispensing of envelopes were enforced by an on-site research coordinator. Patients were randomly assigned to either the smartphone app or the handwritten logbook study groups.

The Baseline Data Questionnaire was administered to obtain information on the participants’ sociodemographics, economic status, clinical characteristics, and exposure to technology. Baseline BP was recorded as the average of the last 2 of 3 consecutive BP measurements taken at 1-min intervals after 5 min of initial rest. The Hypertension Self-Care Profile questionnaire [18] is a validated tool to assess self-care behaviors in the domains of Behavior, Motivation, and Self-Efficacy. The version adapted to the Singaporean population was administered with permission from the authors [19].

At the end of the 3-week HBPM period, participants returned to the clinic for a single (final) follow-up visit. During the visit, the electronic log on the app used by the smartphone app group was exported to the study database for further analysis; similarly, a copy of the logbooks from the patients in the logbook group was uploaded into the study database. For security precautions, all participant data were anonymized and stored in password-protected computers or in locked cabinets.

Upon completion of the study, each participant received a Singapore $30 grocery store voucher.

A total of 928 patients were approached during the recruitment and follow-up period (03/15/2017–07/15/2017). As shown in the Consolidated Standards of Reporting Trials flow diagram (Figure 1), of the 102 patients undergoing screening, 83 were eligible for enrollment. Of those eligible, 80 were randomized. One patient randomized to the smartphone app arm was reallocated to the logbook arm at the baseline visit before commencing HBPM owing to unexplained smartphone incompatibility with the study app during the initial set-up process. One participant in the smartphone app arm withdrew from the study. In the smartphone app arm, 38 participants were included in the final analysis and 11 had home BP recording fidelity >80% at the end of the 3-week follow-up period. For the logbook arm, 41 were included in the analysis and 7 had home BP recording fidelity >80% at the end of 3 weeks.

The smartphone app and logbook arms were comparable in all baseline characteristics with the exception of systolic BP (SBP; Table 1). There was no evidence of SBP as a confounder.

In the primary analysis on 79 participants (smartphone app: 38 and logbook: 41), higher median fidelity was achieved with the use of the smartphone app (66.7%) compared with the logbook (52.4%), although the difference was not statistically significant (Wilcoxon rank-sum, P=.21; Table 2). Similar results were obtained from comparisons of mean fidelity using the t test (P=.21) and the general linear mixed-model (P=.14).

In comparison with the primary outcome, the assessment of time-independent fidelity exhibited higher fidelity in both study arms, although the difference between arms was smaller and not statistically significant (P=.30; Table 2).

For smartphone app–assisted HBPM, a multivariate analysis identified 5 independent baseline variables that exhibited statistically significant associations with fidelity: (1) The number of people working in the household was associated with decreased fidelity (absolute percent and 95% CI) by –8.18% (−16.3 to −0.08) per additional working person (P=.048); (2) Years of hypertension increased fidelity by 1.56% (0.03 to 3.09) per year (P=.046); (3) Years of current smartphone use increased fidelity on an average of 10.5% (0.83 to 20.2) per year of use (P=.03); (4) The number of apps on the current smartphone decreased fidelity by –0.32% (−0.58 to −0.05) per app (P=.02); (5) Mean fidelity had a significant association (P=.03) with ethnicity, with Chinese (n=24) and Indians (n=4) exhibiting higher fidelity by 35.6% (4.27 to 66.9) and 45.0% (8.69 to 81.3), respectively, in comparison with other ethnicities (n=3) composed of Pakistani, Ceylonese, and Eurasian ethnicities.

In the logbook arm, a multivariate analysis identified 3 independent baseline variables that exhibited significant associations with fidelity: (1) Working participants had significantly lower fidelity by 18.6% (1.51 to 35.8; P=.03) compared with nonworking participants; (2) Age was a positive factor with an average increase in fidelity by 1.21% (0.05 to 2.36) per year increase in age (P=.04); (3) For every additional child, fidelity declined by an average of –9.20% (–16.4 to –2.04; P=.01).

Self-care capacity, both in individual domains and the cumulative score of Hypertension Self-Care Profile, exhibited no evidence of association with fidelity.

In assessing the weekly fidelity trend over the 3-week period within each study arm, fidelity was highest in the first week and lower in subsequent weeks (Figure 2). This difference was most pronounced among the app users, whose fidelity in the first week (64.4%) was significantly higher than that of the second (55.1%; P=.001) and third (58.2%; P=.03) weeks of monitoring. Though a similar trend was seen in the logbook arm, no change in fidelity among weeks was significant.

The participants using the smartphone app had higher weekly fidelity than their counterparts using the logbook in the first week, with the least-squares mean difference approaching significance (P=.06). Although mean fidelity for the smartphone app arm continued to outperform the logbook arm in the second and third weeks, these differences were less pronounced (Table 3).

In a posthoc assessment of the elderly participants (60≤age<70), there was no significant difference in overall mean fidelity between app and logbook users (P=.10; Table 4). However, in the assessment of weekly recording fidelities, app users exhibited significantly higher fidelity (79.0%) than logbook users (58.8%) during the first week of monitoring (P=.048; Table 5).

Our study compared the recording fidelity of an app-mediated electronic record versus a handwritten logbook in performing HBPM. Although higher fidelity was observed in the smartphone app arm compared with the logbook arm, indicating the potential for improved fidelity with the use of the app, statistical significance was not achieved. Meanwhile, the findings of our secondary aims have provided valuable information on the fidelity of the app-assisted method by identifying a unique set of potential predictors and describing an attenuation pattern over time. Finally, the post hoc subgroup analysis by age group showed significantly higher recording fidelity with the app versus the handwritten logbook in the elderly during week 1 of monitoring, suggesting that app-assisted HBPM is feasible across a wide range of ages. These findings have promising implications for the expanding use of mHealth technology in hypertension management and warrant further investigation in future studies.

There remains a scarcity of the literature addressing a direct comparison of home BP recording fidelity between app-mediated and handwritten methods. Although some studies do report outcome measures similar to fidelity as we define it (eg, level of adherence to a given HBPM regimen), heterogeneity in regimen frequency and duration make comparisons of study results difficult [3]. In addition to reliable recording fidelity, adequate patient support may also be important to achieving effective clinical outcomes. The findings from a recent systematic review and meta-analysis showed that HBPM alone did not have a higher association with BP lowering and BP control compared with no-home monitoring (usual care), whereas HBPM with additional patient support via feedback, education, and counselling demonstrated a significantly higher association [20]. This emphasizes the importance of future studies to further explore smartphone app use, not only as a high-fidelity recording tool for HBPM but also as a platform for delivering effective patient adherence support such as health information and reminders [21].

From the secondary aim of our study, we found that the participant factors associated with higher fidelity were not the same for the 2 HBPM modalities, suggesting that determinants of fidelity in manual recording of home BP are different from those in smartphone use. To our knowledge, this is the first study to identify potential predictors of fidelity for app-assisted HBPM, and this novel information could be utilized in future studies to develop a clinically useful predictive model that provides decision support for identifying patients most suitable for mHealth-based monitoring. Although determining the mechanism of how certain patient factors lead to higher HBPM fidelity is beyond the scope of this study, we do offer plausible explanations for these associations. Among app users, longer ownership of a smartphone may mean greater proficiency with one’s own device, thus improving the performance of app-mediated HBPM. The number of apps on the phone exhibited a negative association with fidelity, although we find the clinical relevance of this finding questionable as the impact of an additional app on fidelity was negligible. A longer duration of hypertension and fewer working household members were also associated with higher fidelity, perhaps because patients who have been hypertensive for many years may be more accustomed to following an HBPM regimen, whereas those who have more employed family members may have fewer people available at home to provide family support for hypertension management. For handwritten logbook-mediated HBPM, age was associated with fidelity whereas employment and number of children exhibited negative associations. This may reflect the difficulty in manually recording home BP values on a rigorous regimen while simultaneously being occupied with commitments at work or home. With advancing age and eventual retirement, adherence may improve as demands at home and work decrease.

The significant attenuation of fidelity from weeks 1 to 2 in the smartphone app arm may be explained in part by the duration of BP monitoring. The ESH guidelines specify a 7-day period of monitoring for its HBPM regimen [13,14], and prolonging this rigorous regimen beyond the recommended period of 1 week may have led to study fatigue among participants in the smartphone app arm.

The post hoc subgroup analysis by age suggested that the elderly perform HBPM better with the app than the manual logbook, especially during week 1 of monitoring. This finding cannot be regarded as conclusive but suggests that the elderly do not lack in the smartphone proficiency needed for recording home BP, which challenges the common preconception that mHealth-based HBPM may be too complicated for older patients.

The use of a 2-sample test in the primary analysis with no adjustment for potential confounders was appropriate. Of the 30 covariates measured at baseline, only one (SBP) resulted in a significant difference between study groups, which is consistent with the expected false positive error rate for 30 individual tests performed at the 5% significance level. Moreover, there was no evidence that SBP was associated with fidelity.

This study has a number of limitations. First, as a pilot trial, the sample size was limited by time and available resources. As we had no knowledge of either the expected difference in mean fidelity or the population SD, the sample size calculation was based on a targeted Cohen ES of 0.6, which was the smallest feasible and realistically achievable ES, given the available resources and timeframe of the study. The observed ES for fidelity was 0.33, hence the study was obviously underpowered. Second, recruitment was limited to one polyclinic and the participants were exclusively current smartphone users, which could preclude the generalizability of the results to the greater hypertensive population in Singapore or around the world. Third, this was an unblinded study because participants were required to learn the procedures for their respective HBPM methods; investigators were not masked, but we attempted to reduce bias by defining our primary outcome measure and the primary statistical method a priori. Finally, the Hawthorne effect could not be precluded, and the magnitude of the effect on the 2 study arms may have differed. A future full-scale trial of longer duration may mitigate the Hawthorne effect as patients become accustomed to home BP recording over the course of several months or years.

With regard to the strengths of our study, to our knowledge, this is the first RCT to directly compare the recording fidelity of mHealth and the conventional logbook methods of HBPM within the multi-ethnic Southeast Asian population. In addition, although the HBPM regimen used in research settings can vary widely, the regimen used in our study was adopted from an established clinical guideline and recommendations by ESH, which contributes to the reproducibility of this study and standardizes its outcome to allow comparison with other studies that also use this well-established HBPM regimen.

For the future full-scale trial to be sufficiently powered, the sample SD of approximately 28% and mean difference in fidelity of about 8% observed in this study can be used to ensure that a 2-arm study with n=175 participants per group would provide 80% power to detect an ES of 8/28 (approximately 0.30) at alpha=.05.

Although beyond the scope of this study, there remains a need to categorically quantify and compare the influence of over-reporting, underreporting, and falsifying data on the fidelity of logbook- and app-mediated HBPM to guide future app design to promote accuracy in data transfer from home to clinics. Moreover, other important clinical outcomes in hypertension management, such as the magnitude of BP lowering and proportion of BP control, must be assessed.

As with any novel health care implementation, the sustainability of app-mediated HBPM is another area that deserves the focus of future research. As intuition would suggest, health apps are not immune to attenuation of usage over time, which has direct implications on the home BP recording fidelity as was observed in our results as well. One study on the effect of the usage pattern of a health care app on user retention suggests the frequency of utilizing the self-monitoring function of the app—as opposed to other functions such as outpatient support service or medication functions—significantly increases the probability of the app to stay in use [22]. The authors of that study suggest that the benefit of achieving better health outcomes by modifying health behavior based on tracking one’s own health parameters is the mechanism behind how frequent utilization of the self-monitoring function of a health app promotes its sustained use. The self-monitoring function is at the core of app-mediated HBPM, therefore it is vital for future studies not only to describe attenuation patterns of app use over time but also to explore modifiable factors (including BP recording regimens and the ease of app user interface) that promote frequent self-monitoring, thereby improving the app’s retention rate.

As health technologies continue to find an increasingly profound integration into clinical practice, our study makes a contribution to the sparsity of available knowledge on the fidelity of smartphone app–assisted HBPM. Although our pilot study did not find a significant difference in BP recording fidelity between app-assisted and logbook-mediated HBPM, our results have identified unique determinants of fidelity in the 2 recording methods of home BP, which has not been reported before, and have characterized the attenuation of fidelity in time. Our findings also suggest that app-assisted HBPM is viable across a wide age spectrum that includes the elderly.