A professional information specialist searched MEDLINE and Embase, in-process MEDLINE and PubMed unique content, and the Cochrane Database of Systematic Reviews for systematic reviews or controlled trials published from inception until May 12, 2020. We also searched ClinicalTrials.gov for active studies until June 19, 2020. The review protocol is posted on the PROSPERO website [7].

Textbox 1 shows study eligibility criteria for studies evaluating the effects of BP monitors on hypertension. Device engineers examined the devices from the screened studies (manufacturer and model names) and determined whether each device was available for direct purchase by consumers. Studies that included nonconsumer devices (eg, devices requiring a prescription) were excluded. The technology had to collect and store consumer data without requiring manual input and potentially could be sent to a health care professional, although data transmission was not required for study inclusion. We included both US-marketed and non–US-marketed technologies that met the criteria. However, any technology subject to Food and Drug Administration (FDA) clearance must have received FDA clearance to be included.

We carefully examined the interventions provided to each treatment group and determined whether the study design isolated the effect of the BP monitor. This occurred when the intervention group received the BP monitor whereas other comparison groups did not, and any additional treatments were the same between groups. In cases where clinicians made changes to treatment plans (eg, medication or dose adjustments) based on feedback from the BP monitor, we considered it as part of the BP monitor’s effect because such adjustments were only possible due to the device. The comparison groups commonly received usual care, which would not preclude the clinician’s decisions to modify hypertension treatment plans based on BP measurements in other contexts and settings.

Using DistillerSR (Evidence Partners), 3 reviewers (JRT, BR, and JR) screened the titles, and all 6 screened abstracts and full-text articles. For titles, only 1 reviewer assessed the general relevance to the topic. For abstract screening, 2 reviewers were necessary to exclude an article from further consideration; however, only 1 reviewer was necessary to order the full text. Regarding full texts, 2 reviewers assessed the study against the inclusion criteria, and disagreements were resolved by a (senior-level) third reviewer (JRT or JR). Full-text screening also involved determining which articles were associated with other included articles of the same trial.

For each included trial, 1 reviewer (BR or NM) extracted the general trial information, patient characteristics (eg, baseline BP), treatment details (including specific PGHD devices), risk-of-bias items, and outcome data. We examined data on the following reported health outcomes: mortality, emergency room visits, hospitalization, quality of life (QoL), and adverse events (AEs). Surrogate outcomes for hypertension consisted of systolic BP (SBP) and diastolic BP (DBP). Process outcomes included medication changes, dose adjustments, physician consultations, and office visits. We also extracted data on consumer experience, including device adherence, the number of BP readings taken or transmitted, device alerts, ease of use, patient satisfaction, and technical problems.

We assessed the overall risk of bias based on 9 items, including randomization, allocation concealment, baseline similarity between groups, and masking of outcome assessors. The items were adapted from the AHRQ report titled “Mobile Applications for Self-Management of Diabetes” [8]. In addition, we included an item about whether the device’s effects could be isolated (ie, consumer BP monitor alone vs usual care). After considering all 9 items, we categorized each trial as at low, moderate, or high risk of bias.

Given that the included studies were published as early as 1997, for each BP monitor used within the included studies, device engineers assessed the similarity to devices currently on the market from that manufacturer. They used the following scale: (1) this model is similar to a device available from this manufacturer; (2) this model is somewhat different than any device available from this manufacturer; (3) this model is very different from any device available from this manufacturer; and (4) we could not reliably determine the similarity of this model with the ones currently available from this manufacturer.

For isolated effects on health outcomes, we narratively synthesized the summary effect into one of four categories: (1) likely no effect, (2) unclear, (3) possible positive effect, or (4) likely positive effect. If the results consistently demonstrated the lack of an effect (via narrow CIs around a null effect), we coded it as likely no effect. If the results were inconsistent in the direction of effect or study authors could not reach a conclusion, the findings were coded as unclear for that outcome. If ≥1 outcomes had minor inconsistency in findings, but at least 1 study with moderate or low risk of bias showed a positive effect, the findings were coded as possible positive effect. If the results had a consistent positive effect, we coded it as likely positive effect.

When we categorized health outcome data as unclear, we then examined surrogate outcomes, which for hypertension were SBP and DBP. To help interpret the SBP/DBP outcomes, we used a minimally important difference of 2 mm Hg [9,10].

For studies of multicomponent interventions, we did not attempt to classify the data in the manner described earlier because the effect of BP monitoring in those studies could not be determined.

For the full report (ie, 11 clinical conditions), our searches identified 8667 potentially relevant articles, of which we excluded 5755 (66.40%) at the title level (not relevant) and 2196 (25.33%) at the abstract level (Figure 1). We dual-screened the full texts of the remaining 716 articles (8.26%). The review team included 126 (17.6%) of these studies, but upon further review of the devices by device engineers, 12 studies (1.7%) had used only nonconsumer devices and were therefore excluded from the full report (none of the 12 addressed hypertension). A total of 114 unique studies were described in 166 articles. For the subset of screened studies enrolling patients with hypertension, we included 51 studies reported in 80 articles. This review focuses on 49 (96%; 79 articles) of those 51 studies that used BP monitors to generate PGHD for managing hypertension; 2 studies did not use BP monitors to manage hypertension, 1 evaluated a pedometer [11], and the other compared 2 mobile apps [12]. Of the 49 studies, 18 (36.7%) used designs that isolated the effect of BP monitors (eg, BP monitor alone vs usual care or BP monitor+scale vs scale alone), whereas the other 31 (63.3%) used multicomponent designs that did not permit conclusions about the impact on outcomes specific to BP monitors (eg, BP monitor+scale vs usual care).

Key characteristics of the studies using BP monitors for hypertension are shown in Table 1 (18 isolated-effect studies) and Multimedia Appendix 1 (Table S1; 31 multicomponent studies). Of the 49 studies, 47 (96%) were randomized trials, and 2 (4%) were nonrandomized; 21 (43%) studies were conducted in the United States, and other notable countries included the United Kingdom (n=6 studies, 12%), Canada (n=3 studies, 6%), Denmark (n=2 studies. 4%), Finland (n=2 studies, 4%), and South Korea (n=2 studies, 4%). The median number of patients per study at baseline was 222 (IQR 80-433). Patient enrollment dates were reported in 29 (59%) studies and ranged from May 1999 to June 2017. The median length of follow-up was 6 months (IQR 4-12).

Study group comparisons are shown in Table 2 and Multimedia Appendix 1 (Table S1). Of the 49 studies, 42 (86%) had 2 study groups, 4 (8%) studies had 3 groups, and 3 (6%) studies had 4 groups. A usual care control group was used in 43 (88%) studies, whereas 3 (6%) studies used a consumer device in the control group, and 4 (8%) other studies used active comparators without a consumer device (eg, counseling alone). Statistical power analyses were conducted a priori in 39 of the 49 (80%) studies, and 29 of these 39 (74%) studies were based on SBP, DBP, or BP control. Note that 31 of the 49 (62%) studies used only multicomponent interventions, making it impossible to discern the impact specific to the BP monitor. Among these 31 studies, 25 (81%) used a BP monitor along with nondevice interventions, 3 (10%) studies used a BP monitor along with another device, and the other 4 (12%) studies used a BP monitor along with ≥2 other devices.

Table 2 (isolated-effect studies) and Multimedia Appendix 1 (Table S2; multicomponent studies) show the patient characteristics from the 49 studies. The mean age ranged from 49 to 73 years, and the percentage of females ranged from 5% to 88%. The mean baseline SBP was reported in 44 (90%) studies and ranged from 125 to 161 mm Hg. The mean baseline DBP was reported in 42 (86%) studies and ranged from 71 to 97 mm Hg. Only 3 (6%) studies were conducted in rural populations [25,37,38], whereas 24 (49%) were not of rural populations [22,23,26,29,30,35,36,39-62] and the other 22 (44%) did not specify.

Only 21 of the 49 (43%) studies reported health outcomes, which included mortality (n=3 studies, 6%), hospitalizations or emergency room visits (n=2 studies, 4%), QoL (n=13 studies, 26%), and AEs (n=13 studies, 26%). No studies reported other health outcomes related to hypertension, such as major adverse cardiovascular events. All studies reported SBP, DBP, or BP control.

The included studies used 41 different BP monitoring devices (see specifics in Table 1). Of these, 34 (83%) were arm devices and 2 (5%) were wrist devices, and the wrist or arm was unclear in the other 5 (12%) studies. A total of 38 (93%) BP monitors were judged as similar to devices currently on the market from the corresponding manufacturer, 1 (2%) was judged as somewhat different, and 2 (5%) were of unknown similarity.

Regarding the transmission of data (eg, to a website, to study staff, or to health care providers), 19 of 49 (39%) studies used automatic transmission, 6 (12%) used manual data entry for transmission, 20 (41%) had no electronic data transmission, and the other 4 (8%) did not report whether or how data were transmitted.

The isolated effects of a consumer BP monitor device on health outcomes were evaluated in 6 of the 49 (12%) studies. The consumer BP monitors examined included the iHealth BP 7 Wireless Wrist Monitor, Omron 637, Omron 773AC, Omron HEM-705 CP, Omron HEM-712C, and Omron M10-IT. Only 1 of the 6 (17%) studies reported mortality [22], 1 (17%) reported hospitalization [14], and the other 4 (67%) reported QoL [4,18,20,21,31-33,36].

Of the 49 studies, 18 (37%) [4,13-26,28-36] examined the isolated effects of consumer BP monitors on blood pressure. All evaluated the effects compared with usual care (ie, no BP monitor), except for 2 (11%) studies [25,29], each of which compared BP monitors with automatic data transmission with BP monitors without automatic transmission.

All 16 studies on comparisons with usual care reported the effects of PGHD interventions on SBP (Figure 2). The top 4 points were from studies using automatic transmission of BP data, and the remaining 28 points were from studies that did not use automatic transmission. Six studies [4,15-17,19-21,26,31-34] found a statistically significant reduction in SBP favoring the BP monitoring group compared with the control group. However, the results were somewhat inconsistent. For example, Bosworth et al [15-17] found significant improvement only in non-White patients at 12 months; differences were not statistically significant for White patients at any time point or 24 months for any subgroup. The point estimates for SBP are shown in Figure 2, corresponding to 32 reported outcomes from 16 studies. Moreover, 4 of 32 (13%) SBP outcomes identified a reduction of 6 mm Hg or more favoring the consumer BP monitor group compared with usual care; 12 (38%) identified an SBP reduction between 2 mm Hg and 6 mm Hg favoring the consumer BP monitor, 10 (31%) identified SBP differences from −2 mm Hg to +2 mm Hg, and the remaining 3 (9%) found an SBP reduction ≥2 mm Hg favoring the usual care groups. Whether the BP monitor automatically transmitted data (comparing the top 4 points with the other points) did not appear to modify the effect on SBP.

The overall findings for DBP were similar to those for SBP; 5 (31%) [4,15,19,30-34] studies found that consumer BP monitors significantly reduced DBP compared with controls. However, similar to SBP, the results were inconsistent, and statistical significance was found only for particular subgroups or time points in a study. The 32 point estimates for DBP are shown in Figure 3 (restricted to studies with usual care comparison groups). Of these, 1 (3%) identified a DBP reduction of 6 mm Hg or more favoring the consumer BP monitor, 9 (28%) identified a DBP reduction between 2 mm Hg and 6 mm Hg, favoring the consumer BP monitor, and the remaining 19 (59%) identified DBP differences from −2 mm Hg to +2 mm Hg. Whether the BP monitor automatically transmitted data did not appear to modify its effect on DBP.

Regarding the 2 studies examining the effect of data transmission (eg, BP monitor with vs without data transmission), Kaihara et al [25] found that data transmission resulted in an estimated 6 mm Hg lower SBP but no statistically significant effect on DBP. Lakshminarayan et al [29] found a statistically nonsignificant difference of 3.7 mm Hg in favor of data transmission and did not report data on DBP.

BP control was examined in 9 (15%) studies of the isolated effects of consumer BP monitors [13,14,19-23,27,28,30,34]. Most defined BP control as SBP <140 mm Hg and DBP <90 mm Hg, but 1 study [23] used <135/<85 mm Hg; 2 [14,23] studies included a separate definition of <130/80 mm Hg for patients with diabetes. Only 2 of the 9 (22%) studies [19,34] reported statistically significantly higher rates of BP control with BP monitors than with controls.

The remaining 7 (78%) studies found nonsignificant differences in BP control rates between BP-monitored and control patients.

Of the 18 studies on isolated effects of BP monitors, 5 (28%) reported process outcomes, and the results were mixed. For medication prescribing, McManus et al [4,31-33] found that those in the BP monitor group were prescribed statistically significantly more antihypertensive drugs than those in the usual care group (difference 0.11; 95% CI 0.02-0.19), and 3 other studies found no statistically significant impact of BP monitoring on prescriptions.

However, these studies were not statistically powered to detect such effects, so they did not rule out the possibility of an impact on prescriptions.

In addition, Bosworth et al (2009) [14] found no between-group differences in the number of outpatient encounters (medians ranged from 13 to 15).

With regard to data transmission, 5 of the 18 (28%) studies used automatic data transmission, 2 (11%) used manual entry, 10 (31%) had no electronic data transmission, and 1 (3%) did not report whether or how data were transmitted. Of those using automatic data transmission, in Hoffmann-Petersen et al [23], data were transmitted using a Tunstall RTX3371 or Numera telehealth monitor to a study database or an electronic health record after BP measurements.

In Kaihara et al [25], the BP monitor wirelessly transmitted data to a study database over the internet.

In Kim et al [27], the BP monitor readings were wirelessly transmitted via the HealthCircles app on a smartphone to a website.

In Lakshminarayan et al [29], a smartphone transmitted daily BP measurements to a study database. Participants in the PGHD group transmitted data on an average of 89% of the study days and rated the ease of use of the system favorably.

In Zha et al [36], the wireless BP wrist monitor would transmit data to a website using the iHealth MyVitals app on a smartphone.

Of the 18 studies, 2 (18%) studies used manual data transmission [4,35]. In these 2 studies, participants sent BP readings via an SMS text message service or web-based form to a website [4] or entered their BP measurements on a BP-tracking website [35].

Of the 49 studies, 12 (24%) reported on AEs and generally found them to occur infrequently, and 4 [4,14,20,21,31-33,35] of the 18 (22%) studies on isolated effects of BP monitoring reported on AEs; 2 (17%) studies reported that no AEs occurred during the course of the study. A study [20,21] found that serious AEs, including nonfatal cardiovascular events, were rare and not substantially different between the BP monitoring and control groups. Another study [4,31-33] reported on various other AEs, including pain, fatigue, and dry mouth. Only dry mouth occurred significantly more frequently in the BP monitor group than in the usual care group. Of the 49 studies, 11 (22%) [4,14,20,21,31-33,38,44-46,50-56,60,63-74] reported on AEs in studies with multicomponent device groups. Only one of those studies [69-73] reported a significant increase of an AE, swelling of legs, in a multicomponent intervention group that included a BP monitor compared with usual care.

Of the 49 studies, 26 (53%) reported the outcomes of consumer experience. Adherence to the use of BP monitors ranged from 38% to 89%, but device adherence had variable definitions. For example, Logan et al [47] defined adherence as a minimum of 8 readings per week. Zaleski et al [35] only determined whether patients said they were still monitoring their BP. Zha et al [36] measured adherence by dividing the number of received readings by expected readings.

Some studies reported that adherence declined throughout the study. For example, Bosworth et al [14] reported that during the first 2 months, 91% of those using a BP monitor were adherent, whereas 64% were adherent during the last 2 months. The studies also measured BP monitor use in various ways, including the total number of transmissions during the study and the average number of transmissions per week.

Studies measuring the ease of use or satisfaction with consumer BP monitors found favorable ratings. For example, Magid et al [49] reported that 68% of patients using the monitor found it very or extremely easy to use. Rifkin et al [75] reported that 96% of patients would continue to use the BP monitor.

Only 2 studies reported problems with BP monitors. Bosworth et al [63-65] found that 35 alerts were triggered by the monitoring system due to BP monitor problems, which represented 5% of the total alerts that occurred during the study. Lakshminarayan et al [29] found that some patients experienced issues with the BP monitor and the smartphone provided to transmit BP data, including an inability to hold a charge and difficulty using the phone app to see BP data.

Of the 31 multicomponent studies [38-87], 11 (35%) examined the multicomponent effect of BP monitors on health outcomes, and all 31 evaluated multicomponent PGHD for surrogate outcomes including SBP, DBP, and BP control. These study designs did not permit any determination of the effectiveness specific to BP monitors.

Of the 18 studies of isolated effects, we rated 6 (33%) as low risk of bias, 9 (50%) as moderate risk of bias, and 3 (17%) as high risk of bias. In contrast, of the 31 studies of multicomponent effects, we rated 6 (19%) as low risk of bias, 13 (42%) as moderate risk of bias, and 12 (39%) as high risk of bias. The full AHRQ report (in its Appendix Table C-26) contains the item-level and overall risk-of-bias ratings for each study [7].

This systematic review summarizes 49 comparative studies that used consumer BP monitors for hypertension management. However, the effects of these devices on health outcomes remain unclear. Only 18 studies were designed to isolate the BP monitor effect, and only 6 of these 18 (33%) studies reported any health outcome, such as mortality, hospitalization, and QoL. One study [36] found a statistically significant difference in QoL at follow-up favoring usual care over BP monitoring, but QoL also favored usual care at baseline (suggesting a problem in the randomization process). None of the 5 remaining studies found statistically significant effects on health outcomes, possibly because they were powered to detect differences in BP measurements and not necessarily differences in health outcomes. Many studies had only 6 months of follow-up, which may also explain the uncertain effect of BP monitors on health outcomes.

We found consistent benefits of BP monitoring on both surrogate outcomes, SBP and DBP. SBP reductions typical of included studies ranged between 2 and 4 mm Hg, and DBP reductions ranged from 1 to 3 mm Hg. It is unclear whether these modest changes in BP related to consumer BP monitors lead to lower risks of hypertension-related complications or mortality. Many factors may have potentially modified BP reduction in these studies. BP self-monitoring may support behavioral changes or reminder strategies to assist with lifestyle changes or medication adherence [2-4]. In addition, select BP monitors transmit data to health care providers and can improve BP control by facilitating timely recommendations from providers to patients to better manage their BP [87,88]. However, only 5 [23,25,27-29,36] of the 18 (27%) studies on isolated effects of BP monitors used automatic data transmission, and the effects on provider behavior change were rarely described among the included studies. This indicates that many studies did not use the advanced capabilities of modern BP monitors and may explain the unclear impact on health outcomes.

Most studies reported adherence to BP monitor use that ranged from 38% to 89%, but adherence was inconsistently measured. There was also a large gap between self-reported and measured adherence, such as a set number of recordings per week, as self-reported information is not always reliable. In addition, adherence can be affected by a variety of factors, such as daily access to the device, consumer comfort with the device, or self-motivation factors [89]. Spillover to other adherence factors, such as medication adherence or compliance with lifestyle behavior changes to manage hypertension, were not reported but may ultimately be a mechanism by which consumers of BP monitors improve their hypertension. Another consumer experience outcome, overall satisfaction, was reported as highly favorable among the included studies, thus validating the current rising consumer market for these devices.

Many studies evaluated multicomponent interventions, with BP monitors representing only 1 component, and did not separately evaluate the impact of the BP monitor. In our evidence base, only 18 of the 49 (37%) studies permitted such a direct assessment of BP monitor impact. Many PGHD technologies are intended to be used in combination with other interventions for chronic disease management, such as additional devices, exercise sessions, or health education sessions with medical personnel. These interventions may also influence outcomes; therefore, studies should be designed to measure the impact of isolated PGHD technology when added to other components.

This systematic review has several strengths. To our knowledge, this is the first systematic review to synthesize the patient-centered health effects of consumer BP monitors for hypertension management, in addition to their effects on BP. We closely followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) reporting standards and used robust AHRQ Evidence-based Practice Center systematic review methodology, including duplicate literature screening and data extraction. The findings of our review mirror those from 2 recent meta-analyses of systematic reviews of individual patient data [90,91] and contribute summary-level data on health effects as well as key data on medication management and consumer experience. Furthermore, in this review, we used device engineers to verify the consumer availability of BP monitors used in studies and their similarity to currently available models.

This systematic review has limitations related to both the review methodology and the generalizability of the available literature. We judged the overall risk of bias using an adapted tool designed for mobile apps in managing diabetes [8] and therefore may not have detected some biases. We did not assess the possibility of publication bias, which may be a key problem in studies funded by manufacturers of devices that collect PGHD. The included studies rarely provided sufficient detail to delineate the contributions of cointerventions to outcomes, particularly those related to changes in BP. This limits the generalizability of our findings to patients with limited access to care or underserved patient populations. This may also further limit the confidence in the validity of our findings not otherwise captured in our risk-of-bias assessment. Studies with usual care groups often provided few details about what happened with these patients, which may potentially explain the wide variation in BP results among studies. The inclusion criteria of multiple studies were specific to consumers who had access to and familiarity with technology, which could include using the internet, smartphones or computers, arm or wrist devices, or access to electricity. Less technically adept consumers may not experience the same benefits as those enrolled in these studies. In addition, only 3 [25,37,38] of the 49 (6%) studies focused on rural populations, suggesting that these populations are underrepresented. Only 19 of the 49 (39%) studies used automatic data transmission from PGHD devices to health care providers.

Future studies are needed to examine the effectiveness of BP monitors that transmit data to health care providers (which are then used to inform medical decisions). Additional data from implementation studies may help determine which components are critical for sustained BP improvement, which in turn may improve prescription decisions by clinicians and coverage decisions by policy makers. In addition, challenges related to data accuracy, interoperability, privacy, and security should be explored as this field continues to grow.