BACKGROUND AND SIGNIFICANCE
Mobile devices and health technologies have created a new paradigm in delivering care. Mobile health (mHealth) technologies have been created to assess the health status of patients, as well as to diagnose, help with medication adherence and even treat patients through wellness interventions . Patient-provider coordination and better information sharing can be enabled through informatics, information and communication technologies, and in particular, mHealth applications (apps).
We focus on physician-supervised mHealth apps (in contrast to purely self-managed apps) designed to assist blood pressure (BP) management. mHealth apps for managing BP can be helpful in the following ways. First, mHealth apps can collect and communicate data on patient vitals in a structured manner both to patients and to the provider. Second, instead of an infrequent, office-based treatment approach, they enable continuous monitoring of patient condition, provide reliable data, and enable interventions in the form of timely communication, thereby reducing the likelihood of therapeutic inertia. Third, health vitals data self-collected by the patient naturally increases self-awareness, promotes patient health literacy and improves patient self-management of their disease. Fourth, mHealth apps are aligned with Patient-Centred Care (PCC) as the care originates from patients who provide important information through mobile apps. A detailed discussion of pros and cons of using mHealth is provided by Agnihothri et al.[2,3].
Extant research on the benefits of using telemonitoring is based on randomized controlled trials which, by necessity, typically last only for a short duration [4–8]. For a patient-driven intervention such as mHealth, actual patient usage of technology would be a key driver for the magnitude of benefits. It is not clear if the results from randomized studies are generalizable, as actual patient behaviour may be quite different from that of a controlled setting . Hence, there is a need to investigate if the benefits of telemonitoring remain in a clinical practice setting if patients make their own decisions on how frequently or how long to use the app.
We evaluated improvement in BP at a hypertension clinic that uses an mHealth app for some of its patients. The physician (who is a co-author) is a practicing endocrinologist and has been using the proprietary mHealth app for over 8 years to treat patients with diabetes and hypertension. Patients opt in to use the app but whether the patient chooses to do so, and how often the patient uploads a reading, are solely decided by the patient. It should be noted that while some other apps can even measure BP, due to reliability concerns our app is only used to upload the measurement taken through validated devices.
Objective and hypotheses
Our primary hypothesis is that the mHealth app would improve the quality of hypertension care by managing BP (as measured in physician office). Our secondary hypothesis is that the improvement would be higher for patients with more severe hypertension.
We also analysed home blood pressure measurement (HBPM) data uploaded through the app. Using more granular HBPM data allowed us to characterize the type of patient behaviour that leads to better health outcomes.
MATERIALS AND METHODS
The physician co-author worked closely with a software developer and developed a robust, flexible, user-friendly, web-based, proprietary mHealth app. In its current form, the fully EMR-integrated app is being regularly used in clinical practice for over four years. The app was released at the end of 2014. Patients can install the app on their phone or computer, enabling them to upload readings.
mHealth app intervention: description and capabilities
The web-based app enabled patients to continuously communicate data on their vitals while the provider monitored, intervened and gave timely feedback. Patient cuff size was first determined and all patients owned a physician-recommended, cuff-based, BP measurement device. Support staff calibrated the device in the beginning and every 6 months. Patients were trained in using the device and were encouraged to take several readings in a row to get a better estimate. Abnormal readings were checked by a validation coordinator to ensure accuracy (i.e. she asked the patients to take multiple readings).
The app had a built-in decision support system and can be implemented on a large population of patients, was independent of operating systems, location and access to internet, communicated instantly with the provider to make immediate treatment modifications, if needed, allowed broadcasting of chats and connected providers in real time with patients, sent alerts to patients reminding them to enter vitals on time, kept track of patient history, demographics (smoking status, height, weight and so on), and archived data when needed, allowed multiple providers in the group to communicate instantaneously through one portal to create a single continuum of care model for the patients, allowed patients to request refills and medication changes, and sent a summary document automatically to the patient's electronic medical record so that patients can have a macro view of their readings. An example of what a patient sees is presented in the online supplement (Appendix Figure 1, https://links.lww.com/HJH/B671).
Whether or not the patient chooses to opt-in to the app, and how often the patient uploads a reading, was solely decided by the patient and 56% of patients managed by the clinic did not adopt the app. Of the patients who did, 55% adopted in 2014, 24% adopted in 2015 and 20% adopted in 2016. In total, more than 2200 patients have used the app.
We evaluated improvement in patient health, indicated by office-based BP measurement (OBPM), at the hypertension clinic. We had two groups of patients and we compared their OBPM against each other: the treatment group (adopters) which adopted the app, and the control group (nonadopters) which did not adopt the app. The dataset included 1633 patients who were tracked continuously between 2014 and 2016. We excluded patients who were with the practice in 2014 but not in 2016 and vice versa, as we cannot observe their change in OBPM over this time period. For each patient, we had information on their age, sex, race, BMI, smoking status, alcohol consumption status and SBP and DBP readings measured at the office. A summary of demographic information and summary statistics for SBP and DBP, for both the treatment and control groups, can be found in Table 1.
TABLE 1 -
Demographic information and average blood pressure in 2014 and 2016 for treatment and control groups
||Treatment group (n = 726)
||Control group (n = 907)
|No. (%) Malea
|Average (SD) agea
|No. (%) White
|No. (%) Black
|No. (%) Asian
|No. (%) Other Racea
|Average (SD) BMIa
|No. (%) nonsmokera
|No. (%) former smoker
|No. (%) light smoker (<10 per day)
|No. (%) heavy smokera (>10 per day)
|No. (%) smoking status unknown
|No. (%) nondrinker
|No. (%) former drinker
|No. (%) light drinker (<2 per day)
|No. (%) heavy drinker (>2 per day)
|No. (%) unknown drinker
|Average (SD) 2014 office systolic
|average (SD) 2016 office systolic
|average (SD) 2014 office diastolic
|average (SD) 2016 office diastolic
Column 1 and Column 2 present demographic information and average BP between 2014 and 2016 for treatment and control group (standard deviation within parentheses when applicable). Column 3 statistically tests for differences between the two groups: we test for either the difference in means for numerical variables, or difference in proportions for categorical variables. We report the resulting P-values.
aStatistically significant difference at the 5% level.
We used a difference-in-differences (DID) multivariable regression model to address endogeneity (including self-selection) problems present in observational studies [10–12]. For our DID identification strategy, we analysed the OBPM in 2014 (before anyone adopted the app) and 2016 (at least 1 year after adoption for most adopters). Although we had access to both OBPM and home BP measurements (HBPM) uploaded through the app, we used only OBPM data to keep the comparison between the app adopters and nonadopters reasonable. The HBPM uploads were indirectly accounted for, as the interventions by the physician were based on both HBPM and OBPM. Roughly speaking, DID gives an estimate of the treatment effect of using the app, compared with no app use at all.
We ran the DID multivariable regression using SBP as the dependent variable. An indicator to denote app adoption at least one calendar year before is used as an independent variable. We used both patient and year-fixed effects. This design allowed us to control for time-invariant patient characteristics, which may otherwise affect BP, such as sex, race, BMI, income level and education level (the latter two variables are not present in our dataset). We used two observations for each patient: one based on the patient's 2014 OBPM (averaged across all readings in 2014), and one on the patient's 2016 OBPM (averaged across all readings in 2016). We also repeated the analysis using DBP as the dependent variable.
We repeated the DID study restricting our attention to patients whose systolic (or diastolic) BP in 2014 exceeded some threshold. We used three different systolic thresholds (130, 140, and 150 mmHg) and two different diastolic thresholds (80 and 90 mmHg). We note that according to the American Heart Association guidelines, stage 1 hypertension is a SBP ranging from 130 to 139 mmHg or DBP ranging from 80 to 89 mmHg, and stage 2 hypertension is a SBP of 140 mmHg or higher or a DBP of 90 mmHg or higher.
Difference-in-differences robustness tests
We checked for the presence of individual and time effects using Gourieroux, Holly and Monfort (1982) test and F test . We also used the Hausman (1978) test to compare between a fixed effects model and a random effects model . To test the robustness of our results, we made two modifications to our DID regression: instead of using the SBP as the dependent variable, we used the log-transformed SBP, as there is a slight right skew in the distribution of SBP, and we set the dependent variable to be a dummy variable (1 if the SBP is 130 mmHg or above and 0 otherwise, to denote stage 1 hypertension or higher).
We checked the parallel trends assumption by examining the difference in BP from 2012 to 2013 between the adopters and the nonadopters (since no one adopted the app before 2014). To address false positives from DID models [12,15], we also performed a ‘placebo’ test (please see online supplement for more details, https://links.lww.com/HJH/B671).
As a final robustness test, we used matching techniques to preprocess our dataset before repeating our earlier DID analysis. Roughly speaking, matching removes patients from the dataset to ensure that there is minimal difference in some prespecified demographic variables between the treatment and control groups . We matched patients on their baseline BP (in 2014) in addition to other demographic variables (age, sex, BMI, race, smoking status and alcohol status), and then compared their BP in 2016. We used genetic matching method. We then repeated the earlier DID analysis using the matched dataset.
We used two-sided hypothesis testing and used R (version 3.6.2) for our analysis (including the MatchIt package to perform matching). We also used patient-clustered robust standard errors to estimate significance of our estimates.
mHealth app uploads analysis
Our app dataset consisted of 242 437 BP readings uploaded by 960 different adopters between 2014 and 2016. For each reading, we recorded the patient ID, date and time of the upload, and the SBP and DBP of that upload. We investigated if patients who were more engaged with the technology (i.e. higher use of the app) exhibit a larger improvement in their BP. To do so, we considered all the readings uploaded at least 1 year after adopting the app. We defined a new frequency measure, the percentage of weeks uploaded, associated with each reading. It is the number of weeks in which the patient uploaded their BP prior to that reading, divided by 52 (to consider the past one year).
We analysed the relationship between frequency and BP reading (systolic) after the patient had been using the app for at least a year. We ran a multivariable regression with BP reading as the dependent variable, percentage of weeks uploaded as the independent variable and patient-fixed effects to control for individual time-invariant characteristics. This allowed us to estimate the impact of consistent mHealth app use over a prolonged period of time.
We present the percentage of people who belonged to each of the three categories (Normotension and Elevated BP, Stage 1, and Stage 2 hypertension) in Table 2. Among the adopters whose 2014 BP belonged to Normotension (SBP < 120 mmHg and DBP < 80 mmHg) or elevated BP (SBP in the range of 120–129 mmHg and DBP < 80 mmHg), the percentage who belonged to Stage 2 hypertension (SBP ≥140 mmHg or DBP ≥90 mmHg) in 2016 was 7.9% (n = 252) as against 12.0% (n = 391) among nonadopters (the percentages were almost the same for Stage 1–22.2 vs. 22.5%). Among the adopters whose 2014 BP belonged to Stage 1 hypertension (SBP in the range of 130–139 mmHg or DBP in the range of 80–89 mmHg or both), the percentage who belonged to Stage 2 in 2016 was 25.4% (n = 260) as against 28.2% (n = 252) for nonadopters (the percentages were almost the same for Normotension or elevated-29.2 vs. 30.6%). Among the adopters whose 2014 BP belonged to Stage 2 hypertension (SBP ≥140 mmHg or DBP ≥90 mmHg), the percentage who improved their BP to Elevated or normotension in 2016 was 16.5% (n = 200) as against 10.2% (n = 264) among the nonadopters (the percentages were almost the same for Stage 1–26.0 vs. 26.1%).
TABLE 2 -
Overall Impact of mHealth
||Normo and Elevated
||Normotension and Elevated BP (systolic <130 and diastolic <80)
||Stage 1 Hypertension (systolic 130–139 or diastolic 80–89 or both)
||Stage 2 Hypertension (systolic ≥140 or diastolic ≥90 or both)
aTreatment group excluded 14 patients who adopted the app in 2017.
We present the estimates of our DID specification model in Column (1) and Column (5) of Table 3 for SBP and DBP, respectively. In the period from 2014 to 2016, patients in the treatment group reduced their SBP by an additional 1.89 mmHg [P = 0.005; 95% confidence interval (95% CI), 0.58–3.2] and their DBP by an additional 0.87 mmHg (P = 0.015; 95% CI, 0.17–1.57) compared with the patients in the control group. This difference was significant at the 5% level. The regression results when we vary the form of the dependent variable are reported in columns (1) and (2) of online supplement Table 1, https://links.lww.com/HJH/B671. In all the cases, the DID estimate is still negative and significant. The parallel trends assumption was satisfied and the ‘placebo’ test allayed concerns of false positives (see online supplement, https://links.lww.com/HJH/B671).
TABLE 3 -
Difference-in-differences specification for SBP and DBP from 2014 to 2016
||SBP < 130 (2)
||S BP≥150 (4)
||DBP < 80 (6)
||DBP ≥90 (8)
|mHealth adoption estimate
| 95% CI
|Deg. of freedom
|No. patients in control
|No. patients in treatment
Columns (1) and (5) gives DID regression estimates of mHealth adoption for SBP and DBP, respectively. Patient-clustered robust standard errors are in parentheses. Columns (2)-(4) and (6)-(7) selects only the patients whose SBP and DBP satisfied a certain threshold.
Our results on high severity patients are summarized in Table 3 (columns 2–4 and columns 7–8). Among patients whose 2014 SBP is less than 130 mmHg, the adopters exhibited a decrease of 1.06 mmHg (P = 0.2; 95% CI, -0.57 to 2.69) over the nonadopters. The DID estimate was higher and significant as we increased the threshold. For patients above 140 mmHg, the DID estimate was a 4.01 mmHg (P = 0.007; 95% CI, 1.11–6.91) drop for adopters over nonadopters, and for patients above 150 mmHg, the estimate was 6.23 mmHg (P = 0.02; 95% CI, 0.87–11.59). Thus, the improvement in Stage 2 hypertension patients was significant when compared with patients with SBP less than 130 mmHg. For patients above 90 mmHg diastolic, the DID estimate was a 4.37 mmHg (P = 0.01; 95% CI, 1.06–7.68) drop for adopters over nonadopters.
Results from matching analysis
Summary statistics of baseline measurements belonging to the treatment and control group are presented in Appendix Table 3, https://links.lww.com/HJH/B671 of our online supplement and the distribution of all covariates between the two groups both before and after matching is presented in Appendix Figure 2 of the online supplement, https://links.lww.com/HJH/B671.
Our regression results are reported in Appendix Table 4 of our online supplement, https://links.lww.com/HJH/B671. The treatment group reduced their SBP by an additional 1.65 mmHg (P = 0.02; 95% CI, 0.3–3.00) compared to the patients in the control group. The improvement in Stage 2 Hypertension patients (3.68 mmHg, P = 0.02, 95% CI 0.66–6.70) was again significant when compared with patients with SBP less than 130 mmHg (0.818 mmHg, P = 0.34, 95% CI −0.87 to 2.5). As our two groups are well balanced postmatching, it is also unlikely that regression to the mean significantly biases our results [16,17].
Results from mHealth app analysis
Our regression results are reported in Table 4. Our results show that heavier app use is associated with a reduced BP. A one-unit increase in percentage of weeks that had an upload (in the last 52 weeks) is associated with a reduction of 0.026 mmHg in SBP (P < 0.01, 95% CI 0.02–0.03). The drop was higher for sicker patients.
TABLE 4 -
The effect of increased app use on blood pressure
||Dependent variable: SBP (mmHg)
|Percentage of weeks uploaded
|No. of observations
This study assessed the association between using a physician-supervised mHealth app and improvement in BP in a hypertension clinic setting. Patients who adopted the app, compared with those who did not, reduced their SBP by 1.89 mmHg and DBP by 0.87 mmHg on average and stage 2 hypertension patients by 4.01 mmHg systolic and 4.37 mmHg diastolic on average. Milani et al. matched 156 digital-medicine patients with 400 usual care patients and found the mean decrease in BP (after 90 days) among the former group to be higher (14/5 mmHg in digital medicine vs. 4/2 mmHg in usual care). Tucker et al. conducted a systematic review of 25 randomized control trials (RCTs) and found that self-monitoring with clinician support reduced clinic SBP compared with usual care at 12 months by 6.1 (CI −9.0, −3.2) mmHg. McManus et al. conducted a randomized clinical trial with 393 participants and found that self-monitored BP with telemonitoring reduced clinic SBP compared with usual care at 12 months by 4.7 (CI −7.0, −2.4) mmHg.
The main issue in the extant medical literature is their focus on the main effect and the tendency to ignore actual usage in a clinical practice setting . Two popular ways to establish the link are RCTs and observational studies. We used an observational study because of the nature of the data. Alessa et al. conducted a systematic review of 21 studies with a total of 3112 participants (range of 19--1012 participants). Lu et al. conducted a systematic review of 11 RCT studies with a total of 4271 participants. Included in the review of Parati et al. were two large meta-analyses studies that included 69 RCTs with 20 912 cases in total. There are three main concerns in their design we compensated for in our study:
- (1) Methodology: Because of several limitations, most studies in the literature had an average sample size of around 150 patients , while our sample size was larger with 1633 patients. Even large studies in literature with thousands of patients, focused only on the main effects and called for formal studies to evaluate outcomes . Although we were not able to improve upon the blinding and randomization issues, we used sophisticated econometric techniques, such as DID, to account for lack of blinding. We also used matching techniques which are intended to simulate randomized experiments under certain assumptions.
- (2) Generalization of results: The RCT studies in the literature measured outcomes in a tightly controlled environment. The estimated treatment effect of mHealth app usage on health outcomes may not be applicable for mHealth use in general (e.g. in a private clinic setting). Yang et al. expressed concerns about trial protocols (e.g. requirements that may influence clinician-patient face time) confounding results from clinical trials. For instance, in the work of Logan et al., to measure the effect of telemonitoring on BP, ‘all of the eligible subjects were asked to monitor their BP at home daily for 7 days, taking 2 readings in the morning and 2 readings in the evening using a validated Bluetooth-enabled home BP device’. It is unclear whether the treatment effect described here applies to mHealth use in general, because if patients were not part of the experiment, they may go weeks, or even months, between readings. In contrast, the physician in our study let the patients decide how often and when they use the app. Thus, our dataset reflects what one might observe in a noncontrolled setting, allowing for individual variations in actual usage. The few observational studies reported do not use sophisticated methodology to control for self-selection issues .
- (3) Duration: Our study tracked patient measurements for a duration of 3 years, compared with an average duration of around 6 months or a maximum of 12 months in other studies . Parati et al. in their review article call for large, long-term studies before mobile applications can be more widely adopted in practice.
Using an observational study, we clearly established a significant positive association between adopters using the mHealth app and improvement in BP. A reduction in SBP has been shown to reduce major health risks and an intensive BP management is found to be very cost-effective, having a cost of $23 777 per QALY gained . Our result withstood several robustness tests. We also matched the two groups based on age, sex and other patient-individual characteristics.
Second, the reduction in BP (over 2 years) was higher for sicker patients who adopted the app (4.01 mmHg drop in SBP for patients with stage 2 hypertension vs. 1.1 mmHg drop for patients with SBP < 130 mmHg). The higher reduction with app usage among sicker patients may be due to several reasons: patients with high BP may have poor diet and exercise habits, and may be motivated to change them after reviewing them daily on the app, and patients with high BP may benefit more from frequent communication with the doctor.
Third, higher upload frequency was associated with greater reduction in BP. If a patient uploads their readings periodically into the app, that patient enables more frequent monitoring and enables more opportunities to receive communication from the physician and his support staff. Patient engagement driving potential interventions may play an important role in explaining why the app is able to improve their health.
The app improved information flow and provided more opportunities for the provider to intervene. The app filled-in for the sporadic office visits and provided a more complete picture of the patient profile since the previous office visit. Currently (as of 2021), the app can also track and visually present mean arterial BP, mean SBP and DBP and pulse with Kalman filter and age adjuster. In order to detect variability, the app can present longitudinal information on pulse variability, stiffening index and pulse stiffening ratio. The usefulness of such measures can be explored in future research. Patient engagement and chances for potential interventions possibly explain why the app benefits patients. BP measurements through mHealth can complement OBPM and enable personalized interventions, contributing to a more precise medicine .
One of the barriers to wider adoption of mHealth apps by providers is the uncertainty around operational implications of using such an app in practice; we showed that there is significant benefit to both patients and physicians, a fact that when publicized, may serve to mitigate this barrier. Our preliminary data indicated that most of the interventions (97% during a 3-month study period) can be handled by dietitians and medical assistants. Hence, the responsibility of patient monitoring shifts from the physician to the support staff. Therefore, redesigning the care delivery process using mHealth technology can potentially reduce the cost of treating chronic patients and increase physician capacity. Further, easy-to-use apps can help underserved populations, particularly low-income families, and adults over 65 years of age.
Technology can be used to reduce asymmetry and improve coordination, deliver patient-centred care, improve quality of care and reduce healthcare cost. Dishman  identified the three pillars of personal health to be ‘care anywhere, care customization, and care networking’. The concept of care anywhere represents the shift from institutions to mobile, home-based and community-based care. In this context, care must occur at home as a default model unless the patient is extremely sick. Care customization pertains to the shift from population-based to patient-centred treatment (with the help of predictive models for individuals considering patient preference), thus achieving mass-customization of care. Care networking includes the technology infrastructure, business models and organizational models that allow care to be shifted from solo to team-based practice, along with the information technology systems that connect all these people and devices. mHealth technology therefore helps in reducing information asymmetry, improves coordination between the patient and the providers, and improves patient engagement.
Our research focused on investigating the benefits of this specific app in a clinical practice without focusing on a specific functionality of the app. Although we did establish a significant association, we did not establish which of the different factors in the app benefit patients. Our anecdotal evidence strongly points to provider-supervision based on extended longitudinal data as the main driver behind the benefit of mHealth apps. Empirically proving the effect of supervision is however beyond the scope of the data we had access to. Our estimates on the overall improvement (population average) may be lower, as not all adopters upload their readings regularly.
Using mHealth in a private clinical practice can improve patient-provider coordination, enable customized care and could be instrumental in managing patient BP. In this observational study of patients seen at a hypertension clinic, use of a physician-supervised mHealth app was significantly associated with improved health outcomes (reduced BP). The improvement was significant for high-severity patients. Therefore, the use of an mHealth app enables a healthcare system to be proactive (especially for a silent killer such as hypertension) and can significantly improve quality of care.
We would like to thank the two reviewers for their constructive comments. We also would like to acknowledge the staff of the Diabetic Care Associates for their unconditional support.
Conflicts of interest
There are no conflicts of interest.
1. Kumar S, Nilsen WJ, Abernethy A, Atienza A, Patrick K, Pavel M, et al. Mobile health technology evaluation: the mHealth evidence workshop. Am J Prev Med
2. Agnihothri S, Agnihothri R. Application of evidence-based management to chronic disease healthcare: a framework. Manag Decis
3. Agnihothri S, Cui L, Delasay M, Rajan B. The value of mHealth for managing chronic conditions. Healthcare Manag Sci
4. Alessa T, Abdi S, Hawley MS, de Witte L. Mobile apps to support the self-management of hypertension: systematic review of effectiveness, usability, and user satisfaction. JMIR mHealth uHealth
5. Margolis KL, Asche SE, Bergdall AR, Dehmer SP, Groen SE, Kadrmas HM, et al. Effect of home blood pressure telemonitoring and pharmacist management on blood pressure control: a cluster randomized clinical trial. J Am Med Assoc
6. McLean G, Band R, Saunderson K, Hanlon P, Murray E, Little P, et al. Digital interventions to promote self-management in adults with hypertension systematic review and meta-analysis. J Hypertens
7. Lu X, Yang H, Xia X, Lu X, Lin J, Liu F, Gu D. Interactive mobile health intervention and blood pressure management in adults: a meta-analysis of randomized controlled trials. Hypertension
8. Choi W, Wang S, Lee Y, Oh H, Zheng Z. A systematic review of mobile health technologies to support self-management of concurrent diabetes and hypertension. J Am Med Informatics Assoc
9. Shaw RJ, Steinberg DM, Bonnet J, Modarai F, George A, Cunningham T, et al. Mobile health devices: will patients actually use them? J Am Med Informatics Assoc
10. Angrist JD, Pischke J-S. Mostly harmless econometrics: an empiricist's companion. Princeton, NJ: Princeton University Press; 2008.
11. Wooldridge JM. Econometric analysis of cross section and panel data. Cambridge, MA: MIT Press; 2010.
12. Pierce L, Snow DC, McAfee A, Devaraj S, Ow TT, Kohli R. Crossing the quality chasm: a new health system for the 21st century. J Oper Manag
13. Gourieroux C, Holly A, Monfort A. Likelihood ratio test, Wald test, and Kuhn-Tucker test in linear models with inequality constraints on the regression parameters. Econometrica
14. Elsevier, Hausman JA. Chapter 7 Specification and estimation of simultaneous equation models. Handbook of Econometrics. vol. 1. 1983; 391–448.
15. Song H, Tucker AL, Murrell KL, Vinson DR. Closing the productivity gap: improving worker productivity through public relative performance feedback and validation of best practices. Manage Sci
16. Stuart EA. Matching methods for causal inference: a review and a look forward. Stat Sci
17. Linden A. Assessing regression to the mean effects in healthcare initiatives. BMC Med Res Methodol
18. Milani RV, Lavie CJ, Bober RM, Milani AR, Ventura HO. Improving hypertension control and patient engagement using digital tools. Am J Med
19. Tucker KL, Sheppard JP, Stevens R, Bosworth HB, Bove A, Bray EP, et al. Self-monitoring of blood pressure in hypertension: a systematic review and individual patient data meta-analysis. PLOS Med
20. McManus R, Mant J, Franssen M A. N.-T. Lancet, and undefined 2018. Efficacy of self-monitored blood pressure, with or without telemonitoring, for titration of antihypertensive medication (TASMINH4): an unmasked randomised. https://www.sciencedirect.com/science/article/pii/S014067361830309X
. [Accessed 27 February 2021]
21. Yue M, Ho Yu C, Li C, Janet SW. The effectiveness of electronic health interventions on blood pressure control, self-care behavioural outcomes and psychosocial well being in patients with hypertension: a systematic review and meta-analysis. Int J Nurs Stud
22. Parati G, Dolan E, McManus RJ, Omboni S. Home blood pressure telemonitoring in the 21st century. J Clin Hypertens
23. Darkins A, Ryan P, Kobb R, Foster L, Edmonson E, Wakefield B, Lancaster AE. Care Coordination/Home Telehealth: the systematic implementation of health informatics, home telehealth, and disease management to support the care of veteran patients with chronic conditions, Telemed. e-Health
24. Yang WE, Shah LM, Spaulding EM, Wang J, Xun H, Weng D, et al. The role of a clinician amid the rise of mobile health technology. J Am Med Inform Assoc
25. Logan AG, Irvine MJ, McIsaac WJ, Tisler A, Rossos PG, Easty A, et al. Effect of home blood pressure telemonitoring with self-care support on uncontrolled systolic hypertension in diabetics. Hypertension
26. Parati G, Torlasco C, Omboni S, Pellegrini D. Smartphone applications for hypertension management: a potential game-changer that needs more control. Curr Hypertens Rep
27. Ettehad D, Emdin CA, Kiran A, Anderson SG, Callender T, Emberson J, et al. Blood pressure lowering for prevention of cardiovascular disease and death: a systematic review and meta-analysis. Lancet
28. Parati G, Ochoa JE, Bilo G. Moving beyond office blood pressure to achieve a personalized and more precise hypertension management: which way to go? Hypertension
29. Dishman E. Toward personal health: going home and beyond. Washington (DC): National Academies Press (US); 2015.