Introduction

The importance of hypertension (HTN) as a risk factor for cardiovascular diseases, which cause the highest mortality, is well known. In a meta-analysis with individual data from one million adults [13], they found an association of Blood Pressure (BP) with stroke, ischemic heart disease, and other vascular diseases.

According to the Global Burden of Disease Study, hypertension is the number one risk factor, responsible for the deaths of 10.8 million people worldwide in 2019, almost 20% of all deaths [16]. The treatment of HTN has shown a benefit in reducing cardiovascular risk [18]. However, it is well known that only half of hypertensive patients have their numbers under control [9]. BP measurements taken outside the medical centre, either with automated devices (ambulatory BP monitoring -ABPM-) or home measurements taken by the patient at home (home BP monitoring -HBPM-) show a stronger association with cardiovascular disease risk than measurement by the doctor or nurse in the doctor’s office [15]. Indeed, it has been shown that the use of telemonitoring of BP through home measurements can improve HTN control and reduce the numbers compared to conventional treatment [6]. It is therefore of great interest to improve the BP control at home of hypertensive patients.

Patients with high BP are often asked to measure their BP values at home. To carry out this task, patients must fill in a table, manually, to record all data (date, time, systolic BP - top number-, diastolic BP - bottom number) and remember to bring the data record to the next medical appointment. Some problems arise from this procedure. For example, some patients forget to check their BP every day, or do not do so at the stipulated times. Other patients do not write down the correct values or forget to bring the data record to the doctor’s appointment. In addition, the doctor and nurses do not know the current BP values until the day of the medical appointment. Besides, when patients bring their notes to their medical appointments, the doctor has to transcribe them all into a digital file. This is a time-consuming and error-prone task.

Here, we propose the development of TensioBot, a Telegram chatbot that can alleviate those problems. A chatbot is a computer program that can simulate conversations, acting as a human being. Our work aims at answering the following research questions:

  • RQ1) Is it feasible to develop a chatbot that helps patients with high blood pressure to check their BP at home?

  • RQ2) If so, how effective is the TensioBot in terms of BP checking (number of successful measurements, number of missed measurements)?

  • RQ3) Does TensioBot improve knowledge on best practices related to self-managed BP measurement procedures?

  • RQ4) What are the advantages of using TensioBot to record BP measurements compared to the traditional method of writing them down on paper?

We hypothesise that patients can use TensioBot without problems, because they are familiar with the instant messaging app Telegram (or alternatively, WhatsApp). Chatbots like TensioBot could ease the workload of an already overburdened medical system, while preventing errors and promoting more accurate and complete recording of patients’ blood pressure values.

Our hypothesis has been tested and evaluated in a randomized controlled clinical trial with 112 patients (Section 3). We start by framing our work within the area of mHealth interventions.

Helping patients to control their BP: A review of mobile based solutions

There have been some attempts to use technology to solve the aforementioned problems and improve patient adherence to the BP monitoring procedure. An interesting study is that of Vargas et al., who found that SMS text messaging between physicians and patients was effective in aiding BP control, especially in racial minority populations, as two-way and personalized communication was achieved [21]. However, SMS messages are quite limited as they can not deliver multimedia content like videos, graphics or audio messages and, more importantly, can not be used for interacting with the patient. With the aim of overcoming these limitations, efforts have been directed in recent years to develop smartphone apps to support self-management of hypertension. In this line of work, Alessa et al. [1] reviewed 186 mobile apps for BP self-management, and found that, in general, these apps support BP tracking and self-monitoring of medication-related data. While some of the apps allowed BP readings to be received automatically from the BP measuring device, in general, manual data entry was required.

Providing treatments through mobile apps is an interesting approach but the effort could be hindered by the difficulty to install the app and instruct the user about how to use the interface. These user-experience problems could be aggravated when the target audience of the app happens to be elderly people. That is the case for HTN patients. Our approach taps into using a familiar user-interface, Instant Messaging (IM) apps’ interface, where all the user has to do is to chat with a chatbot. Chatbots for mobile health interventions are on the rise. They have been successfully used in different health areas to promote self-management of chronic pain [3, 12], neurological disorders and psychiatric conditions [20], like dementia [5] or Alzheimer [11], nutritional-metabolic-disorders like obesity [19] or diabetes [4], and addictions like smoking [2, 7] or alcoholism [8]. However, as far as we know, mobile chatbots have not been used yet for helping patients with hypertension to track and monitor their BP.

Drawing on the potential and popularity of WeChat, an IM application hugely used in China with more than 1.17 billion users in Q4 2019, Li et al. [14] describe an online intervention where patients with HTN meet in a chat group. Participants were periodically sent articles of health knowledge related to their condition, and offered targeted health advice in a chat. As with TensioBot, the aim is to leverage the popularity of IM systems to improve BP and self-management behavior in patients, early detecting HTN complications. Yet, our strategy differs from Li et al’s. Instead of resorting to doctors or researchers to chat with the patients, we target a fully automated intervention using chatbots as a proxy. As far as we know, there is no other study in the realm of self-checked BP that followed this chatbot approach before.

Methods

Overview

A 2-arm, randomized, controlled trial of an intervention based on the TensioBot mobile application was carried out over 2 years (2018–2020). This clinical trial has been approved by the Clinical Research Ethics Committee of the BioAraba Health Research Institute, with approval number/ID 2017–031 and PI LE.

Subjects and recruitment

We recruited 112 patients that needed to check their BP daily, twice, 7 days before the medical appointment. Half of them were randomly assigned to the bot intervention group and the other to the control group. Near 48% of all the patients had some kind of chronic kidney disease. Table 1 summarizes the characteristics of participants.

Table 1 Description of participants
Full size table

Inclusion criteria

  • Patients attending the Nephrology outpatient clinic (CCEE) of the Araba University Hospital (HUA), with a diagnosis of hypertension or suspected primary or secondary hypertension. They must be at least 18 years old and have been previously instructed to perform BP checks at home with an information sheet available at the outpatient hospital.

  • Patients must have an approved and calibrated tensiometer and a smartphone with Internet connection at home, and sufficient ability to send messages via WhatsApp or Telegram.

  • Patients agreed to participate in the study by signing an informed consent form.

Exclusion criteria

Patients were excluded on the basis of these criteria:

  • Previous diagnosis of severe psychiatric disorder.

  • Severe illness suggesting a life expectancy of less than 6 months or high probability of needing haemodialysis or transplantation in less than 3 months.

  • Severe clotting disorders that can cause bruising if frequent BP measurements are performed.

  • Motor or visual disability that makes it difficult to perform self-measurements at home.

TensioBot intervention description

Patients in the intervention group were asked to contact a chatbot called TensioBot on Telegram. We will start describing Telegram and then define the TensioBot commands. Figure 1 shows some screenshots of a session with TensioBot.

Fig. 1
figure 1

TensioBot is a Telegram based chatbot that can be used from smartphones or desktop computers. The user can type any available command or answer to chatbot’s questions. It is also possible to click on any button shown by the bot. From left to right: the user informs the bot about BP numbers, then clicks on /Video command and finally on /Chart button

Full size image

Telegram platform

Telegram is a popular IM application -more than 500 million active users-. In 2014, Telegram was the first IM provider to introduce the ability to develop and interact with chatbots. Chatbots can respond to text or voice commands from the user, display links, images or videos. Users can start a conversation with a bot by searching for its name as if it were a human contact in their contact list. Users can also click on a direct link to automatically open a conversation with a specific bot.

How TensioBot works

One week before the doctor’s appointment, TensioBot asks patients to measure their BP twice a day, usually once in the morning and once in the evening (the alert times can be edited at any time). When the user receives the alert, he/she should proceed to use the tensiometer and write down on a paper the BP values (highest, lowest) answering the questions posed by the chatbot. It then informs patients to wait two minutes and proceed again for a second measurement. After the second set of measurements, TensioBot shows a line chart with the entire BP history. If after the two-minute waiting period the patient enters a BP value higher than 5 mmHg compared to the values of the first measurement, the bot prompts the patient to perform a third measurement 2 min later.

If a patient enters a BP value that is outside normal levels, the bot asks for confirmation. If it was a typing error by the user, it asks the patient to check the measurement again, and if the number was indeed correctly introduced, the bot notifies the doctor. All the pressure-numbers checked during the last seven days are reflected in a line chart that shows the evolution of the data (third picture in Fig. 1).

TensioBot also offers an option to display a helpful video on good BP measurement practices showing tips on how and when to take it, how to adjust the tensiometer, how to adjust the body position, etc. Once a day the bot also sends a message with other tips related to good BP measurement practices. Finally, the bot offers a command to remind the patient of the scheduled time for the next medical appointment, allowing the patient to change or cancel it directly through a conversation with the chatbot.

Administration commands

All of the above functions are available to any regular user (patients), but TensioBot also aims to help the physician with the management of those patients by offering a couple of administration commands, one for downloading a calc sheet with all the blood pressures for each patient and day and another for showing just the data for a specific patient (passed as argument to the command). Table 2 summarizes all the available commands.

Table 2 List and description of available commands in TensioBot. Some of them are only available for the doctor or nurse
Full size table

Protocol

Each control patient attending their first medical appointment receives a written procedure on how to self-monitor their BP. The nurse assesses the patients’ knowledge and skills on BP self-monitoring using a checklist and ensures that the patient knows when their next medical appointment is and when to perform BP measurements with a Holter device. For patients in the intervention group, the procedure is quite similar, except that they do not receive the information document on the procedure. In this case, the nurse also helps the user to download and install the Telegram app, if the patient does not already have it installed, and start using TensioBot, registering the user with the bot.

At the second visit, patients in both groups (control and intervention) undergo a knowledge and skills check of the BP self-monitoring procedure. We compare the results between the groups. Besides, we statistically analyze patient adherence with regard to the number of correct BP checks in both groups and the results of a survey administered to the intervention group after the second medical visit.

Results

Knowledge and skills about BP checking

We obtained 88 cases for patients who completed both visits. Of these, we eliminated two cases that had a 100% checklist score during their first visit (and therefore their scores could not improve at the second visit).

We wanted to analyze the results of the checklist in relation to patients’ knowledge and skills regarding home BP measurement technique. For each question answered, we gave a score of 1 for those answered correctly and a score of 0 for those answered incorrectly. We summed all the questions and divided by the maximum value, expressing the final value as a percentage.

Then we obtained two columns, one representing the knowledge gain for each set of control patients and a similar one for the set of intervention patients. We wanted to know if the differences in knowledge gains between the two groups were significant, so we applied a t-test, obtaining a t = 2.1159, df = 82.34, p value = 0.03737, 95% confidence interval (0.3915, 12.679) and mean in the Bot group = 24.126, mean in the control group = 17.591.

The p value associated with the test is 0.0374, so we can reject the null hypothesis (퐻0) of no difference between the (true) mean, suggesting that the difference in acquired knowledge of both groups is significant (favouring patients from the bot group). We used a non-parametric test, Wilcoxon rank sum test, to compare the ranks between Bot and Paper groups and the difference was also significant, W = 1159.5, p value = 0.04031.

Effectiveness of TensioBot with regard to the BP checking in-site

In the second visit, a nurse helped each patient to put on a Holter device (ABPM) to check BP during 24 h, and we discarded measurements during sleep to calculate the daytime means of systolic and diastolic BP (it is well known that BP values are lower during sleep). Considering the Holter values as the gold standard, we contrasted them with the BP checking values of the control and intervention groups (mean difference, for systolic and diastolic values).

We applied a two-sample t-test to the difference of the means between BP (systolic and diastolic values) in Holter-Control vs Holter-Bot, yielding values of Table 3. There was a smaller difference for the TensioBot group, both in systolic and diastolic values, but not significant (Fig. 2). The difference and scattering of values can be better seen on Figs. 3 and 4.

Table 3 Two sample t-test and Wilcoxon test values (measuring difference of the means between BP, systolic and diastolic, in Holter-Control vs Holter-Bot)
Full size table
Fig. 2
figure 2

Shows a boxplot with the differences (in mmHg) between the mean diurnal systolic (left) and diastolic (right) BP values with a Holter device (ABPM) and the mean self-managed systolic BP values at home (HBPM), for the intervention and control groups. We can see a more scattered boxplot for the control group, and a negative difference (in contrast to a difference close to 0 in the bot group), but again not statistically significant

Full size image
Fig. 3
figure 3

a Bland Altman chart, shows the mean difference between Holter values and self-managed systolic BP checks on the Y axis and the mean of both values on the X axis, colored by control and intervention groups. We can see three outliers, two for the control group and one for the bot group

Full size image
Fig. 4
figure 4

shows a similar chart, focusing on diastolic BP. The results are quite similar, but with less variation because the diastolic values are lower than the systolic ones

Full size image

Results of the satisfaction survey administered to TensioBot users

There were 4 questions in the survey, answered by n = 40 patients from the intervention group: 1) “Do you think that TensioBot is easy to use?”; 2) “Do you think that TensioBot is useful to help you better record your blood pressure values at home?”; 3) “How do you prefer to register your BP?” and 4) “Have you stopped using TensioBot?”

  1. 1)

    Do you think that TensioBot is easy to use?

92.5% of those surveyed patients think it is quite easy or very easy to use. Yet, there are 3 patients who think it is quite difficult

Very easy n = 23, Pretty easy n = 14, Quite difficult n = 3.

  1. 2)

    Do you think that TensioBot is useful to help you better record your BP values at home?

72.5% of respondents think it is very useful to use / all respondents think it is quite useful or very useful

Very useful n = 29. Quite useful n = 11.

  1. 3)

    How do you prefer to register your BP?

92.5% indicate that they prefer to register their BP at home using TensioBot instead of doing the registration on paper

Prefer to use the bot n = 37.

Prefer to write the numbers down on a paper n = 3 (despite the fact that some of these 3 people answered that the bot is useful and easy to use).

  1. 4)

    Have you stopped using TensioBot?

Only 6 people (15%) stopped using TensioBot. The main reason patients give is that they have difficulties in using a mobile phone.

Discussion

With the data at hand, we are now in a good position to answer our original research questions.

  • RQ1) Is it feasible to develop a chatbot that helps patients with high BP to check their BP at home?

More than 55 patients (mean age = 50.2) successfully participated in this experiment and a large majority (92%) stated that the bot was easy to use and useful (72.5%). Among the respondents, 85% continued to use the bot even after the experiment was over. Considering that patients with HTN are, in general, not young people (including in our study patients as old as 80 years of age), where this kind of smartphone-based intervention might be less accepted, these results pave the way for using chatbots for assisting in self-managed BP monitoring procedures at home, easing the burden on healthcare system, which are overwhelmed by this and other procedures, specially in our current pandemic context. This finding aligns with the current positive literature on how to leverage chatbots for healthcare interventions [10]. As far as we know this is the first time that a chatbot is used to help in self-managed BP checking procedures.

  • RQ2) If so, how effective is the TensioBot in terms of BP checking (number of successful measurements, number of missed measurements)?

We have found that the differences in both groups (intervention and control) between their BP values at home (HBPM) and those obtained using a Holter device (ABPM) were not significantly different. Thus, this means that chatbot users are just as good at checking their BP as regular users (using paper and pencil). But what about our initial hypothesis that chatbot reminders might improve treatment adherence? The data show that there is no significant difference in the number of BP checkings between the Bot and the Paper group (mean = 12.0 and 13.4, respectively, t-test t = 1.61, df = 80, p value = 0.109). Therefore, both methods are equivalent in terms of treatment adherence.

This contrasts with previous work in which the application of behaviour change techniques for health improvement using chatbots seemed to obtain better results than ours [17].

However, both methods being equal in terms of treatment adherence, chatbots offer many other advantages over the usual pen-and-pencil way of BP tracking, as we will see in the following RQs.

  • RQ3) Does TensioBot improve knowledge on best practices related to self-managed BP measurement procedures?

Reminders and advice regularly sent to patients by the chatbot, along with the opportunity to review a video about best BP measurement practices, appear to have had a positive impact on the knowledge and skills acquired in the intervention group. Although small, the mean differences in checklist scores (between the first and second visit) appear to favor the bot group (Bot group mean = 24.126, control group mean = 17.591, differences are significant, p = 0.037).

  • RQ4) What are the advantages of using TensioBot to record BP measurements compared to the traditional method of writing them down on paper?

Having the option of using a chatbot to help patients perform a self-managed BP monitoring procedure has many advantages over the usual pen-and-pencil procedure, not only for the patient, but also for healthcare professionals.

The main advantages for the patient are that he/she spends less time in the health centre, something that is especially interesting nowadays due to the COVID pandemic situation. On the other hand, it prevents patients from forgetting to bring the record sheet on the day of the medical appointment, as the data is automatically stored in the bot (in fact, the BP data history is kept for two years). In addition, the bot sends you appointment reminders so that you do not forget to go to the health centre on the appointed day.

Moreover, the bot reminds the patients when to take BP measurements, thus preventing them from forgetting to check it or being delayed in the procedure. It also periodically reminds them of the correct way to check their BP. Finally, if they make a mistake in recording the data, these are automatically corrected. A major advantage of this mobile application is that, due to the integration of the bot with a popular IM application (Telegram), it is usually not necessary to train users, because they are already familiar with the interface. Moreover, being on Telegram, users of our bot do not need to provide their phone number to start using it, which could be a crucial aspect for privacy in a mHealth intervention.

Regarding the benefits for the healthcare professionals, the doctor/nurse does not have to enter the data by hand, saving time and avoiding transcription errors, as well as being informed in real time of possible out-of-range BP values, without waiting for the next appointment. Such alerts can be customized so as not to overwhelm the system (specifying threshold BP values). All these advantages are achieved thanks to an inexpensive and immediately usable solution, as the source code has been provided under a GPLv3 license, so any interested practitioner could use and adapt it to their needs.

Limitations

The main concern is internal validity. Our setting is that of patients with HTN who were asked if they were willing to participate in the TensioBot experiment. This could lead to a situation where patients invest more effort collecting the data correctly than they would do otherwise. However, as this is a health-related intervention, we could not assign a person to the intervention group directly without confirmation.

As for external validity, our setting is that of a country where IM applications such as WhatsApp or Telegram are quite popular among the population. However, that might not be the case in other countries.

TensioBot uses Telegram as its underlying platform owing to its rich chatbot-development API. Nevertheless, it is not the only one where our bot could be deployed. Facebook Messenger, WeChat or Line are other popular IM platforms that offer chatbot features (public API, multimedia support and the ability to interact with the bot by clicking on buttons) compatible with our needs.

As an alternative option it would also be possible to implement a less user-friendly chatbot (without the option to interact via button navigation) by leveraging the WhatsApp Business API, a private automation chatbot development API offered for medium and large enterprises. Unlike Telegram, WhatsApp users need to share their phone numbers to interact with the app. These limitations led us to rule out WhatsApp as the underlying platform for TensioBot.

Conclusions and future work

Based on previous works within the mHealth field, this paper proposes a chatbot to help hypertensive patients to correctly check their BP at home. Strategies include: 1) alerts and notifications; 2) video and text-based advice on good BP measurement practices; 3) reminders and rescheduling of upcoming medical appointments; 4) notifications of data recording errors; 5) graphical tracking of BP values.

For patients, chatbots might help them to keep track of their BP, remind them when to check it, and keep them in contact with the doctor, while improving their knowledge about BP checking best practices. For doctors and nurses, chatbots provide the opportunity to better track patients’ BP numbers, freeing them from routine paperwork (translating data from paper to a digital form), and keeping them informed when noticeable BP numbers occur. We could leverage natural language processing techniques for improving on this area.

Although the investigated intervention seems to yield equal results in both groups (intervention and control) regarding adherence to the procedure (BP check at a fixed schedule), TensioBot users scored better on a subsequent test of knowledge and skills related to self-managed BP, meaning that they improve their technical skills to perform a more correct measurement. In addition, the introduction of chatbots to alleviate paperwork for doctors and nurses could be a great advantage in a currently overburdened healthcare system.

As a future line, we plan to study the advantages of integrating our bot with digital devices already in use for BP monitoring, freeing the patient from the need to manually enter numbers into the bot. Finally, we would like to devise ways to adapt our chatbot to help patients with other conditions to better follow their treatments. In fact, we have already adapted our bot to help pregnant women keep track of their BP. This bot, which we have named GestaBot, is already being tested in a clinical trial at the Araba University Hospital (HUA), approved by the Clinical Research Ethics Committee of the BioAraba Health Research Institute (with approval number/ID2019–002 and PI LE).