Abstract
Aim
Hypertension remains poorly controlled among older adults because self-care support between outpatient visits is rarely sustained. This study evaluated the effectiveness of a chatbot-based hypertension management program on knowledge, health behaviors, medication adherence, blood pressure control, and selected biomarkers among community-dwelling older adults with hypertension.
Design
Quasi-experimental study.
Methods
A two-group pretest–posttest design was conducted in two healthcare settings: the experimental group was recruited from a tertiary hospital outpatient department and the control group from a primary care health promoting hospital. Sixty-eight participants were recruited using consecutive sampling and assigned to an experimental group (n = 34) or a control group (n = 34). The experimental group received a 12-week chatbot-based program combining education, behavioral monitoring, and lifestyle counseling. The control group received usual care using a hypertension education manual. Outcomes were assessed at baseline and after 12 weeks.
Results
The experimental group showed improvements in hypertension-related knowledge, health behaviors, medication adherence, and blood pressure control. The control group showed no improvement in systolic and a rise in diastolic blood pressure. After adjustment for baseline values and for the demographic characteristics that differed at baseline, the experimental group demonstrated better knowledge (p = 0.002), health behaviors (p = 0.018), medication adherence (p< 0.001) and blood pressure control than the control group (adjusted systolic 127.43 compared to 135.80 mmHg, p = 0.009; adjusted diastolic 69.61 compared to 80.36 mmHg, p< 0.001). No significant adjusted between-group differences in biomarkers were observed.
Conclusion
A chatbot-supported hypertension management program was associated with better self-management and blood pressure control among community-dwelling older adults, although the non-randomized design and recruitment from different settings mean that the findings are preliminary. Longer follow-up is needed to determine effects on biological indicators. Randomized designs are recommended in future studies to strengthen the generalizability of the findings.
Keywords: hypertension, chatbot, community-dwelling, older adult, self-care, self-monitoring
Introduction
Hypertension is a common chronic condition defined by persistently elevated blood pressure exceeding 140/90 mmHg and is a major contributor to cardiovascular morbidity and mortality if inadequately controlled.1 Age-related vascular changes, particularly increased arterial stiffness due to the elastin–collagen transition, along with behavioral changes, lead to higher systolic pressure and greater risks of complications among older adults, including stroke, heart disease, and renal impairment.2–4 While behavior changes, drug adherence, disease management, and blood pressure control and monitoring are essential for preventing future complications and adverse events.2
Globally, hypertension affects approximately 1.28 billion adults aged 30–79 years, with prevalence doubling between 1990 and 2019, especially in low- and middle-income countries.5,6 Sex- and age-specific trends indicate higher prevalence among men under 50 years, while postmenopausal women exhibit increasing systolic blood pressure and higher prevalence after age 65.7,8 In Thailand, although awareness and treatment rates have improved over the past decade, only about 30% of hypertensive adults achieve blood pressure control, far below the expected 60%, resulting in substantial preventable morbidity and mortality.2,5,9
Thai national data further demonstrate a high burden of hypertension and circulatory diseases, particularly in the Northeastern region of Thailand, with a continuously rising prevalence among older adults.10 Hypertension-related complications, including renal, cardiovascular, and cerebrovascular conditions, remain common and are strongly associated with modifiable and non-modifiable risk factors, underscoring the importance of monitoring biomarkers such as lipid profiles, fasting blood glucose, renal function, and electrolytes to monitor disease progression and control.11
Older adults, typically defined as individuals aged 60 years and above, experience age-related functional decline that increases vulnerability to frailty and chronic disease complications.10 The cross-linking theory of aging explains progressive vascular stiffening and functional impairment, necessitating more effective and sustained hypertension management strategies tailored to this population10 However, sensory and functional limitations may hinder older adult’s ability to engage with conventional mobile health interventions.
Multicomponent interventions combining screening, education, counseling, monitoring, and lifestyle modification are most effective for blood pressure reduction.11,12 Technology-assisted approaches such as mobile applications and wearable devices enhance self-monitoring, medication adherence, and real-time feedback, yet most interventions remain designed for younger populations and rely on fragmented, one-way communication in clinical settings.13–15
Chatbots offer a promising solution by enabling continuous two-way communication, personalized education, self-monitoring, and behavioral reinforcement within a familiar digital environment. In Thailand, the widespread use of the LINE application among older adults supports its feasibility as a chatbot platform for hypertension management.16 When integrated with evidence-based self-management strategies such as DASH diet, physical activity, stress reduction, blood pressure monitoring, and avoidance of tobacco and alcohol that chatbots can address limitations of traditional care while reducing healthcare workload.17,18
Health behavior modification is central to sustainable hypertension control, particularly among community-dwelling older adults.19,20 Chatbot-based interventions align well with self-regulation and behavior change theories by supporting goal setting, self-monitoring, feedback, and reinforcement within a therapeutic framework.20–22 Despite challenges related to digital literacy and privacy concerns, these barriers can be mitigated through user-centered design, appropriate content delivery, and enhancement of electronic health literacy.23
Evidence on conversational agents for hypertension nevertheless remains limited and has accumulated mainly in other settings and populations. Reported systems have typically been developed for general adult samples, evaluated over short periods, and focused on a single function such as blood pressure logging or usability testing rather than on a coordinated set of self-management behaviors.17,23 Few have been designed specifically for older adults, delivered through a messaging application that this age group already uses in daily life, or evaluated against laboratory as well as behavioral and clinical outcomes. The present study addresses that gap in four respects including it targets community-dwelling Thai older adults rather than general adult populations; it delivers the program through LINE, the platform already dominant among older adults in Thailand;16 it combines education, medication reminders, goal setting, two-way communication, and continuous blood pressure monitoring in one multicomponent program rather than a single function; and it examines routinely collected biomarkers alongside knowledge, health behaviors, medication adherence, and blood pressure. Its contribution therefore lies less in the use of a chatbot as such than in adapting a multicomponent, theory-based self-management program to the digital habits and functional needs of older adults in the Thai community context.
Given the high outpatient burden of hypertension and related chronic diseases among older adults, coupled with limited clinical time for continuous support, a chatbot-based hypertension management program is both necessary and timely. Therefore, this quasi-experimental study aimed to evaluate the effectiveness of a program on knowledge, health behaviors, medication adherence, blood pressure, and biomarkers among community-dwelling older adults.
The Study
Aim
This study aims to evaluate the effectiveness of a hypertension management program using a chatbot on knowledge, health behaviors, medication adherence, and blood pressure, and to test the preliminary possible biomarkers among community-dwelling older adults with hypertension.
Setting
Older adults who receiving health follow-ups and treatment at outpatient department 9, Srinagarind hospital was recruited into experimental group, and Sila health promotion hospital are managed separately for non-communicable disease care were recruited into control group. In addition, the research team will emphasize and instruct samples and setting not to distribute program content to control groups within community to prevent contamination.
Methods
Design
This study used a two-group pretest–posttest quasi-experimental design to evaluate the effectiveness of a chatbot-based hypertension management program on knowledge, health behaviors, medication adherence, blood pressure, and selected biomarkers among community-dwelling adults aged 60 years and older with diagnosed hypertension. A quasi-experimental approach was adopted due to ethical and logistical constraints in routine clinical settings, while baseline and post-intervention measurements enabled valid within- and between-group comparisons. Selection bias and threats to internal validity were minimized by applying identical eligibility criteria, standardized screening instruments, comparable timelines, and a control group to account for maturation effects. Moreover, an adjusted analysis was used for general characteristics or baseline outcomes. The intervention group received a chatbot-supported program incorporating education, self-monitoring, medication reminders, and lifestyle counseling, whereas the control group received usual care with a standard hypertension management manual. The study is reported in accordance with the CONSORT 2025 checklist (Supplementary Material 1). Blinding was not feasible in this study: participants necessarily knew whether they had received the chatbot, and the personnel who delivered the program, assessed the outcomes and analyzed the data were not blinded to group allocation, which may have introduced detection bias.
Sample
The sample size was calculated using Chanabun’s formula for independent group mean comparison, based on systolic blood pressure as the primary outcome, drawing on effect size data from a comparable self-management intervention by Sornkestrin et al (2023).24 Using a one-tailed significance level of 0.05 (Zα = 1.65), a power of 80% (Zβ = 0.84), a pooled variance of 142.26, and an expected mean difference of 7.70 mmHg (squared difference 59.29), the required sample size, n = 2(Zα + Zβ)2σ2/Δ2, was 29.76 participants per group, rounded to 30. Allowing for an anticipated 10% attrition gave 33.3 participants per group, rounded up to 34, for a final target sample of 68 participants, equally allocated to the experimental (n = 34) and control (n = 34) groups.
Inclusion and Exclusion Criteria
Participants were recruited using consecutive sampling, according to predefined eligibility criteria. Inclusion criteria comprised community-dwelling adults aged ≥ 60 years of both sexes with physician-diagnosed grade 1–2 hypertension who were receiving antihypertensive medication and attending the outpatient department or geriatric clinics, with or without comorbid diabetes mellitus or dyslipidemia. Eligible participants were required to have adequate functional independence (Barthel-ADL ≥ 12), cognitive function (Thai Mental State Examination [TMSE] > 23), and digital health literacy (Thai-eHEALS ≥ 28), own a smartphone, and be able to use the LINE application. Individuals with severe or unstable medical conditions, uncontrolled hypertension (>180/110 mmHg), acute illness, or major chronic diseases were excluded, as were those concurrently using other hypertension management programs, chatbots, or wearable devices. Participants were withdrawn if they developed unstable health conditions, requested withdrawal, or demonstrated insufficient engagement with the intervention, defined as using the chatbot for less than 80% of the intervention period (Figure 1).
Figure 1.
CONSORT flow diagram of participant progress through the study.
Intervention
The intervention comprised a chatbot-based hypertension management program designed to promote self-management among community-dwelling older adults through structured and interactive care. At baseline, participants received standardized hypertension education and hands-on training in using the LINE-based chatbot. The chatbot delivered evidence-based educational content, multimedia materials (an immediate-response message, a health-education video, and a manual), and medication reminders tailored to individual treatment plans. The content was evaluated for validity and reliability by five experts, including one medical doctor, two medical registered nurses, and two gerontological nurse lecturers, who demonstrated excellent methodological rigor, with a content validity index (CVI) of 0.98 and an item–objective congruence (IOC) index of 0.97. Throughout the intervention, it facilitated two-way communication, supported goal setting, provided motivational feedback, and continuously monitored health behaviors and blood pressure, enabling ongoing self-monitoring and behavior adjustment while reducing reliance on in-person clinical visits, as shown in Figure 2. The chatbot was rule-based and operated automatically for almost all interactions, drawing on a bank of responses that had been prepared in advance and reviewed by the expert panel. When a participant reported a blood pressure reading, the chatbot interpreted the value, explained what it meant and what the participant should do next, and gave initial self-care advice; for markedly elevated readings it displayed a button to call Thailand’s 1669 emergency medical hotline so that the participant could reach the nearest hospital. Questions unrelated to blood pressure, such as those about diet, exercise, medication use, or clinic attendance, were answered from the same pre-approved content. A research assistant trained in counseling people with hypertension monitored every participant’s conversation daily through the LINE Official Account and replied personally whenever the chatbot could not answer, so that all information remained within the scope of the program.
Figure 2.
The 12-week chatbot-based hypertension management program.
Control
Participants in the control group received a hypertension management manual and the standard nursing care provided by Sila Health Promoting Hospital, which already included monthly home visits or telephone follow-up and health education at the facility, none of these contacts was added for the study. The only difference from the experimental group was that the control group did not receive the chatbot for information, counseling, and follow-up. Participants were offered the chatbot after the study was completed.
Outcome Measures
A demographic questionnaire consisting of 15 items was used to collect background information, including gender, age, weight, height, education level, marital status, occupation, duration of hypertension diagnosis and treatment, current antihypertensive medication, systolic and diastolic blood pressure, underlying diseases, smoking, alcohol consumption, and sleep quality.
Hypertension knowledge was assessed using an 18-item true–false interview questionnaire, adapted from Namrinchuenchitsai (2022).25 Each correct response was scored 1 point, yielding a total score range of 0–18, with higher scores indicating greater knowledge of hypertension. Content validity was evaluated by three experts, resulting in a content validity index (CVI) of 0.80. Reliability testing with 30 participants demonstrated acceptable internal consistency (KR-20 = 0.76).
Health behaviors were measured using an 8-item checklist adapted from Prerprao and Sriwasom (2023)26 that covered healthy food consumption, DASH diet adherence, sodium restriction, exercise, alcohol consumption, smoking, stress management, and regular antihypertensive medication use. Each item was rated on five frequency categories ranging from less than 10 days per month to more than 25 days per month, reflecting increasing levels of behavior adherence. Content validity was assessed by three specialists, with CVI values ranging from 0.67 to 1.00. The checklist demonstrated excellent inter-rater reliability (IRR = 1.00) and high internal consistency (KR-20 = 0.80).
Medication adherence was assessed using the MAM-O scale,27 consisting of 8 Likert-type items measuring non-adherent behaviors over the past month. Each item was scored from 0 to 3, producing a total score range of 0–24. Scores were reverse-coded so that higher total scores reflected better medication adherence. CVI was 0.87. Pilot testing in 30 patients demonstrated acceptable reliability (Cronbach’s alpha = 0.71).
Blood pressure was measured using a calibrated OMRON automatic blood pressure monitor, calibrated by staff at the study setting Hospitals. Measurement reliability was confirmed using a test–retest method in 30 individuals with comparable characteristics, yielding a high reliability coefficient (r = 0.95). The same device was used consistently at the pretest and posttest. Blood pressure was measured on the left arm with participants seated and at rest; two readings were obtained 5 minutes apart, and the average was used for analysis.
Biomarker data were obtained from routine hospital medical records without additional blood sampling. Pre-intervention values were the results recorded within three to six months before enrollment, and post-intervention values were the first results recorded after completion of the 12-week program, which coincided with the hospitals’ scheduled follow-up blood tests for this patient group. Because blood tests were ordered according to each hospital’s own follow-up policy rather than by the study, the interval between tests and the number of participants with a result differed between biomarkers and between groups; the number of participants contributing to each analysis is therefore shown for every biomarker in the corresponding tables. Collected biomarkers included lipid profile, fasting blood glucose, serum electrolytes (sodium and potassium), blood urea nitrogen, serum creatinine, estimated glomerular filtration rate, and uric acid. These data were analyzed descriptively and comparatively.
Program engagement was monitored using a weekly participation checklist over the 12-week intervention. Participation in each activity was recorded as Yes or No. Participants with ≥ 80% participation were included in the final analysis, while those with < 80% were excluded to ensure adequate intervention exposure. However, all participants met the criteria and were included for final analysis.
Validity, Reliability, and Rigor
This study was conducted by a multidisciplinary research team comprising members from multiple institutions and areas of expertise. The identification of research problems and implementation of the intervention were undertaken in collaboration with an older adult clinic and a health promotion hospital to ensure contextual relevance and practical feasibility. Methodological development was supported by research specialists and academic advisors from the faculty of nursing and the faculty of medicine at Khon Kaen University, the Johns Hopkins University School of Nursing, and Sirindhorn College of Public Health at Khon Kaen, who guided study design, research strategies, and statistical analysis. The principal investigator prospectively registered the study protocol and adhered strictly to the predefined procedures to minimize bias and enhance the validity and reliability of the research process. All study procedures were continuously monitored and cross-checked against the approved protocol by the research team. When operational challenges arose, collaborative discussions were held to reach consensus and maintain methodological rigor.
Statistical Analysis
Data were verified for completeness and analyzed using IBM SPSS Statistics version 28.0 under Khon Kaen University license, with statistical significance set at p < 0.05. Descriptive statistics were used to summarize participant characteristics. At the same time, Chi-square or Fisher’s exact tests and independent t-tests or Mann–Whitney U-tests were applied to compare baseline categorical and continuous variables, respectively. Outcome variables, including knowledge, health behaviors, medication adherence, blood pressure, and biomarkers, were assessed as continuous data, with normality evaluated using the Shapiro–Wilk test. Paired t-tests or Wilcoxon signed-rank tests were used for within-group comparisons, and independent t-tests or Mann–Whitney U-tests for between-group comparisons. All hypothesis tests of program effectiveness were one-tailed, consistent with the directional hypothesis pre-specified in the registered protocol (TCTR20260118005); within-group changes in the control group, which were not directional hypotheses of the program, are reported two-tailed. No adjustment was made for multiple comparisons because the five primary outcomes were pre-specified; the biomarker comparisons are regarded as exploratory. All 68 participants completed both assessments. Therefore, the analyses of the primary outcomes had no missing data; biomarker analyses used every participant with the relevant value, and the number analyzed is reported for each comparison. Finally, analysis of covariance (ANCOVA) was used to compare post-intervention outcomes between groups. For each outcome the model included group as the fixed factor and, as covariates, the baseline value of that same outcome together with the six demographic characteristics that differed significantly between groups at baseline including sex, education level, marital status, occupation, smoking, and alcohol consumption. Adjusting for the corresponding baseline value addresses the imbalance present at entry, while restricting the remaining covariates to the variables that were actually imbalanced avoids over-specifying the model with 68 participants a larger covariate set would have provided fewer than five observations per estimated parameter and would have required listwise deletion of participants with incomplete laboratory data. Homogeneity of regression slopes was satisfied for health behaviors and for systolic and diastolic blood pressure, but not for knowledge or medication adherence; the adjusted estimates for those two outcomes, and a sensitivity analysis across alternative covariate sets showed that both remained significant under every specified examination. Parametric tests are reported throughout because the adjusted means they produce are directly interpretable on the original measurement scale; a sensitivity analysis against the corresponding non-parametric tests is provided in Supplementary Table S1. Medication adherence was scored with the MAM-O scale, on which higher raw scores indicate poorer adherence; item-level responses were reverse-coded before the total score was formed. Therefore, higher totals denote better adherence.
Results
Demographic Data
A total of 68 participants were recruited, with 34 in each group. The experimental group was predominantly male (21; 61.8%), with a mean age of 68.62 ± 5.43 years, whereas the control group was predominantly female (28; 82.4%), with a mean age of 67.91 ± 5.95 years. The experimental group had higher socioeconomic and educational status: 17 (50.0%) were employed as government officials, and 11 (32.4%) held a bachelor’s degree, compared with the control group, in which 18 (52.9%) were unemployed, and 23 (67.6%) had primary education. Marital status also differed, with a higher proportion of married participants in the experimental group (30, 88.2%) and a higher proportion of widowed participants in the control group (12, 35.3%). Health behaviors varied significantly between groups, with smoking (p = 0.011) and alcohol consumption (p = 0.005) more common in the experimental group. In contrast, the control group had a higher proportion of non-smokers (30, 88.2%) and non-drinkers (25, 73.5%). Despite these differences, both groups exhibited similarly high functional independence, cognitive function, and digital health literacy, with comparable ADL scores (experimental 19.62 ± 0.70, control 19.59 ± 0.70), TMSE scores (experimental 33.94 ± 1.39, control 33.50 ± 1.52), and Thai-eHEALS scores (experimental 35.62 ± 1.65, control 35.97 ± 1.78). There were significant baseline imbalances in sex, education, marital status, occupation, smoking, and alcohol consumption, as shown in Table 1.
Table 1.
Baseline Characteristics of Community-Dwelling Older Adults with Hypertension by Group
| Characteristic | Experimental (n = 34) | Control (n = 34) | p-value |
|---|---|---|---|
| Sex | <0.001C | ||
| Male | 21 (61.8) | 6 (17.6) | |
| Female | 13 (38.2) | 28 (82.4) | |
| Age (years) | 68.62 ± 5.43 (60–79) | 67.91 ± 5.95 (60–88) | 0.611T |
| Body weight (kg) | 69.88 ± 10.99 (50–105) | 69.03 ± 11.53 (42–93) | 0.756T |
| Body mass index (kg/m2) | 27.08 ± 5.07 (19–47) | 28.16 ± 4.24 (19–37) | 0.344T |
| Education level | 0.006C | ||
| Illiterate | 0 (0.0) | 1 (2.9) | |
| Primary school | 12 (35.3) | 23 (67.6) | |
| Secondary school | 6 (17.6) | 6 (17.6) | |
| Associate degree | 4 (11.8) | 4 (11.8) | |
| Bachelor’s degree | 11 (32.4) | 0 (0.0) | |
| Master’s degree or higher | 1 (2.9) | 0 (0.0) | |
| Marital status | 0.007C | ||
| Single | 2 (5.9) | 3 (8.8) | |
| Married | 30 (88.2) | 19 (55.9) | |
| Widowed | 2 (5.9) | 12 (35.3) | |
| Occupation | <0.001C | ||
| Farmer | 6 (17.6) | 9 (26.5) | |
| Merchant | 8 (23.5) | 7 (20.6) | |
| Private company employee | 0 (0.0) | 0 (0.0) | |
| Government service | 17 (50.0) | 0 (0.0) | |
| Unemployed | 3 (8.8) | 18 (52.9) | |
| Smoking | 0.011C | ||
| No | 26 (76.5) | 30 (88.2) | |
| Yes | 1 (2.9) | 4 (11.8) | |
| Quit | 7 (20.6) | 0 (0.0) | |
| Alcohol consumption | 0.005C | ||
| No | 23 (67.6) | 25 (73.5) | |
| Yes | 11 (32.4) | 3 (8.8) | |
| Quit | 0 (0.0) | 6 (17.6) | |
| Sleeping problem | 0.784F | ||
| No | 26 (76.5) | 24 (70.6) | |
| Yes | 8 (23.5) | 10 (29.4) | |
| Activities of daily living | 19.62 ± 0.70 (17–20) | 19.59 ± 0.70 (17–20) | 0.814M |
| Thai Mental State Examination | 33.94 ± 1.39 (30–36) | 33.50 ± 1.52 (30–35) | 0.193M |
| Electronic health literacy | 35.62 ± 1.65 (33–39) | 35.97 ± 1.78 (33–39) | 0.422M |
Notes: Categorical variables are shown as n (%) and continuous variables as mean ± SD (minimum–maximum). All comparisons are two-tailed.
Abbreviations: C, chi-square test; F, Fisher’s exact test; M, Mann–Whitney U-test; T, independent-samples t-test.
Knowledge, Behavior, Medication Adherence, and Physiological Changes (Within Groups)
Within-group analyses using paired t-tests showed statistically significant improvements in all primary outcomes in the experimental group after the program. Mean knowledge scores increased from 15.12 ± 1.63 at baseline to 15.74 ± 1.69 after the program (t = −3.78, df = 33, 95% CI −0.95 to −0.28, p < 0.001). Health behaviors improved from 22.53 ± 4.48 to 23.85 ± 4.78 (t = −3.09, df = 33, 95% CI −2.19 to −0.45, p = 0.002). Medication adherence increased from 20.79 ± 2.75 to 23.15 ± 1.60 (t = −4.63, df = 33, 95% CI −3.39 to −1.32, p < 0.001). Mean systolic blood pressure decreased from 138.59 ± 11.73 mmHg to 130.65 ± 11.73 mmHg (t = 3.88, df = 33, 95% CI 3.78 to 12.11, p < 0.001), and mean diastolic blood pressure decreased from 73.47 ± 9.26 mmHg to 70.12 ± 7.48 mmHg (t = 1.99, df = 33, 95% CI −0.08 to 6.79, p = 0.028). In the control group (n = 34), two-tailed tests showed no significant change in knowledge (14.85 ± 1.83 to 14.94 ± 1.65, p = 0.753), health behaviors (19.24 ± 3.46 to 19.38 ± 3.77, p = 0.809), medication adherence (21.79 ± 2.67 to 22.12 ± 2.60, p = 0.148), or systolic blood pressure (130.44 ± 9.98 to 132.59 ± 12.54 mmHg, p = 0.327), whereas diastolic blood pressure rose significantly from 74.79 ± 12.71 mmHg to 79.85 ± 9.30 mmHg (t = −3.19, df = 33, 95% CI −8.29 to −1.83, p = 0.003). Effect sizes for every comparison are given in Tables 2–5.
Table 2.
Knowledge Before and After the Program Within Each Group
| Group | n | Before Mean ± SD |
After Mean ± SD |
t | df | 95% CI of MD |
dz [95% CI] |
p-value |
|---|---|---|---|---|---|---|---|---|
| Experimental | 34 | 15.12 ± 1.63 | 15.74 ± 1.69 | −3.78 | 33 | −0.95, −0.28 | −0.65 [−1.02, −0.28] | <0.001 |
| Control | 34 | 14.85 ± 1.83 | 14.94 ± 1.65 | −0.32 | 33 | −0.65, 0.48 | −0.05 [−0.39, 0.28] | 0.753 |
Notes: Mean difference and t are calculated as before minus after; a negative value indicates an increase. p is one-tailed for the experimental group and two-tailed for the control group.
Abbreviations: CI, confidence interval; df, degrees of freedom; dz, Cohen’s d for paired data; MD, mean difference; SD, standard deviation.
Table 3.
Health Behaviors Before and After the Program Within Each Group
| Group | n | Before Mean ± SD |
After Mean ± SD |
t | df | 95% CI of MD |
dz [95% CI] |
p-value |
|---|---|---|---|---|---|---|---|---|
| Experimental | 34 | 22.53 ± 4.48 | 23.85 ± 4.78 | −3.09 | 33 | −2.19, −0.45 | −0.53 [−0.89, −0.17] | 0.002 |
| Control | 34 | 19.24 ± 3.46 | 19.38 ± 3.77 | −0.24 | 33 | −1.38, 1.08 | −0.04 [−0.38, 0.29] | 0.809 |
Notes: Mean difference and t are calculated as before minus after; a negative value indicates an increase. p is one-tailed for the experimental group and two-tailed for the control group.
Abbreviations: CI, confidence interval; df, degrees of freedom; dz, Cohen’s d for paired data; MD, mean difference; SD, standard deviation.
Table 4.
Medication Adherence Before and After the Program Within Each Group
| Group | n | Before Mean ± SD |
After Mean ± SD |
t | df | 95% CI of MD |
dz [95% CI] |
p-value |
|---|---|---|---|---|---|---|---|---|
| Experimental | 34 | 20.79 ± 2.75 | 23.15 ± 1.60 | −4.63 | 33 | −3.39, −1.32 | −0.79 [−1.18, −0.41] | <0.001 |
| Control | 34 | 21.79 ± 2.67 | 22.12 ± 2.60 | −1.48 | 33 | −0.77, 0.12 | −0.25 [−0.60, 0.09] | 0.148 |
Notes: Mean difference and t are calculated as before minus after; a negative value indicates an increase. p is one-tailed for the experimental group and two-tailed for the control group.
Abbreviations: CI, confidence interval; df, degrees of freedom; dz, Cohen’s d for paired data; MD, mean difference; SD, standard deviation.
Table 5.
Blood Pressure Before and After the Program Within Each Group
| Outcomes | n | Before Mean ± SD |
After Mean ± SD |
t | df | 95% CI of MD |
dz [95% CI] |
p-value |
|---|---|---|---|---|---|---|---|---|
| Systolic blood pressure | ||||||||
| Experimental | 34 | 138.59 ± 11.73 | 130.65 ± 11.73 | 3.88 | 33 | 3.78, 12.11 | 0.67 [0.29, 1.04] | <0.001 |
| Control | 34 | 130.44 ± 9.98 | 132.59 ± 12.54 | −1.00 | 33 | −6.53, 2.24 | −0.17 [−0.51, 0.17] | 0.327 |
| Diastolic blood pressure | ||||||||
| Experimental | 34 | 73.47 ± 9.26 | 70.12 ± 7.48 | 1.99 | 33 | −0.08, 6.79 | 0.34 [−0.01, 0.69] | 0.028 |
| Control | 34 | 74.79 ± 12.71 | 79.85 ± 9.30 | −3.19 | 33 | −8.29, −1.83 | −0.55 [−0.91, −0.19] | 0.003 |
Notes: Mean difference and t are calculated as before minus after; a negative value indicates an increase. p is one-tailed for the experimental group and two-tailed for the control group.
Abbreviations: CI, confidence interval; df, degrees of freedom; dz, Cohen’s d for paired data; MD, mean difference; SD, standard deviation.
Knowledge, Behavior, Medication Adherence, and Physiological Changes (Between Groups)
Between-group comparisons are shown in Tables 6–9. Knowledge did not differ at baseline (15.12 ± 1.63 compared with 14.85 ± 1.83; MD = 0.26, SE = 0.42, p = 0.531) but was higher in the experimental group after the program (15.74 ± 1.69 compared with 14.94 ± 1.65; MD = 0.79, SE = 0.41, t = 1.96, p = 0.027). Health behaviors were already better in the experimental group at baseline (22.53 ± 4.48 compared with 19.24 ± 3.46; MD = 3.29, SE = 0.97, p = 0.001), and the difference widened after the program (23.85 ± 4.78 compared with 19.38 ± 3.77; MD = 4.47, SE = 1.04, t = 4.28, p < 0.001). Medication adherence did not differ at baseline (20.79 ± 2.75 compared with 21.79 ± 2.67; MD = −1.00, SE = 0.66, p = 0.133) but was higher in the experimental group after the program (23.15 ± 1.60 compared with 22.12 ± 2.60; MD = 1.03, SE = 0.52, t = 1.97, p = 0.027). Systolic blood pressure was higher in the experimental group at baseline (138.59 ± 11.73 compared with 130.44 ± 9.98 mmHg; MD = 8.15, SE = 2.64, p = 0.003) and did not differ after the program (130.65 ± 11.73 compared with 132.59 ± 12.54 mmHg; MD = −1.94, SE = 2.94, t = −0.66, p = 0.256). Diastolic blood pressure did not differ at baseline (73.47 ± 9.26 compared with 74.79 ± 12.71 mmHg; MD = −1.32, SE = 2.70, p = 0.625) but was lower in the experimental group after the program (70.12 ± 7.48 compared with 79.85 ± 9.30 mmHg; MD = −9.74, SE = 2.05, t = −4.76, p < 0.001).
Table 6.
Knowledge Compared Between Groups Before and After the Program
| Time | n | Experimental Mean ± SD |
Control Mean ± SD |
MD (SE) | 95% CI of MD | t | Cohen d [95% CI] | p-value |
|---|---|---|---|---|---|---|---|---|
| Before | 34 | 15.12 ± 1.63 | 14.85 ± 1.83 | 0.26 (0.42) | −0.57, 1.10 | 0.63 | 0.15 [−0.32, 0.63] | 0.531 |
| After | 34 | 15.74 ± 1.69 | 14.94 ± 1.65 | 0.79 (0.41) | −0.02, 1.60 | 1.96 | 0.47 [−0.01, 0.96] | 0.027 |
Notes: Mean difference is calculated as experimental minus control. p is two-tailed at baseline and one-tailed after the program.
Abbreviations: CI, confidence interval; Cohen d, standardized mean difference based on the pooled SD; MD, mean difference; SD, standard deviation; SE, standard error.
Table 7.
Health Behaviors Compared Between Groups Before and After the Program
| Time | n | Experimental Mean ± SD |
Control Mean ± SD |
MD (SE) | 95% CI of MD | t | Cohen d [95% CI] | p-value |
|---|---|---|---|---|---|---|---|---|
| Before | 34 | 22.53 ± 4.48 | 19.24 ± 3.46 | 3.29 (0.97) | 1.35, 5.23 | 3.39 | 0.82 [0.33, 1.32] | 0.001 |
| After | 34 | 23.85 ± 4.78 | 19.38 ± 3.77 | 4.47 (1.04) | 2.39, 6.55 | 4.28 | 1.04 [0.53, 1.55] | <0.001 |
Notes: Mean difference is calculated as experimental minus control. p is two-tailed at baseline and one-tailed after the program.
Abbreviations: CI, confidence interval; Cohen d, standardized mean difference based on the pooled SD; MD, mean difference; SD, standard deviation; SE, standard error.
Table 8.
Medication Adherence Compared Between Groups Before and After the Program
| Time | n | Experimental Mean ± SD |
Control Mean ± SD |
MD (SE) | 95% CI of MD | t | Cohen d [95% CI] | p-value |
|---|---|---|---|---|---|---|---|---|
| Before | 34 | 20.79 ± 2.75 | 21.79 ± 2.67 | −1.00 (0.66) | −2.31, 0.31 | −1.52 | −0.37 [−0.85, 0.11] | 0.133 |
| After | 34 | 23.15 ± 1.60 | 22.12 ± 2.60 | 1.03 (0.52) | −0.02, 2.08 | 1.97 | 0.48 [−0.01, 0.96] | 0.027 |
Notes: Mean difference is calculated as experimental minus control. p is two-tailed at baseline and one-tailed after the program.
Abbreviations: CI, confidence interval; Cohen d, standardized mean difference based on the pooled SD; MD, mean difference; SD, standard deviation; SE, standard error.
Table 9.
Blood Pressure Compared Between Groups Before and After the Program
| Time | n | Experimental Mean ± SD |
Control Mean ± SD |
MD (SE) | 95% CI of MD | t | Cohen d [95% CI] | p-value |
|---|---|---|---|---|---|---|---|---|
| Before | 34 | 138.59 ± 11.73 | 130.44 ± 9.98 | 8.15 (2.64) | 2.87, 13.42 | 3.08 | 0.75 [0.26, 1.24] | 0.003 |
| After | 34 | 130.65 ± 11.73 | 132.59 ± 12.54 | −1.94 (2.94) | −7.82, 3.94 | −0.66 | −0.16 [−0.64, 0.32] | 0.256 |
| Before | 34 | 73.47 ± 9.26 | 74.79 ± 12.71 | −1.32 (2.70) | −6.71, 4.06 | −0.49 | −0.12 [−0.59, 0.36] | 0.625 |
| After | 34 | 70.12 ± 7.48 | 79.85 ± 9.30 | −9.74 (2.05) | −13.82, −5.65 | −4.76 | −1.15 [−1.67, −0.64] | <0.001 |
Notes: Mean difference is calculated as experimental minus control. p is two-tailed at baseline and one-tailed after the program.
Abbreviations: CI, confidence interval; Cohen d, standardized mean difference based on the pooled SD; MD, mean difference; SD, standard deviation; SE, standard error.
Within Group and Between Group Analyses of Observable Selected Biomarkers
Within the experimental group, paired analyses of participants with a value both before and after the program showed that most biomarkers remained statistically stable Total cholesterol fell from 155.56 ± 40.63 to 147.06 ± 35.91 mg/dL (n = 16), a change that was not statistically significant (p = 0.106), whereas triglycerides fell significantly from 141.60 ± 54.70 to 119.27 ± 53.40 mg/dL (n = 15; t = 2.37, 95% CI 2.11 to 42.55, p = 0.016, dz = 0.61). No significant pre–post differences were observed for HDL cholesterol, LDL cholesterol, fasting blood glucose, blood urea nitrogen, creatinine, estimated glomerular filtration rate, serum sodium, or serum potassium, and uric acid could not be tested because only two participants had both values. In the control group, paired values were available only for LDL cholesterol, fasting blood glucose, creatinine, estimated glomerular filtration rate, and serum potassium, none of which changed significantly. The number of participants contributing to each comparison and the corresponding effect sizes are shown in Table 10.
Table 10.
Biomarkers Before and After the Program within Each Group
| Biomarker | n | Before Mean ± SD |
After Mean ± SD |
t | df | 95% CI of MD | dz [95% CI] |
p-value |
|---|---|---|---|---|---|---|---|---|
| Experimental group | ||||||||
| Total cholesterol (mg/dL) | 16 | 155.56 ± 40.63 | 147.06 ± 35.91 | 1.31 | 15 | −5.37, 22.37 | 0.33 [−0.18, 0.83] | 0.106 |
| Triglycerides (mg/dL) | 15 | 141.60 ± 54.70 | 119.27 ± 53.40 | 2.37 | 14 | 2.11, 42.55 | 0.61 [0.06, 1.16] | 0.016 |
| HDL cholesterol (mg/dL) | 15 | 49.93 ± 15.55 | 47.67 ± 13.39 | 0.71 | 14 | −4.61, 9.14 | 0.18 [−0.33, 0.69] | 0.491a |
| LDL cholesterol (mg/dL) | 17 | 90.88 ± 34.89 | 88.24 ± 31.56 | 0.45 | 16 | −9.96, 15.25 | 0.11 [−0.37, 0.58] | 0.331 |
| Fasting blood glucose (mg/dL) | 20 | 121.10 ± 28.90 | 112.95 ± 17.67 | 1.57 | 19 | −2.72, 19.02 | 0.35 [−0.10, 0.80] | 0.067 |
| Blood urea nitrogen (mg/dL) | 23 | 17.06 ± 6.88 | 16.11 ± 4.65 | 0.84 | 22 | −1.38, 3.27 | 0.18 [−0.24, 0.59] | 0.205 |
| Creatinine (mg/dL) | 25 | 1.20 ± 0.41 | 1.17 ± 0.31 | 0.54 | 24 | −0.07, 0.12 | 0.11 [−0.29, 0.50] | 0.298 |
| eGFR (mL/min/1.73 m2) | 25 | 59.47 ± 16.74 | 61.73 ± 15.59 | −1.39 | 24 | −5.63, 1.11 | −0.28 [−0.68, 0.12] | 0.089 |
| Serum sodium (mmol/L) | 22 | 138.64 ± 2.56 | 139.27 ± 3.40 | −1.48 | 21 | −1.53, 0.26 | −0.32 [−0.74, 0.11] | 0.076 |
| Serum potassium (mmol/L) | 24 | 4.22 ± 0.50 | 4.35 ± 0.45 | −1.32 | 23 | −0.33, 0.07 | −0.27 [−0.68, 0.14] | 0.100 |
| Control group | ||||||||
| LDL cholesterol (mg/dL) | 28 | 118.43 ± 26.46 | 110.96 ± 21.08 | 1.26 | 27 | −4.68, 19.61 | 0.24 [−0.14, 0.61] | 0.218 |
| Fasting blood glucose (mg/dL) | 34 | 117.91 ± 33.60 | 118.76 ± 31.83 | −0.12 | 33 | −14.88, 13.18 | −0.02 [−0.36, 0.31] | 0.902 |
| Creatinine (mg/dL) | 21 | 0.79 ± 0.19 | 0.79 ± 0.27 | −0.04 | 20 | −0.16, 0.15 | −0.01 [−0.44, 0.42] | 0.965 |
| eGFR (mL/min/1.73 m2) | 33 | 82.36 ± 18.20 | 82.20 ± 15.23 | 0.04 | 32 | −7.58, 7.90 | 0.01 [−0.33, 0.35] | 0.966 |
| Serum potassium (mmol/L) | 6 | 4.89 ± 1.07 | 4.75 ± 0.31 | 0.41 | 5 | −0.77, 1.07 | 0.17 [−0.64, 0.97] | 0.699 |
Notes: Only participants with a recorded value both before and after the program are included in each paired comparison, so n differs between biomarkers. Mean difference and t are calculated as before minus after; a negative value indicates an increase. p is one-tailed for the experimental group and two-tailed for the control group. aChange opposite to the hypothesised direction; two-tailed p reported.
Abbreviations: CI, confidence interval; df, degrees of freedom; dz, Cohen’s d for paired data; eGFR, estimated glomerular filtration rate; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MD, mean difference; SD, standard deviation.
Between-group comparisons using the Mann–Whitney U-test showed that the groups differed in several biomarkers already at baseline. LDL cholesterol was lower in the experimental group both at baseline (Z = −3.59, p < 0.001, r = 0.44) and after the program (Z = −3.61, p < 0.001, r = 0.53). Renal indices were less favorable in the experimental group at both time points: creatinine was higher (Z = 4.72 at baseline and 3.93 after the program, both p < 0.001) and estimated glomerular filtration rate lower (Z = −4.75 and −4.28, both p < 0.001), so the post-program differences in these two markers ran opposite to the hypothesized direction and are reported two-tailed. Serum potassium was lower in the experimental group at baseline (Z = −2.32, p = 0.021) but did not differ significantly after the program (Z = −1.87, p = 0.061). No significant between-group differences were found for total cholesterol, triglycerides, HDL cholesterol, fasting blood glucose, or uric acid at either time point, and blood urea nitrogen and serum sodium could not be compared because the control group had at most one recorded value. The number of participants in each group, mean ranks, and effect sizes (r) are shown in Table 11.
Table 11.
Biomarkers Compared Between Groups Before and After the Program
| Biomarker | Time | Experimental (n) | Control (n) | Mean Rank Experimental | Mean Rank Control | Z | r | p-value |
|---|---|---|---|---|---|---|---|---|
| Cholesterol (mg/dL) | Before | 32 | 9 | 19.89 | 24.94 | −1.12 | 0.17 | 0.263 |
| Triglycerides (mg/dL) | Before | 31 | 7 | 18.10 | 25.71 | −1.64 | 0.27 | 0.101 |
| Triglycerides (mg/dL) | After | 17 | 2 | 9.88 | 11.00 | −0.27 | 0.06 | 0.395 |
| HDL cholesterol (mg/dL) | Before | 31 | 7 | 20.92 | 13.21 | 1.66 | 0.27 | 0.097 |
| HDL cholesterol (mg/dL) | After | 17 | 2 | 9.59 | 13.50 | −0.93 | 0.21 | 0.352a |
| LDL cholesterol (mg/dL) | Before | 32 | 34 | 24.75 | 41.74 | −3.59 | 0.44 | <0.001 |
| LDL cholesterol (mg/dL) | After | 19 | 28 | 15.24 | 29.95 | −3.61 | 0.53 | <0.001 |
| Fasting blood glucose (mg/dL) | Before | 34 | 34 | 35.87 | 33.13 | 0.57 | 0.07 | 0.568 |
| Fasting blood glucose (mg/dL) | After | 20 | 34 | 27.07 | 27.75 | −0.15 | 0.02 | 0.439 |
| Creatinine (mg/dL) | Before | 34 | 34 | 45.81 | 23.19 | 4.72 | 0.57 | <0.001 |
| Creatinine (mg/dL) | After | 25 | 21 | 30.62 | 15.02 | 3.93 | 0.58 | <0.001a |
| eGFR (mL/min/1.73 m2) | Before | 34 | 34 | 23.12 | 45.88 | −4.75 | 0.58 | <0.001 |
| eGFR (mL/min/1.73 m2) | After | 25 | 33 | 18.60 | 37.76 | −4.28 | 0.56 | <0.001a |
| Serum potassium (mmol/L) | Before | 34 | 30 | 27.44 | 38.23 | −2.32 | 0.29 | 0.021 |
| Serum potassium (mmol/L) | After | 24 | 6 | 14.00 | 21.50 | −1.87 | 0.34 | 0.061a |
| Uric acid (mg/dL) | Before | 8 | 10 | 8.62 | 10.20 | −0.62 | 0.15 | 0.534 |
| Uric acid (mg/dL) | After | 3 | 4 | 4.33 | 3.75 | 0.35 | 0.13 | 0.724a |
Notes: n is the number of participants with a recorded value at that time point. p is two-tailed at baseline and one-tailed after the program. aDifference opposite to the hypothesised direction; two-tailed p reported.
Abbreviations: eGFR, estimated glomerular filtration rate; HDL, high-density lipoprotein; LDL, low-density lipoprotein; r, effect size (|Z|/√N); Z, standardized Mann–Whitney statistic.
Adjusted Analysis of Covariance (ANCOVA)
After adjustment for the corresponding baseline value and the demographic characteristics that differed at baseline, the chatbot-based program was associated with significantly better outcomes than usual care across all five primary measures. Adjusted mean knowledge scores were higher in the experimental group (15.90 ± 0.23 compared with 14.78 ± 0.23; F = 9.48, p = 0.002, partial η2 = 0.13), as were health behaviors (22.66 ± 0.61 compared with 20.57 ± 0.61; F = 4.63, p = 0.018, partial η2 = 0.07) and medication adherence (23.52 ± 0.33 compared with 21.75 ± 0.33; F = 11.55, p < 0.001, partial η2 = 0.16). The experimental group also showed lower adjusted systolic blood pressure (127.43 ± 2.16 compared with 135.80 ± 2.16 mmHg; F = 5.97, p = 0.009, partial η2 = 0.09) and lower adjusted diastolic blood pressure (69.61 ± 1.39 compared with 80.36 ± 1.39 mmHg; F = 24.32, p < 0.001, partial η2 = 0.29). Among the biomarkers, no significant adjusted between-group differences were found for LDL cholesterol (p = 0.341), creatinine (p = 0.862), estimated glomerular filtration rate (p = 0.807) or serum potassium (p = 0.993). Fasting blood glucose was lower in the experimental group under the pre-specified one-tailed test (106.05 ± 7.09 compared with 122.83 ± 5.05 mg/dL; F = 2.96, p = 0.046), although this difference was not significant on a two-tailed test (p = 0.092) and is therefore regarded as exploratory. The adjusted analyses are shown in Table 12, and their sensitivity to the choice of covariates is shown in Table 13, and the conclusions for all five primary outcomes were unchanged under the alternative covariate specifications.
Table 12.
Effectiveness of the Hypertension Management Program Applying a Chatbot Between the Experimental and Control Groups After Adjustment for Covariates (ANCOVA)
| Outcome | n (E/C) | Mean ± SD Experimental |
Mean ± SD Control |
Adjusted Mean (SE) Ex. |
Adjusted Mean (SE) Con. |
F | pa | η2 |
|---|---|---|---|---|---|---|---|---|
| Knowledged | 34/34 | 15.74 ± 1.69 | 14.94 ± 1.65 | 15.90 (0.23) | 14.78 (0.23) | 9.48 | 0.002 | 0.138 |
| Health behaviors | 34/34 | 23.85 ± 4.78 | 19.38 ± 3.77 | 22.66 (0.61) | 20.57 (0.61) | 4.63 | 0.018 | 0.073 |
| Medication adherenced | 34/34 | 23.15 ± 1.60 | 22.12 ± 2.60 | 23.52 (0.33) | 21.75 (0.33) | 11.55 | <0.001 | 0.164 |
| Systolic blood pressure (mmHg) | 34/34 | 130.65 ± 11.73 | 132.59 ± 12.54 | 127.43 (2.16) | 135.80 (2.16) | 5.97 | 0.009 | 0.092 |
| Diastolic blood pressure (mmHg) | 34/34 | 70.12 ± 7.48 | 79.85 ± 9.30 | 69.61 (1.39) | 80.36 (1.39) | 24.32 | <0.001 | 0.292 |
| LDL cholesterol (mg/dL)b | 17/28 | 86.47 ± 30.42 | 110.96 ± 21.08 | 99.50 (7.79) | 104.12 (5.47) | 0.17 | 0.341 | 0.005 |
| Fasting blood glucose (mg/dL)bc | 20/34 | 112.95 ± 17.67 | 118.76 ± 31.83 | 106.05 (7.09) | 122.83 (5.05) | 2.96 | 0.046 | 0.062 |
| Creatinine (mg/dL)b | 25/21 | 1.17 ± 0.31 | 0.79 ± 0.27 | 1.05 (0.06) | 0.94 (0.07) | 1.23 | 0.862 | 0.032 |
| eGFR (mL/min/1.73 m2)b | 25/33 | 61.73 ± 15.59 | 82.20 ± 15.23 | 70.40 (3.84) | 75.64 (3.13) | 0.77 | 0.807 | 0.015 |
| Serum potassium (mmol/L)b | 24/6 | 4.35 ± 0.45 | 4.75 ± 0.31 | 4.30 (0.08) | 4.96 (0.21) | 7.36 | 0.993 | 0.259 |
Notes: Each model included group as the fixed factor with the baseline value of the same outcome and the six demographic characteristics that differed significantly between groups at baseline (sex, education level, marital status, occupation, smoking, and alcohol consumption) as covariates. Homogeneity of regression slopes was met for health behaviors (p = 0.069), systolic blood pressure (p = 0.959) and diastolic blood pressure (p = 0.145), but not for knowledge (p = 0.040) or medication adherence (p < 0.001). aOne-tailed analysis. bLaboratory values were obtained from routine medical records and were therefore available for a subset of participants; the number analysed is given for each biomarker. cThe difference in fasting blood glucose was significant under the pre-specified one-tailed test (p = 0.046) but not under a two-tailed test (p = 0.092). dHomogeneity of regression slopes was not met for this outcome.
Abbreviations: C, control group; E, experimental group; eGFR, estimated glomerular filtration rate; F, ANCOVA test statistic; LDL, low-density lipoprotein; SD, standard deviation; SE, standard error; η2, partial eta squared.
Table 13.
Sensitivity of the Adjusted Between-Group Comparison to the Choice of Covariates
| Outcome | Model 1: Baseline Only F (p)a |
Model 2: Baseline and Demographics F (p)a |
Model 3: Full Covariate Set F (p)a |
n Model 1/2/3 |
Slopes pb |
|---|---|---|---|---|---|
| Knowledge | 4.38 (0.020) | 9.48 (0.002) | 4.50 (0.020) | 68/68/62 | 0.040 |
| Health behaviors | 5.86 (0.009) | 4.63 (0.018) | 1.77 (0.095) | 68/68/62 | 0.069 |
| Medication adherence | 11.69 (<0.001) | 11.55 (<0.001) | 6.89 (0.006) | 68/68/62 | <0.001 |
| Systolic blood pressure | 4.30 (0.021) | 5.97 (0.009) | 2.17 (0.074) | 68/68/62 | 0.959 |
| Diastolic blood pressure | 28.39 (<0.001) | 24.32 (<0.001) | 8.22 (0.003) | 68/68/62 | 0.145 |
| LDL cholesterolc | 2.36 (0.066) | 0.17 (0.341) | 0.00 (0.515) | 45/45/42 | 0.018 |
| Fasting blood glucosec | 0.81 (0.187) | 2.96 (0.046) | 0.00 (0.525) | 54/54/49 | 0.604 |
| Creatininec | 4.25 (0.977) | 1.23 (0.862) | 0.05 (0.415) | 46/46/41 | 0.008 |
| eGFRc | 6.73 (0.994) | 0.77 (0.807) | 1.10 (0.150) | 58/58/52 | 0.001 |
| Serum potassiumc | 0.93 (0.828) | 7.36 (0.993) | 4.52 (0.973) | 30/30/28 | 0.313 |
Notes: aOne-tailed analysis. bHomogeneity of regression slopes (group × baseline interaction). cLaboratory values were available for a subset of participants. Therefore, model 3 for these outcome rests on few observations per estimated parameter and is shown for completeness only.
Abbreviations: F, ANCOVA test statistic; p, p value.
Discussions
This quasi-experimental study, with a two-group, pre- and post-test design, found a statistically significant difference in gender distribution (p < 0.001), with the experimental group predominantly male (61.80%) and the control group predominantly female (82.40%). This disparity reflects epidemiological trends where hypertension in early older males is often associated with risk factors such as smoking and alcohol consumption.7,10 In contrast, hypertension prevalence rises in post-menopausal women due to estrogen withdrawal and arterial stiffening,6,8 consistent with Thai morbidity data.10
The mean age was in the experimental: 68.62 ± 5.43 and control: 67.91 ± 5.95, representing the early older adult stage where blood pressure control is often hindered by persistent habits formed during middle age due to work obligations.9 Body mass index was also similar in the experimental group (27.08 ± 5.07) and the control group (28.16 ± 4.24), indicating an overweight status in both groups and comparable baseline metabolic risk, a key factor associated with hypertension.11 The age variable also mirrors the population, confirming that participants in this age range can engage with mHealth interventions despite being in a life stage where long-standing habits formed in middle age often persist.9,11
Alcohol consumption was significantly higher in the experimental group, placing the experimental group at a physiological disadvantage due to alcohol’s known association with elevated blood pressure6,9 and its role as a culturally ingrained barrier to hypertension control in Thailand.9 However, the experimental group achieved superior blood pressure control despite an elevated baseline risk, suggesting an opportunity to use the chatbot-assisted intervention. The program’s frequent digital reminders and health literacy support successfully motivated patients to moderate consumption and neutralize this specific behavioral risk.28 The high prevalence of alcohol users in the experimental group contrasts with where participants in hypertension education programs typically exhibited lower alcohol use rates. This specific demographic profile mirrors the hard-to-reach male population described by who are often resistant to lifestyle modification.29 Nevertheless, the success of this intervention aligns with who demonstrated that structured self-management training can effectively modify deep-seated health behaviors and improve clinical outcomes, even among older adults with established risk factors, such as alcohol consumption.30
The improvements observed in the experimental group may be explained in part by the chatbot’s micro-learning design, which delivers hypertension content through short videos and infographics to reduce cognitive load and accommodate age-related decline.31,32 This approach addresses the limitations of text-heavy messaging, which is less effective for older adults. It aligns with evidence that multimedia education improves health literacy and self-efficacy in hypertension management. Integration within the familiar LINE platform minimized the learning curve and promoted active engagement, consistent with findings from other interactive chatbot interventions.33,34 Despite known digital literacy barriers among some older adults, the observed improvements indicate that simplified content, visual aids, and large fonts effectively tailor education to users’ literacy levels.29 In particular, the interactive prompts with immediate visual or video feedback likely enhanced knowledge retention by encouraging active retrieval and application, reinforcing learning through repeated, manageable cognitive engagement rather than passive information exposure.17,29
The chatbot acted as an external self-regulation tool, shifting health monitoring from episodic clinic visits to continuous engagement through timely warnings and suggested messages, consistent with self-management theory.21,22 Its interactive design that using quick replies and multimedia that addressed limitations of static long-message services and promoted active participation, which is critical for older adults.31 This two-way interaction enhanced self-efficacy, a key determinant of health behavior among Thai older adults, by reinforcing habits through immediate feedback and positive messaging35 However, sustaining engagement remains a challenge, as digital health use may decline after an initial novelty effect, underscoring the need for personalized, adaptive content in future iterations.36
Medication adherence improved significantly in the experimental group, while no statistically significant change was observed in the control group. This improvement may reflect targeted strategies that address older adults’ barriers, including timely, intrusive LINE reminders that reduce memory burden.30,36 In addition, timely educational messages clarified the long-term benefits of medication despite asymptomatic status, strengthening outcome expectancy and motivation, as outlined in Kanfer’s framework. Unlike reminder-only approaches, which are insufficient when medication beliefs are weak, combining reminders with a clear educational rationale supports sustained adherence.37 Finally, looped feedback and reinforcement fostered a sense of achievement and social accountability, with the chatbot functioning as a virtual health partner that reduced adherence lapses.38,39
The experimental group showed statistically significant reductions in both systolic and diastolic blood pressure, whereas diastolic pressure in the control group rose significantly over the same period. A rise of this kind in a usual-care group may reflect regression toward the mean, seasonal variation, or the less intensive follow-up available at a primary care unit; it also widens the adjusted between-group difference in diastolic pressure and should be interpreted with that in mind. These improvements paralleled positive behavioral changes, including reduced sodium intake facilitated by ongoing education, reminders, and motivational prompts, which may decrease fluid retention and vascular volume in older adults with hypertension.30,36 Enhanced medication adherence likely reduced dose omissions, which are known to contribute to blood pressure variability. In addition, regular self-monitoring allowed participants to link daily behaviors with blood pressure trends through immediate feedback, reinforcing self-regulatory learning, consistent with established self-management frameworks.34,36,38 Therefore, the reductions of this size are clinically meaningful over the longer term cannot be established from a 12-week study in which site and intervention were confounded.
Biomarker findings were exploratory, with a significant reduction observed only in triglycerides, which likely reflects sensitivity to short-term lifestyle changes such as alcohol intake.34 Stable markers, including LDL cholesterol and renal indices, showed no significant change, consistent with evidence that long-standing metabolic and renal adaptations require longer intervention in older adults.39–41
After covariate adjustment, the program was associated with significantly better knowledge (F = 9.48, p = 0.002), health behaviors (F = 4.63, p = 0.018), medication adherence (F = 11.55, p < 0.001) and blood pressure control (systolic F = 5.97, p = 0.009; diastolic F = 24.32, p < 0.001) than usual care, despite baseline demographic differences. Adjusted between-group differences in the selected biomarkers were not statistically significant, apart from an exploratory reduction in fasting blood glucose that did not hold under a two-tailed test, suggesting that, while behavioral and clinical outcomes respond in the short term, longer intervention and follow-up periods are required to sustain metabolic changes in older adults.40–42 Randomized designs are recommended in future studies to strengthen the generalizability of the findings.
These findings are applicable to hypertension care delivered by at least two disciplines. Nurses can use the chatbot to sustain education, self-monitoring and behavioral counseling between clinic visits and to identify participants whose readings warrant earlier review, reducing reliance on in-person follow-up. Physicians gain a continuous record of home blood pressure and medication-taking that informs treatment decisions at the next consultation, and they remain responsible for ordering and interpreting the laboratory monitoring reported here. Community health personnel at primary care units can extend the same program to older adults who attend health promotion hospitals rather than tertiary outpatient clinics. Implementation therefore depends on a multidisciplinary team rather than on nursing alone.
Limitations and Recommendations
The most important limitation of this study is that the two groups were recruited from different healthcare settings. The experimental group attended outpatient department 9 of Srinagarind Hospital, a tertiary university hospital, whereas the control group received non-communicable disease care at Sila Health Promoting Hospital, a primary care unit. These settings differ in the intensity of routine follow-up, in the composition of the care team, and in the populations, and the groups also differed at baseline in sex, education, marital status, occupation, smoking, and alcohol consumption. Site and intervention are therefore confounded by design. Therefore, the observed differences cannot be attributed to the chatbot program alone and may in part reflect the setting in which care was received. Adjusting for baseline values and for the imbalanced characteristics reduces but cannot eliminate this confounding, because setting-level factors that were not measured remain unaccounted. The quasi-experimental design further limits causal inference and may introduce selection bias and residual confounding. Self-reported measures of health behaviors and medication adherence are subject to recall and social desirability bias. Generalizability is constrained to digitally literate older adults with smartphones and internet access who attend these two services. The 12-week follow-up captured short-term behavioral and blood pressure changes but was likely insufficient to detect delayed biomarker effects, and contamination between groups could not be fully excluded. For these reasons the results are best regarded as preliminary and hypothesis-generating. For future studies, the program should strengthen digital health capacity by training and supporting nurses, promoting active patient co-management with caregiver involvement, and using simple, user-friendly technologies. Policies should address digital inequity, ensure data privacy, and integrate chatbot-supported care into routine services. Methodologically, randomized controlled trials that draw both groups from the same setting, with longer follow-up and economic evaluation, are needed to confirm effectiveness and assess sustainability.
Conclusions
After adjustment for baseline values and for the characteristics that differed between groups, older adults who received the chatbot-based program showed better health behaviors and blood pressure control than those who received usual care, together with higher knowledge and medication adherence. Adjusted between-group differences in the selected biomarkers were not statistically significant. Because the groups were drawn from different healthcare settings and were not randomly allocated, these results indicate association rather than established efficacy, and the clinical importance of the observed differences requires confirmation. The program is a promising but preliminary approach to supporting hypertension self-management in this population and should now be tested in a randomized trial with longer follow-up.
Acknowledgments
We would like to thank those who contributed to developing the hypertension management program applying a chatbot. A special thanks is extended to the community-dwelling older adult who participated in the study. Lastly, we would like to acknowledge the registered nurses at the two study settings, as well as other healthcare professionals for their cooperation and participation throughout the research process.
Funding Statement
This study was financially supported by the Fundamental Fund of Khon Kaen University, receiving funding from the National Science, Research, and Innovation Fund (NSRF) (Grant Number NSRF69-006). Appreciation is extended to this research fund for making this study possible.
Clinical Trial Registration
The study protocol was registered with the Thai Clinical Trials Registry (TCTR20260118005).
Data Sharing Statement
The deidentified individual participant data underlying the results reported in this article are available from the corresponding author at sbanharak@kku.ac.th on reasonable request from the date of publication. Under the informed consent approved by the Khon Kaen University Office of the Ethics and Human Research Committee, participant records are held under a numeric code on a password-protected computer accessible only to the principal investigator, are reported only in aggregate, and are retained for one year after completion of the study, after which the data and source documents are destroyed. The approved consent permits deidentified data to be placed in a repository specified by the journal for sharing with other researchers, provided that the data cannot identify or be linked to any participant. Each request will accordingly be referred to the Committee, and any release of data or extension of the approved retention period will follow its decision.
Ethical Approval
Ethical approval was obtained from the Office of the Ethics and Human Research Committee, Khon Kaen University (No. HE681006). This study was conducted in accordance with the Declaration of Helsinki.
Consent for Publication
This article does not contain any images, videos, recordings or other material relating to an identifiable individual, and all results are reported in aggregate form. Written informed consent covering participation in the study and the publication of anonymized findings was obtained from every participant before data collection.
Author Contributions
All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work. http://www.icmje.org/recommendations/
Khanisorn Ransinyo:
Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Writing original draft.
Samoraphop Banharak:
Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Supervision, Writing original draft, Writing review and editing.
Ladawan Panpanit:
Conceptualization, Formal analysis, Methodology, Writing review and editing.
Supavadee Thiengtham:
Conceptualization, Formal analysis, Methodology, Writing review and editing.
Pimchaya Visetsittikul:
Conceptualization, Formal analysis, Methodology, Writing review and editing.
Chakkarin Sommana:
Conceptualization, Formal analysis, Methodology, Writing review and editing.
Panita Limpawattana:
Conceptualization, Formal analysis, Methodology, Writing review and editing.
Nada Lukkahatai:
Conceptualization, Formal analysis, Methodology, Writing review and editing.
Sutin Chanabun:
Conceptualization, Formal analysis, Methodology, Validation, Writing review and editing.
Janjira Duangornnam:
Conceptualization, Formal analysis, Methodology, Writing review and editing.
Disclosure
The authors declare that there is no conflict of interest regarding the publication of this study.
References
- 1.American Heart Association. What is high blood pressure? 2024. Available from: https://www.heart.org/en/health-topics/high-blood-pressure/the-facts-about-high-blood-pressure/what-is-high-blood-pressure. Accessed July 1, 2024.
- 2.Mancia G, Kreutz R, Brunström M, et al. 2023 ESH guidelines for the management of arterial hypertension: the task force for the management of arterial hypertension of the European society of hypertension, endorsed by the European society of hypertension and the European renal association. J Hypertens. 2023;41(12):1874–19. doi: 10.1097/HJH.0000000000003480 [DOI] [PubMed] [Google Scholar]
- 3.David GH, Thomas MC, Christopher SW. Pathophysiology of hypertension: the mosaic theory and beyond. Circ Res. 2021;128(7):847–863. doi: 10.1161/CIRCRESAHA.121.318082 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Christopher CK. Hypertension in the Older person: is age just a number? Intern Med J. 2022;52(11):1877–1883. doi: 10.1111/imj.15949 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.World Health Organization. Global report on hypertension: the race against a silent killer. Geneva, Switzerland: World Health Organization; 2023. [Google Scholar]
- 6.NCD Risk Factor Collaboration. Worldwide trends in hypertension prevalence and progress in treatment and control from 1990 to 2019: a pooled analysis of 1201 population-representative studies with 104 million participants. Lancet. 2021;398(10304):957–980. doi: 10.1016/S0140-6736(21)01330-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Benetos A, Petrovic M, Strandberg T. Hypertension management in older and frail older patients. Circ Res. 2019;124(7):1045–1060. doi: 10.1161/CIRCRESAHA.118.313236 [DOI] [PubMed] [Google Scholar]
- 8.Ji H, Kim A, Ebinger JE, et al. Sex differences in blood pressure trajectories over the life course. JAMA Cardiol. 2020;5(3):19–26. doi: 10.1001/jamacardio.2019.5306 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Aekplakorn W, Pakcharoen H, Satheannoppakao W. Thai national health examination survey, 6th, 2019–2020. Nonthaburi: Health Systems Research Institute; 2021. Available from: https://www.hiso.or.th/hiso/picture/reportHealth/report/sreport6/sreport6_full.pdf. Accessed July 29, 2024. [Google Scholar]
- 10.Department of Disease Control. Summary of illness 2024. Strategy and planning division of office of the permanent secretary ministry of public health. 2024. Available from: https://spd.moph.go.th/wp-content/uploads/2024/01/ill_2023_full_30012567.pdf. Accessed July 2, 2024.
- 11.Sakboonyarat B, Rangsin R. Characteristics and clinical outcomes of people with hypertension receiving continuous care in Thailand: a cross-sectional study. Lancet Reg Health Southeast Asia. 2024;21(1):1–17. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Xia T, Zhou F, Nianogo AR. Interventions in hypertension: systematic review and meta-analysis of natural and quasi-experiments. Clin Hypertens. 2022;28(13):1–25. doi: 10.1186/s40885-022-00198-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Almasi S, Hosseini A, Emami H, Sabahi A. Mobile health technology for hypertension management: a systematic review. Acta Med Iran. 2020;58(6):249–259. [Google Scholar]
- 14.Shen Z, Ding S, Shi S, Zhong Z. Association between social support and medication literacy in older adults with hypertension. Front Public Health. 2022;10. doi: 10.3389/fpubh.2022.987526 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Hou G, Anicetus U, He J. How to design font size for older adults: a systematic literature review with a mobile device. Front Psychol. 2022;13(1):1–19. doi: 10.3389/fpsyg.2022.931646 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Kemp S. Social statistics in Thailand 2024. 2024. Available from: https://datareportal.com/reports/digital-2024-thailand. Accessed July 7, 2024.
- 17.Griffin AC, Khairat S, Bailey SC, Chung AE. A chatbot for hypertension self-management support: user-centered design, development, and usability testing. JAMIA Open. 2023;6(3):1–10. doi: 10.1093/jamiaopen/ooad073 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Timsina YP, Pandey P, Mondal IH, Dar AH. Non-pharmacological management of hypertension: a systematic review. Food Chem Adv. 2023;3(1):100406. doi: 10.1016/j.focha.2023.100406 [DOI] [Google Scholar]
- 19.Sattayasomboon T, Sattayasomboon Y. Hypertension and type 2 diabetes mellitus patients: a systematic review on the chronic care model (CCM) and self-management support (SMS). South Coll Netw J Nurs Public Health. 2020;7(2):232–243. [Google Scholar]
- 20.Kanfer FH, Gaelick-Buys L. Self-management methods. In: Kanfer F, Goldtein A, editors. Helping People Change: A Textbook of Methods. 4th ed. New York: Pergamonpress; 1991:305–360. [Google Scholar]
- 21.Kanfer FH. Self-management therapy: orchestration of basic components for individual clients. Eur Psychother. 2000;1(1):10–14. [Google Scholar]
- 22.Briesch AM, Chafoulea SM. Review and analysis of literature on self-management interventions to promote appropriate classroom behaviors (1988–2008). Sch Psychol Q. 2009;24(2):106–118. doi: 10.1037/a0016159 [DOI] [Google Scholar]
- 23.Islam J, Hasan M, Hasan MM. A systematic review of chatbot-enables chronic disease management interventions and mHealth. TechRxiv. 2023. [Google Scholar]
- 24.Sornkestrin A, Thongpetch J, Duangjun J, Rungnoey N. The effectiveness of health promotion and self-management support program on self-management behavior. J Health Sci. 2023;32(2):299–311. [Google Scholar]
- 25.Namrinchuenchitsai S. Knowledge, attitude and behavior of hypertension patients in the area of responsibility of sena-de-lu district health promotion hospital, mae usuu subdistrict, tha song yang district, tak province. 2022.
- 26.Prerprao S, Sriwasom N. Discharge planning model for elderly patients with high blood pressure Khao Wong Hospital, Kalasin Province. J Health Environ Educ. 2023;8(3):482–494. [Google Scholar]
- 27.Somprasong R, Saengcharoen W. Development of the medication adherence measure for oral medicines: testing in patients with hypertension. Thai J Pharm Pract. 2024;16(1):274–282. [Google Scholar]
- 28.Phetmanee B, Amnatsatsu K, Kerdmongkol P, Kittipimpanon K. Effects of self-management support program on self-management behavior and blood pressure among the older adults with uncontrolled hypertension. J Public Health Nurs. 2022;36(1):86–101. [Google Scholar]
- 29.Delavar F, Pashzad S, Negarandeh R. The effects of self-management education tailored to health literacy on medication adherence and blood pressure control among elderly people with primary hypertension. Patient Educ Couns. 2020;103(2):336–342. doi: 10.1016/j.pec.2019.08.028 [DOI] [PubMed] [Google Scholar]
- 30.Bumrungsuk S. Effects of the self-management training program on self-management behavior and blood pressure levels among elderly people with hypertension. Front Nurs. 2022;9(1):71–80. doi: 10.2478/fon-2022-0009 [DOI] [Google Scholar]
- 31.Kim M. Effect of customized long-message service and phone-based health-coaching on elderly people with hypertension. Iran J Public Health. 2019;48(4):655–663. [PMC free article] [PubMed] [Google Scholar]
- 32.Ongkulna K, Pothiban L, Panuthai S, Chintanawat R. Enhancing self-management through geragogy-based education in older adults with uncontrolled hypertension: a randomized controlled trial. Pac Rim Int J Nurs Res. 2022;26(4):690–705. [Google Scholar]
- 33.Echezarra L, Pereira J, Saracho R. TensioBot: a chatbot assistant for self-managed in-house blood pressure checking. J Med Syst. 2021;45(4):54. [DOI] [PubMed] [Google Scholar]
- 34.Sun T, Xu X, Ding Z, et al. Development of a health behavioral digital intervention for patients with hypertension based on an intelligent health promotion system and WeChat: randomized controlled trial. JMIR mHealth uHealth. 2024;12:e53006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Audthiya P, Pothiban L, Panuthai S, Chintanawat R. Enhancing autonomy and self-management behaviors through a patient-centered communication program for older adults with hypertension. Pac Rim Int J Nurs Res. 2021;25(4):525–538. [Google Scholar]
- 36.Sukpattanasrikul S, Monkong S, Leelacharas S, Krairit O, Viwatwongkasem C. Comparison of hypertensive outcomes after the implementation of self-management program for older adults with uncontrolled hypertension in Krabi, Thailand. J Health Res. 2021;36(4):641–651. doi: 10.1108/JHR-12-2020-0626 [DOI] [Google Scholar]
- 37.Kolcu M, Ergun A. Effect of a nurse-led hypertension management program on quality of life, medication adherence and hypertension management in older adults. Geriatr Gerontol Int. 2020;20(12):1182–1189. doi: 10.1111/ggi.14068 [DOI] [PubMed] [Google Scholar]
- 38.Zhang J, Guo L, Mao J, et al. The effects of nursing of Roy adaptation model on the elderly hypertensive. Ann Palliat Med. 2021;10(12):12149–12158. doi: 10.21037/apm-21-2803 [DOI] [PubMed] [Google Scholar]
- 39.Lau D, Ringrose J, McAlister FA, et al. Telemonitoring and protocolized case management for hypertensive community dwelling older adults (TECHNOMED): a randomized controlled trial. J Hypertens. 2022;40(9):1702–1712. doi: 10.1097/HJH.0000000000003202 [DOI] [PubMed] [Google Scholar]
- 40.Soltani S, Arablou T, Jayedi A, Salehi AA. Adherence to the dietary approaches to stop hypertension (DASH) diet in relation to all-cause and cause-specific mortality: a systematic review and dose-response meta-analysis of prospective cohort studies. Nutr J. 2020;19(1):37. doi: 10.1186/s12937-020-00554-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Filippini T, Malavolti M, Whelton PK, Naska A, Orsini N, Vinceti M. Blood pressure effects of sodium reduction: dose-response meta-analysis of experimental studies. Circulation. 2021;143(16):1542–1567. doi: 10.1161/CIRCULATIONAHA.120.050371 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Knight J, Nigam Y. Anatomy and physiology of ageing 1: the cardiovascular system. Nurs Times. 2017;113(2):22–24. [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The deidentified individual participant data underlying the results reported in this article are available from the corresponding author at sbanharak@kku.ac.th on reasonable request from the date of publication. Under the informed consent approved by the Khon Kaen University Office of the Ethics and Human Research Committee, participant records are held under a numeric code on a password-protected computer accessible only to the principal investigator, are reported only in aggregate, and are retained for one year after completion of the study, after which the data and source documents are destroyed. The approved consent permits deidentified data to be placed in a repository specified by the journal for sharing with other researchers, provided that the data cannot identify or be linked to any participant. Each request will accordingly be referred to the Committee, and any release of data or extension of the approved retention period will follow its decision.