An adaptive text message intervention to promote well-being and health behavior adherence for patients with cardiovascular disease: intervention design and preliminary results

Abstract

Background:

Most individuals with heart disease struggle to adhere to cardiovascular health behaviors, despite their known health benefits. Text message interventions (TMIs) are a promising treatment modality for health behavior promotion, but existing TMIs typically deliver a fixed set of messages and do not target well-being constructs associated with adherence and cardiovascular health.

Methods:

A four-week TMI, which delivers daily messages to promote well-being and adherence to health behaviors and dynamically adapts based on participant feedback to deliver increasingly customized messages, was developed by the study team. Then, its feasibility, acceptability, and preliminary efficacy were assessed in a single-arm, proof-of-concept trial in 14 individuals with coronary artery disease (age M = 67.9, SD = 8.7). Participants received daily text messages related to well-being, physical activity, or diet, rated each message’s utility, and these ratings informed the TMI’s choice of future text messages. Feasibility was assessed by the proportion of messages successfully sent, and acceptability was assessed by participant ratings of intervention burden and text message utility. Finally, the intervention’s preliminary efficacy was explored by measuring pre-post changes in psychological and behavioral outcomes.

Results:

The TMI was both feasible (93% of participants received all messages) and well-accepted (mean text message utility: 7.0/10 [SD 2.5]; mean intervention utility: 6.4/10 [SD 0.9]; mean intervention burden: 0.5/10 [SD 0.9]). Participants reported that messages related to well-being were particularly helpful and that most messages led to an action (e.g., eating more vegetables, being kind to others). The TMI led to non-significant, small-to-medium effect size improvements in happiness, optimism, determination, depression, anxiety, self-rated health, and diet (d = .19 to .48), and, unexpectedly, small reductions in activity and physical function (d = −.20 and −.32).

Discussion:

The adaptive TMI was feasible, well-accepted, and associated with non-significant improvements in psychological outcomes and mixed effects on behavioral outcomes. Larger, well-powered studies are needed to determine whether this TMI is able to improve well-being and health-related outcomes in this high-risk population.

Introduction

Among the 120 million Americans with cardiovascular disease (1), adherence to health behaviors (e.g., heart-healthy diet, physical activity) is associated with a reduced risk of cardiovascular disease progression and complications (2). Despite these benefits, a majority of patients with heart disease struggle to adhere to recommended guidelines for health behaviors (3, 4). While multicomponent interventions (e.g., cardiac rehabilitation) to promote health behavior adherence can be beneficial (5), they are resource-intensive, poorly attended (6), and not available in many areas of the U.S. (7, 8). Furthermore, existing health behavior interventions often do not address psychological factors (e.g., low self-efficacy, reduced well-being, depression) that are independently associated with adherence and cardiovascular health outcomes (9).

Text message interventions (TMIs) address many of the limitations of existing health behavior programs. TMIs are highly accessible (i.e., 96% of Americans have access to a cell phone (10)) and can be implemented with fewer resources than mobile application-based or in-person programs. TMIs also can be customized to individual participants (e.g., based on health behavior goals, personal preferences, and medical factors), which can have substantial benefits, as personalized health interventions tend to be more effective than generic, non-personalized ones (11, 12). Finally, TMIs can also deliver content that focuses specifically on the promotion of well-being constructs (e.g., optimism, positive affect) that have been linked to cardiovascular health (9).

Existing health behavior TMIs, most of which deliver a fixed set of educational or motivational text messages, lead to small- to medium-effect-sized beneficial effects on adherence in persons with heart disease and other populations (12–14). Adaptive TMIs, which select text messages based on regular, ongoing patient feedback, have the potential to have even greater effects on adherence and subsequent health outcomes than interventions customized based solely on static factors. However, to our knowledge, adaptive TMIs that deliver health behavior and well-being-focused messages have never been studied in patients with heart disease.

To address this gap, a four-week, adaptive TMI that delivers daily messages to promote well-being and adherence to cardiac health behaviors was designed. In this manuscript, the development of this TMI is described, and the results of a single-arm, proof-of-concept trial to assess the feasibility, acceptability, and preliminary efficacy of this novel TMI in patients with coronary artery disease are reported.

Material and methods

Intervention design

Text messages

Two-hundred fifty-four text messages were created and labeled with attributes based on their primary topic (well-being, physical activity, or diet), directiveness (educational or directive), and the presence of social content (yes/no) (see Appendix 1). Health behavior messages were based on the American Heart Association’s impact goals for cardiovascular health promotion (15), while well-being messages were based largely on the literature related to positive psychology (16), a field that utilizes psychological activities (e.g., using strengths, expressing gratitude) to promote well-being-related constructs (17).

Learning algorithm

Participants received one text message each day for 28 days and were asked to rate its utility (helpfulness) on a 0–10 Likert scale. During the first 14 days, a fixed set of messages with a variety of attributes were delivered to obtain initial participant preferences. Then, daily adaptation began through an algorithm that weighted the participants’ utility ratings from all previous messages (see Figure 1).

Figure 1.

Text message algorithm

* Utility ratings from participants were transformed from a 0–10 scale to a −5 to +5 scale, so that scores below 5 reduced the likelihood of receiving messages with an attribute, and scores above 5 increased the likelihood of receiving messages with that attribute. In this example, the participant’s rating of 8 was transformed to a score of +3.

Each day, the learning algorithm reviewed each participant’s ratings for all prior messages and calculated an overall utility rating for the attributes associated with those messages. The algorithm then selected a set of messages for potential delivery that day, in which 70% were chosen based on the combination of attributes that the participant rated most highly, and 30% were chosen at random (to prevent the algorithm from learning too quickly and to ensure that participants continued to receive a variety of message types). Then, one of these messages was randomly chosen for delivery.

The adaptive algorithm was maintained and executed in a secure, online DynamoDB database (18) that was hosted on Amazon Web Services (19). All text messages were sent and received using Twilio (20), a secure text messaging platform.

Proof-of-concept trial

The feasibility, acceptability and preliminary efficacy of the intervention then was assessed in a single-arm, proof-of-concept trial. These platforms all utilize security measures to maintain participant confidentiality, and the procedures for this study were reviewed and approved by the institution’s Research Computing Compliance team.

Participants

Participants were eligible if they had a history of acute coronary syndrome (ACS) (21) and access to a cellular phone that could receive text messages. Exclusion criteria included cognitive impairment (22), inability to communicate in English, a medical condition precluding activity, or terminal illness. The study was registered prospectively on ClinicalTrials.gov (NCT04077229), and study procedures were approved by the Partners Healthcare Institutional Review Board (IRB; #2016P001472) prior to initiation of recruitment. All contact with participants was carried out remotely. All patients provided written informed consent prior to participation.

Measures

Feasibility, acceptability, and participants’ feedback

The intervention’s feasibility and acceptability were considered the study’s primary outcomes. Feasibility was measured by the percentage of text messages that were successfully delivered, through a review of the DynamoDB database and confirmation with participants at the 4-week follow-up assessment.

Intervention acceptability was measured by mean ratings of the intervention burden on a 0–10 Likert scale (0=‘not burdensome at all’, to 1=‘extremely burdensome’). A priori, feasibility was defined as successful delivery of over 90% of planned messages and acceptability as a mean burden score of < 4/10.

Additionally, as secondary outcomes, throughout the TMI program, participants rated the utility of each individual text message (0=‘not helpful at all’ to 10=‘very helpful’) on a daily basis, and preferences for specific message types were assessed by calculating the mean utility rating of all messages delivered with that attribute. Following the intervention, participants rated the intervention’s utility (using the same 0–10 Likert scale).

Finally, participants provided feedback of the intervention through six open-ended questions (see Table 1) related to the intervention’s burden, utility, and impact, along with their preferences for future intervention delivery.

Table 1.

Participants’ opinion about the intervention

Measure	Question	Responses	N (%)
Burden	Were the texts bothersome or annoying?	Yes	N = 0 (0.00 %)
		No	N = 12 (85.71%)
		Sometimes	N = 2 (14.29%)
Usefulness	Were they helpful?	Yes	N = 12 (85.71%)
		No	N = 2 (14.29%)
Relative usefulness	Were some more helpful than others?	PA	N = 3 (21.43%)
		PA and diet	N = 4 (28.57%)
		Diet	N = 0 (0.00 %)
		Well-being	N = 6 (42.86%)
		All equally helpful	N = 1 (7.14%)
Impact	Did a text ever specifically lead to an action (e.g., activity, a kind act)?	Yes	N = 10 (71.43%)
		No	N = 4 (28.57%)
Frequency	Were they not frequent enough? Too frequent? Just right?	Just right	N = 14 (100%)
Duration	How long would such a program be useful? Would it be useful to have occasional reminders?	Just right	N = 8 (57.14%)
		Longer	N = 5 (35.71%)
		NR/DK	N = 1 (7.14%)

Notes: NR/DK = No response / Do not know; PA = physical activity.

Preliminary efficacy

The preliminary impact of the intervention was assessed through pre-post changes in psychological and health-related measures from baseline to 4- and 8-week timepoints. Consistent with previous studies (23, 24), psychological outcomes were assessed through participant ratings of happiness, optimism, determination, depression, and anxiety over the past week using a 0–10 Likert-style scale for each item/construct. Self-rated health similarly was assessed by a Likert-scale item (1=‘very good’, to 5=‘very bad’) about their general health. Moderate physical activity was assessed as the number of minutes of moderate activity per week, an approach that has been validated and used in clinical practice (25). Dietary adherence was measured with the diet item from the Medical Outcomes Study Specific Adherence Scale (26), and functional status was measured with the well-validated Duke Activity Status Index (27).

Data analysis

Statistical analyses were carried out using Stata (version 15.1; College Station, TX). The proportion of messages received was calculated to assess feasibility, and descriptive statistics (mean, SD) were utilized to describe the burden and utility of the intervention and individual messages. To compare the mean score of participants’ perceived utility of each type of message (well-being, PA or diet), the Kruskal-Wallis H test was used. Regarding feedback related to the intervention, two study team members (AC and CC) systematically reviewed participants’ responses to the six structured questions in the exit interviews, derived themes from this raw data, and then reported the results using descriptive statistics. Paired t-tests were performed to measure the preliminary impact of the intervention (from baseline to post-intervention [4 weeks], and from baseline to follow-up [8 weeks]) on psychological and health-related outcomes. Given the small sample size and preliminary nature of this study, rather than focus on statistical significance, effect size was examined to explore the magnitude of the intervention’s effect. Effect sizes (Cohen’s d) were calculated by subtracting the pre-intervention mean from the post-intervention mean (or follow-up intervention mean), divided by the pooled standard deviation of the pre- and post-intervention assessment (or pre-intervention and follow-up assessment) (28).

Results

Participants

A total of 35 participants were screened for the study, and 2 were excluded (lack of interest, limited mobility; see Figure 2). Consent forms were sent to 33 individuals, 17 enrolled, 16 completed the baseline measures, and 15 started the TMI. Of those 15, 5 (33.3%) were women, all were non-Latino White, and they had a mean age of 67.9 (SD 8.7) years. Follow-up information was obtained from 13 participants at 4 weeks and 14 participants at 8 weeks (see Figure 2).

Figure 2.

Flow of participants

Dr. Celano has received salary support from BioXcel Pharmaceuticals and honoraria for talks to Sunovion Pharmaceuticals on topics unrelated to this research. The authors report no proprietary or commercial interest in any product mentioned or concept discussed in this article.

Feasibility, acceptability, and participants’ feedback

All participants except for one (N=14) received all 28 text messages (successful delivery = 93.3%), and 344 out of the 392 sent messages received a rating from participants (87.8%). Participants rated the intervention as minimally burdensome (M = 0.5, SD = 0.9) and moderately useful (M = 6.4, SD = 0.9)

Similarly, participants rated individual text messages as being moderately useful (M = 7.0, SD = 2.5). On analysis via the Kruskal-Wallis H test, there was a statistically significant difference in the utility ratings across the different message types (χ²(2) = 21.471, p < 0.001). Pairwise comparisons with Bonferroni-adjusted p-values revealed that participants rated the utility of well-being messages (M = 7.7, SD = 2.5) more highly than diet (M = 6.4, SD = 2.6, p < .001, r = −4.08) and physical activity messages (M = 6.7, SD = 2.3, p < .001, r = −3.90).

In exit interviews (see Table 1), 86% of participants reported that the text message intervention was helpful, and none found the text messages to be bothersome. When asked about individual preferences for specific message types, 43% of individuals identified the well-being text messages as being most helpful, followed by diet and physical activity messages (29%). Most participants (71%) reported that the text messages had a direct impact that led to an action (e.g., being kind to others, taking a walk, eating more vegetables), and all considered the frequency of text message delivery to be appropriate. Regarding intervention duration, 57% of participants liked the 4-week duration, while 36% felt that it should be longer.

Preliminary efficacy

The intervention led to small- to medium-sized pre-post improvements in happiness, optimism, determination, depression, and anxiety (Cohen’s d range from .19 to .38) and moderate-sized improvements in self-rated health and dietary adherence (d range from .45 to .48) from baseline to Week 4 (see Table 2). In contrast, physical activity and function scores showed small decreases from baseline to Week 4 (d = −.20 and d = −.32, respectively). At 8 weeks, the impact of the intervention was somewhat attenuated. While participants maintained a change in optimism at 8 weeks (d = .51), the intervention’s impact on other psychological and functional outcomes was reduced at that time point (d range from 0 to .22) (See Table 2). No statistically significant differences emerged in the paired t-test analyses in any of the variables and assessment times.

Table 2.

Means (M), standard deviations (SD), t-test results and effect sizes (d) from baseline to post-intervention (4 weeks), and from baseline to follow-up (8 weeks)

			Pre-post intervention (4 weeks)					Follow-up (8 weeks)
Measure	N	Baseline	N	Δ1	t	p	d	N	Δ2	t	p	d
Happiness	16	7.6 (2.2)	13	0.5	1.15	.28	0.32	14	0.4	0.73	.48	0.22
Optimism	16	7.1 (1.7)	13	0.5	1.10	.29	0.33	14	0.7	1.38	.19	0.51
Determination	16	7.5 (2.1)	13	0.5	0.55	.59	0.19	14	0.1	0.00	1.0	0.00
Depression	16	3.0 (2.4)	13	−0.6	−1.11	.29	−0.28	14	−0.1	0.00	1.0	0.00
Anxiety	16	3.4 (1.9)	13	−0.8	−1.18	.26	−0.38	14	−0.6	−0.62	.55	−0.23
Physical activity	16	255.0 (292.5)	13	−99.2	−0.88	.40	−0.32	14	−78.9	−1.56	.14	−0.37
Dietary adherence	16	3.3 (1.9)	13	0.8	1.68	.12	0.45	14	0.5	1.30	.22	0.33
Health	16	1.9 (0.6)	13	0.4	1.81	.10	0.48	14	0.2	0.76	.46	0.27
Activity status	16	46.3 (13.1)	13	−3.5	1.18	.26	−0.20	14	−2.7	−0.76	.46	−0.20

Notes: Δ₁ = mean difference (post-intervention – baseline); Δ₂ = mean difference (follow-up – baseline).

Discussion

This manuscript details the successful creation of an adaptive TMI to promote well-being and health behavior adherence in patients with coronary artery disease. To our knowledge, this is the first work that presents a TMI designed to promote well-being and health behavior adherence for patients with cardiovascular disease that dynamically adapts to participants’ preferences. In addition, the proof-of-concept trial showed that the intervention was both feasible and well-accepted. More than 90% of participants received all messages, and they rated the TMI as moderately helpful and not burdensome, in line with other health TMI studies (11, 29, 30). In addition, the text message frequency (i.e., daily) and intervention duration (4 weeks) were appreciated by most participants, and most expressed that the messages led to at least one direct action (e.g., activity, kind act).

In both message ratings and exit interviews, participants’ identified well-being messages as being particularly helpful. These messages were most highly rated in terms of utility, and in exit interviews participants reported them as being most helpful. Additionally, the TMI led to small to moderate, non-significant increases in happiness, optimism, and determination, and non-significant reductions in depression and anxiety levels. Though not statistically significant, the magnitude of the intervention’s effects are consistent with recent meta-analyses on the efficacy of positive psychology interventions, which incorporate specific activities designed to increase positive emotions, cognitions or behaviors (17, 31). This trial suggests that positive activities can be delivered effectively using text messages and raises the possibility of incorporating text messages into more formal positive psychology interventions or attempting to deliver positive psychology interventions exclusively using text messages. Further research may clarify how to utilize this technology most effectively in the future.

Previous studies have shown that TMIs can increase engagement in health behaviors, including physical activity, a heart-healthy diet, and medication adherence (12, 14). In this study, results on health-related outcomes were mixed: while the intervention led to non-significant, small-to-moderate effect-size increases in self-reported dietary adherence and health, it led to small, non-significant reductions in physical activity at follow-up. It is difficult to identify a clear reason for the reduction in activity, which was contrary to what was expected and inconsistent with prior studies (14). Possible reasons for this finding includes the lack of consistent focus on activity (e.g., only 38% of delivered messages were related to physical activity), the short duration of the intervention, and the lack of coaching from study staff, as coaching may be an important adjunct to text messages to promote behavior change and improve health outcomes (32). It also is possible that the intervention was not personalized or specific enough to impact this important outcome.

This study has several limitations. It has no control group, a small sample, and a homogeneous population, which makes it difficult to generalize the results to other populations or to establish causal relationships among the variables. Additionally, a substantial portion of eligible participants (48%) did not return the consent form, and it is unclear if these individuals were different from those who ultimately enrolled. Finally, psychological outcomes were measured using single-item Likert-style ratings, rather than validated questionnaires. The effects obtained in this initial trial should therefore be considered with caution. A next step will be to investigate whether these results can result in psychological and health behavior outcomes in a randomized study with adequate power to detect such effects and allow comparison to a control condition.

Conclusions

This first approach showed promising results worth highlighting. The TMI was feasible and well accepted, and nearly 90% of messages received feedback, highlighting the willingness of participants to engage with the program. Additionally, most participants had a favorable opinion about its content and duration, and over 70% expressed that the messages led to specific actions and health behaviors. Well-being messages in particular were rated as helpful by a great number of participants, which supports the potential benefits of including this type of content in health behavior interventions. If effective in larger, controlled studies, this highly scalable intervention has the potential to improve well-being and health-related outcomes in patients with heart disease.