Nudge me: tailoring text messages for prescription adherence through N-of-1 interviews

Abstract

Almost 50% of patients with cardiovascular diseases face challenges in taking medications and increased morbidity and mortality. Text messaging may impact medication refill behavior and can be delivered at scale to patients by texting mobile phones. To obtain feedback from persons with chronic conditions on the design of interactive text messages and determine language of message for making messages that can motivate patients to refill medications on time. We purposively sampled 35 English and Spanish speaking patients with at least one chronic condition from three large healthcare delivery systems to participate in N-of-1 video-based synchronous interviews. Research assistants shared ideas for theory-informed text messages with content intended to persuade patients to refill their medication. We transcribed recorded interviews and conducted a content analysis to identify strategies to employ generating a dynamic interactive text message library intended to increase medication refill. Those interviewed were of diverse age and race/ethnicity and typical of persons with multiple chronic conditions. Several participants emphasized that personally tailored and positively framed messages would be more persuasive than generic and/or negative messages. Some patients appreciated humor and messages that could evoke a sense of social support from their providers and rejected the use of emojis. Messages to remind patients to refill medications may facilitate improvements in adherence, which in turn can improve chronic care. Designing messages that are persuasive and can prompt action is feasible and should be considered given the ease with which such messages can be delivered automatically at scale.

Implications.

Practice: Using N-of-1 interviews to create health messaging is a cost-effective and real-time way to create dynamic and impactful text messaging campaigns.

Policy: Research using N-of-1 interviews works to create text message campaigns focusing on medication adherence.

Research: The N-of-1 method is a cost-effective and impactful way to create a message library for texting campaigns in real time.

BACKGROUND

Interventions to improve medication adherence have historically been resource intensive, limiting their generalizability and scalability across large patient populations which has led to small benefits with variable cost-effectiveness, as supported by a 2014 Cochrane review [1]. Medication adherence interventions that rely on tailored support from allied health professionals, commonly a pharmacist can be costly, difficult to replicate and scale, and produce only variable and often small effects [2]. Other approaches to increase medication adherence focus on using automated telephone communication systems to deliver voice messages and collect health-related information from patients using either the telephone touch-tone keypad or voice recognition software [3]. While this strategy is less intensive and reduces cost, it still produces only modest improvement in adherence to statin medications and small reductions in LDL cholesterol. Interventions comparing strategies to facilitate medication reminders such as a pill bottle strip with toggles, digital timer cap, or standard pillbox do not demonstrate a significant difference in adherence [4]. Using N-of-1 interviews we plan on creating meaningful and impactful text messages that not only encourage the behavior of medication adherence but can be utilized and cost effective scalable across multiple health care systems.

The increasing popularity and widespread use of mobile phones provides a great opportunity for researchers and health practitioners to find solutions that can capitalize on mobile technologies to deliver more complex health behavior change interventions at scale. Text message-based interventions offer users a convenient and relatively unobtrusive way to receive health information, enable researchers to design customized messages based on user’s needs and individual characteristics, serve as an interactive platform facilitating communication between users and health practitioners, and are useful in collecting real-time data on a large number of participants. There has not been a study to compare genetic text versus more complex tailored text messaging to change behavior.

Behavioral interventions delivered by text messages have been found effective in promoting a variety of health behaviors, including diet, weight loss, medication adherence, diabetes self-management, and smoking cessation and evidence demonstrates that personalized text messages have greater effects than those that do not [5–7]. The recent meta-analysis of mobile telephone text messaging specific to medication adherence for chronic diseases showed a two-fold increase in medication adherence served as a foundation to build the text messaging resource that patients would respond to via text message which we describe here [8]. While these meta analyses demonstrate the potential for text messaging interventions, the underlying studies were markedly heterogeneous, leaving questions about the best strategies for message design, the impact of generic vs. tailored approaches, optimal message timing and intensity and whether bi-directional messaging is useful. Authors of both reviews concluded the results should be interpreted with caution given the short duration of studies and reliance on self-reported measures. These data on the influence of text messaging interventions are cause for optimism. However, it is not yet known how this type of intervention might be optimized at scale and whether doing so will improve outcomes.

METHODS

Patient recruitment and enrollment

Participants were patients in one of three large health care systems in the Rocky Mountain region of the USA, including a Veteran’s Administration hospital (VA), a large urban safety-net hospital (Denver Health, DH) and an urban hospital serving a mix of private and publicly insured patients (University of Colorado, UCH). Patients eligible for the study had one or more of the following cardiovascular conditions (hypertension, hyperlipidemia, diabetes, coronary artery disease, and/or atrial fibrillation) and with a prescription for one or more of the classes of medications to treat the cardiovascular conditions (b-blockers, calcium channel blockers, angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, HMG-CoA reductase inhibitor (statins), thiazide diuretics, PGY-2 inhibitors (anti-platelets), direct oral anticoagulants, oral diabetes medications). Exclusions included patients who: (i) have neither a landline or cellphone; or (ii) were enrolled in hospice or palliative care; or (iii) were non-English or Spanish speaking; or (iv) were enrolled in another clinical trial if denoted in the electronic health record (EHR). We identified patients with a refill gap of at least 7 days within the past year utilizing pharmacy refill data obtained from EHRs from each of the three hospital systems partnering on this research. We specifically sought patients from the following patient groups: (i) Age > and <= 65 years of age; (ii) male and female patients; (iii) one versus multiple cardiovascular condition of interest; and (4) English- and Spanish-speaking patients.

We purposively sampled 35 participants to help review and generate dynamic and engaging English and Spanish text-messages through multiple N-of-1 (i.e., within subject) assessments that conform to evidence-based strategies for persuasive message design. This approach is ideal for rapidly iterating a user informed program with input from a range of stakeholders, as it offers a way to quickly respond and iterate new versions of messages until consensus across participants can be reached, and an opportunity to expose participants to multiple messages and understand whether and the extent to which each intervention may be superior to usual care [9, 10].

Once potential participants were identified, we sent them a letter informing them of the study and their eligibility and asked them to opt-out of participation if they did not want to be contacted. If they did not return the opt out card within two weeks, we included them in a database of potential participants to contact and recruit for this N-of-1 investigation of text message content. We telephoned potential participants and invited their participation in a 1-hr qualitative interview held in person at a clinic location in one of the health care sites or held over a video-conferencing platform. During the initial phone call patients were asked if they preferred to meet in person or if they would like to utilize video-conferencing. For the video-conference, a PowerPoint file was sent to the patient via email and ensured that the patient was comfortable accessing the file themselves for the interview. In person, visits were held in a private clinic room with the same power point presentation. The power point was used as a tool to graphically present slides of potential text messages (see Fig. 1). Research assistants also went to the clinic location and recruited patients following routine visits; the same options of in person interviews or video conferences were offered at this time. All research visits were audio recorded for further analysis. We approached 79 patients and invited them to participate in interviews with research assistants either in person or via phone (31 at DH 34 at UCH 14 at the VA). There were 35 patients who agreed and 35 (14 at UCH, 8 at DH, 13 at VA) completed interviews. Reasons for declining included time commitment and not feeling comfortable with text messaging for health care related communication.

Fig 1.

An example of one slide used in the N of 1 interviews.

Implementation

We created multiple options for text message content as an a-priori text messaging library related to three types of messages: (i) increasing norms for adherence, (ii) commitment to medication adherence, and (iii) salience of messages related to adherence in order to increase medication refills. As a research team we created variations on messages per our theoretical framework to utilize narrative, humor, practicality, etc., that participants could react to. Sample a-priori messages are shown in Fig. 1. Typically, in one session, a participant would comment on their preference for messages among 10 sets of two or three messages discussing what they liked or disliked in each message.

In each N-of-1 interview, a study professional research assistant shared multiple versions of text messages and elicited reactions to content. They documented preferences for specific message versions and suggestions to change or alter content for each participant. Thus, for each message, participants would iteratively declare preferences for a specific message version. We recorded preferences within and across participants.

Analysis

Data were analyzed using an iterative content analysis. After completing three to five interviews, the study team would convene and discuss the findings to identify themes and patterns. Where there was consensus about message preference or content, the team would make changes and conduct subsequent interviews using the changed and updated messages. Figure 2 shows the tracking of comments as well as totally of positive and negative responses to each message. They repeated this across five iterations, at which time we determined that we reached a saturation of information such that participants were no longer generating new suggestions for messaging, nor were they identifying new reasons for preferring particular messages. We then created a final library of messages that represented shared preferences for content and language.

Fig 2.

How each message was tracked and comments documented following each N of 1 interview.

We documented participant demographics (Table 1) to ensure an accurate reflection of the populations at each of the three hospital systems. When research assistants observed recruitment of more men than expected we increased efforts to recruit more women. Patients at these clinics were older but researchers made an effort to recruit younger patients as well because in initial interviews we observed differences in text message preferences between age groups with younger patients using more abbreviations and shorter messages while older patients preferred the longer narrative messages.

Table 1.

Demographics of participants

	N = 35
Race and ethnicity
White	20 (57%)
African American	10 (28%)
Asian	1 (2%)
Hispanic/Latino	5 (14%)
Reported gender
Male	22 (62%)
Female	13 (37%)
Age
<50	5 (14%)
50–59	11 (31%)
60–69	11 (31%)
>70	7 (20%)

RESULTS

Almost two thirds of participants were male, and over half were white, with 28% identifying as African American and 14% as Latino. Just over 30% of the sample was aged 50–59 and another 31% 60–69, typical of persons with multiple chronic conditions (Table 1).

Messages evolved as each group of interview feedback was analyzed and feedback was incorporated into the message library. Table 2 shows the evolution of messages. A number of messages started with using emojis but both the younger and older interviews disliked the use of these icons; younger participants shared that it seemed like researchers were “trying too hard” and older participants told researchers that they get confused and frustrated with emojis. Messages that brought up the topic of death resulted in strong reactions often negative so wording was changed to be more positive around avoiding hospital stays rather than bringing up the possibility of death. We used generic names of people to normalize behavior but the names caused confusion, with participants asking who those people were rather than thinking of the message as a normalizing of behavior. To address this, we replaced the specific name with a general “your neighbor” but patients did not like having other people involved, especially associated with health behaviors; many people responded that the responsibility was on them so they wanted the messages to only address themselves. One message used a popular sports slogan but during the process of the interviews that slogan became politicized; we removed that message wording completely.

Table 2.

Examples of message evolution

Original	Intermediate	Final
Tell us your best strategy to make getting refills a habit! Text 1 = set my alarm; 2 = rely on my family; 3 = make it part of my weekly routine; 4 = other or unknown.	We noticed you didn’t refill some of your meds. Tell us why! Text 1 = too expensive; 2 = I forgot; 3 = I don’t like taking them; 4 = Other.	Hi (FIRST NAME) We noticed you haven’t refilled your (DRUG NAME). Reply 1 = you’ll get them refilled in the next 2 days 2 = I’m still working on a plan to get this done.
Hi, me again. I know I’m needy, but I’d feel better if you refilled your meds [TS: Inline image present. Retain.].	Your pharmacist misses you!	Hi (FIRST NAME) It’s easy to forget to get your meds – that’s what we’re here for! Reply 1 = I have a plan to get your prescription Reply 2 = I’ll get to it later this week.
Joe always remembers his meds – he makes a habit of going every Friday since the pharmacy is right near his favorite menudo spot! Make a healthy habit by planning your regular medication pick up.	Your neighbor always remembers their meds – they make a habit of going every Friday since the pharmacy is right near their favorite menudo spot! Make a healthy habit by planning your regular medication pick up.	Hi (FIRST NAME) I care about my well-being. I will get to the pharmacy by: Reply 1 = I’ll do it today! Reply 2 = I’ll do it later this week.
Lots of people have bad side effect from their meds. The side effects from NOT taking them (example: death) can be much worse.	Lots of people have bad side effect from their meds. The side effects from NOT taking them can be much worse.	Hi FIRST NAME Patients like you often take a number of medications, but are happy they help them stay out of the hospital and at home. Can you commit to refilling your meds today? Reply 1 = YES I can commit today. Reply 2 = I will commit later this week.
Myth or Fact: heart disease is the leading cause of death in Colorado. Text 1=True, 2=False.	Fact: heart disease is a leading cause of death in the state of Colorado.	Message deleted due to confusion and strongly held believes around heart disease.
Just do it.	Studies show that when you make a promise you are more likely to follow through. Make a promise to pick up your meds today.	Studies show that when you make a commitment you are more likely to follow through. Make a commitment to pick up your meds today.

DISCUSSION

How can we optimize messages so that they can help facilitate an improvement in medication adherence when delivered at scale? We know from decades of research in health behavior that when we explicitly include content in interventions shaped by social and behavioral science theory, we can realize greater program effects [11]. Moreover, there are key theoretical constructs that have been consistently shown to be effective when used in the design of health promotion and chronic illness self-management interventions, specifically, increasing knowledge, shaping attitudes, building skills and supporting healthy social norms [12, 13]. Designing this type of information in a text-message can be challenging given their obligatory brevity. One perspective to facilitate development of text messages that can improve knowledge, attitudes, norms and skills in a brief format is to consider behavioral “nudges” from the fields of behavioral economics and cognitive psychology. A nudge is defined as a small change that “alters people’s behavior in a predictable way without forbidding any options or significantly changing their economic incentives”; people make decisions either “intuitively,” quickly drawing on emotion and past experiences or “reasonably” using a thoughtful, analytic approach [14]. Nudges take advantage of the intuitive aspects of decision-making. A technology-delivered nudge should positively influence individuals’ behaviors through the use of non-intrusive education, social norm setting, and reciprocity expectation [15]. Three types of nudge interventions supported by prior literature that could be feasibly implemented as text messages within the context of medication adherence include (i) Communicating social norms that activate and guide behavior in positive ways when a message normalizes positive behaviors (e.g., using the stairs vs. elevators); (ii) eliciting behavioral commitments, such as committing to filling one’s prescription, shown to be effective at improving a range of behaviors, including judicious use of antibiotics among clinicians; and (iii) relying on narrative stories to increase vividness and comprehension of medical outcomes, which may be particularly effective at helping patients concretely understand potential risks of non-adherence, spurring them to take action (improving medication adherence) to prevent negative outcomes [6, 7, 16, 17].

Even if we are able to generate content using the behavioral nudge framework, we must also compete with a deluge of information and distractions that phone users have on a daily basis to make text messages engaging so participants will pay attention, act on and derive intended benefit. Persuasive communication strategies with demonstrated efficacy for engaging participants in the social media realm that are potentially adaptable for text-messaging include: (i) messages that convey the sender’s status of being intelligent or wise, (ii) evoke emotions, (iii) trigger a reflexive response, (iv) increase identity with a group or community, and (v) use a narrative structure [18]. Health communication theory further considers the importance of framing messages with a positive or negative outcome (gain/loss framing) and of helping people process messages so they become more identified with the outcome [19, 20]. Critical to this process is the consideration of message design to maximize access, which requires attention to user literacy and numeracy. In order to keep patients engaged during the next phase of research the messages will be evaluated yearly and updated as needed from the library of messages created during this initial process. The need to keep patients engaged is certainly important, and we will be assessing engagement with messages regularly with the intent to update and adapt messages to include those that resonate the most in the library moving forward.

The Integrated Theory of mHealth pulls all these theoretical elements together, suggesting that for mobile and digital health interventions we must simultaneously consider integrating traditional theory into programs that can increase access to content and employ know strategies to increase engagement with content [21, 22]. We adapted this framework (see Fig. 3) to guide development of behavioral nudge messages to support adherence with medication refills.

Fig 3.

Integrated theory of mHealth.

Although text messaging has been used with positive effect to influence medication adherence [23–25] we know of no systematic effort to design theory-based, engaging and persuasive messages for a texting environment. Our hypothesis is that using behavioral nudges delivered via text messages can offer greater benefit than more generic messaging for medication refill behavior [25], and that these tailored and optimized messages can then be delivered at scale to many thousands of patients across health care systems. In order to test this hypothesis, we first had to develop these optimized messages. This study reports on qualitative efforts to design optimized message content to support adherence to cardiovascular medication refills that are now being evaluated in a large-scale randomized trial to assess their impact in comparison to generic text messages.

Creating text messages that are optimized for specific populations targeted for an intervention may continue to improve medication refill adherence and is being testing in our current trial. Medication refill adherence is an attainable way to increase health outcomes by empowering patients to take control. By utilizing text messaging this intervention can be delivered to a maximum number of patients through an already ubiquitous technology that is commonly used. This low barrier approach gives patients a familiar way to interact with their health care throughout their daily routines. With text messaging, patients do not have to remember passwords to access messages and can easily respond as well. By tailoring messages we anticipate patients have more incentive both to interact with messages that resonate for them and remember messages even in the face of all the interactions they have with their mobile device on a daily basis. We hypothesize that tailored messages that resonate will facilitate a commitment to medication refill behavior. This N-of-1 approach can easily be replicated to adapt and create optimized messages for different populations or different medication adherence goals. The ability to change and adapt messages quickly based on feedback and then continue a process to refine messages based on feedback can be used as many times as necessary in as many different contexts as needed.

We are currently engaged in evaluating the efficacy of these messages for improved medication adherence across three healthcare delivery systems in a randomized controlled pragmatic trial. We have successfully identified patients with chronic cardiovascular conditions taking medications to treat hypertension, atrial fibrillation, coronary artery disease, diabetes and/or hyperlipidemia within each of our three health systems. We have leveraged real-time pharmacy refill data to identify episodes of non-adherence through gaps in medication refills. Eligible patients with medication refill gaps are randomized into the following study arms: standard care with no messages, generic text message reminder; text message behavioral nudges only; or text message behavioral nudge plus a pre-programmed artificially intelligent interactive chatbot designed to identify and resolve barriers to medication refill and adherence. When a patient has a seven-day gap from when they should have refilled a medication based on their EHR they will start receiving the intervention. Messages are sent via text to their cell phones on day 1, 3, 5, 7, and 10 after a gap in refilling medications is documented. If a patient does not have a mobile number or if the message fails to send as a text the patient will receive the same message via an interactive voice response robotic message. The intervention will continue to follow and send messages to patients as they demonstrate delays in refilling medications for the next two years at which point analysis and dissemination will continue.

Funding: This study was funded by the National Institutes of Health (NIH) Health Care Systems Research Collaboratory by cooperative agreement UG3HLD144163 from the National Heart, Lung, and Blood Institute. This work also received logistical and technical support from the NIH Collaboratory Coordinating Center through cooperative agreement U24AT009676. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Compliance with Ethical Standards

Authors’ Statement of Conflict of Interest and Adherence to Ethical Standards: Michael Ho is supported by grants from NHLBI, VA HSR&D, and University of Colorado School of Medicine. He has a research agreement with Bristol-Myers Squibb through the University of Colorado. He serves as the Deputy Editor for Circulation: Cardiovascular Quality and Outcomes. Nothing to disclose: Joy Waughtal; Phat Luong; Lisa M Sandy; Sheana Bull.

Human Rights: “All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.” COMIRB # 18–0630.

Informed Consent: Informed consent was obtained from all individual participants included in the study.

Welfare of Animals: This article does not contain any studies with animals performed by any of the authors.