Adjuvant endocrine therapy (AET) is routinely prescribed for 5 to 10 years to women with early-stage estrogen receptor–positive (ER+) breast cancer to reduce recurrence and mortality [1-4]. It is an oral tablet taken once daily. Despite its benefits, up to three-fourths (75%) of women do not take AET as prescribed [5-7], with up to half of women discontinuing it within 5 years [6,8]. Nonadherence to AET increases the risk of breast cancer–related recurrence and mortality and increases health care costs [4,9,10].

Nonadherence to AET can be intentional or unintentional. Intentional reasons for nonadherence include difficulties with side effects, concerns about AET, and psychological distress [11-15]. Unintentional reasons include forgetfulness, being away from home, and difficulty refilling a prescription [16]. Unintentional nonadherence is reported more frequently than intentional nonadherence in women with breast cancer taking AET [16,17]. Subjective cognitive impairment, including impaired memory, is commonly reported by breast cancer survivors following chemotherapy [18,19], which may contribute to unintentional nonadherence [7,12,20,21].

A Cochrane review of 182 randomized controlled trials on medication adherence interventions across multiple long-term conditions highlighted the need for more effective and novel interventions to support medication adherence [22]. Existing interventions supporting AET adherence tend to focus on educating women about AET and breast cancer and often solely use written information [23-26]. Most of these interventions do not focus on unintentional barriers to adherence, despite their prevalence [26]. Most existing interventions have been minimally effective in improving adherence [23-26].

Making medication-taking behaviors habitual could address unintentional nonadherence, as habits reduce reliance on memory. A habitual behavior is an action triggered by exposure to a contextual cue [27]. Habitual behaviors are learned through a process of context-consistent repetition: consistently repeating a behavior (eg, taking medication) in the presence of a specific context cue (eg, a time of day) strengthens a specific cue-behavior association [28-33]. Habit theory suggests that although initial performances may require conscious effort, as the association is reinforced, the behavior can be activated automatically with minimal dependence on conscious memory or attention [30]. The formation of a habit is initially rapid, plateauing over time as the habit is formed [33,34]. Although there is individual variation in the timescale in which a behavior becomes habitual, habit formation tends to occur most notably within the first 2 weeks of attempting to change the behavior [33,35,36]. This highlights the need for support in the early phases of a behavior change intervention aimed at supporting habit formation [31,34].

According to habit theory, if habits for medication-taking behaviors are formed, taking medication could become more automatic, thereby reducing reliance on memory or the motivation to take AET [37]. In a meta-analysis including 771 interventions supporting medication adherence in a variety of clinical contexts, larger effect sizes were observed for habit-based interventions than for those using simple prompts or educational materials alone [38,39]. However, a recent review highlighted that many medication adherence interventions described as “habit based” are not theoretically informed and do not promote the process of context-consistent repetition, which is fundamental to habit formation [40].

SMS text messaging interventions are a low-cost method for supporting habit formation. They can serve as an initial reminder to take the medication and could lead to sustained behavior change through prompting medication taking in the same context repeatedly to support habit formation [41]. Meta-analyses of SMS text messaging interventions in other long-term conditions have reported positive effects on adherence (odds ratios 1.39-2.11) [42,43]. However, to date, no habit-based interventions to support medication adherence have exclusively used SMS text messages [40].

Outside of habit-based interventions, SMS text messaging interventions aimed at improving adherence to AET among women with breast cancer have shown mixed results [44-47]. In these interventions, the same messages are often repeated, potentially causing response fatigue [45,46]. The use of simple prompts in some of these interventions fails to make use of the range of known behavior change strategies and, therefore, may be unlikely to produce sustained behavior change once the messages have ceased [46]. Behavior change techniques (BCTs) are described as irreducible, active ingredients of an intervention that can be used to label the content of behavioral interventions [48]. Existing SMS text messaging interventions based on simple prompts use only 1 of the 93 BCTs identified in version 1 of the behavior change taxonomy (BCTTv1) [48]. Additional BCTs, beyond prompts and cues, should be used to promote sustained behavior change.

The message development processes for existing SMS text messaging–based interventions targeting AET adherence generally lack detail and transparency, and there is rarely a theoretical justification for the message content. Intervention development guidance highlights the importance of having a theoretical basis for an intervention and advocates for detailed reporting of intervention development to improve transparency [49-52]. Therefore, we sought to address these limitations in the field by developing SMS text messages based on specific, theory-based BCTs chosen to target habit formation, in line with habit theory [34].

Using an established process [53,54], we aimed to develop a pool of brief SMS text messages that promote habitual medication taking, are acceptable to patients, and show fidelity to explicit BCTs. This process involved (1) an expert workshop to create the messages, (2) a patient focus group to determine message acceptability, (3) a web-based patient survey to assess message acceptability, and (4) a web-based expert survey to assess the fidelity of the final messages to the intended BCTs (Figure 1). The studies in this paper form part of a broader research program, which aimed to develop and optimize a multicomponent intervention to support adherence to AET in women with breast cancer [26,55]. The multicomponent intervention includes 4 components targeting a range of known barriers to AET adherence. SMS text messages targeting memory and forgetfulness are one of these components [26,55].

Using the Target, Action, Context, Time framework for behavior specification [56], the target behavior is defined as women with breast cancer prescribed AET (target), taking their AET tablet (action) daily (time) in a consistent context (context). To achieve the behavior of adherence (ie, taking AET daily), ordering and collecting prescriptions is also required. Therefore, we targeted these associated behaviors to facilitate the performance of the overall behavior of taking medication daily.

On the basis of evidence suggesting that habit interventions could be effective for improving medication adherence [38,39] and the potential for habitual medication taking to reduce forgetfulness of taking AET, SMS text messages were designed based on habit theory. The process of habit formation involves multiple stages: deciding to undertake a behavior, initiating and repeating the behavior, and acting consistently in the same context [28-33]. Although the first 2 stages are necessary for any long-term behavior change attempt, the final stage promotes the formation of cue-behavior associations, which is unique to habit formation [29,32]. On the basis of published guidance, we selected 6 BCTs from the BCTTv1 to be used in the intervention [29,32,34,57]. Using BCTTv1 enabled us to develop a theoretically informed intervention specified in a way that will improve generalizability and replicability. The 6 chosen BCTs included a BCT explicitly promoting context-dependent repetition (habit formation [also known as context-dependent repetition]), BCTs promoting the use of feasible environmental cues for medication taking (prompts/cues, restructuring the physical environment, and adding objects to the environment [32,34,57]), a BCT specifying behavioral responses to these cues (action planning) [34,57,58], and a BCT promoting monitoring successful implementation of these responses (self-monitoring of behavior [34,48,57]). The BCTs were identified by 1 author (SMCG) and discussed and agreed by consensus with 4 other members of the research team (SGS, DPF, YKB, and ER).

We followed an established process using a series of studies to develop the pool of SMS text messages [53,54]. This approach involved developing SMS text messages based on theory with experts in behavior change before gaining qualitative and quantitative feedback from the patient population. The final stage involved checking whether the messages still targeted the BCTs they intended to target after any adaptations had been made.

A selection of 20 messages were sent to a consultant clinical pharmacist with experience of AET to ensure advice in the messages was appropriate and not conflicting with general medical advice. The consultant pharmacist was a member of the trial management group for the larger program of research that this study is part of. The 20 messages to be reviewed included all messages in which there could be any risk of conflicting or dangerous advice. The review occurred at this stage to ensure that the final versions of the messages were checked.

Ethics approval was granted by the University of Leeds School of Medicine Ethics Committee (MREC 20-038 July 2021). All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. Informed consent was obtained from all the participants included in this study. All data from the study were deidentified. Participants were compensated in the following ways: study 1: £100, study 2: £37.50, study 3: small incentive provided directly from the market research company, and study 4: £25 Amazon voucher.

We removed 2 messages and amended 3 messages following the advice of a consultant pharmacist (Table 5). Messages related to taking the AET tablet with hot drinks were removed, as this is not recommended.

After all the studies, a pool of 66 messages remained (examples in Multimedia Appendix 4). The full pool of messages is available to research teams upon reasonable request. All messages are 1-way and are designed to be sent by an automated message system. Overall, the messages have a readability score of 8.2 on the Flesch-Kincaid reading grade level scale, which corresponds to the reading age of an eighth grader (aged 13 years) [61].

In 4 linked studies involving behavior change experts and women with breast cancer, we developed a pool of 66 SMS text messages to promote habitual medication taking. The text messages were considered acceptable to women with breast cancer and had adequate fidelity to the target BCTs. The BCTs were chosen based on habit theory to promote context-dependent repetition of medication-taking behaviors so that habits may form. Next, we will examine the extent to which these messages can support adherence to AET in women affected by breast cancer [55]. If SMS text messages are effective in improving AET, they could form part of a multicomponent support program for women with breast cancer.

This study builds on previous attempts to use SMS text messaging interventions in women taking AET by developing explicitly theory-based messages. Previous SMS text messaging interventions aimed at improving AET adherence have produced equivocal findings [45-47]. One trial reported statistically significant effects at 6-month follow-up (P=.03), but these effects were not sustained at 12 months (P=.62) [46]. This null result at longer follow-up may be explained by the use of atheoretical, simple text prompts. Although simple prompts may be effective for short-term behavior change, they may create reliance on the text message as a prompt. In turn, behavior change may not be sustained upon message cessation. To address this issue, our messages encourage taking medication in the same context repeatedly to form cue-behavior associations, in line with habit theory [28-33]. Forming context-dependent associations for medication taking can lead to sustained behavior change even if the messages are stopped [28-33].

Our iterative text message development process using 4 interlinked studies enabled us to continually optimize the pool of messages throughout the development process. We used the key steps recommended by the intervention development guidelines [51]. For example, we included stakeholder involvement at multiple points throughout the development process, drew on existing theories, and continually refined the intervention. This advances on previous interventions that tend to be limited in their reported development process and therefore lack justification for the content of the intervention. Owing to our process, our pool of text messages quantitatively demonstrated acceptability in our target population and adequate fidelity to the intended BCTs. Consequently, the messages could be used to test habit theory in future evaluations.

The final pool of text messages demonstrated prospective acceptability in women with breast cancer. This was based on a single assessment of acceptability. Our approach was effective in ensuring that no messages considered unacceptable to women with breast cancer were included in the final pool of text messages. Previous studies involving SMS text messages to support adherence to diabetes medications found prospective and experienced acceptability to be correlated [54]. In the subsequent evaluation of these messages, our focus will turn to assessing experienced acceptability, including the satisfaction and usefulness of the messages when delivered over a prolonged period [54]. The theoretical framework of acceptability conceptualizes acceptability as a multifaceted construct composed of 7 components and could therefore provide a useful basis for the assessment of experienced acceptability [62].

Despite our rigorous intervention development process, some key uncertainties remain. First, there are uncertainties surrounding how these text messages should be used in an intervention; which messages to use, the frequency with which messages should be sent, and the duration for which they need to be sent to support habit formation [63]. Such uncertainties surrounding implementation could be explored further to build an optimal intervention using our pool of SMS text messages. Our approach to implementation, including justifications for our proposed frequency of messages, is explained elsewhere [26]. Second, some messages may show fidelity to other BCTs in the BCTTv1 outside of the 6 specified BCTs they were generated to target, for example, “problem-solving.” To aid transparency, the messages chosen to be used in any intervention should be coded using the BCTTv1 to reflect any additional BCTs that the messages may target [26,48].

We adapted an established approach previously used to develop SMS text messages to support diabetes self-management and demonstrated how the process can be conducted remotely [53]. Our approach can act as a guide for other researchers conducting remote codevelopment work. However, our study had limitations. In the web-based workshop in study 1, it was difficult to facilitate conversations between behavior change experts to collaboratively generate SMS text messages as planned. Consequently, the messages were largely developed individually. To encourage more collaboration during the workshop, researchers could consider allocating a period devoted to anonymously editing other participants’ messages on the web-based platform (eg, Padlet), which may facilitate collaboration better than discussion alone. In addition, all women (5/5, 100%) in study 2 and most women (49/60, 82%) in study 3 were of White ethnicity. This limits generalizability, as the acceptability of digital health interventions may be influenced by ethnicity and cultural norms [64]. Further developmental studies are needed to explore the acceptability and appropriateness of these messages in a wider range of sociodemographic groups.

In conclusion, we conducted a series of 4 linked studies using mixed methods to develop a pool of 66 brief messages to support medication adherence to AET in women with breast cancer. The messages were based on 6 BCTs theorized to support habit formation. The messages were rated as acceptable to women with breast cancer and showed fidelity to the intended BCTs. Further evaluation of these messages is needed to establish whether they can support medication adherence behaviors.