Inclusion criteria.

The trial will recruit participants from primary care ambulatory clinic setting who meet the following criteria: (1) Aged between 21–84 years old; (2) Prescribed or intent to restart statins with or without ezetimibe for lipid management; (3) Medication non-adherence as defined by the “Extent of Non-adherence” sub-scale of the DOSE Non-Adherence Measure, with a score ≥ 1 (range from 0–3); (4) Singapore residents (citizens, permanent residents, or long-term pass holders); (5) In possession of a smartphone or tablet with Android or iOS operating systems; (6) Have internet access on their mobile devices.

Exclusion criteria.

Participants will be excluded if they meet any of the following: (1) Does not read or understand English; (2) Current use of smartphone medication adherence app(s) that include statins; (3) Concurrent use of PCSK9 inhibitors in addition to statins and/or ezetimibe; (4) Participation in another study that uses medications that could affect lipid levels; (5) Severe renal impairment defined as chronic kidney disease stage 4 and above; (6) Severe liver disease (e.g., Child-Pugh Class C); (7) Existing musculoskeletal-related complaints or diagnoses which may confound adverse event reporting; (8) Uncorrected thyroid conditions, especially poorly-controlled hypothyroidism; (9) Documented psychiatric diagnosis, history of mental illness, or deemed as unable to give informed consent; (10) Currently pregnant, breastfeeding, or expecting to get pregnant during the course of the study; (11) Guarded prognosis with expectant mortality within 12 months or less.

Health coach-supported mobile health app.

The CADENCE digital personal health assistant (D-PHA) is a multi-component mobile health platform integrated with virtual health coaching. Developed by the study team, the D-PHA platform was designed to improve adherence to medications and lifestyle modifications through health education, improving their medication-taking behaviors, and adopting more healthy lifestyle behaviors. The behavioral change driven by the app was designed based on the established behaviour change technique taxonomy (v1) of 93 techniques hierarchically organized into 16 higher-order categories, including goals and planning, feedback and monitoring, social support, shaping knowledge, and self-belief [35]. In addition, two newly proposed categories used in digital therapeutic interventions, which are personalization and gamification, were also incorporated. To augment participants’ lifestyle behaviours and promote adherence to medication, the app consists key features such as educational modules, medication reminders, conversational agent (rules-based chatbot), and in-app communication with health coaches.

Educational modules cover content related to medication therapy for lipid management and the sequelae of poor lipid management, developed by a multidisciplinary team of medical doctor, pharmacist, and health psychologist. The content was structured based on Health Belief Model and Transtheoretical Model of Change [36,37]. To ensure clinical accuracy and contextual relevance, an expert review panel consisting of primary care physicians, cardiologist, pharmacists, dietitian, health psychologist, and patient advocate was convened to review the content, which covers six themes: (1) Introduction to heart health; (2) Lipids and Cardiovascular Disease; (3) Lipid-loweing medications; (4) Importance of medication adherence; (5) Heart-healthy behaviours and mindsets; and (6) Towards sustainable heart-healthy living. Each theme is made up of several modules, which comprise multiple units. Each unit is made up of a video recording or animation accompanied with transcripts, followed by a poll question or knowledge quiz to help participants reflect their understanding of the content.

Reminders and logs features are available on the D-PHA app, allowing participants to record food diaries and receive reminders for medications and clinic appointments. The platform also sends reminders via in-app text messages and push notifications to promote app usage and medication adherence.

Rule-based chatbot was designed to engage participants in conversations that uncover their perceived barriers to medication adherence and assists them with setting their personal health goals. Their interactions with the chatbot are tracked and made available to health coach to inform personalized coaching.

Health coaching.

The health coaching component of the intervention is delivered digitally via the D-PHA app. Health coaches are given access to a designated dashboard where they can schedule health coaching sessions, review app usage, and interacts with their assigned participants through asynchronous (text messaging) and synchronous communication (video-call). A special focus will be devoted to medication adherence through gathering information on participants’ knowledge regarding medication-taking behaviours and associated side effects, as well as identifying discrepancies between the list of prescribed lipid-lowering medications and participants’ self-reported use of the medications.

The sessions are delivered by health coaches who underwent two training programs. They completed a core coach training and certification program approved by the National Board for Health & Wellness Coaching (NBHWC) [38]. Additionally, they participated in a focused program on health coaching for older adults, offered by a higher education institution in Singapore [39]. This dual training approach ensures that health coaches acquire core coaching competencies and are capable of applying them effectively within the local context.

Although there is a lack of consensus on the intensity and frequency of coaching sessions, existing literature suggests the duration of effective health coaching programs to be between 6–12 months to achieve behaviour change, with 6 months being reported most frequently [30,40]. The number of sessions ranged from 1 to 12 [41]. Based on these findings, the minimum number of coaching sessions between participants and health coaches is set at four, with each session taking place once a month and lasting up to 60 minutes. After completion of four health coaching sessions, participants may request for two additional sessions if needed. Participants are encouraged to text their health coaches via the D-PHA app for additional support in between health coaching sessions. Health coaches may also initiate additional interactions with the participant if the app’s usage analytics suggest low engagement.

During the session, health coaches will follow a pre-specified health coaching guidelines customized for the trial to minimize discrepancies among health coaches. They will review adherence to prescription instructions, explore reasons for non-adherence, assist participants in setting SMART (Specific, Measurable, Achievable, Realistic, and Time-bound) behavior goals, and review goals and app usage. Health coaches will not provide any medical advice or instruction, including starting, stopping, or titrating medications. The scope of health coaching will be strictly related to medication adherence, lifestyle behaviours, and general health and well-being. Health coaches will keep a log detailing the communication with the participant.

Usage analytics for health coaching.

These information provide insights into participants’ engagement with the various features of the app. Passively collected data such as the number of logins, modules completion, time spent online, and other click-level data are captured and analysed. These data are presented on the dashboard of a central management system (i.e., administration portal). Additionally, each health coach has access to a designated dashboard to arrange coaching session, chat with patients, and monitor their app usage.

Before each session, health coaches will review the information gathered from usage analytics to guide and plan the upcoming session. The information is displayed on the health coach dashboard and includes medication logs, food diary entries, learning progress in the educational component, interactions with chatbot, and reflection journal entries. A weekly action score summarizing participant’s weekly activity in the app will also be shown to health coach. Information collected by the rule-based chatbot, such as the user’s readiness to change, perceived barriers and misbeliefs may also be used to guide a targeted behavioural intervention.

Economic evaluation.

A cost-effectiveness study will be conducted to investigate the expected changes to total costs and health benefits, as measured by Quality Adjusted Life Years, from a decision to adopt the human coach-supported digital personal health assistant. A Markov model, to achieve this, will be defined by the natural history of a patient with hyperlipidemia. The analysis will consider only the health system perspective. The structure of the model will be validated with cincial experts. The following information is needed: transition probabilities, costs, and effectiveness measures.

The transition probabilities will be sourced from the data collected in the trial (e.g., risk categories according to the Singapore-modified Framingham Risk Score 2023, adherence rate), and from the literature.

Intervention costs will be estimated using the information available within the trial budget. The cost of developing the app as a one-off cost, training the personnel, and maintenance cost will be calculated. Furthermore, the costs of statin treatment and the costs of cardiovascular events will be estimated and will be used as model inputs. Cost-effectiveness, healthcare utilisation, and direct medical costs will be collected.

The study will use several effectiveness measures. For health economic evaluations, effectiveness measures must be either in natural measurements, such as hospital avoidance, or patient-reported outcome measures, such as utility estimates. The effectiveness measures that will be used are EQ-5D-5L utility estimates (quality-adjusted life years – QALY), the number of hospital admissions, ED presentations and unplanned primary care visits avoided.

Model evaluation.

During analysis, statistical distributions will be used to describe uncertainty in model parameters. The normal, uniform, beta, and gamma distributions will be employed depending on the type of parameter. Fitted distributions will be randomly re-sampled and the economic outcomes of ‘change to total costs’ (ΔC) and ‘change to total effectiveness’ (ΔE) will be simulated 10,000 times. This approach propagates uncertainty in prior parameters forward to the total cost and outcomes. A decision rule for cost-effectiveness is given by (ΔC/ ΔE) <λ, where λ is the decision maker’s maximum willingness to pay (WTP) for health benefits. In the absence of a WTP value for Singapore, for this study, we use a value equalivent to three times that of the per capita GDP in Singapore. The flexibility of a Markov model will be used, with different scenarios being analysed to produce meaningful results for different stakeholders.

Baseline.

Participants in both control and intervention groups will be asked to complete a series of self-reported questionnaires. Lipid profile, anthropometric measurements, and Singapore-modified Framingham risk score will be obtained from electronic health records. Participants assigned to the intervention group will be assisted by the CRCs to install the D-PHA app into their mobile devices and register a user account using a pre-specified User ID. The CRC will introduce different features of the D-PHA app through an “on-boarding” encounter.

Follow-up Visits.

The study has three follow-up visits at month 3, month 6, and month 12. Attending PCT will plan the follow-up visits to coincide with their regular clinic visits. Before each follow-up visit, CRCs will send reminders to participants. Data on clinician-reported outcome measures and patient-reported outcome measures will be obtained during each follow-up visit (Fig 1). Additionally, participants in the intervention group will complete the mHealth app usability questionnaire (MAUQ) and Chatbot Usability Questionnaire at month 6 and month 12.