The goal of this study is to find out if adding electronic medical record (EMR) prompts helps prevent people with pre-diabetes from developing diabetes. It will also look at how these prompts affect doctor and patient behaviors.

The main questions are:

Does it improve follow-up care, such as blood tests, referrals, and medication? Does the EMR prompt reduce the number of patients who progress to diabetes within six months?

Researchers will compare clinics that use EMR prompts with clinics that do not.

Participants will:

Receive usual care for pre-diabetes at their polyclinic In some clinics, doctors will see EMR prompts suggesting tests, referrals, and medication Complete surveys about their health and lifestyle at different time points

This is a two-year cluster-randomized controlled trial conducted across eight primary care polyclinics within the National University Polyclinics (NUP) network in Singapore. These clinics provide multidisciplinary family medicine and chronic disease management services to a large and diverse population. All sites use a unified Electronic Medical Record (EMR) system (Epic, National University Health System cluster), which supports standardized clinical workflows, integrates decision-support tools, and enables secure extraction of de-identified data for research.

The study targets adults aged 21-59 years with prediabetes. All clinicians in both intervention and control clinics will receive standardized clinical training on updated prediabetes clinical practice guidelines and patient education materials. The updated workflow emphasizes lifestyle modification and behavioural counselling as the foundation of diabetes prevention. Clinicians are guided to refer patients to a dietitian or structured lifestyle programme if body mass index (BMI) is 23 kg/m² or above, counsel patients on nutrition and physical activity, schedule a six-month follow-up review, order HbA1c testing prior to the next review, and consider metformin initiation if HbA1c exceeds 6.5% after six months of lifestyle intervention, particularly in adults under 60 years with BMI ≥ 23 kg/m². Training will be conducted virtually during protected lunchtime sessions.

The study consists of three sequential phases. Phase 0 (Baseline) involves no workflow intervention, during which baseline EMR and survey data are collected. Phase 1 (Workflow Phase) introduces the standardized prediabetes clinical workflow across all clinics. Phase 2 (Prompt Phase) introduces EMR-based smart-set prompts only in intervention clinics to evaluate whether prompts further increase referrals, follow-up scheduling, HbA1c testing, and metformin prescribing beyond the workflow alone. Control clinics continue to use the standardized workflow without EMR prompts. The smart-set prompts are designed to be non-intrusive and provide decision support without interrupting workflow or overriding clinical judgment. Clinicians retain full autonomy to accept, modify, or dismiss suggested actions.

A sub-sample of approximately 300 patients will complete questionnaires assessing lifestyle behaviours and patient activation using the Consumer Health Activation Index (CHAI) to complement EMR-derived outcomes. Approximately 80-100 clinicians are expected to complete voluntary, anonymous surveys assessing knowledge, confidence, and clinical behaviours using the COM-B framework. Baseline clinical and survey data will be collected prior to intervention implementation, with follow-up data collected at multiple time points to evaluate short- and longer-term outcomes.

Intervention components include standardized workflow implementation and clinician education across all clinics, with additional EMR-based prompts implemented only in intervention clinics. Smart-set prompts integrated within Epic display automated reminders at the point of care, with options to facilitate orders for laboratory tests, referrals, medications, and follow-up scheduling. Prompts are non-mandatory to preserve clinician autonomy. The intervention is informed by the COM-B model to enhance clinician capability (through training and guidelines), opportunity (through EMR-enabled workflows and referral pathways), and motivation (through feedback and reinforcement). The Transtheoretical Model (TTM) will be used to monitor stages of change among both patients and clinicians.

Survey data will be collected electronically using FormSG, a secure, government-hosted platform approved for research use. All study data will be stored on institution-approved, PDPA-compliant servers with access restricted to authorized study personnel. Identifiable and de-identified datasets will be stored separately. De-identified datasets will be transferred to analysts using encrypted, password-protected channels. Only the Principal Investigator will have access to the linkage file containing study identifiers and personal identifiers. Hard-copy consent forms will be stored in locked cabinets accessible only to the Principal Investigator. Study data will be retained for six years following study completion in accordance with institutional policy, after which electronic data will be securely deleted and physical records destroyed.

Analyses will follow the intention-to-treat principle, with participants analysed according to their assigned clinic groups. Baseline characteristics will be summarized descriptively. Changes over time and differences between intervention and control clinics will be examined using regression models appropriate to outcome type, including mixed-effects logistic regression to account for clustering at the clinic level. Time-to-event analyses using Cox proportional hazards regression will be used to assess progression to diabetes while accounting for variable follow-up durations. Missing data will be addressed using multiple imputation, and sensitivity analyses will be conducted to assess robustness of findings. All analyses will use two-sided tests with a significance level of 0.05.

Outcome data will be collected at baseline, 6 months, 12 months, and at 18 and 24 months to assess short- and longer-term effects of workflow and EMR-based decision-support implementation.

This study will contribute evidence on the effectiveness of a non-intrusive, EMR-embedded clinical decision-support system for improving guideline-concordant prediabetes care in primary care and inform scalable strategies for diabetes prevention.

Inclusion Criteria for study population (EMR based analytic cohort):

• All adults aged 21 to 59 years with prediabetes who attend any of the eight participating polyclinics during the study period will be included in the EMR-based analytic cohort.
• Prediabetes is defined as impaired fasting glucose or impaired glucose tolerance according to standard clinical criteria, with either diagnosis documented in the EMR problem list or visit diagnosis.
• Individuals with a prior diagnosis of diabetes will be excluded.

Inclusion Criteria for patient surveys:

• Adults aged 40-59 years
• Diagnosed with prediabetes (impaired fasting glucose or impaired glucose tolerance)
• Receiving routine follow-up care at NUP polyclinics
• Able to provide written informed consent
• Able to complete questionnaires in English

Exclusion criteria for patient surveys:

• Prior diagnosis of diabetes mellitus or gestational diabetes
• Severe acute or chronic liver or kidney disease
• Pregnancy
• Cognitive impairment
• Inability to communicate in English

Inclusion Criteria for clinician surveys:

• All clinicians that consult patients in National University Polyclinics
• No exclusion criteria specified for clinicians

Privacy and Confidentiality: Our study involves some sensitive health information. Even with de-identification, there remains a risk of re-identification, particularly given the detailed nature of our data and the specific population studied.

Ethical Constraints: Our informed consent process and ethics approval did not explicitly include provisions for broad data sharing beyond the immediate research team and oversight committees.

Regulatory Compliance: Local health data protection regulations place strict limitations on the sharing of individual-level health data, even in anonymized form.

However, we are committed to scientific transparency and reproducibility. Therefore, we will make aggregate data, statistical codes, and detailed study protocols available upon reasonable request, subject to approval by our institutional review board and data access committee.