AI Diabetes Clinic: AI-First Clinic Model for Type 2 Diabetes Diagnosis, Therapy Optimisation, and Complication Surveillance, with a Saudi Arabia Implementation Blueprint

1) Diagnostic Confirmation and Phenotyping

For newly diagnosed or suspected type 2 diabetes, the system validates diagnostic criteria, checks for red flags (eg, possible type 1 diabetes or ketosis risk), and classifies phenotype features relevant to therapy choice: obesity, CKD, ASCVD risk, heart failure risk, hypoglycaemia risk, and frailty. Outputs are evidence-linked and explanation-aware to reduce automation bias [6,18].

2) Guideline-Concordant Treatment Optimisation

The treatment engine is a constrained optimisation system: it proposes therapy options and titration plans consistent with current standards, patient preferences, comorbidities, contraindications, and local formulary constraints. Recommendations should be traceable to guideline logic (eg, prioritising cardiovascular and kidney protection where indicated) and include monitoring plans and safety checks [6,7,18].

3) AI-Enabled Screening for Complications

Retinopathy: The clinic deploys AI-assisted fundus imaging to triage and refer patients for ophthalmology review. Autonomous AI models have demonstrated potential to expand screening capacity when embedded into clinical workflows with clear escalation and quality assurance [8].

Kidney disease: Automated interpretation of eGFR and albuminuria trends supports early detection of CKD progression, medication safety checks, and risk-based follow-up intervals. The clinic should follow evidence-based evaluation and management principles and explicitly manage nephrotoxic exposure and dosing adjustments [7].

Neuropathy and foot disease: AI-enabled foot assessment can combine structured clinical exam inputs (eg, monofilament, vibration, pulse status) with smartphone imaging and tele-podiatry to identify ulcers and high-risk feet. Large multicentre studies have reported promising performance for AI-based ulcer detection using smartphone and cloud workflows [9].

4) Remote Monitoring and Early Warning

Remote monitoring integrates glucometer/CGM data, blood pressure, weight, activity, and medication adherence signals. Temporal models flag patterns suggestive of deterioration (eg, rising glucose variability, recurrent hypoglycaemia, rapid weight loss, declining eGFR), prompting nurse outreach or clinician review. Alert thresholds must be calibrated to avoid alarm fatigue, and monitoring performance must be continuously audited [6,18,21].

5) Psychological, Neurological, and Sexual Health Triage

Diabetes care requires systematic detection of depression and anxiety and proactive discussion of sexual dysfunction and neuropathy symptoms. The clinic integrates validated short instruments (PHQ-9, GAD-7, IIEF-5) with AI-assisted triage to behavioural health, sexual medicine, or neurology services, with culturally sensitive workflows and clinician oversight [26,27,28].

Erectile dysfunction (ED) is frequent in men with diabetes; a large meta-analysis reported a pooled prevalence of 52.5% and substantially higher odds compared with controls [29]. Because ED can precede overt cardiovascular events and may identify men with high cardiometabolic risk, evaluation should incorporate cardiovascular risk assessment and guideline-based triage (including consideration of occult coronary disease when clinically appropriate) [30]. In the AI diabetes clinic, ED risk prediction combines longitudinal glycaemic exposure, blood pressure, lipids, obesity, smoking, medication profiles, and neuropathy markers with brief patient-reported measures to trigger early counselling, optimisation of cardiometabolic therapy, and timely initiation of first-line ED treatments or referral to urology/andrology as needed [31,32].